Integrating spectral, texture, soil and fertilization information for plot-level prediction of sugarcane yield, millable stalk population and Brix from Jilin-1 imagery 

Integrating spectral, texture, soil and fertilization information for plot-level prediction of sugarcane yield, millable stalk population and Brix from Jilin-1 imagery | Research Square window.SnipcartSettings = { analytics: { enabled: false } }; (function() { var accessVector = localStorage.getItem('access_vector') || ''; window.dataLayer = window.dataLayer || []; if (accessVector) { window.dataLayer.push({ user: { profile: { profileInfo: { snid: accessVector } } } }); } })(); (function(w,d,s,l,i){w[l]=w[l]||[];w[l].push({'gtm.start':new Date().getTime(),event:'gtm.js'});var f=d.getElementsByTagName(s)[0],j=d.createElement(s),dl=l!='dataLayer'?'&l='+l:'';j.async=true;j.src='https://www.googletagmanager.com/gtm.js?id='+i+dl;f.parentNode.insertBefore(j,f);})(window,document,'script','dataLayer','GTM-K279D39R'); Browse Preprints In Review Journals COVID-19 Preprints AJE Video Bytes Research Tools Research Promotion AJE Professional Editing AJE Rubriq About Preprint Platform In Review Editorial Policies Our Team Advisory Board Help Center Sign In Submit a Preprint Cite Share Download PDF Research Article Integrating spectral, texture, soil and fertilization information for plot-level prediction of sugarcane yield, millable stalk population and Brix from Jilin-1 imagery Yun Zhang, Guangtao Xu, Dan Li, Yinglin Lu, Hao Jiang, Shasha Luo, and 8 more This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-9371174/v1 This work is licensed under a CC BY 4.0 License Status: Posted Version 1 posted You are reading this latest preprint version Abstract Purpose The primary objective of this study was to evaluate the potential of high spatial resolution Jilin-1 satellite imagery for plot-level prediction of sugarcane yield, millable stalk population, and Brix, and to assess whether integrating spectral, texture, soil, and fertilization information could improve prediction performance for precision sugarcane management. Methods Jilin-1 satellite imagery acquired at four growth stages, from seedling to maturity, was used to derive vegetation indices (VIs) and texture indices (TIs), including the normalized difference texture index (NDTI), enhanced vegetation texture index (EVTI), and double-difference ratio texture index (DDRTI). Soil chemical properties (SCPs) and fertilization information (FI) were further incorporated with the remotely sensed variables. Machine learning models were developed for plot-level prediction of sugarcane traits across plant cane and first ratoon cane, and texture window size was optimized to improve TI extraction and model performance. Results For yield, the combination of VIs and TIs outperformed VIs alone at the tillering stage (R 2 CV = 0.65, RMSECV = 15.06 t/ha, RPDCV = 1.68). Adding SCPs and FI further improved yield prediction across plant cane and first ratoon cane (R 2 CV = 0.70, RMSECV = 13.84 t/ha, RPDCV = 1.83). Millable stalk population was best predicted at the maturation stage by VIs and Tis, achieving the best performance (R 2 CV = 0.63, RMSECV = 6602 stalks/ha, RPDCV = 1.66). The best Brix model integrated VIs, TIs, SCPs, and FI at the maturation stage (R 2 CV = 0.44, RMSECV = 0.53 °Bx, RPDCV = 1.33). SHAP analysis identified VIs as the dominant features for sugarcane traits prediction. And, DDRTI contributed more than NDTI and EVTI in yield and Brix prediction. Conclusion It is concluded that integrating spectral, texture, soil, and fertilization information from high spatial resolution Jilin-1 imagery is a promising approach for improving plot-level prediction of key sugarcane traits. Sugarcane phenotypic traits Jilin-1 Texture indices Machine learning SHAP interpretability Figures Figure 1 Figure 2 Figure 3 Figure 4 Figure 5 Figure 6 Figure 7 Figure 8 1. Introduction Sugarcane ( Saccharum spp.) is one of the most important crops in tropical and subtropical regions, with a global planting area of more than 27 million hectares (FAO, 2025 ). It contributes more than 80% of global raw sugar production and plays a major role in bioenergy development, accounting for approximately 40% of global fuel ethanol production (Canton, 2021 ; Rosa et al., 2026 ). Stable high-yield and high-quality sugarcane production is important for the sustainable development of both the sugar and bioenergy industries. Sugarcane is a semi-perennial crop with a growth cycle of approximately 12–18 months, typically including the seedling, tillering, elongation, and maturation stages (Som-ard et al., 2021 ). Across these stages, substantial changes occur in canopy structure, biomass accumulation, leaf area index (LAI), and physiological status, which lead to pronounced variation in remote sensing signals (Yang et al., 2025 ). During the seedling stage, new ratoons emerge, and biomass and LAI remain low. At the tillering stage, canopy cover expands rapidly and biomass accumulation accelerates. During the elongation stage, stalk elongation and leaf expansion dominate, leading to rapid biomass accumulation, high canopy closure and an increasingly complex vertical canopy structure. At the maturation stage, leaves progressively senesce and chlorophyll content declines, while sucrose accumulates in the stalks (Li et al., 2019 ). This phenological transition may weaken the direct relationship between canopy greenness and yield-related traits, which makes conventional greenness-based indicators less reliable for monitoring final production status (Som-ard et al., 2021 ). Sugarcane traits, such as millable stalk population, stalk weight, plant height, biomass and sucrose content, directly determine yield and economic return (Meier et al., 2016; Li et al., 2019 ). Early season acquisition of these traits is important. It enables timely management interventions, supports early screening in breeding programs, and improves decision making before harvest (Akbarian et al., 2024 ). However, conventional sugarcane phenotyping still relies heavily on manual field surveys and destructive sampling. Estimating yield and the millable stalk population usually requires large-scale field harvesting and manual counting (Zhu et al., 2026 ). Determining sugar content depends on destructive crushing of mature stalks (de França e Silva et al., 2024 ). These approaches are laborious-intensive, costly and destructive, which limits the application of precision agriculture. Remote sensing provides a non-destructive and large-scale alternative for crop monitoring and has been widely used in sugarcane mapping, growth monitoring, stress detection, and yield estimation (Jimenez-Sierra et al., 2020 ; Xu et al., 2025 ; Kambli & Palkar, 2025 ). Many studies have used vegetation indices such as NDVI, GNDVI and EVI, to characterize canopy greenness and vigor (Akbarian et al., 2022 ). However, some commonly used broadband vegetation indices (e.g., NDVI) may become less sensitive under medium to high vegetation cover (Gao et al., 2023 ). In addition, many inversion models were originally developed based on medium and low spatial resolution imagery in which each pixel generally represents the average condition of a relatively large canopy area (Luciano et al., 2021 ; Amorim et al., 2022 ). As spatial resolution increases from medium or low resolution to the meter scale, the physical meaning of a pixel changes. Canopy heterogeneity, geometric effects, and local radiometric fluctuations become much more pronounced (Fan et al., 2025 ). The applicability of these models to high spatial resolution data has not been thoroughly evaluated. Thus, traditional models may lead to greater retrieval uncertainty. Furthermore, canopy saturation is prone to occur when vegetation indices are applied to medium to high vegetation coverage (Zhao et al., 2012 ). Relying solely on spectral information is insufficient to capture the spatial complexity of sugarcane canopies, which further constrains the prediction of canopy traits (Liu et al., 2023a ). High spatial resolution remote sensing data offer spectral information and spatial details (including texture, shape, edge, and canopy arrangement patterns). Texture features derived from high-resolution imagery have the potential to characterize canopy spatial organization and structural variability (Zhu et al., 2026 ). The influences of canopy gaps, shadows, row structure, and adjacency-related radiometric interactions become more evident (Fan et al., 2025 ). Some studies have shown that integrating spectral and texture features from UAV imagery, can improve the estimation of crop biomass (Xie & Yang, 2020 ; Liu et al., 2023a ), leaf area index (Yuan et al., 2023 ) and yield (Yu et al., 2025 ). However, the performance of texture based methods is also sensitive to feature extraction settings, such as window size, and direction, which may substantially affect parameter inversion accuracy and model robustness (Ozkan & Demirel, 2021 ). Moreover, the limited spatial coverage and operational scalability of UAV imagery restrict its use in routine large-area applications (Phang et al., 2023 ). Meter-level satellite imagery is a critical observational scale that bridges ultra fine observations and operational agricultural monitoring (Aleman-Montes et al., 2023; Fan et al., 2025 ). Open-access satellite constellations such as Sentinel have played an important role in crop monitoring (Luciano et al., 2021 ). However, these data are often constrained by revisit intervals and cloud contamination in tropical and subtropical regions. Increasing attention has been directed toward high resolution commercial imagery for more detailed agricultural monitoring and improved yield estimation (Wang et al., 2025 ). Nevertheless, systematic investigations of texture features under high spatial resolution conditions, including the optimal parameters for texture feature extraction and effective combination strategies remain limited. The potential of integrating spectral and texture data has not yet been fully explored for sugarcane phenotyping. Among these platforms, the Chinese Jilin-1 satellite constellation provides a balance among spatial resolution, revisit frequency and cost-effectiveness. Its application potential has been evaluated in leaf area index inversion for corn and rice (Du et al., 2023 ), environmental quality assessment in flood-prone areas (Chen et al., 2023 ), and wetland vegetation classification using UAV-satellite fusion technology (Fu et al., 2022 ). Although Jilin-1 imagery perform effectively in these fields, its potential in field-level crop phenotyping monitor has not been fully investigated. For sugarcane yield estimation, previous studies have adopted process-based crop models combined with data assimilation to improve prediction accuracy at regional scales (Xu et al., 2025 ). However, applying these process-based framework to meter level imagery is often data- and parameter-intensive. Acquiring such fine scale inputs is challenging to, which limits their practical applicability in fine-scale crop monitoring (de França e Silva et al., 2024 ; Rosa et al., 2026 ). Machine learning (ML) algorithms are good at modeling complex non-linear relationships. They have shown great potential in dealing with multi-source heterogeneous data (Paudel et al., 2021 ; Li et al., 2026 ). In addition, explainable artificial intelligence has also attracted increasing attention due to the transparency of model decision-making processes (Xuan et al., 2025 ; Wu et al., 2025 ). Integrating high-resolution spectral features, texture information, and multi-source agricultural auxiliary data for plot level sugarcane trait prediction by ML algorithm has not yet been fully established. This study aims to evaluate the potential of meter-level Jilin-1 imagery for plot-level sugarcane phenotyping. Plot-level field observations were collected for key sugarcane traits, including yield, millable stalk population, and Brix. Three objectives were addressed: 1) trait estimation, assessing the ability of spectral and texture indices to estimate key sugarcane traits; 2) data integration, evaluating whether the incorporation of multi-source agricultural auxiliary data improves prediction accuracy; and 3) model interpretation, identifying the most important predictors for different traits using explainable machine learing. 2. Materials and methods 2.1. Study area and experimental design A two-year trial was conducted at the Zhanjiang Research Center (Nanfan Seed Industry Research Institute of Guangdong Academy of Sciences) in Suixi County, Guangdong Province (Fig. 1 ). This location has a tropical monsoon climate, with an annual average temperature of about 23°C and precipitation exceeding 1400 mm. There is sufficient sunlight, heat and water, which are highly suitable for sugarcane cultivation. The study spanned two years, focusing on plant cane in 2024 and first ratoon cane in 2025. The experimental field consisted of 32 plots, and the sugarcane variety planted was ‘Yutang 03373’. Each 100-square-meter plot contained 9 rows, with each row approximately 10 m in length. Basal and topdressing fertilizers were applied at the seedling and tillering stages, respectively. The experiment comprised eight fertilization regimes (T1–T8) using a compound fertilizer (N: P 2 O 5 : K 2 O = 15: 15: 15), as detailed in Table S1 . 2.2. Data acquisition and preprocessing 2.2.1. Field data acquisition Phenotypic data were collected on November 19, 2024 (plant cane) and December 2, 2025 (first ratoon cane), at the early maturity stage. In addition, for each plot, 20 plants were randomly selected to measure plant height (cm), stem diameter (cm) and single stem weight (g). To minimize edge effects, sampling was restricted to the three central rows (rows 4–6) of each plot. Twenty stalks were randomly selected from this zone for yield component analysis. Brix was measured from the juice of middle internodes using a handheld refractometer. Millable stalk population were defined as healthy, mature stalks meeting agronomic standards for height and diameter. The total number of stalks was estimated by counting the number of stalks in the middle row and extrapolating the density to the plot level. Yield was calculated based on the formula for stem fresh weight per hectare. The calculation formulas are as follows: $$\:W=\pi\:\times\:{\left(\frac{D}{2}\right)}^{2}\times\:L\times\:\rho\:$$ 1 $$\:Y=W\times\:MS$$ 2 where Y is the yield (t/ha), W is the theoretical single stalk weight (kg), D and L denote stalk diameter (cm) and length (cm), respectively, MS is the number of millable stalks, and ρ is the stalk density (constant at 1.05 g/cm³). Soil samples were collected from the 0–20 cm soil layer of each plot prior to planting. The soil chemical properties were analyzed in the laboratory (Table S2). 2.2.2. Remote sensing data acquisition and preprocessing Jilin-1 satellite imagery for the seedling, tillering, elongation and maturation stages was obtained via the official data portal ( https://www.jl1mall.com/ ). Each scene included a panchromatic band and four multispectral bands (Blue, Green, Red, and Near-Infrared). Table S3 details the sensor specifications and acquisition dates. We digitized plot boundaries using QGIS 3.34. The preprocessing workflow was conducted in ENVI 5.6, involving radiometric calibration, atmospheric correction, geometric registration, and image fusion to derive surface reflectance. The spectral response functions for atmospheric correction were obtained from the website ( https://www.jl1mall.com/resrepo/ ). 2.2.3. Calculation of vegetation and texture indices The preprocessed Jilin-1 imagery and sugarcane plot vector boundaries were imported into the GEE platform. Twenty widely used VIs were computed, as listed in Table S4. Texture features (TFs) were extracted using the GLCM algorithm (Haralick et al., 1973 ). This process employed three moving window sizes (3×3, 5×5, and 7×7 pixels), and the texture features were averaged across four directional angles (0°, 45°, 90°, and 135°) for each window size. Consequently, eight standard TFs (Table S5) were generated for each of the four spectral bands Furthermore, three advanced indices, namely the normalized difference texture index (NDTI), enhanced vegetation texture index (EVTI) and double difference ratio texture index (DDRTI), were constructed by combining these TFs. Table 1 lists the calculation formulas for the TIs. Table 1 Summary of texture indices used in this study. Parameters Formulas Reference Normalized Difference Texture Index (NDTI) \(\:NDTI=\frac{{T}_{1}-{T}_{2}}{{T}_{1}+{T}_{2}}\) (Zheng et al., 2019 ) Enhanced Vegetation Texture Index (EVTI) \(\:EVTI=\frac{2.5\times\:\left({T}_{1}-{T}_{2}\right)}{\left({T}_{1}+6\times\:{T}_{2}-7.5\times\:{T}_{3}+1\right)}\) (Yuan et al., 2023 ) Double Difference Ratio Texture Index (DDRTI) \(\:DDRTI=\frac{{T}_{1}-{T}_{2}}{{T}_{3}-{T}_{4}}\) This study Note: T 1 –T 4 denote the same texture feature derived from arbitrary spectral bands. 2.3. Feature selection Pearson correlation analysis was performed to investigate the relationships between input variables, revealing the presence of high multicollinearity and redundancy. To address this issue, SVR-RFE (Guyon et al., 2002 ) was used to identify the optimal feature subset. This method iteratively removed the features with the lowest model contribution, retaining only those with the highest predictive power for each growth stage. 2.4. Machine learning models This study evaluated the ability of Jilin-1 satellite imagery to predict phenotypic traits of four growth stages of sugarcane (seedling, tillering, elongation, and maturation stages). Four regression models were compared, namely ridge regression (RR) (Hoerl & Kennard, 1970 ), Gaussian process regression (GPR) (Rasmussen and Williams, 2006 ), categorical boosting (CatBoost) (Friedman, 2002 ; Zhang et al., 2020 ) and support vector regression (SVR) (Drucker et al., 1997 ; Durbha et al., 2007 ). All analyses and modeling were conducted using Python 3.12. 2.5. Model optimization and evaluation Grid search with five-fold cross-validation was used to optimize the model hyperparameters. The specific parameters and search ranges are listed in Table S6. Leave-one-out cross-validation (LOOCV) (Brovelli et al., 2008 ) was used to evaluate the model. Subsequently, the bootstrap method (Efron, 1979 ) was applied to further verify the robustness and reliability of the model. Model performance was evaluated using the coefficient of determination (R 2 ), root mean square error (RMSE) and ratio of performance to deviation (RPD). The RPD values were interpreted based on the classification criteria proposed by Gaston et al. ( 2010 ): 1.0–1.4 indicates poor prediction; 1.4–1.8, fair; 1.8–2.0, good; and 2.0–2.5, very good. The calculation formulas are as follows: $$\:{R}^{2}=1-\frac{{\sum\:}_{i=1}^{n}{\left({y}_{i}-{\widehat{y}}_{i}\right)}^{2}}{{\sum\:}_{i=1}^{n}{\left({y}_{i}-{\stackrel{-}{y}}_{i}\right)}^{2}}$$ 3 $$\:RMSE=\sqrt{\frac{{\sum\:}_{i=1}^{n}{\left({y}_{i}-{\widehat{y}}_{i}\:\right)}^{2}}{n}}$$ 4 $$\:RPD=\frac{SD}{RMSE}$$ 5 $$\:SD=\sqrt{\frac{{\sum\:}_{i\:=1}^{n}{\left({y}_{i}-{\stackrel{-}{y}}_{i}\right)}^{2}}{n}}$$ 6 where \(\:{y}_{i}\) is the actual value; \(\:{\widehat{y}}_{i}\) is the predicted value; \(\:{\stackrel{-}{y}}_{i}\) is the mean of the actual values; n is the number of samples; and SD is the standard deviation of the target variable. 2.6. Feature importance assessment SHAP (Lundberg and Lee, 2017 ) was used to interpret model predictions. It assigned each input feature an importance value and ranked the features at a global level. All calculations were performed using the SHAP library in Python 3.12. 2.7. Research framework Figure 2 illustrates the workflow of this study, which consists of the following key steps: (a) Data Acquisition and Preprocessing. Obtain high-quality surface reflectance data from Jilin-1 satellite imagery through radiometric calibration, atmospheric correction, geometric correction, and band fusion; collect field survey data including soil chemical properties (SCPs), fertilization information (FI), and sugarcane phenotypic data, and complete data preprocessing and accuracy verification to form a multi-source dataset. (b) Optimization of Optimal Texture Window Size. Calculate multiple vegetation indices (VIs) and extract texture features using the Gray Level Co-occurrence Matrix (GLCM). Construct texture indices (TIs) under different window sizes (3×3, 5×5 and 7×7), then build machine learning models (SVR, RR, GPR and CatBoost) and evaluate performance via Leave-One-Out Cross Validation (LOOCV) with RPD CV , to determine the optimal texture window size for sugarcane phenotype prediction. (c) Determination of the Optimal Prediction Model. Construct multi-dimensional input feature schemes (VIs, VIs + TIs and VIs + TIs + SCPs + FI) based on the optimal texture window size, train multiple machine learning models, and assess model performance using LOOCV and Bootstrap resampling with evaluation metrics (R 2 CV , RMSE CV , and RPD CV ) to select the optimal prediction model. (d) Model Interpretability Analysis. Apply the SHAP interpretable method to the optimal prediction model, quantify feature importance, analyze the response patterns of key features, and compare the contribution of multi-source features to reveal the driving mechanisms of model predictions for sugarcane phenotypes. 3. Results and analysis 3.1. Variations of sugarcane traits, NDVI and DDRTI_MEA (B,R,G,NIR) The statistical results and scatter plots for yield, millable stalk population, and brix data for 2024 and 2025 are presented in Fig. S1 and Table 2 . The mean values of yield, millable stalk population and brix in 2024 were higher than those in 2025. Variation analysis results indicated that there were no significant differences between the two years of yield data and brix ( p > 0.05), whereas significant differences were observed for millable stalk population ( p < 0.05). The first ratoon maintained yield with fewer stalks by increased individual stalk weight. The stalk quality and yield were preserved even the stalk number declined. The coefficient of variation (CV) for brix remained below 4% across two years. The CV for millable stalk population increased from 14.93% in 2024 to 20.50% in 2025, and the minimum value decreased from 36900 to 22500 stalks/ha. It seems that this variety compensated for the loss of millable stalks through increased individual stalk weight in ratoon cane with relatively stable yield and Brix. In this study, there are a large number of VIs and TIs were used. NDVI and DDRTI_MEA (B,R,G,NIR) are presented as representative examples for seasonal pattern analysis. The seasonal variation of NDVI in 2024 and 2025 was presented in Fig. S2a. It is clear that NDVI increased more rapidly in 2025 than 2024 from the seedling to the tillering. During the subsequent growth period, NDVI continued to rise and reached its maximum in early September 2025, whereas the peak occurred in mid-September 2024. In addition, during maturation, NDVI declined less and remained higher in 2025. The seasonal dynamics of DDRTI_MEA (B,R,G,NIR) in 2024 and 2025 were given in Fig S2b. In 2024, DDRTI_MEA (B,R,G,NIR) increased slightly from the seedling stage to the tillering stage and then gradually declined after elongation. In 2025, this index was low during the early stages, followed by a rapid from late tillering to elongation with a pronounced peak in early September before dropping during maturation. Table 2 Descriptive statistics of the measured sugarcane traits. Sugarcane Traits Year Mean SD Min Max CV(%) Yield (t/ha) 2024 70.53 27.45 25.67 136.01 38.92 2025 66.54 23.77 25.47 106.17 35.73 Millable Stalk Population (stalks/ha) 2024 57825 8636 36900 74700 14.93 2025 44803 9186 22500 58500 20.50 Brix (°Bx) 2024 20.87 0.77 19.54 22.64 3.73 2025 20.61 0.61 19.28 21.84 2.97 3.2. Correlation analysis of input features and sugarcane traits 3.2.1 Correlation between vegetation indices and sugarcane traits Twenty VIs were evaluated by Pearson correlation analysis across four growth stages (Fig. S3). Correlation coefficients ( r ) and significance levels ( p ) were calculated. For sugarcane yield (Fig. S3a), the strongest correlations were observed during the tillering and elongation stages. At the tillering stage, GLI exhibited the highest positive correlation ( r = 0.60, p < 0.001), followed by CVI and GRVI ( r = 0.52). At the elongation stage, NGRDI and WDRVI showed the strongest correlations ( r = 0.57, p < 0.001). Correlations at the seedling and maturation stages were generally weaker. The correlation between VIs and millable stalks showed distinct temporal patterns (Fig. S3b). The seedling stage was the most critical window for monitoring stalk number. Almost all VIs showed uniformly strong positive correlations (approx. r = 0.60) with high statistical significance ( p < 0.001). Interestingly, an inverse relationship was observed in later stages. During the elongation and maturation stages, most VIs were significantly negatively correlated with millable stalk population (| r| values ranging from 0.50 to 0.60). In contrast to yield and millable stalk population, Brix showed no significant correlation with any of the tested VIs across all growth stages (Fig. S3c). The correlation coefficients remained low ( |r| < 0.25) and statistically insignificant, suggesting that standard VIs may not be sensitive enough to detect variations in sugarcane Brix. 3.2.2 Correlation between texture indices and phenotypic traits To evaluate TIs suitable for modeling, Pearson correlation analyses were conducted between TIs (calculated using different window sizes of 3×3, 5×5, and 7×7) and sugarcane phenotypic traits across four growth stages (seedling, tillering, elongation, and maturation). In this study, TIs were derived by combining the same TFs across different spectral bands. Eight TFs were selected, and the three major categories of TIs (NDTI, EVTI, and DDRTI) were divided into eight sub-categories, respectively. From each sub-category, the texture index exhibiting the highest absolute correlation coefficient (| r |) was selected. Figs. S4, S5, and S6 illustrate the Pearson correlation coefficients between TIs and yield, millable stalk population, and Brix. TIs calculated with different window sizes performed differently for the three phenotypic traits. For sugarcane yield, the correlation of texture indices first increased and then decreased during growth. It rose from 0.40 ( p < 0.001) at the seedling stage to 0.67 ( p < 0.001) at the tillering stage, then dropped to 0.53 ( p < 0.001) at the maturation stage. For millable stalk population, the correlation of texture indices increased gradually. It rose from 0.47 ( p < 0.001) at the seedling stage to 0.68 ( p < 0.001) at the maturation stage. For Brix, the correlation showed a downward trend. The highest correlation coefficient was 0.33 ( p < 0.05) at the seedling stage. 3.2.3 Correlation between agronomic factors and phenotypic traits Pearson correlation analysis was conducted to evaluate the relationship between SCPs, FI, and sugarcane phenotypic traits (Fig. S7). Significant correlations were observed between yield and various SCPs as well as FI. In terms of SCPs, AP showed the strongest negative correlation ( r = -0.43, p < 0.001), while soil pH exhibited a significant positive correlation ( r = 0.40, p < 0.001). Regarding FI, there was a significant positive correlation between yield and variables such as BF_CF, BF_KCl, TD_CF, and TD_KCl ( r = 0.46, p < 0.001). Millable stalk population, a key component of yield, showed the highest correlation with FI ( r = 0.35, p < 0.05), while its correlation with SCPs was low. In contrast, the correlations between Brix and these indicators were all weak and statistically insignificant. 3.3. Determination of optimal texture calculation window size Four machine learning models were employed to determine the optimal window size, with RPD CV scores illustrated in Fig. 3 . Table S7 summarized the best parameter combinations for each stage. The results indicate that the optimal texture combination varied across growth stages. 3.4. Prediction of sugarcane traits based on multi-source features 3.4.1. Yield estimation model Table S8 summarizes the yield prediction results. The performance of four models (SVR, GPR, RR, and CatBoost) with the SVR-RFE feature selection method across four growth stages were compared. Figure 4 (a1–c1) presents the prediction results of the best models based on LOOCV for the three different input data combinations. When using only VIs, the SVR model at the tillering stage achieved the highest accuracy. The R 2 CV was 0.60, the RMSE CV was 16.02 t/ha, and the RPD CV was 1.58 (Fig. 4a1). After adding TIs, the RR model performed the best at the tillering stage. The R 2 CV increased to 0.65, with an RMSE CV of 15.06 t/ha and an RPD CV of 1.68 (Fig. 4b1). Finally, with the addition of SCPs and FI, the SVR model at the tillering stage reached the highest accuracy. The R 2 CV rose to 0.70, the RMSE CV decreased to 13.84 t/ha, and the RPD CV reached 1.83 (Fig. 4c1). The RPD value shifted from the “fair” category (1.4–1.8) to the “good” category (1.8–2.0). To further verify the stability and robustness of the prediction models, the distribution of RPD values were also obtained from the bootstrap method (Fig. 4 (a2–c2)). The mean RPD in Fig. 4 c2 was 1.56, slightly lower than the 1.58 in Fig. 4 b2. However, the 75th percentile in Fig. 4 c2 reached 1.75, surpassing the 1.73 in Fig. 4 b2. 3.4.2. Millable stalk population estimation model For millable stalk population, the performance of tour models (SVR, GPR, RR, and CatBoost) is presented in Table S9. Three input combinations were compared. Figure 5 (a1–c1) presents the prediction results of the best models based on LOOCV for the three different input data combinations. When using only VIs, the SVR model at the maturation stage achieved the highest accuracy. The R 2 CV was 0.57, the RMSE CV was 7181 stalks/ha, and the RPD CV was 1.52 (Fig. 5a1). After adding TIs, the SVR model still performed the best at the maturation stage. The R 2 CV increased to 0.63, with an RMSE CV of 6602 stalks/ha and an RPD CV of 1.66 (Fig. 5b1). Finally, with the addition of SCPs and FI, the optimal prediction window shifted to the elongation stage. However, the overall accuracy decreased compared to the previous step. The R 2 CV dropped to 0.62, the RMSE CV increased to 6733 stalks/ha, and the RPD CV fell to 1.62 (Fig. 5c1). The model performance remained within the “fair” category (1.4–1.8) and did not reach the “good” standard (1.8–2.0). To further verify the stability and robustness of the prediction models, the distribution of RPD values was obtained from the bootstrap method. As shown in Fig. 5 (a2–c2), the model in Fig. 5 b2 performed best. Its 25% percentile (1.37), mean (1.51), and 75% percentile (1.63) were higher than those of the other two models. This indicates that the model with VIs + TIs (Fig. 5 b2) achieved the highest accuracy and stability. 3.4.3. Brix estimation model Table S10 shows the Brix prediction results. Preliminary results indicated that the SVR model consistently outperformed the other three models across all growth stages. Figure 6 (a1–c1) presents the prediction results of the SVR model based on LOOCV for three different input data combinations. Using VIs alone, the SVR model at the seedling stage achieved the highest accuracy. The R 2 CV was 0.21, the RMSE CV was 0.63 °Bx, and the RPD CV was 1.12 (Fig. 6a1). After adding TIs, the SVR model still performed the best at the seedling stage. The R 2 CV increased to 0.40, with an RMSE CV of 0.55 °Bx and an RPD CV of 1.29 (Fig. 6b1). Finally, with the addition of SCPs and FI, the optimal prediction window shifted to the maturation stage. Notably, the overall accuracy improved compared to the previous step. The R 2 CV rose to 0.44, the RMSE CV decreased to 0.53 °Bx, and the RPD CV reached 1.33 (Fig. 6c1). The model performance remained within the “poor” category (1.0–1.4) and did not reach the “fair” standard (1.4–1.8). To further verify the stability and robustness of the SVR prediction model, the distribution of RPD CV values was obtained from the bootstrap method (Fig. 6a2–c2). The model in Fig. 6b2 exhibited slightly higher stability metrics (a mean RPD CV of 1.18). However, the model in Fig. 6c1 achieved the highest explanatory power (R 2 CV = 0.44) in the LOOCV assessment. Considering the priority of prediction accuracy, the SVR model developed by VIs + TIs + SCPs + FI was identified as the final model. 3.5. Model interpretation based on SHAP SHAP analysis was used to explain feature contributions to sugarcane trait prediction. Figure 7 presents the SHAP summary and feature importance for the optimal models. For yield prediction, Fig. 7a1 ranks the features by their importance. GNDVI was the most important feature, contributing 24.7% to the model. CI and RVI were the next most important features, contributing 22.7% and 22.5%, respectively. BF_CMF ranked fourth (8.1%), followed by CVI (7.9%) and DDRTI_CON (G,NIR,B,R) (4.7%) occupied the fifth and sixth positions. The inset pie chart shows the contribution of different categories (Fig. 7a2). VIs were the dominant drivers, accounting for 77.8% of the total contribution, which was significantly higher than that of FI (11.7%) and TIs (10.5%). Figure 7 a3 illustrates how these features affect yield; red dots represent high feature values, while blue dots represent low values. Yield prediction was positively correlated with GNDVI (ranked first) and CVI (ranked fifth). Conversely, CI and RVI (ranked second and third) showed a negative correlation, where lower values corresponded to higher yield predictions. Finally, BF_CMF (ranked fourth) had a positive impact on yield, whereas DDRTI_CON (G,NIR,B,R) (ranked sixth) showed a negative correlation. For the prediction of millable stalk population, Fig. 7b1 ranks the features by importance. CVI was the most critical feature, contributing 24.7% to the model. NDTI_HOM (B_G) , EVTI_VAR (NIR_R_B) , and EVTI_MEA (G,NIR,R) followed, contributing 18.6%, 18.0%, and 16.9%, respectively. NDTI_VAR(G,NIR) (9.9%) and EVTI_ASM (G,B,NIR,G) (6.1%) ranked fifth and sixth. The inset pie chart (Fig. 7b2) shows that TIs were the dominant drivers, accounting for 69.5% of the total contribution, which was much higher than that of VIs (30.5%). Figure 7b3 displays the feature effects on millable stalk population. The prediction was positively correlated with CVI (ranked first), NDTI_VAR (G,NIR) (ranked fifth), and EVTI_ASM (G,B,NIR,G) (ranked sixth). In contrast, NDTI_HOM (B,G) , EVTI_VAR (NIR,R,B) , and EVTI_MEA (G,NIR,R) (ranked second, third, and fourth) showed negative correlations, as lower values corresponded to higher millable stalk population predictions. For Brix prediction, Fig. 7c1 ranks the features by importance. GDVI was the top feature, contributing 35.9% to the model, followed by BNDVI with 27.8%. DDRTI_VAR (G,R,B,NIR) (8.9%) and DDRTI_HOM (G,R,B,NIR) (4.5%) ranked third and fourth, while WDRVI took the fifth spot (4.1%). Notably, soil pH also played a role in the model. The inset pie chart (Fig. 7c2) indicates that VIs were the dominant drivers (67.9%), far exceeding the contribution of TIs (30.8%) and SCPs (1.3%). Figure 7c3 shows the feature effects on Brix. Brix prediction was negatively correlated with GDVI (ranked first), DDRTI_VAR (G,R,B,NIR) (ranked third), and DDRTI_HOM (G,R,B,NIR) (ranked fourth). Conversely, BNDVI and WDRVI (ranked second and fifth) showed positive correlations, where higher values corresponded to higher Brix predictions. Soil pH also showed a positive correlation with Brix prediction. 3.6. Prediction mapping based on optimal models The respective models were applied to predict the phenotypic traits of sugarcane. Figure 8 illustrates the spatial distribution of sugarcane yield, millable stalk population, and Brix across the study area for the plant cane (year 2024) and ratoon cane (year 2025) seasons. There was a marked decline in sugarcane yield from the plant cane to first ratoon cane. In 2024 (plant cane), the majority of the field exhibited high yields, ranging between 63 and 105 t/ha (Fig. 8a1). Conversely, in 2025 (first ratoon cane), yields decreased significantly, with most areas shifting to a lower range of 21–63 t/ha (Fig. 8a2). A similar downward trend was observed for the number of millable stalks. The plant cane showed high stalk density, predominantly exceeding 45000 stalks/ha (Fig. 8b1). However, in 2025, the density dropped visibly with a large portion of the field recording values below 45000 stalks/ha (Fig. 8b2). Finally, Brix levels slightly declined from plant cane to first ratoon cane. As illustrated in Fig. 8c1, the 2024 map displays a mix of moderate to high Brix levels (mostly 20–23 °Bx). In contrast, the 2025 map (Fig. 8c2) is dominated by lower values (mostly 19–21 °Bx), suggesting a reduced sucrose accumulation potential in the ratoon year. 4. Discussion 4.1. Comparison of sugarcane traits estimation models Based on the plot level seasonal variation analysis in 2024, NDV of plant cane was generally lower than that of the ratoon during the seedling and tillering stages, which may contribute to the advantage of ratoon cane in canopy establishment. Singla et al. ( 2018 ) indicated that the temporal values of NDVI data of Landsat 8 during the first week of April and last week of August to the end of September can be used to discriminate of ratoon sugarcane in local area. Ratoon sugarcane has a relatively high seedling density, especially in the late emergence stage (Xu et al., 2021 ). The canopies overlap completely, which makes it difficult to clearly distinguish the stem-leaf boundaries of individual plants (Zhu et al., 2026 ). From late tillering to the elongation stage, DDRTI_MEA (B,R,G,NIR) was markedly higher in ratoon cane than in plant cane, and this difference was more pronounced in the mid-to-late season. This is likely associated with the greater variation of first ratoon to accumulated below-ground pests and diseases, drought in seedling, as well as missing stools, and uneven emergence establishment (Xu et al., 2021 ; Dlamini et al., 2024 ). As the sugarcane enters the rapid growth phase, within-plot differences in canopy closure, shadowing, and exposed soil patches tend to be amplified, which resulted in increased spatial heterogeneity of the canopy in high spatial resolution imagery (Li et al., 2025 ). DDRTI_MEA (B,R,G,NIR) is constructed from multi-band GLCM mean texture features and may be more sensitive to within plot spatial structural differences, which makes it better suited to capturing texture changes driven by heterogeneous growth in ratoon cane (Zhu et al., 2026 ). Based on 20 plants randomly selected from each plot at early maturation stage, the plant height, stem diameter, and single stem weight were compared between two years. Plant height did not differ significantly with mean values of 197.25cm in 2024 and 199.42 cm in 2025. Mean stem diameter increased from 2.66cm in 2024 to 2.80cm in 2025. Single stem weight differed significantly with the mean value of 1199 g in 2024 and 1334 g in 2025 (Table S11). Four algorithms (SVR, RR, GPR, and CatBoost) were evaluated in this study. Across all input features, SVR achieved the highest accuracy. This aligns with our previous findings for rice (Yuan et al., 2023 ) and other crops (Shi et al., 2022 ). Indeed, SVR consistently demonstrates superior stability and fitting capabilities in agricultural applications. This is primarily attributed to the principle of structural risk minimization, which ensures that SVR maintains strong generalization ability even with high-dimensional and small-sample datasets. While GPR is also an established approach for such problems, and showed similar performance to RR, both were outperformed by SVR in this study. In contrast, CatBoost exhibited instability and yielded poorer results. Unlike SVR, CatBoost fits residuals via iterative boosting. In small sample scenarios, this approach can be susceptible to noise and prone to overfitting (Friedman, 2002 ). Consequently, lacking the robust regularization of SVR, CatBoost demonstrated lower stability, highlighting the importance of selecting models tailored to the specific properties of the data. During the early growth stage, the R 2 CV s of sugarcane traits prediction model using VIs alone were relatively low. The R 2 CV s for yield varied from 0.14 to 0.29. Those for millable stalk population ranged from 0.27 to 0.33, and the Brix prediction using the SVR model reached only 0.21. Akbarian et al. ( 2024 ) reported the R 2 of 0.53 in training of SVR yield model in early growth stage of sugarcane and R 2 of 0.57 in middle stage based on UAV multispectral data. Vasconcelos et al. ( 2025 ) achieved a R 2 of 0.89 in the local area sugarcane yield prediction based on high spatial resolution data (PlantScope), genetic information and metrological data via the heteroskedastic gamma regression. And the mid-term yield prediction model based on VIs outperformed the early-stage model, which is consistent with the findings of Akbarian et al. ( 2022 ). When VIs were used alone, yield prediction accuracy showed no significant improvement at the elongation and maturity stages, which may indirectly reflect the saturation effect of vegetation indices under dense canopy conditions (Yue et al., 2019 ; Sofonia et al., 2019 ). In the middle and late growth periods, sugarcane canopy coverage approaches saturation, and conventional vegetation indices exhibit reduced sensitivity to biomass, making it difficult to detect subtle canopy variations (Shendryk et al., 2020 ). Sofonia et al. ( 2019 ) reported that the VI based model had the R 2 of 0.322 compared to the R 2 of 0.707 of LiDAR based model in sugarcane yield estimation. This results also indicates that relying solely on VIs is insufficient to characterize the complex process of yield formation. The combination of LiDAR and VIs could improve the yield model performance (Sofonia et al., 2019 ; Shendryk et al., 2020 ). Integrating auxiliary data, such as meteorological variables, genetic data, phenology info, soil texture info, with vegetation indices (VIs) also had the potential to enhance the accuracy of sugarcane yield estimation (Amorim et al., 2022 ; Aleman-Montes et al., 2023). For millable stalk population, the model’s predictive accuracy was highest at the maturation stage, achieving an R 2 CV value of 0.57. This finding aligns closely with the R 2 CV of approximately 0.60 reported by Khuimphukhieo et al. ( 2025 ). Sofinia et al. (2019) reported the R 2 of 0.502 for millable stalk population by UAV multispectral data-based model. With respect to Brix, the associations with VIs were generally weak or statistically non-significant. Although Some studies reported the detectable relationship between VIs and Brix (Chea et al. 2020 ; Aleman-Montes et al., 2023), it varied considerably across sugarcane cultivars. In this study, VIs alone model failed in Brix estimation, while in the maturation stage, the including of soil properties improved explanation of Brix model. Previous studies have demonstrated that texture features of high spatial resolution image and UAV data could partly capture differences in crop canopy characteristics (Zhang et al., 2021 ; Liu et al., 2023a , b ; Zhu et al., 2026 ). The optimum texture window sizes and types of texture parameters depended on data source, crop types, growth stage and agronomic indicators (Liu et al., 2023a , b ). Furthermore, TIs were more useful than individual TFs. Yu et al. ( 2024 ) reported that the correlation between TIs and LAI in apple orchards was stronger than that of TFs. TIs could reflect more detailed features and richer information (Guo et al., 2023 ) and enhance the accuracy of growth simulation (Yuan et al., 2023 ). In this study, good predictive accuracy for sugarcane traits were obtained by the fusion of VIs and TIs. The maximum R 2 CV of the yield model achieved to 0.65, and those of the millable stalk population and Brix reached 0.63 and 0.40, respectively. These results demonstrate that the integration of VIs and TIs derived from Jilin-1 satellite images outperforms the use of VIs alone for predicting sugarcane traits. At the field level, crop growth is closely related to factors such as cultivar, soil fertility, and fertilization management (Li et al., 2026 ). In this study, by including these variables, the R 2 CV for yield increased to 0.70, and the tillering stage remained the optimal prediction period. For Brix, the R 2 CV improved to 0.44, and the optimal prediction stage shifted to the maturity stage. Sugar content is strongly influenced by cultivar, late season crop status, temperature and management related information. In contrast, for millable stalk population, the R 2 CV slightly decreased with the inputs of SCPs and FI. The additional variables may introduce redundant or less informative signals for millable stalk population (Yu et al., 2025 ). 4.2. SHAP analysis of multi-source feature contributions Khosravirad et al. ( 2019 ) and Rahman & Robson ( 2020 ) successfully employed GNDVI to predict sugarcane yield, achieving favorable accuracy. Akbarian et al. ( 2022 ) obtained similar results using CI. In this study, SHAP analysis revealed that GNDVI was the most influential feature, with CI ranking second in contribution for yield estimation. The higher GNDVI values tended to increase the model prediction. However, higher CI and RVI values tended to reduce the yield prediction, which may present nonlinear response patterns with yield. TIs were less dominant than VIs, which may provide complementary information on plot level spatial structural variation. Zhu et al. ( 2026 ) indicated that the combination of VIs and texture features improved the discrimination of sugarcane ratoon. Compared with NDTI (Zheng et al., 2019 ; Yu et al., 2024 ) and EVTI (Yuan et al., 2023 ), DDRTIs showed a greater contribution to yield estimation. DDRTI i s derived by four band GLCM texture features, which may amplify spatial variation in canopy structure across multiple spectral responses. In this study, first ratoon cane exhibited larger within plot heterogeneous (with larger CV values of stem parameters and single stem weight compared to those in 2024). In terms of fertilization, base fertilizer calcium magnesium fertilizer (BF_CMF) and base fertilizer urea (BF_Urea) were identified as important features. However, topdressing was not selected as a final predictive feature. This result indicates that early nutrient supply is critical for sugarcane yield formation. Base fertilizer served as the primary nutrient source for young plants, supporting root growth, tillering, and early canopy development. Similar results were also reported in corn growth (Li et al., 2022 ; Li et al., 2026 ). For the prediction of millable stalk population, EVTI and NDTI contributed more than DDRTI. Although DDRTI was proved effective for yield estimation, it was less suitable for predicting the number of millable stalk population. Increased index complexity did not guarantee improved performance, and it was important to select texture indices for specific traits (Guo et al., 2023 ). Conversely, for Brix prediction, the contribution of DDRTI exceeded that of EVTI and NDTI. Additionally, soil pH showed a positive effect on Brix prediction. Dao et al. ( 2022 ) suggested that pH should be optimized to maximize sugarcane sugar accumulation. Soil pH influences nutrient availability and enzyme activity, such as protein kinase activity, which regulates sucrose metabolism (Huber et al., 1996). For the prediction of sugarcane yield, millable stalk population, and Brix, VIs were among the most important features. This indicates that canopy spectral reflectance characteristics can characterize photosynthetic physiological status and population structure differences during sugarcane growth and development, serving as key remote sensing indicators of crop vigor and yield formation potential. The inclusion of TIs provided complementary plot level spatial information related to within canopy heterogeneity and structural organization, which could not be fully captured by VIs alone. The formulations of NDTI, EVTI, and DDRTI in this study differed from those reported in some studies, where texture indices were derived either by combining multiple texture features from a single band (Zheng et al., 2019) or through a random texture-feature-cross of multiple band combination (Yu et al., 2024 ). In this study, the texture indices were constructed in a form analogous to conventional VIs. This design may enhance interpretability and reduce inconsistencies. 4.3. Limitations and future prospects This study investigated the potential of Jilin-1 multispectral imagery for plot level sugarcane traits estimation. The promising results were reported for yield and quality related traits (millable stalk population and Brix). However, several limitations remain due to constraints in data availability, study scope and methodological design. Firstly, this study only analyzed data from 2024 and 2025, and included only plant cane and first ratoon cane. The results showed that they did not differ significantly in yield or Brix, but significant differences were observed in stalk weight, millable stalk population, and stalk diameter (Table S11). Second ratoon cane data was not included in this study. The conclusions remain limited in the applicability across ratoon generations. Secondly, the texture indices used in this study were able to improve the performance in sugarcane traits estimation, which may be able to capture differences in canopy structure at the plot level. Some texture indices appeared to be more sensitive to overall canopy conditions (e.g., NDRTI). Others seemed more responsive to structural variation at the individual plant level. However, the causes of these differences were not further analyzed. Thirdly, due to the limited sample size, this study mainly relied on traditional machine learning methods. And SVR was identified as the best model. Such methods are practical under small sample size. More data driven algorithms (Akbarian et al., 2024 ), and hybrid framework combining crop growth models with ML (de Oliveira et al., 2020; Xu et al., 2025 ), were not explored. In addition, although high spatial resolution multispectral imagery (Jilin-1) was used, the lack of accurate regional-scale soil property information limited the analysis and map to the experimental plot scale. Large-area mapping and regional validation were not conducted, which constrains the application of the findings. Moreover, under high spatial resolution conditions, canopy reflectance may be affected by factors such as soil background, adjacency effects, planting density, and viewing geometry (Fan et al., 2025 ; Li et al., 2026 ?). In this study, plot-level mean values of reflectance were used for analysis, but the influence of within-plot soil background variation and related factors on reflectance was not assessed. Finally, although the models achieved acceptable accuracy in yield and millable stalk population estimation, the performance for Brix estimation remained relatively low. This sugar-related quality trait is more difficult to monitor using the current remote sensing and modeling framework. Future research should incorporate multi-source data, such as temperature, precipitation, soil properties, varietal characteristics, management practices, thermal, and LiDAR to improve the accuracy and robustness of sugarcane trait estimation (Xie &Yang, 2020 ). 5. Conclusion This study integrated VIs, TIs, SCPs, and FI to predict sugarcane phenotypic traits. SVR had the best performance in the estimation of yield, millable stalk population, and Brix. The inclusion of TIs with different window sizes in different growth stages improved the model accuracy. For sugarcane yield, combining VIs, TIs, SCPs, and FI allowed the model to reach the best prediction accuracy during the tillering period across plant cane and first ratoon cane (R 2 CV = 0.70, RMSE CV = 13.84 t/ha, RPD CV = 1.83). The yield model had the potential to predict sugarcane yield in early growth stage. For millable stalk population, the inclusion of VIs and TIs achieved the highest accuracy at the maturation stage (R 2 CV = 0.63, RMSE CV = 6602 stalks/ha, RPD CV = 1.66). Regarding Brix, the inclusion of SCPs and FI together with VIs and TIs achieved the highest prediction accuracy at the maturation stage (R 2 CV = 0.44, RMSE CV = 0.53 °Bx, RPD CV = 1.33). These results demonstrate the contribution of VIs and texture indices based on optimal texture windows, as well as multi-source data fusion, in sugarcane traits estimation. SCPs suitable for sugarcane growth and appropriate fertilization ratios contributed to high sugarcane yield and Brix. This study validated the applicability of Jilin-1 high-resolution satellite imagery for field-scale sugarcane traits estimation and provided an important insight for remote sensing phenotyping of other crops. Declarations Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. Springer Nature or its licensor (e.g. a society or other partner) holds exclusive rights to this article under a publishing agreement with the author(s) or other rightsholder(s); author self-archiving of the accepted manuscript version of this article is solely governed by the terms of such publishing agreement and applicable law. Declaration of Competing Interest The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper. Author Contribution Yun Zhang: Conceptualization, Data curation, Investigation, Methodology, Writing – original draft. Guangtao Xu: Data curation, Investigation, Methodology, Writing – original draft. Dan Li: Conceptualization, Methodology, Writing – original draft, Writing – review and funding acquisition. Yinglin Lu: Investigation, Data curation, Writing – review and funding acquisition. Hao Jiang: Investigation, Data curation. Shasha Luo: Investigation, Data curation. Mingfu Wen: Investigation, Data curation.Jing Zhang: Investigation, Data curation. Xingda Chen: Investigation, Data curation. Qiong Zheng: Methodology, Data curation. Chongyang Wang: Investigation, Data curation. Zihan Liu: Investigation, Data curation. Mengjun Ku: Investigation, Data curation. Junliang Chen: Investigation, Data curation. Acknowledgements This work was supported by Guangzhou Science and Technology Project (No. 2024B03J1321), GDAS’Project of Science and Technology Development (2022GDASZH-2022010202) and the China Association for Science and Technology (CAST) Special Talent Cultivation Project for PhD Students (2025). Data availability The data that support the findings of this study are available in the main text and the supplementary materials. Part of the code and test data for model training are available at https://github.com/guangtaoxu08-dev/SPT_JL . Additional data are available from the corresponding author upon reasonable request. References Akbarian, S., Xu, C., Wang, W., Ginns, S., & Lim, S. (2022). Sugarcane yields prediction at the row level using a novel cross-validation approach to multi-year multispectral images. Computers and Electronics in Agriculture , 198 , 107024. https://doi.org/10.1016/j.compag.2022.107024 Akbarian, S., Jamnani, M. R., Xu, C., Wang, W., & Lim, S. (2024). Plot level sugarcane yield estimation by machine learning on multispectral images: a case study of Bundaberg, Australia. Information Processing in Agriculture , 11 (4), 476–487. https://doi.org/10.xxxx/j.inpa.2024.xxxx Alemán-Montes, B., Zabala, A., Henríquez, C., & Serra, P. (2023). Modelling two sugarcane agro-industrial yields using Sentinel/Landsat time-series data and their spatial validation at different scales in Costa Rica. Remote Sensing , 15 , 5476. https://doi.org/10.3390/rs15235476 Amorim, M. T. A., Silvero, N. E. Q., Bellinaso, H., Gómez, A. M. R., Greschuk, L. T., Campos, L. R., & Demattê, J. A. M. (2022). Impact of soil types on sugarcane development monitored over time by remote sensing. Precision Agriculture , 23 , 1532–1552. https://doi.org/10.1007/s11119-022-09893-7 Brovelli, A. M., Crespi, M., Fratarcangeli, F., Giannone, F., & Realini, E. (2008). Accuracy assessment of high resolution satellite imagery orientation by leave-one-out method. ISPRS Journal of Photogrammetry and Remote Sensing , 63 (4), 427–440. https://doi.org/10.1016/j.isprsjprs.2008.01.006 Canton, H. (2021). Food and Agriculture Organization of the United Nations—FAO. The Europa Directory of International Organizations, 297–305. ISBN: 978-1-00-317990-0. Chea, C., Saengprachatanarug, K., Posom, J., Wongphati, M., & Taira, E. (2020). Sugar yield parameters and fiber prediction in sugarcane fields using a multispectral camera mounted on a small unmanned aerial system (UAS). Sugar Tech , 22 , 605–621. https://doi.org/10.1007/s12355-020-00802-5 Chen, Z., Zhang, Y., Zang, P., Liu, Z., & Zhao, W. (2023). Mapping the grassland quality of the flooded plain based on Jilin-01 images with a red-edge spectral band. Frontiers in Earth Science , 10 , 1024680. 10.3389/feart.2022.1024680 Dao, J., Mao, J., Li, R., Yang, S., Deng, J., Fan, X., & Zhang, Y. (2022). Effects of key climate and soil factors on sugarcane production in different ecological regions in China. Soil and Fertilizer Sciences in China , (12), 102–113. 10.11838/sfsc.1673-6257.21593 de Silva, F., Chaves, N. R., Luciano, M. E. D., Sanches, A. C. S., de Almeida, I. D., C.M. and, & Adami, M. (2024). Sugarcane yield estimation using satellite remote sensing data in empirical or mechanistic modeling: a systematic review. Remote Sensing , 16 , 863. https://doi.org/10.3390/rs16050863 de Oliveira, M. P. G., & Rodrigues, L. H. A. (2020). How good are the models available for estimating sugar content in sugarcane? European Journal Of Agronomy , 113 , 125992. https://doi.org/10.1016/j.eja.2019.125992 Dlamini, N. E., Franke, A. C., & Zhou, M. (2024). Indices for measuring ratooning ability of sugarcane varieties. Crop Science , 64 , 667–677. https://doi.org/10.1002/csc2.21191 Drucker, H., Burges, C. J. C., Kaufman, L., Smola, A. J., & Vapnik, V. (1997). Support vector regression machines. Advances in Neural Information Processing Systems 9 (pp. 155–161). MIT Press. Du, Y., Zhu, R., Gong, J., Wang, D., & Zhong, X. (2023). Retrieval of crop leaf area index based on Jilin-1GP image. Remote Sensing Technology and Application , 38 (4), 816–826. https://doi.org/10.11873/j.issn.1004-0323.2023.4.0816 Durbha, S. S., King, R. L., & Younan, N. H. (2007). Support vector machines regression for retrieval of leaf area index from multiangle imaging spectroradiometer. Remote Sensing of Environment , 107 (1–2), 348–361. https://doi.org/10.1016/j.rse.2006.07.032 Efron, B. (1979). Bootstrap methods: another look at the jackknife. The Annals of Statistics , 7 (1), 1–26. https://doi.org/10.1214/aos/1176344552 Fan, W. J., Peng, N. J., Cao, B., Mu, X. H., Yang, S. Q., He, Q. C., Zhai, D. C., Ren, H. Z., Cui, Y. K., & Yan, G. J. (2025). Progress in high spatial resolution vegetation quantitative remote sensing. National Remote Sensing Bulletin , 29 (6), 1365–1387. https://doi.org/10.11834/jrs.20255036 FAO (2025). FAOSTAT: crops and livestock products. Available at: http://www.fao.org/faostat/en/#data/QCL (accessed 18 December 2025). Friedman, J. H. (2002). Stochastic gradient boosting. Computational Statistics and Data Analysis , 38 , 367–378. https://doi.org/10.1016/S0167-9473(01)00065-2 Fu, B., Zuo, P., Liu, M., Lan, G., He, H., Lao, Z., Zhang, Y., Fan, D., & Gao, E. (2022). Classifying vegetation communities in karst wetland by synergistic use of image fusion and object-based machine learning algorithm with Jilin-1 and UAV multispectral images. Ecological Indicators , 140 , 108989. https://doi.org/10.1016/j.ecolind.2022.108989 Gao, S., Zhong, R., Yan, K., Ma, X., Chen, X., Pu, J., Gao, S., Qi, J., Yin, G., & Myneni, R. B. (2023). Evaluating the saturation effect of vegetation indices in forests using 3D radiative transfer simulations and satellite observations. Remote Sensing of Environment , 295 , 113665. https://doi.org/10.1016/j.rse.2023.113665 Gaston, E., Frías, J. M., Cullen, P. J., O'Donnell, C. P., & Gowen, A. A. (2010). Prediction of polyphenol oxidase activity using visible near-infrared hyperspectral imaging on mushroom (Agaricus bisporus) caps. Journal of Agricultural and Food Chemistry , 58 (10), 6226–6233. https://doi.org/10.1021/jf100501q Guo, Y., Xiao, Y., Hao, F., Zhang, X., Chen, J., de Beurs, K., He, Y., & Fu, Y. H. (2023). Comparison of different machine learning algorithms for predicting maize grain yield using UAV-based hyperspectral images. International Journal of Applied Earth Observation and Geoinformation , 124 , 103528. https://doi.org/10.1016/j.jag.2023.103528 Guyon, I., Weston, J., Barnhill, S., & Vapnik, V. (2002). Gene selection for cancer classification using support vector machines. Machine Learning , 46 (1–3), 389–422. https://doi.org/10.1023/A:1012487302797 Haralick, R. M., Shanmugam, K., & Dinstein, I. (1973). Textural features for image classification. IEEE Transactions on Systems, Man, and Cybernetics, SMC-3(6), 610–621. https://doi.org/10.1109/TSMC.1973.4309314 Hoerl, A. E., & Kennard, R. W. (1970). Ridge regression: biased estimation for nonorthogonal problems. Technometrics , 12 , 55–67. https://doi.org/10.1080/00401706.1970.10488634 Huber, S. C., & Huber, J. L. (1996). Role and regulation of sucrose-phosphate synthase in higher plants. Annual Review of Plant Physiology and Plant Molecular Biology , 47 , 431–444. https://doi.org/10.1146/annurev.arplant.47.1.431 Jimenez-Sierra, D. A., Benítez-Restrepo, H. D., Vargas-Cardona, H. D., & Chanussot, J. (2020). Graph-based data fusion applied to change detection and biomass estimation in rice crops. Remote Sensing , 12 (17), 2683. https://doi.org/10.3390/rs12172683 Kambli, M., & Palkar, B. (2025). Remote sensing and machine learning methods to analyse the vegetation of sugarcane crop. Multimedia Tools and Applications , 84 , 45297–45319. https://doi.org/10.1007/s11042-025-20950-8 Khosravirad, M., Omid, M., Sarmadian, F., & Hosseinpour, S. (2019). Predicting sugarcane yields in Khuzestan using a large time-series of remote sensing imagery region. Int Arch Photogramm Remote Sens Spat Inf Sci XLII , -4/W18 , 645–648. https://doi.org/10.5194/isprs-archives-xlii-4-w18-645-2019 Khuimphukhieo, I., Bhandari, M., Enciso, J., & da Silva, J. A. (2025). Estimating sugarcane yield and its components using unoccupied aerial systems (UAS)-based high throughput phenotyping (HTP). Computers and Electronics in Agriculture, 237 (Part B), 110658. https://doi.org/10.1016/j.compag.2025.110658 Li, D., Miao, Y., Ransom, C. J., Bean, G. M., Kitchen, N. R., Fernández, F. G., Sawyer, J. E., Camberato, J. J., Carter, P. R., Ferguson, R. B., et al. (2022). Corn nitrogen nutrition index prediction improved by integrating genetic, environmental, and management factors with active canopy sensing using machine learning. Remote Sensing , 14 , 394. https://doi.org/10.3390/rs14020394 Li, D., Wang, C., Miao, Y., et al. (2026). Developing an innovative stacking ensemble machine learning and multi-source data fusion-based precision nitrogen management strategy for corn. Precision Agriculture , 27 , 41. https://doi.org/10.1007/s11119-026-10344-7 Li, J., Duan, T. T., Deng, K., Jing, Z. X., Lin, Z., Liang, Y. Q., Deng, C. N., Gao, Y., & Zheng, C. (2019). Responses of sugarcane varieties growth to nitrogen fertilizer application and related use efficiency. Chinese Journal of Tropical Crops , 40 (4), 649–656. https://doi.org/10.3969/j.issn.1000-2561.2019.04.001 Li, Q., Song, X., Meng, X., Zhang, J., Zhang, M., Chen, L. Y., Li, J., & Wang, B. (2025). Shaping future sugarcane: ideal plant architecture and breeding strategies. Molecular Plant , 18 , 725–728. https://doi.org/10.1016/j.molp.2025.04.004 Liu, Y., Feng, H., Yue, J., Fan, Y., Bian, M., Ma, Y., Jin, X., Song, X., & Yang, G. (2023a). Estimating potato above-ground biomass by using integrated unmanned aerial system-based optical, structural, and textural canopy measurements. Computers and Electronics in Agriculture , 213 , 108229. https://doi.org/10.1016/j.compag.2023.108229 Li, Y. R. (2022). Growth and development of sugarcane (Saccharum spp. hybrid) and its relationship with environmental factors. In K. K. Verma, X. P. Song, V. D. Rajput, S. Solomon, Y. R. Li, & G. P. Rao (Eds.), Agro-industrial Perspectives on Sugarcane Production under Environmental Stress . Springer. https://doi.org/10.1007/978-981-19-3955-6_1 Liu, J., Zhu, Y., Song, L., Su, X., Li, J., Zheng, J., Zhu, X., Ren, L., Wang, W., & Li, X. (2023b). Optimizing window size and directional parameters of GLCM texture features for estimating rice AGB based on UAV multispectral imagery. Frontiers in Plant Science , 14 , 1284235. https://doi.org/10.3389/fpls.2023.1284235 Luciano, A. C. S., Picoli, M. C. A., Duft, D. G., Rocha, J. V., Leal, M. R. L. V., & le Maire, G. (2021). Empirical model for forecasting sugarcane yield on a local scale in Brazil using Landsat imagery and random forest algorithm. Computers and Electronics in Agriculture , 184 , 106063. https://doi.org/10.1016/j.compag.2021.106063 Lundberg, S. M., & Lee, S. I. (2017). A unified approach to interpreting model predictions. In: Proceedings of the 31st International Conference on Neural Information Processing Systems (NIPS’17), pp. 4768–4777. https://doi.org/10.1016/j.compag.2021.106063 Meier, E. A., & Thorburn, P. J. (2016). Long term sugarcane crop residue retention offers limited potential to reduce nitrogen fertilizer rates in Australian wet tropical environments. Frontiers in Plant Science , 7 , 1017. https://doi.org/10.3389/fpls.2016.01017 Ozkan, U. Y., & Demirel, T. (2021). The influence of window size on remote sensing-based prediction of forest structural variables. Ecological Processes , 10 , 60. https://doi.org/10.1186/s13717-021-00330-4 Phang, S. K., Chiang, T. H. A., Happonen, A., & Chang, M. M. L. (2023). From satellite to UAV-based remote sensing: a review on precision agriculture. Ieee Access : Practical Innovations, Open Solutions , 11 , 127057–127076. https://doi.org/10.1109/ACCESS.2023.3330886 Paudel, D., Boogaard, H., de Wit, A., Janssen, S., Osinga, S., Pylianidis, C., & Athanasiadis, I. N. (2021). Machine learning for large-scale crop yield forecasting. Agricultural Systems , 187 , 103016. https://doi.org/10.1016/j.agsy.2020.103016 Rasmussen, C. E., & Williams, C. K. I. (2006). Gaussian Processes for Machine Learning . The MIT Press. 978-0-262-18253-9. Rahman, M. M., & Robson, A. (2020). Integrating Landsat-8 and Sentinel-2 time series data for yield prediction of sugarcane crops at the block level. Remote Sensing , 12 , 1313. https://doi.org/10.3390/rs12081313 Rosa, J. M., Fattori Junior, I. M., Melo, M. L. A., & Marin, F. R. (2026). Remote sensing-driven ensemble smoother assimilation of LAI for regional sugarcane yield estimation. Remote Sensing Applications: Society and Environment , 41 , 101952. https://doi.org/10.1016/j.rsase.2026.101952 Shendryk, Y., Sofonia, J., Garrard, R., Rist, Y., Skocaj, D., & Thorburn, P. (2020). Fine-scale prediction of biomass and leaf nitrogen content in sugarcane using UAV LiDAR and multispectral imaging. International Journal of Applied Earth Observation and Geoinformation , 92 , 102177. https://doi.org/10.1016/j.jag.2020.102177 Singla, S. K., Garg, R. D., & Dubey, O. P. (2018). Sugarcane ratoon discrimination using LANDSAT NDVI temporal data. Spatial Information Research , 26 , 415–425. https://doi.org/10.1007/s41324-018-0184-0 Shi, Y., Gao, Y., Wang, Y., Luo, D., Chen, S., Ding, Z., & Fan, K. (2022). Using unmanned aerial vehicle-based multispectral image data to monitor the growth of intercropping crops in tea plantation. Frontiers in Plant Science , 13 , 820585. https://doi.org/10.3389/fpls.2022.820585 Sofonia, J., Shendryk, Y., Phinn, S., Roelfsema, C., Kendoul, F., & Skocaj, D. (2019). Monitoring sugarcane growth response to varying nitrogen application rates: a comparison of UAV SLAM LiDAR and photogrammetry. International Journal of Applied Earth Observation and Geoinformation , 82 , 101878. https://doi.org/10.1016/j.jag.2019.05.011 Som-ard, J., Atzberger, C., Izquierdo-Verdiguier, E., Vuolo, F., & Immitzer, M. (2021). Remote sensing applications in sugarcane cultivation: a review. Remote Sensing , 13 , 4040. https://doi.org/10.3390/rs13204040 Vasconcelos, J. C. S., Arantes, C. S., Speranza, E. A., Antunes, J. F. G., Barbosa, L. A. F., Cançado, G., & M.d.A. (2025). Predicting sugarcane yield through temporal analysis of satellite imagery during the growth phase. Agronomy , 15 , 793. https://doi.org/10.3390/agronomy15040793 Wang, H., Dai, Y., Yao, Q., Ma, L., Zhang, Z., & Lv, X. (2025). Multi-task learning model driven by climate and remote sensing data collaboration for mid-season cotton yield prediction. Field Crops Research , 333 , 110070. https://doi.org/10.1016/j.fcr.2025.110070 Wu, N., Zhou, Y., Yin, S., Gong, H., & Zhang, C. (2025). Revealing the nonlinear impact of environmental regulation on ecological resilience using the XGBoost-SHAP model: evidence from the Yangtze River Delta region, China. Journal of Cleaner Production , 514 , 145700. https://doi.org/10.1016/j.jclepro.2025.145700 Xie, C., & Yang, C. (2020). A review on plant high-throughput phenotyping traits using UAV-based sensors. Computers and Electronics in Agriculture , 178 , 105731. https://doi.org/10.1016/j.compag.2020.105731 Xu, F., Wang, Z., Lu, G., Zeng, R., & Que, Y. (2021). Sugarcane ratooning ability: research status, shortcomings, and prospects. Biology , 10 (10), 1052. https://doi.org/10.3390/biology10101052 Xu, J., Du, X., Dong, T., Li, Q., Zhang, Y., Wang, H., Liu, M., Zhu, J., & Yang, J. (2025). Estimation of sugarcane biomass from Sentinel-2 leaf area index using an improved SAFY model (SAFY-Sugar). International Journal of Applied Earth Observation and Geoinformation, 140, 104570. https://doi.org/xxxxx. Xuan, F., Su, W., Chen, Z., Huang, X., Zhai, W., Li, X., Zeng, Y., Li, Z., Li, J., & Huang, J. (2025). Performance of stacking machine learning and volume model for improving corn above ground biomass prediction. Plant Phenomics , 7 (3), 100068. Yang, Y., Wu, Z., Wang, D., Wang, C., Yang, X., Wang, Y., Wang, J., Huang, Q., Hou, L., Wang, Z., et al. (2025). Sugarcane phenology retrieval in heterogeneous agricultural landscapes based on spatiotemporal fusion remote sensing data. Agriculture , 15 , 1578. https://doi.org/10.3390/agriculture15151578 Yu, J., Zhang, Y., Song, Z., Jiang, D., Guo, Y., Liu, Y., & Chang, Q. (2024). Estimating leaf area index in apple orchard by UAV multispectral images with spectral and texture information. Remote Sensing , 16 (17). 3237.https://doi.org/10.3390/rs16173237 Yu, Z., Chen, Q., Shan, Q., Lv, Y., Chen, Z., Zhou, Y., Zhang, P., Jiang, H., & Cao, W. (2025). Developing soybean yield prediction model based on multicolor space and texture features of UAV images. Computers and Electronics in Agriculture , 235 , 110345. https://doi.org/10.1016/j.compag.2025.110345 Yuan, W., Meng, Y., Li, Y., Ji, Z., Kong, Q., Gao, R., & Su, Z. (2023). Research on rice leaf area index estimation based on fusion of texture and spectral information. Computers and Electronics in Agriculture , 211 , 108016. https://doi.org/10.1016/j.compag.2023.108016 Yue, J., Yang, G., Tian, Q., Feng, H., Xu, K., & Zhou, C. (2019). Estimate of winter-wheat above-ground biomass based on UAV ultrahigh-ground-resolution image textures and vegetation indices. ISPRS Journal of Photogrammetry and Remote Sensing , 150 , 226–244. https://doi.org/10.1016/j.isprsjprs.2019.02.022 Zhang, J., Qiu, X., Wu, Y., Zhu, Y., Cao, Q., Liu, X., & Cao, W. (2021). Combining texture, color, and vegetation indices from fixed-wing UAS imagery to estimate wheat growth parameters using multivariate regression methods. Computers and Electronics in Agriculture , 185 , 106138. https://doi.org/10.1016/j.compag.2021.106138 Zhang, Y., Zhao, Z., & Zheng, J. (2020). CatBoost: a new approach for estimating daily reference crop evapotranspiration in arid and semi-arid regions of Northern China. Journal of Hydrology , 588 , 125087. https://doi.org/10.1016/j.jhydrol.2020.125087 Zhao, J., Li, J., Liu, Q., & Yang, L. (2012). A preliminary study on mechanism of LAI inversion saturation. The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, XXXIX-B1, 77–81. https://doi.org/10.5194/isprsarchives-XXXIX-B1-77-2012 Zheng, H., Cheng, T., Zhou, M., Li, D., Yao, X., Tian, Y., Cao, W., & Zhu, Y. (2019). Improved estimation of rice aboveground biomass combining textural and spectral analysis of UAV imagery. Precision Agriculture , 20 , 611–629. https://doi.org/10.1007/s11119-018-9600-7 Zhu, H., Dong, Z., Wei, L., Qin, S., Qin, X., & He, Y. (2026). From canopy segmentation to accurate prediction: an UAV-based multi-feature fusion framework for plot-scale ratoon sugarcane seedling counting. International Journal of Applied Earth Observation and Geoinformation , 147 , 105183. https://doi.org/10.1016/j.jag.2026.105183 Additional Declarations No competing interests reported. Supplementary Files SupplementaryMaterial.docx Cite Share Download PDF Status: Posted Version 1 posted You are reading this latest preprint version Research Square lets you share your work early, gain feedback from the community, and start making changes to your manuscript prior to peer review in a journal. As a division of Research Square Company, we’re committed to making research communication faster, fairer, and more useful. We do this by developing innovative software and high quality services for the global research community. Our growing team is made up of researchers and industry professionals working together to solve the most critical problems facing scientific publishing. Also discoverable on Platform About Our Team In Review Editorial Policies Advisory Board Help Center Resources Author Services Accessibility API Access RSS feed Manage Cookie Preferences © Research Square 2026 | ISSN 2693-5015 (online) Privacy Policy Terms of Service Do Not Sell My Personal Information {"props":{"pageProps":{"initialData":{"identity":"rs-9371174","acceptedTermsAndConditions":true,"allowDirectSubmit":true,"archivedVersions":[],"articleType":"Research Article","associatedPublications":[],"authors":[{"id":627919444,"identity":"8520a9ef-692a-4c10-a726-19e98f508cea","order_by":0,"name":"Yun Zhang","email":"","orcid":"","institution":"Anhui Normal University","correspondingAuthor":false,"prefix":"","firstName":"Yun","middleName":"","lastName":"Zhang","suffix":""},{"id":627919450,"identity":"711f0c15-43ce-4028-a1e2-f64e7f7ec2b2","order_by":1,"name":"Guangtao Xu","email":"","orcid":"","institution":"Anhui Normal University","correspondingAuthor":false,"prefix":"","firstName":"Guangtao","middleName":"","lastName":"Xu","suffix":""},{"id":627919452,"identity":"f546d658-1880-4519-97d1-6e5035688bc0","order_by":2,"name":"Dan Li","email":"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAZAAAAAyAQMAAABI0h/eAAAABlBMVEX///8AAABVwtN+AAAACXBIWXMAAA7EAAAOxAGVKw4bAAAAvklEQVRIiWNgGAWjYBACPmYGAwaGCgbGBqK1sIG1nCFJCwNQC2MbSVrYmTd+LpxXK9vff4Dxww8GuzwiHMZWLD1z23HjGTcSmCV7GJKLidDCYyDNu+1YYsMNBgZpBoYDiQRdCNRi/Jt3zrHE+ecPMP8mVouZNG9DTeKGAwlsxNrCVmbNc+yA8cYbiW2WPQbJhLXw8x/efJunpk523vnDh2/8qLAjrAUKDgMxKGoMiFQPBHXEKx0Fo2AUjIKRBwAgJzev8A9RzQAAAABJRU5ErkJggg==","orcid":"","institution":"Guangzhou Institute of Geography, Guangdong Academy of Sciences","correspondingAuthor":true,"prefix":"","firstName":"Dan","middleName":"","lastName":"Li","suffix":""},{"id":627919454,"identity":"71780dbc-d5e3-4f46-a393-6b81bd40fef6","order_by":3,"name":"Yinglin Lu","email":"","orcid":"","institution":"Institute of Nanfan and Seed Industry, Guangdong Academy of Sciences","correspondingAuthor":false,"prefix":"","firstName":"Yinglin","middleName":"","lastName":"Lu","suffix":""},{"id":627919462,"identity":"60c7e713-16be-40f6-81a9-4864975943f4","order_by":4,"name":"Hao Jiang","email":"","orcid":"","institution":"Guangzhou Institute of Geography, Guangdong Academy of Sciences","correspondingAuthor":false,"prefix":"","firstName":"Hao","middleName":"","lastName":"Jiang","suffix":""},{"id":627919465,"identity":"d869d961-8511-4c12-ab49-c72c90e94f6c","order_by":5,"name":"Shasha Luo","email":"","orcid":"","institution":"Institute of Nanfan and Seed Industry, Guangdong Academy of Sciences","correspondingAuthor":false,"prefix":"","firstName":"Shasha","middleName":"","lastName":"Luo","suffix":""},{"id":627919472,"identity":"fcbf37be-e447-4acd-92cf-194f64f43bf9","order_by":6,"name":"Mingfu Wen","email":"","orcid":"","institution":"Zhanjiang Research Center, Institute of Nanfan and Seed Industry, Guangdong Academy of Sciences","correspondingAuthor":false,"prefix":"","firstName":"Mingfu","middleName":"","lastName":"Wen","suffix":""},{"id":627919478,"identity":"fe393506-f6c5-4b7d-8346-613392ff2875","order_by":7,"name":"Jing Zhang","email":"","orcid":"","institution":"Guangzhou Institute of Geography, Guangdong Academy of Sciences","correspondingAuthor":false,"prefix":"","firstName":"Jing","middleName":"","lastName":"Zhang","suffix":""},{"id":627919479,"identity":"dd44232a-97d6-4fcd-9c5a-ba2b1b931a33","order_by":8,"name":"Xingda Chen","email":"","orcid":"","institution":"Guangzhou Institute of Geography, Guangdong Academy of Sciences","correspondingAuthor":false,"prefix":"","firstName":"Xingda","middleName":"","lastName":"Chen","suffix":""},{"id":627919480,"identity":"33bc0e84-aed0-4809-9083-a7dccaf34aee","order_by":9,"name":"Qiong Zheng","email":"","orcid":"","institution":"Changsha University of Science and Technology","correspondingAuthor":false,"prefix":"","firstName":"Qiong","middleName":"","lastName":"Zheng","suffix":""},{"id":627919483,"identity":"b669f9c7-7d81-4b18-90a5-562693a2855f","order_by":10,"name":"Chongyang Wang","email":"","orcid":"","institution":"Guangzhou Institute of Geography, Guangdong Academy of Sciences","correspondingAuthor":false,"prefix":"","firstName":"Chongyang","middleName":"","lastName":"Wang","suffix":""},{"id":627919485,"identity":"2d0f4128-f00c-4895-b4d0-95c6caf9c8de","order_by":11,"name":"Zihan Liu","email":"","orcid":"","institution":"Anhui Normal University","correspondingAuthor":false,"prefix":"","firstName":"Zihan","middleName":"","lastName":"Liu","suffix":""},{"id":627919486,"identity":"568f5429-2d02-43c5-a556-4a58cce313bd","order_by":12,"name":"Mengjun Ku","email":"","orcid":"","institution":"Guangzhou Institute of Geography, Guangdong Academy of Sciences","correspondingAuthor":false,"prefix":"","firstName":"Mengjun","middleName":"","lastName":"Ku","suffix":""},{"id":627919487,"identity":"1e095343-a9c7-4442-a820-e07dde7a5539","order_by":13,"name":"Junliang Chen","email":"","orcid":"","institution":"Guangzhou Institute of Geography, Guangdong Academy of Sciences","correspondingAuthor":false,"prefix":"","firstName":"Junliang","middleName":"","lastName":"Chen","suffix":""}],"badges":[],"createdAt":"2026-04-09 16:54:02","currentVersionCode":1,"declarations":"","doi":"10.21203/rs.3.rs-9371174/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-9371174/v1","draftVersion":[],"editorialEvents":[],"editorialNote":"","failedWorkflow":false,"files":[{"id":107900358,"identity":"34b4ca8c-78e9-4e2e-aeda-e3aadcd830a1","added_by":"auto","created_at":"2026-04-27 11:31:48","extension":"png","order_by":1,"title":"Figure 1","display":"","copyAsset":false,"role":"figure","size":524118,"visible":true,"origin":"","legend":"\u003cp\u003eGeographical location of the study area. (a) Inset map showing the location of Guangdong Province; (b) Location of Suixi County; (c) Specific distribution of the experimental plots.\u003c/p\u003e","description":"","filename":"floatimage1.png","url":"https://assets-eu.researchsquare.com/files/rs-9371174/v1/11337bba81bbfcb9f1f89edd.png"},{"id":108007262,"identity":"00d8d639-a3d3-4cc4-92af-27589c587a5b","added_by":"auto","created_at":"2026-04-28 12:59:12","extension":"png","order_by":2,"title":"Figure 2","display":"","copyAsset":false,"role":"figure","size":420441,"visible":true,"origin":"","legend":"\u003cp\u003eWorkflow of this study.\u003c/p\u003e","description":"","filename":"floatimage2.png","url":"https://assets-eu.researchsquare.com/files/rs-9371174/v1/0df97944b1ed0620473d03dd.png"},{"id":107900359,"identity":"88e5ff4b-5e41-4b04-a640-5709d7e32c20","added_by":"auto","created_at":"2026-04-27 11:31:48","extension":"png","order_by":3,"title":"Figure 3","display":"","copyAsset":false,"role":"figure","size":396010,"visible":true,"origin":"","legend":"\u003cp\u003eModel performance (RPD) under different texture window sizes across growth stages. (a) Yield; (b) Millable Stalk Population; (c) Brix.\u003c/p\u003e","description":"","filename":"floatimage3.png","url":"https://assets-eu.researchsquare.com/files/rs-9371174/v1/4f7e7e223666b9eeada7066e.png"},{"id":108006779,"identity":"6559ac8b-281e-4a65-8712-a74e4f607ef8","added_by":"auto","created_at":"2026-04-28 12:57:05","extension":"png","order_by":4,"title":"Figure 4","display":"","copyAsset":false,"role":"figure","size":382989,"visible":true,"origin":"","legend":"\u003cp\u003ePrediction results of yield using the best performing models for three different input feature combinations. Top row (LOOCV results): (a1) SVR with VIs (Tillering stage); (b1) RR with VIs + TIs (Tillering stage); (c1) SVR with VIs + TIs + SCPs + FI (Tillering stage). Bottom row (Bootstrap results): (a2) SVR with VIs (Tillering stage); (b2) RR with VIs + TIs (Tillering stage); (c2) SVR with VIs + TIs + SCPs + FI (Tillering stage).\u003c/p\u003e","description":"","filename":"floatimage4.png","url":"https://assets-eu.researchsquare.com/files/rs-9371174/v1/9ea398853c978b1954ff43e5.png"},{"id":107900361,"identity":"fa0e6848-1051-4ab2-8155-070079533b20","added_by":"auto","created_at":"2026-04-27 11:31:48","extension":"png","order_by":5,"title":"Figure 5","display":"","copyAsset":false,"role":"figure","size":410439,"visible":true,"origin":"","legend":"\u003cp\u003ePrediction results of millable stalk population using the best performing models for three different input feature combinations. Top row (LOOCV results): (a1) SVR with VIs (Maturation stage); (b1) SVR with VIs + TIs (Maturation stage); (c1) SVR with VIs + TIs + SCPs + FI (Elongation stage). Bottom row (Bootstrap results): (a2) SVR with VIs (Maturation stage); (b2) SVR with VIs + TIs (Maturation stage); (c2) SVR with VIs + TIs + SCPs + FI (Elongation stage).\u003c/p\u003e","description":"","filename":"floatimage5.png","url":"https://assets-eu.researchsquare.com/files/rs-9371174/v1/6e8205e94db8d2d5bc9a816c.png"},{"id":108006880,"identity":"ad2c0776-fefb-46c9-b154-747298218409","added_by":"auto","created_at":"2026-04-28 12:57:45","extension":"png","order_by":6,"title":"Figure 6","display":"","copyAsset":false,"role":"figure","size":406049,"visible":true,"origin":"","legend":"\u003cp\u003ePrediction results of Brix by the SVR model using three different input feature combinations. Top row (LOOCV results): (a1) VIs (Seedling stage); (b1) VIs + TIs (Seedling stage); (c1) VIs + TIs + SCPs + FI (Maturation stage). Bottom row (Bootstrap results): (a2) VIs (Seedling stage); (b2) VIs + TIs (Seedling stage); (c2) VIs + TIs + SCPs + FI (Maturation stage).\u003c/p\u003e","description":"","filename":"floatimage6.png","url":"https://assets-eu.researchsquare.com/files/rs-9371174/v1/328911571ebbc2fbef42d9a3.png"},{"id":108007081,"identity":"7fdbd57b-f074-431f-91cf-1959c87e0497","added_by":"auto","created_at":"2026-04-28 12:58:27","extension":"png","order_by":7,"title":"Figure 7","display":"","copyAsset":false,"role":"figure","size":562976,"visible":true,"origin":"","legend":"\u003cp\u003eSHAP feature importance plots for the optimal prediction models. (a1)–(a3) Yield (SVR: VIs + TIs + SCPs + FI, Tillering stage); (b1)–(b3) Millable Stalk Population (SVR: VIs + TIs, Maturation stage); (c1)–(c3) Brix (SVR: VIs + TIs + SCPs + FI, Maturation stage).\u003c/p\u003e","description":"","filename":"floatimage7.png","url":"https://assets-eu.researchsquare.com/files/rs-9371174/v1/82266933dbad3aed42f4c4dc.png"},{"id":107900362,"identity":"4bddc7e0-17e4-47ef-a79e-e8f193eb4ce0","added_by":"auto","created_at":"2026-04-27 11:31:48","extension":"png","order_by":8,"title":"Figure 8","display":"","copyAsset":false,"role":"figure","size":575461,"visible":true,"origin":"","legend":"\u003cp\u003ePrediction results of sugarcane phenotypic traits based on the optimal model. (a1) Yield (plant cane)-2024; (a2) Yield (first ratoon cane)-2025; (b1) Millable Stalk Population-2024; (b2) Millable Stalks-2025; (c1) Brix-2024; (c2) Brix-2025.\u003c/p\u003e","description":"","filename":"floatimage8.png","url":"https://assets-eu.researchsquare.com/files/rs-9371174/v1/df2c36c38b7cd6a929b7c27d.png"},{"id":108491162,"identity":"b57b4cd0-ca16-475f-b013-3dd855ae77b5","added_by":"auto","created_at":"2026-05-05 09:52:37","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":4245838,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-9371174/v1/a5bb07a0-b067-4f29-8c2f-b4dcc7931c50.pdf"},{"id":107900357,"identity":"53aaf3d1-65b2-4825-99c7-a8d65f04de53","added_by":"auto","created_at":"2026-04-27 11:31:48","extension":"docx","order_by":0,"title":"","display":"","copyAsset":false,"role":"supplement","size":6629172,"visible":true,"origin":"","legend":"","description":"","filename":"SupplementaryMaterial.docx","url":"https://assets-eu.researchsquare.com/files/rs-9371174/v1/c607966a4648fa7f95980f6d.docx"}],"financialInterests":"No competing interests reported.","formattedTitle":"Integrating spectral, texture, soil and fertilization information for plot-level prediction of sugarcane yield, millable stalk population and Brix from Jilin-1 imagery ","fulltext":[{"header":"1. Introduction","content":"\u003cp\u003eSugarcane (\u003cem\u003eSaccharum\u003c/em\u003e spp.) is one of the most important crops in tropical and subtropical regions, with a global planting area of more than 27\u0026nbsp;million hectares (FAO, \u003cspan citationid=\"CR18\" class=\"CitationRef\"\u003e2025\u003c/span\u003e). It contributes more than 80% of global raw sugar production and plays a major role in bioenergy development, accounting for approximately 40% of global fuel ethanol production (Canton, \u003cspan citationid=\"CR6\" class=\"CitationRef\"\u003e2021\u003c/span\u003e; Rosa et al., \u003cspan citationid=\"CR48\" class=\"CitationRef\"\u003e2026\u003c/span\u003e). Stable high-yield and high-quality sugarcane production is important for the sustainable development of both the sugar and bioenergy industries. Sugarcane is a semi-perennial crop with a growth cycle of approximately 12\u0026ndash;18 months, typically including the seedling, tillering, elongation, and maturation stages (Som-ard et al., \u003cspan citationid=\"CR53\" class=\"CitationRef\"\u003e2021\u003c/span\u003e). Across these stages, substantial changes occur in canopy structure, biomass accumulation, leaf area index (LAI), and physiological status, which lead to pronounced variation in remote sensing signals (Yang et al., \u003cspan citationid=\"CR61\" class=\"CitationRef\"\u003e2025\u003c/span\u003e). During the seedling stage, new ratoons emerge, and biomass and LAI remain low. At the tillering stage, canopy cover expands rapidly and biomass accumulation accelerates. During the elongation stage, stalk elongation and leaf expansion dominate, leading to rapid biomass accumulation, high canopy closure and an increasingly complex vertical canopy structure. At the maturation stage, leaves progressively senesce and chlorophyll content declines, while sucrose accumulates in the stalks (Li et al., \u003cspan citationid=\"CR35\" class=\"CitationRef\"\u003e2019\u003c/span\u003e). This phenological transition may weaken the direct relationship between canopy greenness and yield-related traits, which makes conventional greenness-based indicators less reliable for monitoring final production status (Som-ard et al., \u003cspan citationid=\"CR53\" class=\"CitationRef\"\u003e2021\u003c/span\u003e).\u003c/p\u003e \u003cp\u003eSugarcane traits, such as millable stalk population, stalk weight, plant height, biomass and sucrose content, directly determine yield and economic return (Meier et al., 2016; Li et al., \u003cspan citationid=\"CR35\" class=\"CitationRef\"\u003e2019\u003c/span\u003e). Early season acquisition of these traits is important. It enables timely management interventions, supports early screening in breeding programs, and improves decision making before harvest (Akbarian et al., \u003cspan citationid=\"CR2\" class=\"CitationRef\"\u003e2024\u003c/span\u003e). However, conventional sugarcane phenotyping still relies heavily on manual field surveys and destructive sampling. Estimating yield and the millable stalk population usually requires large-scale field harvesting and manual counting (Zhu et al., \u003cspan citationid=\"CR70\" class=\"CitationRef\"\u003e2026\u003c/span\u003e). Determining sugar content depends on destructive crushing of mature stalks (de Fran\u0026ccedil;a e Silva et al., \u003cspan citationid=\"CR10\" class=\"CitationRef\"\u003e2024\u003c/span\u003e). These approaches are laborious-intensive, costly and destructive, which limits the application of precision agriculture. Remote sensing provides a non-destructive and large-scale alternative for crop monitoring and has been widely used in sugarcane mapping, growth monitoring, stress detection, and yield estimation (Jimenez-Sierra et al., \u003cspan citationid=\"CR28\" class=\"CitationRef\"\u003e2020\u003c/span\u003e; Xu et al., \u003cspan citationid=\"CR59\" class=\"CitationRef\"\u003e2025\u003c/span\u003e; Kambli \u0026amp; Palkar, \u003cspan citationid=\"CR29\" class=\"CitationRef\"\u003e2025\u003c/span\u003e).\u003c/p\u003e \u003cp\u003eMany studies have used vegetation indices such as NDVI, GNDVI and EVI, to characterize canopy greenness and vigor (Akbarian et al., \u003cspan citationid=\"CR1\" class=\"CitationRef\"\u003e2022\u003c/span\u003e). However, some commonly used broadband vegetation indices (e.g., NDVI) may become less sensitive under medium to high vegetation cover (Gao et al., \u003cspan citationid=\"CR21\" class=\"CitationRef\"\u003e2023\u003c/span\u003e). In addition, many inversion models were originally developed based on medium and low spatial resolution imagery in which each pixel generally represents the average condition of a relatively large canopy area (Luciano et al., \u003cspan citationid=\"CR40\" class=\"CitationRef\"\u003e2021\u003c/span\u003e; Amorim et al., \u003cspan citationid=\"CR4\" class=\"CitationRef\"\u003e2022\u003c/span\u003e). As spatial resolution increases from medium or low resolution to the meter scale, the physical meaning of a pixel changes. Canopy heterogeneity, geometric effects, and local radiometric fluctuations become much more pronounced (Fan et al., \u003cspan citationid=\"CR17\" class=\"CitationRef\"\u003e2025\u003c/span\u003e). The applicability of these models to high spatial resolution data has not been thoroughly evaluated. Thus, traditional models may lead to greater retrieval uncertainty. Furthermore, canopy saturation is prone to occur when vegetation indices are applied to medium to high vegetation coverage (Zhao et al., \u003cspan citationid=\"CR68\" class=\"CitationRef\"\u003e2012\u003c/span\u003e). Relying solely on spectral information is insufficient to capture the spatial complexity of sugarcane canopies, which further constrains the prediction of canopy traits (Liu et al., \u003cspan citationid=\"CR37\" class=\"CitationRef\"\u003e2023a\u003c/span\u003e).\u003c/p\u003e \u003cp\u003eHigh spatial resolution remote sensing data offer spectral information and spatial details (including texture, shape, edge, and canopy arrangement patterns). Texture features derived from high-resolution imagery have the potential to characterize canopy spatial organization and structural variability (Zhu et al., \u003cspan citationid=\"CR70\" class=\"CitationRef\"\u003e2026\u003c/span\u003e). The influences of canopy gaps, shadows, row structure, and adjacency-related radiometric interactions become more evident (Fan et al., \u003cspan citationid=\"CR17\" class=\"CitationRef\"\u003e2025\u003c/span\u003e). Some studies have shown that integrating spectral and texture features from UAV imagery, can improve the estimation of crop biomass (Xie \u0026amp; Yang, \u003cspan citationid=\"CR57\" class=\"CitationRef\"\u003e2020\u003c/span\u003e; Liu et al., \u003cspan citationid=\"CR37\" class=\"CitationRef\"\u003e2023a\u003c/span\u003e), leaf area index (Yuan et al., \u003cspan citationid=\"CR64\" class=\"CitationRef\"\u003e2023\u003c/span\u003e) and yield (Yu et al., \u003cspan citationid=\"CR63\" class=\"CitationRef\"\u003e2025\u003c/span\u003e). However, the performance of texture based methods is also sensitive to feature extraction settings, such as window size, and direction, which may substantially affect parameter inversion accuracy and model robustness (Ozkan \u0026amp; Demirel, \u003cspan citationid=\"CR43\" class=\"CitationRef\"\u003e2021\u003c/span\u003e). Moreover, the limited spatial coverage and operational scalability of UAV imagery restrict its use in routine large-area applications (Phang et al., \u003cspan citationid=\"CR44\" class=\"CitationRef\"\u003e2023\u003c/span\u003e). Meter-level satellite imagery is a critical observational scale that bridges ultra fine observations and operational agricultural monitoring (Aleman-Montes et al., 2023; Fan et al., \u003cspan citationid=\"CR17\" class=\"CitationRef\"\u003e2025\u003c/span\u003e). Open-access satellite constellations such as Sentinel have played an important role in crop monitoring (Luciano et al., \u003cspan citationid=\"CR40\" class=\"CitationRef\"\u003e2021\u003c/span\u003e). However, these data are often constrained by revisit intervals and cloud contamination in tropical and subtropical regions. Increasing attention has been directed toward high resolution commercial imagery for more detailed agricultural monitoring and improved yield estimation (Wang et al., \u003cspan citationid=\"CR55\" class=\"CitationRef\"\u003e2025\u003c/span\u003e). Nevertheless, systematic investigations of texture features under high spatial resolution conditions, including the optimal parameters for texture feature extraction and effective combination strategies remain limited. The potential of integrating spectral and texture data has not yet been fully explored for sugarcane phenotyping.\u003c/p\u003e \u003cp\u003eAmong these platforms, the Chinese Jilin-1 satellite constellation provides a balance among spatial resolution, revisit frequency and cost-effectiveness. Its application potential has been evaluated in leaf area index inversion for corn and rice (Du et al., \u003cspan citationid=\"CR14\" class=\"CitationRef\"\u003e2023\u003c/span\u003e), environmental quality assessment in flood-prone areas (Chen et al., \u003cspan citationid=\"CR8\" class=\"CitationRef\"\u003e2023\u003c/span\u003e), and wetland vegetation classification using UAV-satellite fusion technology (Fu et al., \u003cspan citationid=\"CR20\" class=\"CitationRef\"\u003e2022\u003c/span\u003e). Although Jilin-1 imagery perform effectively in these fields, its potential in field-level crop phenotyping monitor has not been fully investigated. For sugarcane yield estimation, previous studies have adopted process-based crop models combined with data assimilation to improve prediction accuracy at regional scales (Xu et al., \u003cspan citationid=\"CR59\" class=\"CitationRef\"\u003e2025\u003c/span\u003e). However, applying these process-based framework to meter level imagery is often data- and parameter-intensive. Acquiring such fine scale inputs is challenging to, which limits their practical applicability in fine-scale crop monitoring (de Fran\u0026ccedil;a e Silva et al., \u003cspan citationid=\"CR10\" class=\"CitationRef\"\u003e2024\u003c/span\u003e; Rosa et al., \u003cspan citationid=\"CR48\" class=\"CitationRef\"\u003e2026\u003c/span\u003e). Machine learning (ML) algorithms are good at modeling complex non-linear relationships. They have shown great potential in dealing with multi-source heterogeneous data (Paudel et al., \u003cspan citationid=\"CR45\" class=\"CitationRef\"\u003e2021\u003c/span\u003e; Li et al., \u003cspan citationid=\"CR34\" class=\"CitationRef\"\u003e2026\u003c/span\u003e). In addition, explainable artificial intelligence has also attracted increasing attention due to the transparency of model decision-making processes (Xuan et al., \u003cspan citationid=\"CR60\" class=\"CitationRef\"\u003e2025\u003c/span\u003e; Wu et al., \u003cspan citationid=\"CR56\" class=\"CitationRef\"\u003e2025\u003c/span\u003e). Integrating high-resolution spectral features, texture information, and multi-source agricultural auxiliary data for plot level sugarcane trait prediction by ML algorithm has not yet been fully established.\u003c/p\u003e \u003cp\u003eThis study aims to evaluate the potential of meter-level Jilin-1 imagery for plot-level sugarcane phenotyping. Plot-level field observations were collected for key sugarcane traits, including yield, millable stalk population, and Brix. Three objectives were addressed: 1) trait estimation, assessing the ability of spectral and texture indices to estimate key sugarcane traits; 2) data integration, evaluating whether the incorporation of multi-source agricultural auxiliary data improves prediction accuracy; and 3) model interpretation, identifying the most important predictors for different traits using explainable machine learing.\u003c/p\u003e"},{"header":"2. Materials and methods","content":"\u003cdiv id=\"Sec3\" class=\"Section2\"\u003e \u003ch2\u003e2.1. Study area and experimental design\u003c/h2\u003e \u003cp\u003eA two-year trial was conducted at the Zhanjiang Research Center (Nanfan Seed Industry Research Institute of Guangdong Academy of Sciences) in Suixi County, Guangdong Province (Fig.\u0026nbsp;\u003cspan refid=\"Fig1\" class=\"InternalRef\"\u003e1\u003c/span\u003e). This location has a tropical monsoon climate, with an annual average temperature of about 23\u0026deg;C and precipitation exceeding 1400 mm. There is sufficient sunlight, heat and water, which are highly suitable for sugarcane cultivation. The study spanned two years, focusing on plant cane in 2024 and first ratoon cane in 2025. The experimental field consisted of 32 plots, and the sugarcane variety planted was \u0026lsquo;Yutang 03373\u0026rsquo;. Each 100-square-meter plot contained 9 rows, with each row approximately 10 m in length.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003eBasal and topdressing fertilizers were applied at the seedling and tillering stages, respectively. The experiment comprised eight fertilization regimes (T1\u0026ndash;T8) using a compound fertilizer (N: P\u003csub\u003e2\u003c/sub\u003eO\u003csub\u003e5\u003c/sub\u003e: K\u003csub\u003e2\u003c/sub\u003eO\u0026thinsp;=\u0026thinsp;15: 15: 15), as detailed in Table \u003cspan refid=\"MOESM1\" class=\"InternalRef\"\u003eS1\u003c/span\u003e.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec4\" class=\"Section2\"\u003e \u003ch2\u003e2.2. Data acquisition and preprocessing\u003c/h2\u003e \u003cdiv id=\"Sec5\" class=\"Section3\"\u003e \u003ch2\u003e2.2.1. Field data acquisition\u003c/h2\u003e \u003cp\u003ePhenotypic data were collected on November 19, 2024 (plant cane) and December 2, 2025 (first ratoon cane), at the early maturity stage. In addition, for each plot, 20 plants were randomly selected to measure plant height (cm), stem diameter (cm) and single stem weight (g). To minimize edge effects, sampling was restricted to the three central rows (rows 4\u0026ndash;6) of each plot. Twenty stalks were randomly selected from this zone for yield component analysis. Brix was measured from the juice of middle internodes using a handheld refractometer. Millable stalk population were defined as healthy, mature stalks meeting agronomic standards for height and diameter. The total number of stalks was estimated by counting the number of stalks in the middle row and extrapolating the density to the plot level. Yield was calculated based on the formula for stem fresh weight per hectare. The calculation formulas are as follows:\u003cdiv id=\"Equ1\" class=\"Equation\"\u003e\u003cdiv format=\"TEX\" class=\"mathdisplay\" id=\"FileID_Equ1\" name=\"EquationSource\"\u003e\n$$\\:W=\\pi\\:\\times\\:{\\left(\\frac{D}{2}\\right)}^{2}\\times\\:L\\times\\:\\rho\\:$$\u003c/div\u003e\u003cdiv class=\"EquationNumber\"\u003e1\u003c/div\u003e\u003c/div\u003e\u003cdiv id=\"Equ2\" class=\"Equation\"\u003e\u003cdiv format=\"TEX\" class=\"mathdisplay\" id=\"FileID_Equ2\" name=\"EquationSource\"\u003e\n$$\\:Y=W\\times\\:MS$$\u003c/div\u003e\u003cdiv class=\"EquationNumber\"\u003e2\u003c/div\u003e\u003c/div\u003e\u003c/p\u003e \u003cp\u003ewhere \u003cem\u003eY\u003c/em\u003e is the yield (t/ha), \u003cem\u003eW\u003c/em\u003e is the theoretical single stalk weight (kg), \u003cem\u003eD\u003c/em\u003e and \u003cem\u003eL\u003c/em\u003e denote stalk diameter (cm) and length (cm), respectively, \u003cem\u003eMS\u003c/em\u003e is the number of millable stalks, and \u003cem\u003eρ\u003c/em\u003e is the stalk density (constant at 1.05 g/cm\u0026sup3;).\u003c/p\u003e \u003cp\u003eSoil samples were collected from the 0\u0026ndash;20 cm soil layer of each plot prior to planting. The soil chemical properties were analyzed in the laboratory (Table S2).\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec6\" class=\"Section3\"\u003e \u003ch2\u003e2.2.2. Remote sensing data acquisition and preprocessing\u003c/h2\u003e \u003cp\u003eJilin-1 satellite imagery for the seedling, tillering, elongation and maturation stages was obtained via the official data portal (\u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://www.jl1mall.com/\u003c/span\u003e\u003cspan address=\"https://www.jl1mall.com/\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e). Each scene included a panchromatic band and four multispectral bands (Blue, Green, Red, and Near-Infrared). Table S3 details the sensor specifications and acquisition dates. We digitized plot boundaries using QGIS 3.34. The preprocessing workflow was conducted in ENVI 5.6, involving radiometric calibration, atmospheric correction, geometric registration, and image fusion to derive surface reflectance. The spectral response functions for atmospheric correction were obtained from the website (\u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://www.jl1mall.com/resrepo/\u003c/span\u003e\u003cspan address=\"https://www.jl1mall.com/resrepo/\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e).\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec7\" class=\"Section3\"\u003e \u003ch2\u003e2.2.3. Calculation of vegetation and texture indices\u003c/h2\u003e \u003cp\u003eThe preprocessed Jilin-1 imagery and sugarcane plot vector boundaries were imported into the GEE platform. Twenty widely used VIs were computed, as listed in Table S4. Texture features (TFs) were extracted using the GLCM algorithm (Haralick et al., \u003cspan citationid=\"CR25\" class=\"CitationRef\"\u003e1973\u003c/span\u003e). This process employed three moving window sizes (3\u0026times;3, 5\u0026times;5, and 7\u0026times;7 pixels), and the texture features were averaged across four directional angles (0\u0026deg;, 45\u0026deg;, 90\u0026deg;, and 135\u0026deg;) for each window size. Consequently, eight standard TFs (Table S5) were generated for each of the four spectral bands Furthermore, three advanced indices, namely the normalized difference texture index (NDTI), enhanced vegetation texture index (EVTI) and double difference ratio texture index (DDRTI), were constructed by combining these TFs. Table\u0026nbsp;\u003cspan refid=\"Tab1\" class=\"InternalRef\"\u003e1\u003c/span\u003e lists the calculation formulas for the TIs.\u003c/p\u003e \u003cp\u003e \u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab1\" border=\"1\"\u003e \u003ccaption language=\"En\"\u003e \u003cdiv class=\"CaptionNumber\"\u003eTable 1\u003c/div\u003e \u003cdiv class=\"CaptionContent\"\u003e \u003cp\u003eSummary of texture indices used in this study.\u003c/p\u003e \u003c/div\u003e \u003c/caption\u003e \u003ccolgroup cols=\"3\"\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e \u003cthead\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\"\u003e \u003cp\u003eParameters\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c2\"\u003e \u003cp\u003eFormulas\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c3\"\u003e \u003cp\u003eReference\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eNormalized Difference Texture Index (NDTI)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e\u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:NDTI=\\frac{{T}_{1}-{T}_{2}}{{T}_{1}+{T}_{2}}\\)\u003c/span\u003e\u003c/span\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e(Zheng et al., \u003cspan citationid=\"CR69\" class=\"CitationRef\"\u003e2019\u003c/span\u003e)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eEnhanced Vegetation Texture Index (EVTI)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e\u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:EVTI=\\frac{2.5\\times\\:\\left({T}_{1}-{T}_{2}\\right)}{\\left({T}_{1}+6\\times\\:{T}_{2}-7.5\\times\\:{T}_{3}+1\\right)}\\)\u003c/span\u003e\u003c/span\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e(Yuan et al., \u003cspan citationid=\"CR64\" class=\"CitationRef\"\u003e2023\u003c/span\u003e)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eDouble Difference Ratio Texture Index (DDRTI)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e\u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:DDRTI=\\frac{{T}_{1}-{T}_{2}}{{T}_{3}-{T}_{4}}\\)\u003c/span\u003e\u003c/span\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eThis study\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003c/tbody\u003e \u003c/colgroup\u003e \u003ctfoot\u003e \u003ctr\u003e\u003ctd colspan=\"3\"\u003eNote: T\u003csub\u003e1\u003c/sub\u003e\u0026ndash;T\u003csub\u003e4\u003c/sub\u003e denote the same texture feature derived from arbitrary spectral bands.\u003c/td\u003e\u003c/tr\u003e \u003c/tfoot\u003e \u003c/table\u003e\u003c/div\u003e \u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv id=\"Sec8\" class=\"Section2\"\u003e \u003ch2\u003e2.3. Feature selection\u003c/h2\u003e \u003cp\u003ePearson correlation analysis was performed to investigate the relationships between input variables, revealing the presence of high multicollinearity and redundancy. To address this issue, SVR-RFE (Guyon et al., \u003cspan citationid=\"CR24\" class=\"CitationRef\"\u003e2002\u003c/span\u003e) was used to identify the optimal feature subset. This method iteratively removed the features with the lowest model contribution, retaining only those with the highest predictive power for each growth stage.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec9\" class=\"Section2\"\u003e \u003ch2\u003e2.4. Machine learning models\u003c/h2\u003e \u003cp\u003eThis study evaluated the ability of Jilin-1 satellite imagery to predict phenotypic traits of four growth stages of sugarcane (seedling, tillering, elongation, and maturation stages). Four regression models were compared, namely ridge regression (RR) (Hoerl \u0026amp; Kennard, \u003cspan citationid=\"CR26\" class=\"CitationRef\"\u003e1970\u003c/span\u003e), Gaussian process regression (GPR) (Rasmussen and Williams, \u003cspan citationid=\"CR46\" class=\"CitationRef\"\u003e2006\u003c/span\u003e), categorical boosting (CatBoost) (Friedman, \u003cspan citationid=\"CR19\" class=\"CitationRef\"\u003e2002\u003c/span\u003e; Zhang et al., \u003cspan citationid=\"CR67\" class=\"CitationRef\"\u003e2020\u003c/span\u003e) and support vector regression (SVR) (Drucker et al., \u003cspan citationid=\"CR13\" class=\"CitationRef\"\u003e1997\u003c/span\u003e; Durbha et al., \u003cspan citationid=\"CR15\" class=\"CitationRef\"\u003e2007\u003c/span\u003e). All analyses and modeling were conducted using Python 3.12.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec10\" class=\"Section2\"\u003e \u003ch2\u003e2.5. Model optimization and evaluation\u003c/h2\u003e \u003cp\u003eGrid search with five-fold cross-validation was used to optimize the model hyperparameters. The specific parameters and search ranges are listed in Table S6.\u003c/p\u003e \u003cp\u003eLeave-one-out cross-validation (LOOCV) (Brovelli et al., \u003cspan citationid=\"CR5\" class=\"CitationRef\"\u003e2008\u003c/span\u003e) was used to evaluate the model. Subsequently, the bootstrap method (Efron, \u003cspan citationid=\"CR16\" class=\"CitationRef\"\u003e1979\u003c/span\u003e) was applied to further verify the robustness and reliability of the model. Model performance was evaluated using the coefficient of determination (R\u003csup\u003e2\u003c/sup\u003e), root mean square error (RMSE) and ratio of performance to deviation (RPD). The RPD values were interpreted based on the classification criteria proposed by Gaston et al. (\u003cspan citationid=\"CR22\" class=\"CitationRef\"\u003e2010\u003c/span\u003e): 1.0\u0026ndash;1.4 indicates poor prediction; 1.4\u0026ndash;1.8, fair; 1.8\u0026ndash;2.0, good; and 2.0\u0026ndash;2.5, very good. The calculation formulas are as follows:\u003cdiv id=\"Equ3\" class=\"Equation\"\u003e\u003cdiv format=\"TEX\" class=\"mathdisplay\" id=\"FileID_Equ3\" name=\"EquationSource\"\u003e\n$$\\:{R}^{2}=1-\\frac{{\\sum\\:}_{i=1}^{n}{\\left({y}_{i}-{\\widehat{y}}_{i}\\right)}^{2}}{{\\sum\\:}_{i=1}^{n}{\\left({y}_{i}-{\\stackrel{-}{y}}_{i}\\right)}^{2}}$$\u003c/div\u003e\u003cdiv class=\"EquationNumber\"\u003e3\u003c/div\u003e\u003c/div\u003e\u003cdiv id=\"Equ4\" class=\"Equation\"\u003e\u003cdiv format=\"TEX\" class=\"mathdisplay\" id=\"FileID_Equ4\" name=\"EquationSource\"\u003e\n$$\\:RMSE=\\sqrt{\\frac{{\\sum\\:}_{i=1}^{n}{\\left({y}_{i}-{\\widehat{y}}_{i}\\:\\right)}^{2}}{n}}$$\u003c/div\u003e\u003cdiv class=\"EquationNumber\"\u003e4\u003c/div\u003e\u003c/div\u003e\u003cdiv id=\"Equ5\" class=\"Equation\"\u003e\u003cdiv format=\"TEX\" class=\"mathdisplay\" id=\"FileID_Equ5\" name=\"EquationSource\"\u003e\n$$\\:RPD=\\frac{SD}{RMSE}$$\u003c/div\u003e\u003cdiv class=\"EquationNumber\"\u003e5\u003c/div\u003e\u003c/div\u003e\u003cdiv id=\"Equ6\" class=\"Equation\"\u003e\u003cdiv format=\"TEX\" class=\"mathdisplay\" id=\"FileID_Equ6\" name=\"EquationSource\"\u003e\n$$\\:SD=\\sqrt{\\frac{{\\sum\\:}_{i\\:=1}^{n}{\\left({y}_{i}-{\\stackrel{-}{y}}_{i}\\right)}^{2}}{n}}$$\u003c/div\u003e\u003cdiv class=\"EquationNumber\"\u003e6\u003c/div\u003e\u003c/div\u003e\u003c/p\u003e \u003cp\u003ewhere \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:{y}_{i}\\)\u003c/span\u003e\u003c/span\u003e is the actual value; \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:{\\widehat{y}}_{i}\\)\u003c/span\u003e\u003c/span\u003e is the predicted value; \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:{\\stackrel{-}{y}}_{i}\\)\u003c/span\u003e\u003c/span\u003e is the mean of the actual values; \u003cem\u003en\u003c/em\u003e is the number of samples; and \u003cem\u003eSD\u003c/em\u003e is the standard deviation of the target variable.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec11\" class=\"Section2\"\u003e \u003ch2\u003e2.6. Feature importance assessment\u003c/h2\u003e \u003cp\u003eSHAP (Lundberg and Lee, \u003cspan citationid=\"CR41\" class=\"CitationRef\"\u003e2017\u003c/span\u003e) was used to interpret model predictions. It assigned each input feature an importance value and ranked the features at a global level. All calculations were performed using the SHAP library in Python 3.12.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec12\" class=\"Section2\"\u003e \u003ch2\u003e2.7. Research framework\u003c/h2\u003e \u003cp\u003eFigure\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003e illustrates the workflow of this study, which consists of the following key steps:\u003c/p\u003e \u003cp\u003e \u003col\u003e \u003cspan\u003e \u003cli\u003e \u003cp\u003e(a) Data Acquisition and Preprocessing. Obtain high-quality surface reflectance data from Jilin-1 satellite imagery through radiometric calibration, atmospheric correction, geometric correction, and band fusion; collect field survey data including soil chemical properties (SCPs), fertilization information (FI), and sugarcane phenotypic data, and complete data preprocessing and accuracy verification to form a multi-source dataset.\u003c/p\u003e \u003c/li\u003e \u003c/span\u003e \u003cspan\u003e \u003cli\u003e \u003cp\u003e(b) Optimization of Optimal Texture Window Size. Calculate multiple vegetation indices (VIs) and extract texture features using the Gray Level Co-occurrence Matrix (GLCM). Construct texture indices (TIs) under different window sizes (3\u0026times;3, 5\u0026times;5 and 7\u0026times;7), then build machine learning models (SVR, RR, GPR and CatBoost) and evaluate performance via Leave-One-Out Cross Validation (LOOCV) with RPD\u003csub\u003eCV\u003c/sub\u003e, to determine the optimal texture window size for sugarcane phenotype prediction.\u003c/p\u003e \u003c/li\u003e \u003c/span\u003e \u003cspan\u003e \u003cli\u003e \u003cp\u003e(c) Determination of the Optimal Prediction Model. Construct multi-dimensional input feature schemes (VIs, VIs\u0026thinsp;+\u0026thinsp;TIs and VIs\u0026thinsp;+\u0026thinsp;TIs\u0026thinsp;+\u0026thinsp;SCPs\u0026thinsp;+\u0026thinsp;FI) based on the optimal texture window size, train multiple machine learning models, and assess model performance using LOOCV and Bootstrap resampling with evaluation metrics (R\u003csup\u003e2\u003c/sup\u003e\u003csub\u003eCV\u003c/sub\u003e, RMSE\u003csub\u003eCV\u003c/sub\u003e, and RPD\u003csub\u003eCV\u003c/sub\u003e) to select the optimal prediction model.\u003c/p\u003e \u003c/li\u003e \u003c/span\u003e \u003cspan\u003e \u003cli\u003e \u003cp\u003e(d) Model Interpretability Analysis. Apply the SHAP interpretable method to the optimal prediction model, quantify feature importance, analyze the response patterns of key features, and compare the contribution of multi-source features to reveal the driving mechanisms of model predictions for sugarcane phenotypes.\u003c/p\u003e \u003c/li\u003e \u003c/span\u003e \u003c/ol\u003e \u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003c/div\u003e"},{"header":"3. Results and analysis","content":"\u003cdiv id=\"Sec14\" class=\"Section2\"\u003e \u003ch2\u003e3.1. Variations of sugarcane traits, NDVI and DDRTI_MEA\u003csub\u003e(B,R,G,NIR)\u003c/sub\u003e\u003c/h2\u003e \u003cp\u003eThe statistical results and scatter plots for yield, millable stalk population, and brix data for 2024 and 2025 are presented in Fig. \u003cspan refid=\"MOESM1\" class=\"InternalRef\"\u003eS1\u003c/span\u003e and Table\u0026nbsp;\u003cspan refid=\"Tab2\" class=\"InternalRef\"\u003e2\u003c/span\u003e. The mean values of yield, millable stalk population and brix in 2024 were higher than those in 2025. Variation analysis results indicated that there were no significant differences between the two years of yield data and brix (\u003cem\u003ep\u003c/em\u003e\u0026thinsp;\u0026gt;\u0026thinsp;0.05), whereas significant differences were observed for millable stalk population (\u003cem\u003ep\u003c/em\u003e\u0026thinsp;\u0026lt;\u0026thinsp;0.05). The first ratoon maintained yield with fewer stalks by increased individual stalk weight. The stalk quality and yield were preserved even the stalk number declined.\u003c/p\u003e \u003cp\u003eThe coefficient of variation (CV) for brix remained below 4% across two years. The CV for millable stalk population increased from 14.93% in 2024 to 20.50% in 2025, and the minimum value decreased from 36900 to 22500 stalks/ha. It seems that this variety compensated for the loss of millable stalks through increased individual stalk weight in ratoon cane with relatively stable yield and Brix.\u003c/p\u003e \u003cp\u003eIn this study, there are a large number of VIs and TIs were used. NDVI and DDRTI_MEA\u003csub\u003e(B,R,G,NIR)\u003c/sub\u003e are presented as representative examples for seasonal pattern analysis. The seasonal variation of NDVI in 2024 and 2025 was presented in Fig. S2a. It is clear that NDVI increased more rapidly in 2025 than 2024 from the seedling to the tillering. During the subsequent growth period, NDVI continued to rise and reached its maximum in early September 2025, whereas the peak occurred in mid-September 2024. In addition, during maturation, NDVI declined less and remained higher in 2025. The seasonal dynamics of DDRTI_MEA\u003csub\u003e(B,R,G,NIR)\u003c/sub\u003e in 2024 and 2025 were given in Fig S2b. In 2024, DDRTI_MEA\u003csub\u003e(B,R,G,NIR)\u003c/sub\u003e increased slightly from the seedling stage to the tillering stage and then gradually declined after elongation. In 2025, this index was low during the early stages, followed by a rapid from late tillering to elongation with a pronounced peak in early September before dropping during maturation.\u003c/p\u003e \u003cp\u003e \u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab2\" border=\"1\"\u003e \u003ccaption language=\"En\"\u003e \u003cdiv class=\"CaptionNumber\"\u003eTable 2\u003c/div\u003e \u003cdiv class=\"CaptionContent\"\u003e \u003cp\u003eDescriptive statistics of the measured sugarcane traits.\u003c/p\u003e \u003c/div\u003e \u003c/caption\u003e \u003ccolgroup cols=\"7\"\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c4\" colnum=\"4\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c5\" colnum=\"5\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c6\" colnum=\"6\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c7\" colnum=\"7\"\u003e\u003c/div\u003e \u003cthead\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\"\u003e \u003cp\u003eSugarcane Traits\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c2\"\u003e \u003cp\u003eYear\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c3\"\u003e \u003cp\u003eMean\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c4\"\u003e \u003cp\u003eSD\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c5\"\u003e \u003cp\u003eMin\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c6\"\u003e \u003cp\u003eMax\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c7\"\u003e \u003cp\u003eCV(%)\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\" morerows=\"1\" rowspan=\"2\"\u003e \u003cp\u003eYield (t/ha)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e2024\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e70.53\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e27.45\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e25.67\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e136.01\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e \u003cp\u003e38.92\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e2025\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e66.54\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e23.77\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e25.47\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e106.17\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e \u003cp\u003e35.73\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\" morerows=\"1\" rowspan=\"2\"\u003e \u003cp\u003eMillable Stalk Population (stalks/ha)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e2024\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e57825\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e8636\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e36900\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e74700\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e \u003cp\u003e14.93\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e2025\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e44803\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e9186\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e22500\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e58500\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e \u003cp\u003e20.50\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\" morerows=\"1\" rowspan=\"2\"\u003e \u003cp\u003eBrix (\u0026deg;Bx)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e2024\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e20.87\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e0.77\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e19.54\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e22.64\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e \u003cp\u003e3.73\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e2025\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e20.61\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e0.61\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e19.28\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e21.84\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e \u003cp\u003e2.97\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003c/tbody\u003e \u003c/colgroup\u003e \u003c/table\u003e\u003c/div\u003e \u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec15\" class=\"Section2\"\u003e \u003ch2\u003e3.2. Correlation analysis of input features and sugarcane traits\u003c/h2\u003e \u003cdiv id=\"Sec16\" class=\"Section3\"\u003e \u003ch2\u003e3.2.1 Correlation between vegetation indices and sugarcane traits\u003c/h2\u003e \u003cp\u003eTwenty VIs were evaluated by Pearson correlation analysis across four growth stages (Fig. S3). Correlation coefficients (\u003cem\u003er\u003c/em\u003e) and significance levels (\u003cem\u003ep\u003c/em\u003e) were calculated. For sugarcane yield (Fig. S3a), the strongest correlations were observed during the tillering and elongation stages. At the tillering stage, GLI exhibited the highest positive correlation (\u003cem\u003er\u003c/em\u003e\u0026thinsp;=\u0026thinsp;0.60, \u003cem\u003ep\u003c/em\u003e\u0026thinsp;\u0026lt;\u0026thinsp;0.001), followed by CVI and GRVI (\u003cem\u003er\u003c/em\u003e\u0026thinsp;=\u0026thinsp;0.52). At the elongation stage, NGRDI and WDRVI showed the strongest correlations (\u003cem\u003er\u003c/em\u003e\u0026thinsp;=\u0026thinsp;0.57, \u003cem\u003ep\u003c/em\u003e\u0026thinsp;\u0026lt;\u0026thinsp;0.001). Correlations at the seedling and maturation stages were generally weaker. The correlation between VIs and millable stalks showed distinct temporal patterns (Fig. S3b). The seedling stage was the most critical window for monitoring stalk number. Almost all VIs showed uniformly strong positive correlations (approx. \u003cem\u003er\u003c/em\u003e\u0026thinsp;=\u0026thinsp;0.60) with high statistical significance (\u003cem\u003ep\u003c/em\u003e\u0026thinsp;\u0026lt;\u0026thinsp;0.001). Interestingly, an inverse relationship was observed in later stages. During the elongation and maturation stages, most VIs were significantly negatively correlated with millable stalk population (|\u003cem\u003er|\u003c/em\u003e values ranging from 0.50 to 0.60). In contrast to yield and millable stalk population, Brix showed no significant correlation with any of the tested VIs across all growth stages (Fig. S3c). The correlation coefficients remained low (\u003cem\u003e|r|\u003c/em\u003e \u0026lt; 0.25) and statistically insignificant, suggesting that standard VIs may not be sensitive enough to detect variations in sugarcane Brix.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec17\" class=\"Section3\"\u003e \u003ch2\u003e3.2.2 Correlation between texture indices and phenotypic traits\u003c/h2\u003e \u003cp\u003eTo evaluate TIs suitable for modeling, Pearson correlation analyses were conducted between TIs (calculated using different window sizes of 3\u0026times;3, 5\u0026times;5, and 7\u0026times;7) and sugarcane phenotypic traits across four growth stages (seedling, tillering, elongation, and maturation). In this study, TIs were derived by combining the same TFs across different spectral bands. Eight TFs were selected, and the three major categories of TIs (NDTI, EVTI, and DDRTI) were divided into eight sub-categories, respectively. From each sub-category, the texture index exhibiting the highest absolute correlation coefficient (|\u003cem\u003er\u003c/em\u003e|) was selected. Figs. S4, S5, and S6 illustrate the Pearson correlation coefficients between TIs and yield, millable stalk population, and Brix. TIs calculated with different window sizes performed differently for the three phenotypic traits. For sugarcane yield, the correlation of texture indices first increased and then decreased during growth. It rose from 0.40 (\u003cem\u003ep\u003c/em\u003e\u0026thinsp;\u0026lt;\u0026thinsp;0.001) at the seedling stage to 0.67 (\u003cem\u003ep\u003c/em\u003e\u0026thinsp;\u0026lt;\u0026thinsp;0.001) at the tillering stage, then dropped to 0.53 (\u003cem\u003ep\u003c/em\u003e\u0026thinsp;\u0026lt;\u0026thinsp;0.001) at the maturation stage. For millable stalk population, the correlation of texture indices increased gradually. It rose from 0.47 (\u003cem\u003ep\u003c/em\u003e\u0026thinsp;\u0026lt;\u0026thinsp;0.001) at the seedling stage to 0.68 (\u003cem\u003ep\u003c/em\u003e\u0026thinsp;\u0026lt;\u0026thinsp;0.001) at the maturation stage. For Brix, the correlation showed a downward trend. The highest correlation coefficient was 0.33 (\u003cem\u003ep\u003c/em\u003e\u0026thinsp;\u0026lt;\u0026thinsp;0.05) at the seedling stage.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec18\" class=\"Section3\"\u003e \u003ch2\u003e3.2.3 Correlation between agronomic factors and phenotypic traits\u003c/h2\u003e \u003cp\u003ePearson correlation analysis was conducted to evaluate the relationship between SCPs, FI, and sugarcane phenotypic traits (Fig. S7). Significant correlations were observed between yield and various SCPs as well as FI. In terms of SCPs, AP showed the strongest negative correlation (\u003cem\u003er\u003c/em\u003e = -0.43, \u003cem\u003ep\u003c/em\u003e\u0026thinsp;\u0026lt;\u0026thinsp;0.001), while soil pH exhibited a significant positive correlation (\u003cem\u003er\u003c/em\u003e\u0026thinsp;=\u0026thinsp;0.40, \u003cem\u003ep\u003c/em\u003e\u0026thinsp;\u0026lt;\u0026thinsp;0.001). Regarding FI, there was a significant positive correlation between yield and variables such as BF_CF, BF_KCl, TD_CF, and TD_KCl (\u003cem\u003er\u003c/em\u003e\u0026thinsp;=\u0026thinsp;0.46, \u003cem\u003ep\u003c/em\u003e\u0026thinsp;\u0026lt;\u0026thinsp;0.001). Millable stalk population, a key component of yield, showed the highest correlation with FI (\u003cem\u003er\u003c/em\u003e\u0026thinsp;=\u0026thinsp;0.35, \u003cem\u003ep\u003c/em\u003e\u0026thinsp;\u0026lt;\u0026thinsp;0.05), while its correlation with SCPs was low. In contrast, the correlations between Brix and these indicators were all weak and statistically insignificant.\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv id=\"Sec19\" class=\"Section2\"\u003e \u003ch2\u003e3.3. Determination of optimal texture calculation window size\u003c/h2\u003e \u003cp\u003eFour machine learning models were employed to determine the optimal window size, with RPD\u003csub\u003eCV\u003c/sub\u003e scores illustrated in Fig.\u0026nbsp;\u003cspan refid=\"Fig3\" class=\"InternalRef\"\u003e3\u003c/span\u003e. Table S7 summarized the best parameter combinations for each stage. The results indicate that the optimal texture combination varied across growth stages.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec20\" class=\"Section2\"\u003e \u003ch2\u003e3.4. Prediction of sugarcane traits based on multi-source features\u003c/h2\u003e \u003cdiv id=\"Sec21\" class=\"Section3\"\u003e \u003ch2\u003e3.4.1. Yield estimation model\u003c/h2\u003e \u003cp\u003eTable S8 summarizes the yield prediction results. The performance of four models (SVR, GPR, RR, and CatBoost) with the SVR-RFE feature selection method across four growth stages were compared. Figure\u0026nbsp;\u003cspan refid=\"Fig4\" class=\"InternalRef\"\u003e4\u003c/span\u003e(a1\u0026ndash;c1) presents the prediction results of the best models based on LOOCV for the three different input data combinations.\u003c/p\u003e \u003cp\u003eWhen using only VIs, the SVR model at the tillering stage achieved the highest accuracy. The R\u003csup\u003e2\u003c/sup\u003e\u003csub\u003eCV\u003c/sub\u003e was 0.60, the RMSE\u003csub\u003eCV\u003c/sub\u003e was 16.02 t/ha, and the RPD\u003csub\u003eCV\u003c/sub\u003e was 1.58 (Fig.\u0026nbsp;4a1). After adding TIs, the RR model performed the best at the tillering stage. The R\u003csup\u003e2\u003c/sup\u003e\u003csub\u003eCV\u003c/sub\u003e increased to 0.65, with an RMSE\u003csub\u003eCV\u003c/sub\u003e of 15.06 t/ha and an RPD\u003csub\u003eCV\u003c/sub\u003e of 1.68 (Fig.\u0026nbsp;4b1). Finally, with the addition of SCPs and FI, the SVR model at the tillering stage reached the highest accuracy. The R\u003csup\u003e2\u003c/sup\u003e\u003csub\u003eCV\u003c/sub\u003e rose to 0.70, the RMSE\u003csub\u003eCV\u003c/sub\u003e decreased to 13.84 t/ha, and the RPD\u003csub\u003eCV\u003c/sub\u003e reached 1.83 (Fig.\u0026nbsp;4c1). The RPD value shifted from the \u0026ldquo;fair\u0026rdquo; category (1.4\u0026ndash;1.8) to the \u0026ldquo;good\u0026rdquo; category (1.8\u0026ndash;2.0).\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003eTo further verify the stability and robustness of the prediction models, the distribution of RPD values were also obtained from the bootstrap method (Fig.\u0026nbsp;\u003cspan refid=\"Fig4\" class=\"InternalRef\"\u003e4\u003c/span\u003e(a2\u0026ndash;c2)). The mean RPD in Fig.\u0026nbsp;\u003cspan refid=\"Fig4\" class=\"InternalRef\"\u003e4\u003c/span\u003ec2 was 1.56, slightly lower than the 1.58 in Fig.\u0026nbsp;\u003cspan refid=\"Fig4\" class=\"InternalRef\"\u003e4\u003c/span\u003eb2. However, the 75th percentile in Fig.\u0026nbsp;\u003cspan refid=\"Fig4\" class=\"InternalRef\"\u003e4\u003c/span\u003ec2 reached 1.75, surpassing the 1.73 in Fig.\u0026nbsp;\u003cspan refid=\"Fig4\" class=\"InternalRef\"\u003e4\u003c/span\u003eb2.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec22\" class=\"Section3\"\u003e \u003ch2\u003e3.4.2. Millable stalk population estimation model\u003c/h2\u003e \u003cp\u003eFor millable stalk population, the performance of tour models (SVR, GPR, RR, and CatBoost) is presented in Table S9. Three input combinations were compared. Figure\u0026nbsp;\u003cspan refid=\"Fig5\" class=\"InternalRef\"\u003e5\u003c/span\u003e(a1\u0026ndash;c1) presents the prediction results of the best models based on LOOCV for the three different input data combinations.\u003c/p\u003e \u003cp\u003eWhen using only VIs, the SVR model at the maturation stage achieved the highest accuracy. The R\u003csup\u003e2\u003c/sup\u003e\u003csub\u003eCV\u003c/sub\u003e was 0.57, the RMSE\u003csub\u003eCV\u003c/sub\u003e was 7181 stalks/ha, and the RPD\u003csub\u003eCV\u003c/sub\u003e was 1.52 (Fig.\u0026nbsp;5a1). After adding TIs, the SVR model still performed the best at the maturation stage. The R\u003csup\u003e2\u003c/sup\u003e\u003csub\u003eCV\u003c/sub\u003e increased to 0.63, with an RMSE\u003csub\u003eCV\u003c/sub\u003e of 6602 stalks/ha and an RPD\u003csub\u003eCV\u003c/sub\u003e of 1.66 (Fig.\u0026nbsp;5b1). Finally, with the addition of SCPs and FI, the optimal prediction window shifted to the elongation stage. However, the overall accuracy decreased compared to the previous step. The R\u003csup\u003e2\u003c/sup\u003e\u003csub\u003eCV\u003c/sub\u003e dropped to 0.62, the RMSE\u003csub\u003eCV\u003c/sub\u003e increased to 6733 stalks/ha, and the RPD\u003csub\u003eCV\u003c/sub\u003e fell to 1.62 (Fig.\u0026nbsp;5c1). The model performance remained within the \u0026ldquo;fair\u0026rdquo; category (1.4\u0026ndash;1.8) and did not reach the \u0026ldquo;good\u0026rdquo; standard (1.8\u0026ndash;2.0).\u003c/p\u003e \u003cp\u003eTo further verify the stability and robustness of the prediction models, the distribution of RPD values was obtained from the bootstrap method. As shown in Fig.\u0026nbsp;\u003cspan refid=\"Fig5\" class=\"InternalRef\"\u003e5\u003c/span\u003e(a2\u0026ndash;c2), the model in Fig.\u0026nbsp;\u003cspan refid=\"Fig5\" class=\"InternalRef\"\u003e5\u003c/span\u003eb2 performed best. Its 25% percentile (1.37), mean (1.51), and 75% percentile (1.63) were higher than those of the other two models. This indicates that the model with VIs\u0026thinsp;+\u0026thinsp;TIs (Fig.\u0026nbsp;\u003cspan refid=\"Fig5\" class=\"InternalRef\"\u003e5\u003c/span\u003eb2) achieved the highest accuracy and stability.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec23\" class=\"Section3\"\u003e \u003ch2\u003e3.4.3. Brix estimation model\u003c/h2\u003e \u003cp\u003eTable S10 shows the Brix prediction results. Preliminary results indicated that the SVR model consistently outperformed the other three models across all growth stages. Figure\u0026nbsp;\u003cspan refid=\"Fig6\" class=\"InternalRef\"\u003e6\u003c/span\u003e(a1\u0026ndash;c1) presents the prediction results of the SVR model based on LOOCV for three different input data combinations.\u003c/p\u003e \u003cp\u003eUsing VIs alone, the SVR model at the seedling stage achieved the highest accuracy. The R\u003csup\u003e2\u003c/sup\u003e\u003csub\u003eCV\u003c/sub\u003e was 0.21, the RMSE\u003csub\u003eCV\u003c/sub\u003e was 0.63 \u0026deg;Bx, and the RPD\u003csub\u003eCV\u003c/sub\u003e was 1.12 (Fig.\u0026nbsp;6a1). After adding TIs, the SVR model still performed the best at the seedling stage. The R\u003csup\u003e2\u003c/sup\u003e\u003csub\u003eCV\u003c/sub\u003e increased to 0.40, with an RMSE\u003csub\u003eCV\u003c/sub\u003e of 0.55 \u0026deg;Bx and an RPD\u003csub\u003eCV\u003c/sub\u003e of 1.29 (Fig.\u0026nbsp;6b1). Finally, with the addition of SCPs and FI, the optimal prediction window shifted to the maturation stage. Notably, the overall accuracy improved compared to the previous step. The R\u003csup\u003e2\u003c/sup\u003e\u003csub\u003eCV\u003c/sub\u003e rose to 0.44, the RMSE\u003csub\u003eCV\u003c/sub\u003e decreased to 0.53 \u0026deg;Bx, and the RPD\u003csub\u003eCV\u003c/sub\u003e reached 1.33 (Fig.\u0026nbsp;6c1). The model performance remained within the \u0026ldquo;poor\u0026rdquo; category (1.0\u0026ndash;1.4) and did not reach the \u0026ldquo;fair\u0026rdquo; standard (1.4\u0026ndash;1.8).\u003c/p\u003e \u003cp\u003eTo further verify the stability and robustness of the SVR prediction model, the distribution of RPD\u003csub\u003eCV\u003c/sub\u003e values was obtained from the bootstrap method (Fig.\u0026nbsp;6a2\u0026ndash;c2). The model in Fig.\u0026nbsp;6b2 exhibited slightly higher stability metrics (a mean RPD\u003csub\u003eCV\u003c/sub\u003e of 1.18). However, the model in Fig.\u0026nbsp;6c1 achieved the highest explanatory power (R\u003csup\u003e2\u003c/sup\u003e\u003csub\u003eCV\u003c/sub\u003e\u0026thinsp;=\u0026thinsp;0.44) in the LOOCV assessment. Considering the priority of prediction accuracy, the SVR model developed by VIs\u0026thinsp;+\u0026thinsp;TIs\u0026thinsp;+\u0026thinsp;SCPs\u0026thinsp;+\u0026thinsp;FI was identified as the final model.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv id=\"Sec24\" class=\"Section2\"\u003e \u003ch2\u003e3.5. Model interpretation based on SHAP\u003c/h2\u003e \u003cp\u003eSHAP analysis was used to explain feature contributions to sugarcane trait prediction. Figure\u0026nbsp;\u003cspan refid=\"Fig7\" class=\"InternalRef\"\u003e7\u003c/span\u003e presents the SHAP summary and feature importance for the optimal models. For yield prediction, Fig.\u0026nbsp;7a1 ranks the features by their importance. GNDVI was the most important feature, contributing 24.7% to the model. CI and RVI were the next most important features, contributing 22.7% and 22.5%, respectively. BF_CMF ranked fourth (8.1%), followed by CVI (7.9%) and DDRTI_CON\u003csub\u003e(G,NIR,B,R)\u003c/sub\u003e (4.7%) occupied the fifth and sixth positions. The inset pie chart shows the contribution of different categories (Fig.\u0026nbsp;7a2). VIs were the dominant drivers, accounting for 77.8% of the total contribution, which was significantly higher than that of FI (11.7%) and TIs (10.5%). Figure\u0026nbsp;\u003cspan refid=\"Fig7\" class=\"InternalRef\"\u003e7\u003c/span\u003ea3 illustrates how these features affect yield; red dots represent high feature values, while blue dots represent low values. Yield prediction was positively correlated with GNDVI (ranked first) and CVI (ranked fifth). Conversely, CI and RVI (ranked second and third) showed a negative correlation, where lower values corresponded to higher yield predictions. Finally, BF_CMF (ranked fourth) had a positive impact on yield, whereas DDRTI_CON\u003csub\u003e(G,NIR,B,R)\u003c/sub\u003e (ranked sixth) showed a negative correlation.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003eFor the prediction of millable stalk population, Fig.\u0026nbsp;7b1 ranks the features by importance. CVI was the most critical feature, contributing 24.7% to the model. NDTI_HOM\u003csub\u003e(B_G)\u003c/sub\u003e, EVTI_VAR\u003csub\u003e(NIR_R_B)\u003c/sub\u003e, and EVTI_MEA\u003csub\u003e(G,NIR,R)\u003c/sub\u003e followed, contributing 18.6%, 18.0%, and 16.9%, respectively. NDTI_VAR(G,NIR) (9.9%) and EVTI_ASM\u003csub\u003e(G,B,NIR,G)\u003c/sub\u003e (6.1%) ranked fifth and sixth. The inset pie chart (Fig.\u0026nbsp;7b2) shows that TIs were the dominant drivers, accounting for 69.5% of the total contribution, which was much higher than that of VIs (30.5%). Figure\u0026nbsp;7b3 displays the feature effects on millable stalk population. The prediction was positively correlated with CVI (ranked first), NDTI_VAR\u003csub\u003e(G,NIR)\u003c/sub\u003e (ranked fifth), and EVTI_ASM\u003csub\u003e(G,B,NIR,G)\u003c/sub\u003e (ranked sixth). In contrast, NDTI_HOM\u003csub\u003e(B,G)\u003c/sub\u003e, EVTI_VAR\u003csub\u003e(NIR,R,B)\u003c/sub\u003e, and EVTI_MEA\u003csub\u003e(G,NIR,R)\u003c/sub\u003e (ranked second, third, and fourth) showed negative correlations, as lower values corresponded to higher millable stalk population predictions.\u003c/p\u003e \u003cp\u003eFor Brix prediction, Fig.\u0026nbsp;7c1 ranks the features by importance. GDVI was the top feature, contributing 35.9% to the model, followed by BNDVI with 27.8%. DDRTI_VAR\u003csub\u003e(G,R,B,NIR)\u003c/sub\u003e (8.9%) and DDRTI_HOM\u003csub\u003e(G,R,B,NIR)\u003c/sub\u003e (4.5%) ranked third and fourth, while WDRVI took the fifth spot (4.1%). Notably, soil pH also played a role in the model. The inset pie chart (Fig.\u0026nbsp;7c2) indicates that VIs were the dominant drivers (67.9%), far exceeding the contribution of TIs (30.8%) and SCPs (1.3%). Figure\u0026nbsp;7c3 shows the feature effects on Brix. Brix prediction was negatively correlated with GDVI (ranked first), DDRTI_VAR\u003csub\u003e(G,R,B,NIR)\u003c/sub\u003e (ranked third), and DDRTI_HOM\u003csub\u003e(G,R,B,NIR)\u003c/sub\u003e (ranked fourth). Conversely, BNDVI and WDRVI (ranked second and fifth) showed positive correlations, where higher values corresponded to higher Brix predictions. Soil pH also showed a positive correlation with Brix prediction.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec25\" class=\"Section2\"\u003e \u003ch2\u003e3.6. Prediction mapping based on optimal models\u003c/h2\u003e \u003cp\u003eThe respective models were applied to predict the phenotypic traits of sugarcane. Figure\u0026nbsp;\u003cspan refid=\"Fig8\" class=\"InternalRef\"\u003e8\u003c/span\u003e illustrates the spatial distribution of sugarcane yield, millable stalk population, and Brix across the study area for the plant cane (year 2024) and ratoon cane (year 2025) seasons.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003eThere was a marked decline in sugarcane yield from the plant cane to first ratoon cane. In 2024 (plant cane), the majority of the field exhibited high yields, ranging between 63 and 105 t/ha (Fig.\u0026nbsp;8a1). Conversely, in 2025 (first ratoon cane), yields decreased significantly, with most areas shifting to a lower range of 21\u0026ndash;63 t/ha (Fig.\u0026nbsp;8a2). A similar downward trend was observed for the number of millable stalks. The plant cane showed high stalk density, predominantly exceeding 45000 stalks/ha (Fig.\u0026nbsp;8b1). However, in 2025, the density dropped visibly with a large portion of the field recording values below 45000 stalks/ha (Fig.\u0026nbsp;8b2). Finally, Brix levels slightly declined from plant cane to first ratoon cane. As illustrated in Fig.\u0026nbsp;8c1, the 2024 map displays a mix of moderate to high Brix levels (mostly 20\u0026ndash;23 \u0026deg;Bx). In contrast, the 2025 map (Fig.\u0026nbsp;8c2) is dominated by lower values (mostly 19\u0026ndash;21 \u0026deg;Bx), suggesting a reduced sucrose accumulation potential in the ratoon year.\u003c/p\u003e \u003c/div\u003e"},{"header":"4. Discussion","content":"\u003cdiv id=\"Sec27\" class=\"Section2\"\u003e \u003ch2\u003e4.1. Comparison of sugarcane traits estimation models\u003c/h2\u003e \u003cp\u003eBased on the plot level seasonal variation analysis in 2024, NDV of plant cane was generally lower than that of the ratoon during the seedling and tillering stages, which may contribute to the advantage of ratoon cane in canopy establishment. Singla et al. (\u003cspan citationid=\"CR50\" class=\"CitationRef\"\u003e2018\u003c/span\u003e) indicated that the temporal values of NDVI data of Landsat 8 during the first week of April and last week of August to the end of September can be used to discriminate of ratoon sugarcane in local area. Ratoon sugarcane has a relatively high seedling density, especially in the late emergence stage (Xu et al., \u003cspan citationid=\"CR58\" class=\"CitationRef\"\u003e2021\u003c/span\u003e). The canopies overlap completely, which makes it difficult to clearly distinguish the stem-leaf boundaries of individual plants (Zhu et al., \u003cspan citationid=\"CR70\" class=\"CitationRef\"\u003e2026\u003c/span\u003e). From late tillering to the elongation stage, DDRTI_MEA\u003csub\u003e(B,R,G,NIR)\u003c/sub\u003e was markedly higher in ratoon cane than in plant cane, and this difference was more pronounced in the mid-to-late season. This is likely associated with the greater variation of first ratoon to accumulated below-ground pests and diseases, drought in seedling, as well as missing stools, and uneven emergence establishment (Xu et al., \u003cspan citationid=\"CR58\" class=\"CitationRef\"\u003e2021\u003c/span\u003e; Dlamini et al., \u003cspan citationid=\"CR12\" class=\"CitationRef\"\u003e2024\u003c/span\u003e). As the sugarcane enters the rapid growth phase, within-plot differences in canopy closure, shadowing, and exposed soil patches tend to be amplified, which resulted in increased spatial heterogeneity of the canopy in high spatial resolution imagery (Li et al., \u003cspan citationid=\"CR36\" class=\"CitationRef\"\u003e2025\u003c/span\u003e). DDRTI_MEA\u003csub\u003e(B,R,G,NIR)\u003c/sub\u003e is constructed from multi-band GLCM mean texture features and may be more sensitive to within plot spatial structural differences, which makes it better suited to capturing texture changes driven by heterogeneous growth in ratoon cane (Zhu et al., \u003cspan citationid=\"CR70\" class=\"CitationRef\"\u003e2026\u003c/span\u003e). Based on 20 plants randomly selected from each plot at early maturation stage, the plant height, stem diameter, and single stem weight were compared between two years. Plant height did not differ significantly with mean values of 197.25cm in 2024 and 199.42 cm in 2025. Mean stem diameter increased from 2.66cm in 2024 to 2.80cm in 2025. Single stem weight differed significantly with the mean value of 1199 g in 2024 and 1334 g in 2025 (Table S11).\u003c/p\u003e \u003cp\u003eFour algorithms (SVR, RR, GPR, and CatBoost) were evaluated in this study. Across all input features, SVR achieved the highest accuracy. This aligns with our previous findings for rice (Yuan et al., \u003cspan citationid=\"CR64\" class=\"CitationRef\"\u003e2023\u003c/span\u003e) and other crops (Shi et al., \u003cspan citationid=\"CR51\" class=\"CitationRef\"\u003e2022\u003c/span\u003e). Indeed, SVR consistently demonstrates superior stability and fitting capabilities in agricultural applications. This is primarily attributed to the principle of structural risk minimization, which ensures that SVR maintains strong generalization ability even with high-dimensional and small-sample datasets. While GPR is also an established approach for such problems, and showed similar performance to RR, both were outperformed by SVR in this study. In contrast, CatBoost exhibited instability and yielded poorer results. Unlike SVR, CatBoost fits residuals via iterative boosting. In small sample scenarios, this approach can be susceptible to noise and prone to overfitting (Friedman, \u003cspan citationid=\"CR19\" class=\"CitationRef\"\u003e2002\u003c/span\u003e). Consequently, lacking the robust regularization of SVR, CatBoost demonstrated lower stability, highlighting the importance of selecting models tailored to the specific properties of the data.\u003c/p\u003e \u003cp\u003eDuring the early growth stage, the R\u003csup\u003e2\u003c/sup\u003e\u003csub\u003eCV\u003c/sub\u003es of sugarcane traits prediction model using VIs alone were relatively low. The R\u003csup\u003e2\u003c/sup\u003e\u003csub\u003eCV\u003c/sub\u003es for yield varied from 0.14 to 0.29. Those for millable stalk population ranged from 0.27 to 0.33, and the Brix prediction using the SVR model reached only 0.21. Akbarian et al. (\u003cspan citationid=\"CR2\" class=\"CitationRef\"\u003e2024\u003c/span\u003e) reported the R\u003csup\u003e2\u003c/sup\u003e of 0.53 in training of SVR yield model in early growth stage of sugarcane and R\u003csup\u003e2\u003c/sup\u003e of 0.57 in middle stage based on UAV multispectral data. Vasconcelos et al. (\u003cspan citationid=\"CR54\" class=\"CitationRef\"\u003e2025\u003c/span\u003e) achieved a R\u003csup\u003e2\u003c/sup\u003e of 0.89 in the local area sugarcane yield prediction based on high spatial resolution data (PlantScope), genetic information and metrological data via the heteroskedastic gamma regression. And the mid-term yield prediction model based on VIs outperformed the early-stage model, which is consistent with the findings of Akbarian et al. (\u003cspan citationid=\"CR1\" class=\"CitationRef\"\u003e2022\u003c/span\u003e). When VIs were used alone, yield prediction accuracy showed no significant improvement at the elongation and maturity stages, which may indirectly reflect the saturation effect of vegetation indices under dense canopy conditions (Yue et al., \u003cspan citationid=\"CR65\" class=\"CitationRef\"\u003e2019\u003c/span\u003e; Sofonia et al., \u003cspan citationid=\"CR52\" class=\"CitationRef\"\u003e2019\u003c/span\u003e). In the middle and late growth periods, sugarcane canopy coverage approaches saturation, and conventional vegetation indices exhibit reduced sensitivity to biomass, making it difficult to detect subtle canopy variations (Shendryk et al., \u003cspan citationid=\"CR49\" class=\"CitationRef\"\u003e2020\u003c/span\u003e). Sofonia et al. (\u003cspan citationid=\"CR52\" class=\"CitationRef\"\u003e2019\u003c/span\u003e) reported that the VI based model had the R\u003csup\u003e2\u003c/sup\u003e of 0.322 compared to the R\u003csup\u003e2\u003c/sup\u003e of 0.707 of LiDAR based model in sugarcane yield estimation. This results also indicates that relying solely on VIs is insufficient to characterize the complex process of yield formation. The combination of LiDAR and VIs could improve the yield model performance (Sofonia et al., \u003cspan citationid=\"CR52\" class=\"CitationRef\"\u003e2019\u003c/span\u003e; Shendryk et al., \u003cspan citationid=\"CR49\" class=\"CitationRef\"\u003e2020\u003c/span\u003e). Integrating auxiliary data, such as meteorological variables, genetic data, phenology info, soil texture info, with vegetation indices (VIs) also had the potential to enhance the accuracy of sugarcane yield estimation (Amorim et al., \u003cspan citationid=\"CR4\" class=\"CitationRef\"\u003e2022\u003c/span\u003e; Aleman-Montes et al., 2023).\u003c/p\u003e \u003cp\u003eFor millable stalk population, the model\u0026rsquo;s predictive accuracy was highest at the maturation stage, achieving an R\u003csup\u003e2\u003c/sup\u003e\u003csub\u003eCV\u003c/sub\u003e value of 0.57. This finding aligns closely with the R\u003csup\u003e2\u003c/sup\u003e\u003csub\u003eCV\u003c/sub\u003e of approximately 0.60 reported by Khuimphukhieo et al. (\u003cspan citationid=\"CR32\" class=\"CitationRef\"\u003e2025\u003c/span\u003e). Sofinia et al. (2019) reported the R\u003csup\u003e2\u003c/sup\u003e of 0.502 for millable stalk population by UAV multispectral data-based model. With respect to Brix, the associations with VIs were generally weak or statistically non-significant. Although Some studies reported the detectable relationship between VIs and Brix (Chea et al. \u003cspan citationid=\"CR7\" class=\"CitationRef\"\u003e2020\u003c/span\u003e; Aleman-Montes et al., 2023), it varied considerably across sugarcane cultivars. In this study, VIs alone model failed in Brix estimation, while in the maturation stage, the including of soil properties improved explanation of Brix model.\u003c/p\u003e \u003cp\u003ePrevious studies have demonstrated that texture features of high spatial resolution image and UAV data could partly capture differences in crop canopy characteristics (Zhang et al., \u003cspan citationid=\"CR66\" class=\"CitationRef\"\u003e2021\u003c/span\u003e; Liu et al., \u003cspan citationid=\"CR37\" class=\"CitationRef\"\u003e2023a\u003c/span\u003e,\u003cspan citationid=\"CR39\" class=\"CitationRef\"\u003eb\u003c/span\u003e; Zhu et al., \u003cspan citationid=\"CR70\" class=\"CitationRef\"\u003e2026\u003c/span\u003e). The optimum texture window sizes and types of texture parameters depended on data source, crop types, growth stage and agronomic indicators (Liu et al., \u003cspan citationid=\"CR37\" class=\"CitationRef\"\u003e2023a\u003c/span\u003e,\u003cspan citationid=\"CR39\" class=\"CitationRef\"\u003eb\u003c/span\u003e). Furthermore, TIs were more useful than individual TFs. Yu et al. (\u003cspan citationid=\"CR62\" class=\"CitationRef\"\u003e2024\u003c/span\u003e) reported that the correlation between TIs and LAI in apple orchards was stronger than that of TFs. TIs could reflect more detailed features and richer information (Guo et al., \u003cspan citationid=\"CR23\" class=\"CitationRef\"\u003e2023\u003c/span\u003e) and enhance the accuracy of growth simulation (Yuan et al., \u003cspan citationid=\"CR64\" class=\"CitationRef\"\u003e2023\u003c/span\u003e). In this study, good predictive accuracy for sugarcane traits were obtained by the fusion of VIs and TIs. The maximum R\u003csup\u003e2\u003c/sup\u003e\u003csub\u003eCV\u003c/sub\u003e of the yield model achieved to 0.65, and those of the millable stalk population and Brix reached 0.63 and 0.40, respectively. These results demonstrate that the integration of VIs and TIs derived from Jilin-1 satellite images outperforms the use of VIs alone for predicting sugarcane traits.\u003c/p\u003e \u003cp\u003eAt the field level, crop growth is closely related to factors such as cultivar, soil fertility, and fertilization management (Li et al., \u003cspan citationid=\"CR34\" class=\"CitationRef\"\u003e2026\u003c/span\u003e). In this study, by including these variables, the R\u003csup\u003e2\u003c/sup\u003e\u003csub\u003eCV\u003c/sub\u003e for yield increased to 0.70, and the tillering stage remained the optimal prediction period. For Brix, the R\u003csup\u003e2\u003c/sup\u003e\u003csub\u003eCV\u003c/sub\u003e improved to 0.44, and the optimal prediction stage shifted to the maturity stage. Sugar content is strongly influenced by cultivar, late season crop status, temperature and management related information. In contrast, for millable stalk population, the R\u003csup\u003e2\u003c/sup\u003e\u003csub\u003eCV\u003c/sub\u003e slightly decreased with the inputs of SCPs and FI. The additional variables may introduce redundant or less informative signals for millable stalk population (Yu et al., \u003cspan citationid=\"CR63\" class=\"CitationRef\"\u003e2025\u003c/span\u003e).\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec28\" class=\"Section2\"\u003e \u003ch2\u003e4.2. SHAP analysis of multi-source feature contributions\u003c/h2\u003e \u003cp\u003eKhosravirad et al. (\u003cspan citationid=\"CR30\" class=\"CitationRef\"\u003e2019\u003c/span\u003e) and Rahman \u0026amp; Robson (\u003cspan citationid=\"CR47\" class=\"CitationRef\"\u003e2020\u003c/span\u003e) successfully employed GNDVI to predict sugarcane yield, achieving favorable accuracy. Akbarian et al. (\u003cspan citationid=\"CR1\" class=\"CitationRef\"\u003e2022\u003c/span\u003e) obtained similar results using CI. In this study, SHAP analysis revealed that GNDVI was the most influential feature, with CI ranking second in contribution for yield estimation. The higher GNDVI values tended to increase the model prediction. However, higher CI and RVI values tended to reduce the yield prediction, which may present nonlinear response patterns with yield. TIs were less dominant than VIs, which may provide complementary information on plot level spatial structural variation. Zhu et al. (\u003cspan citationid=\"CR70\" class=\"CitationRef\"\u003e2026\u003c/span\u003e) indicated that the combination of VIs and texture features improved the discrimination of sugarcane ratoon. Compared with NDTI (Zheng et al., \u003cspan citationid=\"CR69\" class=\"CitationRef\"\u003e2019\u003c/span\u003e; Yu et al., \u003cspan citationid=\"CR62\" class=\"CitationRef\"\u003e2024\u003c/span\u003e) and EVTI (Yuan et al., \u003cspan citationid=\"CR64\" class=\"CitationRef\"\u003e2023\u003c/span\u003e), DDRTIs showed a greater contribution to yield estimation. DDRTI i\u003cspan type=\"Underline\" class=\"Underline\" name=\"Emphasis\"\u003es derived by\u003c/span\u003e four band GLCM texture features, which may amplify spatial variation in canopy structure across multiple spectral responses. In this study, first ratoon cane exhibited larger within plot heterogeneous (with larger CV values of stem parameters and single stem weight compared to those in 2024). In terms of fertilization, base fertilizer calcium magnesium fertilizer (BF_CMF) and base fertilizer urea (BF_Urea) were identified as important features. However, topdressing was not selected as a final predictive feature. This result indicates that early nutrient supply is critical for sugarcane yield formation. Base fertilizer served as the primary nutrient source for young plants, supporting root growth, tillering, and early canopy development. Similar results were also reported in corn growth (Li et al., \u003cspan citationid=\"CR33\" class=\"CitationRef\"\u003e2022\u003c/span\u003e; Li et al., \u003cspan citationid=\"CR34\" class=\"CitationRef\"\u003e2026\u003c/span\u003e).\u003c/p\u003e \u003cp\u003eFor the prediction of millable stalk population, EVTI and NDTI contributed more than DDRTI. Although DDRTI was proved effective for yield estimation, it was less suitable for predicting the number of millable stalk population. Increased index complexity did not guarantee improved performance, and it was important to select texture indices for specific traits (Guo et al., \u003cspan citationid=\"CR23\" class=\"CitationRef\"\u003e2023\u003c/span\u003e). Conversely, for Brix prediction, the contribution of DDRTI exceeded that of EVTI and NDTI. Additionally, soil pH showed a positive effect on Brix prediction. Dao et al. (\u003cspan citationid=\"CR9\" class=\"CitationRef\"\u003e2022\u003c/span\u003e) suggested that pH should be optimized to maximize sugarcane sugar accumulation. Soil pH influences nutrient availability and enzyme activity, such as protein kinase activity, which regulates sucrose metabolism (Huber et al., 1996). For the prediction of sugarcane yield, millable stalk population, and Brix, VIs were among the most important features. This indicates that canopy spectral reflectance characteristics can characterize photosynthetic physiological status and population structure differences during sugarcane growth and development, serving as key remote sensing indicators of crop vigor and yield formation potential. The inclusion of TIs provided complementary plot level spatial information related to within canopy heterogeneity and structural organization, which could not be fully captured by VIs alone. The formulations of NDTI, EVTI, and DDRTI in this study differed from those reported in some studies, where texture indices were derived either by combining multiple texture features from a single band (Zheng et al., 2019) or through a random texture-feature-cross of multiple band combination (Yu et al., \u003cspan citationid=\"CR62\" class=\"CitationRef\"\u003e2024\u003c/span\u003e). In this study, the texture indices were constructed in a form analogous to conventional VIs. This design may enhance interpretability and reduce inconsistencies.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec29\" class=\"Section2\"\u003e \u003ch2\u003e4.3. Limitations and future prospects\u003c/h2\u003e \u003cp\u003eThis study investigated the potential of Jilin-1 multispectral imagery for plot level sugarcane traits estimation. The promising results were reported for yield and quality related traits (millable stalk population and Brix). However, several limitations remain due to constraints in data availability, study scope and methodological design.\u003c/p\u003e \u003cp\u003eFirstly, this study only analyzed data from 2024 and 2025, and included only plant cane and first ratoon cane. The results showed that they did not differ significantly in yield or Brix, but significant differences were observed in stalk weight, millable stalk population, and stalk diameter (Table S11). Second ratoon cane data was not included in this study. The conclusions remain limited in the applicability across ratoon generations. Secondly, the texture indices used in this study were able to improve the performance in sugarcane traits estimation, which may be able to capture differences in canopy structure at the plot level. Some texture indices appeared to be more sensitive to overall canopy conditions (e.g., NDRTI). Others seemed more responsive to structural variation at the individual plant level. However, the causes of these differences were not further analyzed.\u003c/p\u003e \u003cp\u003eThirdly, due to the limited sample size, this study mainly relied on traditional machine learning methods. And SVR was identified as the best model. Such methods are practical under small sample size. More data driven algorithms (Akbarian et al., \u003cspan citationid=\"CR2\" class=\"CitationRef\"\u003e2024\u003c/span\u003e), and hybrid framework combining crop growth models with ML (de Oliveira et al., 2020; Xu et al., \u003cspan citationid=\"CR59\" class=\"CitationRef\"\u003e2025\u003c/span\u003e), were not explored. In addition, although high spatial resolution multispectral imagery (Jilin-1) was used, the lack of accurate regional-scale soil property information limited the analysis and map to the experimental plot scale. Large-area mapping and regional validation were not conducted, which constrains the application of the findings. Moreover, under high spatial resolution conditions, canopy reflectance may be affected by factors such as soil background, adjacency effects, planting density, and viewing geometry (Fan et al., \u003cspan citationid=\"CR17\" class=\"CitationRef\"\u003e2025\u003c/span\u003e; Li et al., \u003cspan citationid=\"CR34\" class=\"CitationRef\"\u003e2026\u003c/span\u003e?). In this study, plot-level mean values of reflectance were used for analysis, but the influence of within-plot soil background variation and related factors on reflectance was not assessed.\u003c/p\u003e \u003cp\u003eFinally, although the models achieved acceptable accuracy in yield and millable stalk population estimation, the performance for Brix estimation remained relatively low. This sugar-related quality trait is more difficult to monitor using the current remote sensing and modeling framework. Future research should incorporate multi-source data, such as temperature, precipitation, soil properties, varietal characteristics, management practices, thermal, and LiDAR to improve the accuracy and robustness of sugarcane trait estimation (Xie \u0026amp;Yang, \u003cspan citationid=\"CR57\" class=\"CitationRef\"\u003e2020\u003c/span\u003e).\u003c/p\u003e \u003c/div\u003e"},{"header":"5. Conclusion","content":"\u003cp\u003eThis study integrated VIs, TIs, SCPs, and FI to predict sugarcane phenotypic traits. SVR had the best performance in the estimation of yield, millable stalk population, and Brix. The inclusion of TIs with different window sizes in different growth stages improved the model accuracy. For sugarcane yield, combining VIs, TIs, SCPs, and FI allowed the model to reach the best prediction accuracy during the tillering period across plant cane and first ratoon cane (R\u003csup\u003e2\u003c/sup\u003e\u003csub\u003eCV\u003c/sub\u003e\u0026thinsp;=\u0026thinsp;0.70, RMSE\u003csub\u003eCV\u003c/sub\u003e = 13.84 t/ha, RPD\u003csub\u003eCV\u003c/sub\u003e = 1.83). The yield model had the potential to predict sugarcane yield in early growth stage. For millable stalk population, the inclusion of VIs and TIs achieved the highest accuracy at the maturation stage (R\u003csup\u003e2\u003c/sup\u003e\u003csub\u003eCV\u003c/sub\u003e\u0026thinsp;=\u0026thinsp;0.63, RMSE\u003csub\u003eCV\u003c/sub\u003e = 6602 stalks/ha, RPD\u003csub\u003eCV\u003c/sub\u003e = 1.66). Regarding Brix, the inclusion of SCPs and FI together with VIs and TIs achieved the highest prediction accuracy at the maturation stage (R\u003csup\u003e2\u003c/sup\u003e\u003csub\u003eCV\u003c/sub\u003e\u0026thinsp;=\u0026thinsp;0.44, RMSE\u003csub\u003eCV\u003c/sub\u003e = 0.53 \u0026deg;Bx, RPD\u003csub\u003eCV\u003c/sub\u003e = 1.33). These results demonstrate the contribution of VIs and texture indices based on optimal texture windows, as well as multi-source data fusion, in sugarcane traits estimation. SCPs suitable for sugarcane growth and appropriate fertilization ratios contributed to high sugarcane yield and Brix. This study validated the applicability of Jilin-1 high-resolution satellite imagery for field-scale sugarcane traits estimation and provided an important insight for remote sensing phenotyping of other crops.\u003c/p\u003e"},{"header":"Declarations","content":"\u003cp\u003e\u003cstrong\u003ePublisher’s note\u0026nbsp;\u003c/strong\u003eSpringer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.\u003c/p\u003e\n\u003cp\u003eSpringer Nature or its licensor (e.g. a society or other partner) holds exclusive rights to this article under a publishing agreement with the author(s) or other rightsholder(s); author self-archiving of the accepted manuscript version of this article is solely governed by the terms of such publishing agreement and applicable law.\u003c/p\u003e\u003cp\u003e \u003ch2\u003eDeclaration of Competing Interest\u003c/h2\u003e \u003cp\u003eThe authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.\u003c/p\u003e \u003c/p\u003e\u003ch2\u003eAuthor Contribution\u003c/h2\u003e\u003cp\u003eYun Zhang: Conceptualization, Data curation, Investigation, Methodology, Writing \u0026ndash; original draft. Guangtao Xu: Data curation, Investigation, Methodology, Writing \u0026ndash; original draft. Dan Li: Conceptualization, Methodology, Writing \u0026ndash; original draft, Writing \u0026ndash; review and funding acquisition. Yinglin Lu: Investigation, Data curation, Writing \u0026ndash; review and funding acquisition. Hao Jiang: Investigation, Data curation. Shasha Luo: Investigation, Data curation. Mingfu Wen: Investigation, Data curation.Jing Zhang: Investigation, Data curation. Xingda Chen: Investigation, Data curation. Qiong Zheng: Methodology, Data curation. Chongyang Wang: Investigation, Data curation. Zihan Liu: Investigation, Data curation. Mengjun Ku: Investigation, Data curation. Junliang Chen: Investigation, Data curation.\u003c/p\u003e\u003ch2\u003eAcknowledgements\u003c/h2\u003e \u003cp\u003eThis work was supported by Guangzhou Science and Technology Project (No. 2024B03J1321), GDAS\u0026rsquo;Project of Science and Technology Development (2022GDASZH-2022010202) and the China Association for Science and Technology (CAST) Special Talent Cultivation Project for PhD Students (2025).\u003c/p\u003e\u003ch2\u003eData availability\u003c/h2\u003e \u003cp\u003eThe data that support the findings of this study are available in the main text and the supplementary materials. Part of the code and test data for model training are available at \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://github.com/guangtaoxu08-dev/SPT_JL\u003c/span\u003e\u003cspan address=\"https://github.com/guangtaoxu08-dev/SPT_JL\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e. Additional data are available from the corresponding author upon reasonable request.\u003c/p\u003e"},{"header":"References","content":"\u003col\u003e\u003cli\u003e\u003cspan\u003eAkbarian, S., Xu, C., Wang, W., Ginns, S., \u0026amp; Lim, S. (2022). Sugarcane yields prediction at the row level using a novel cross-validation approach to multi-year multispectral images. \u003cem\u003eComputers and Electronics in Agriculture\u003c/em\u003e, \u003cem\u003e198\u003c/em\u003e, 107024. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1016/j.compag.2022.107024\u003c/span\u003e\u003cspan address=\"10.1016/j.compag.2022.107024\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eAkbarian, S., Jamnani, M. R., Xu, C., Wang, W., \u0026amp; Lim, S. (2024). Plot level sugarcane yield estimation by machine learning on multispectral images: a case study of Bundaberg, Australia. \u003cem\u003eInformation Processing in Agriculture\u003c/em\u003e, \u003cem\u003e11\u003c/em\u003e(4), 476\u0026ndash;487. https://doi.org/10.xxxx/j.inpa.2024.xxxx\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eAlem\u0026aacute;n-Montes, B., Zabala, A., Henr\u0026iacute;quez, C., \u0026amp; Serra, P. (2023). Modelling two sugarcane agro-industrial yields using Sentinel/Landsat time-series data and their spatial validation at different scales in Costa Rica. \u003cem\u003eRemote Sensing\u003c/em\u003e, \u003cem\u003e15\u003c/em\u003e, 5476. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.3390/rs15235476\u003c/span\u003e\u003cspan address=\"10.3390/rs15235476\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eAmorim, M. T. A., Silvero, N. E. Q., Bellinaso, H., G\u0026oacute;mez, A. M. R., Greschuk, L. T., Campos, L. R., \u0026amp; Dematt\u0026ecirc;, J. A. M. (2022). Impact of soil types on sugarcane development monitored over time by remote sensing. \u003cem\u003ePrecision Agriculture\u003c/em\u003e, \u003cem\u003e23\u003c/em\u003e, 1532\u0026ndash;1552. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1007/s11119-022-09893-7\u003c/span\u003e\u003cspan address=\"10.1007/s11119-022-09893-7\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eBrovelli, A. M., Crespi, M., Fratarcangeli, F., Giannone, F., \u0026amp; Realini, E. (2008). Accuracy assessment of high resolution satellite imagery orientation by leave-one-out method. \u003cem\u003eISPRS Journal of Photogrammetry and Remote Sensing\u003c/em\u003e, \u003cem\u003e63\u003c/em\u003e(4), 427\u0026ndash;440. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1016/j.isprsjprs.2008.01.006\u003c/span\u003e\u003cspan address=\"10.1016/j.isprsjprs.2008.01.006\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eCanton, H. (2021). Food and Agriculture Organization of the United Nations\u0026mdash;FAO. The Europa Directory of International Organizations, 297\u0026ndash;305. ISBN: 978-1-00-317990-0.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eChea, C., Saengprachatanarug, K., Posom, J., Wongphati, M., \u0026amp; Taira, E. (2020). Sugar yield parameters and fiber prediction in sugarcane fields using a multispectral camera mounted on a small unmanned aerial system (UAS). \u003cem\u003eSugar Tech\u003c/em\u003e, \u003cem\u003e22\u003c/em\u003e, 605\u0026ndash;621. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1007/s12355-020-00802-5\u003c/span\u003e\u003cspan address=\"10.1007/s12355-020-00802-5\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eChen, Z., Zhang, Y., Zang, P., Liu, Z., \u0026amp; Zhao, W. (2023). Mapping the grassland quality of the flooded plain based on Jilin-01 images with a red-edge spectral band. \u003cem\u003eFrontiers in Earth Science\u003c/em\u003e, \u003cem\u003e10\u003c/em\u003e, 1024680. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.3389/feart.2022.1024680\u003c/span\u003e\u003cspan address=\"10.3389/feart.2022.1024680\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eDao, J., Mao, J., Li, R., Yang, S., Deng, J., Fan, X., \u0026amp; Zhang, Y. (2022). Effects of key climate and soil factors on sugarcane production in different ecological regions in China. \u003cem\u003eSoil and Fertilizer Sciences in China\u003c/em\u003e, (12), 102\u0026ndash;113. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.11838/sfsc.1673-6257.21593\u003c/span\u003e\u003cspan address=\"10.11838/sfsc.1673-6257.21593\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003ede Silva, F., Chaves, N. R., Luciano, M. E. D., Sanches, A. C. S., de Almeida, I. D., C.M. and, \u0026amp; Adami, M. (2024). Sugarcane yield estimation using satellite remote sensing data in empirical or mechanistic modeling: a systematic review. \u003cem\u003eRemote Sensing\u003c/em\u003e, \u003cem\u003e16\u003c/em\u003e, 863. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.3390/rs16050863\u003c/span\u003e\u003cspan address=\"10.3390/rs16050863\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003ede Oliveira, M. P. G., \u0026amp; Rodrigues, L. H. A. (2020). How good are the models available for estimating sugar content in sugarcane? \u003cem\u003eEuropean Journal Of Agronomy\u003c/em\u003e, \u003cem\u003e113\u003c/em\u003e, 125992. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1016/j.eja.2019.125992\u003c/span\u003e\u003cspan address=\"10.1016/j.eja.2019.125992\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eDlamini, N. E., Franke, A. C., \u0026amp; Zhou, M. (2024). Indices for measuring ratooning ability of sugarcane varieties. \u003cem\u003eCrop Science\u003c/em\u003e, \u003cem\u003e64\u003c/em\u003e, 667\u0026ndash;677. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1002/csc2.21191\u003c/span\u003e\u003cspan address=\"10.1002/csc2.21191\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eDrucker, H., Burges, C. J. C., Kaufman, L., Smola, A. J., \u0026amp; Vapnik, V. (1997). Support vector regression machines. \u003cem\u003eAdvances in Neural Information Processing Systems 9\u003c/em\u003e (pp. 155\u0026ndash;161). MIT Press.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eDu, Y., Zhu, R., Gong, J., Wang, D., \u0026amp; Zhong, X. (2023). Retrieval of crop leaf area index based on Jilin-1GP image. \u003cem\u003eRemote Sensing Technology and Application\u003c/em\u003e, \u003cem\u003e38\u003c/em\u003e(4), 816\u0026ndash;826. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.11873/j.issn.1004-0323.2023.4.0816\u003c/span\u003e\u003cspan address=\"10.11873/j.issn.1004-0323.2023.4.0816\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eDurbha, S. S., King, R. L., \u0026amp; Younan, N. H. (2007). Support vector machines regression for retrieval of leaf area index from multiangle imaging spectroradiometer. \u003cem\u003eRemote Sensing of Environment\u003c/em\u003e, \u003cem\u003e107\u003c/em\u003e(1\u0026ndash;2), 348\u0026ndash;361. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1016/j.rse.2006.07.032\u003c/span\u003e\u003cspan address=\"10.1016/j.rse.2006.07.032\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eEfron, B. (1979). Bootstrap methods: another look at the jackknife. \u003cem\u003eThe Annals of Statistics\u003c/em\u003e, \u003cem\u003e7\u003c/em\u003e(1), 1\u0026ndash;26. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1214/aos/1176344552\u003c/span\u003e\u003cspan address=\"10.1214/aos/1176344552\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eFan, W. J., Peng, N. J., Cao, B., Mu, X. H., Yang, S. Q., He, Q. C., Zhai, D. C., Ren, H. Z., Cui, Y. K., \u0026amp; Yan, G. J. (2025). Progress in high spatial resolution vegetation quantitative remote sensing. \u003cem\u003eNational Remote Sensing Bulletin\u003c/em\u003e, \u003cem\u003e29\u003c/em\u003e(6), 1365\u0026ndash;1387. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.11834/jrs.20255036\u003c/span\u003e\u003cspan address=\"10.11834/jrs.20255036\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eFAO (2025). FAOSTAT: crops and livestock products. Available at: \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttp://www.fao.org/faostat/en/#data/QCL\u003c/span\u003e\u003cspan address=\"http://www.fao.org/faostat/en/#data/QCL\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (accessed 18 December 2025).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eFriedman, J. H. (2002). Stochastic gradient boosting. \u003cem\u003eComputational Statistics and Data Analysis\u003c/em\u003e, \u003cem\u003e38\u003c/em\u003e, 367\u0026ndash;378. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1016/S0167-9473(01)00065-2\u003c/span\u003e\u003cspan address=\"10.1016/S0167-9473(01)00065-2\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eFu, B., Zuo, P., Liu, M., Lan, G., He, H., Lao, Z., Zhang, Y., Fan, D., \u0026amp; Gao, E. (2022). Classifying vegetation communities in karst wetland by synergistic use of image fusion and object-based machine learning algorithm with Jilin-1 and UAV multispectral images. \u003cem\u003eEcological Indicators\u003c/em\u003e, \u003cem\u003e140\u003c/em\u003e, 108989. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1016/j.ecolind.2022.108989\u003c/span\u003e\u003cspan address=\"10.1016/j.ecolind.2022.108989\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eGao, S., Zhong, R., Yan, K., Ma, X., Chen, X., Pu, J., Gao, S., Qi, J., Yin, G., \u0026amp; Myneni, R. B. (2023). Evaluating the saturation effect of vegetation indices in forests using 3D radiative transfer simulations and satellite observations. \u003cem\u003eRemote Sensing of Environment\u003c/em\u003e, \u003cem\u003e295\u003c/em\u003e, 113665. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1016/j.rse.2023.113665\u003c/span\u003e\u003cspan address=\"10.1016/j.rse.2023.113665\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eGaston, E., Fr\u0026iacute;as, J. M., Cullen, P. J., O'Donnell, C. P., \u0026amp; Gowen, A. A. (2010). Prediction of polyphenol oxidase activity using visible near-infrared hyperspectral imaging on mushroom (Agaricus bisporus) caps. \u003cem\u003eJournal of Agricultural and Food Chemistry\u003c/em\u003e, \u003cem\u003e58\u003c/em\u003e(10), 6226\u0026ndash;6233. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1021/jf100501q\u003c/span\u003e\u003cspan address=\"10.1021/jf100501q\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eGuo, Y., Xiao, Y., Hao, F., Zhang, X., Chen, J., de Beurs, K., He, Y., \u0026amp; Fu, Y. H. (2023). Comparison of different machine learning algorithms for predicting maize grain yield using UAV-based hyperspectral images. \u003cem\u003eInternational Journal of Applied Earth Observation and Geoinformation\u003c/em\u003e, \u003cem\u003e124\u003c/em\u003e, 103528. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1016/j.jag.2023.103528\u003c/span\u003e\u003cspan address=\"10.1016/j.jag.2023.103528\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eGuyon, I., Weston, J., Barnhill, S., \u0026amp; Vapnik, V. (2002). Gene selection for cancer classification using support vector machines. \u003cem\u003eMachine Learning\u003c/em\u003e, \u003cem\u003e46\u003c/em\u003e(1\u0026ndash;3), 389\u0026ndash;422. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1023/A:1012487302797\u003c/span\u003e\u003cspan address=\"10.1023/A:1012487302797\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eHaralick, R. M., Shanmugam, K., \u0026amp; Dinstein, I. (1973). Textural features for image classification. IEEE Transactions on Systems, Man, and Cybernetics, SMC-3(6), 610\u0026ndash;621. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1109/TSMC.1973.4309314\u003c/span\u003e\u003cspan address=\"10.1109/TSMC.1973.4309314\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eHoerl, A. E., \u0026amp; Kennard, R. W. (1970). Ridge regression: biased estimation for nonorthogonal problems. \u003cem\u003eTechnometrics\u003c/em\u003e, \u003cem\u003e12\u003c/em\u003e, 55\u0026ndash;67. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1080/00401706.1970.10488634\u003c/span\u003e\u003cspan address=\"10.1080/00401706.1970.10488634\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eHuber, S. C., \u0026amp; Huber, J. L. (1996). Role and regulation of sucrose-phosphate synthase in higher plants. \u003cem\u003eAnnual Review of Plant Physiology and Plant Molecular Biology\u003c/em\u003e, \u003cem\u003e47\u003c/em\u003e, 431\u0026ndash;444. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1146/annurev.arplant.47.1.431\u003c/span\u003e\u003cspan address=\"10.1146/annurev.arplant.47.1.431\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eJimenez-Sierra, D. A., Ben\u0026iacute;tez-Restrepo, H. D., Vargas-Cardona, H. D., \u0026amp; Chanussot, J. (2020). Graph-based data fusion applied to change detection and biomass estimation in rice crops. \u003cem\u003eRemote Sensing\u003c/em\u003e, \u003cem\u003e12\u003c/em\u003e(17), 2683. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.3390/rs12172683\u003c/span\u003e\u003cspan address=\"10.3390/rs12172683\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eKambli, M., \u0026amp; Palkar, B. (2025). Remote sensing and machine learning methods to analyse the vegetation of sugarcane crop. \u003cem\u003eMultimedia Tools and Applications\u003c/em\u003e, \u003cem\u003e84\u003c/em\u003e, 45297\u0026ndash;45319. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1007/s11042-025-20950-8\u003c/span\u003e\u003cspan address=\"10.1007/s11042-025-20950-8\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eKhosravirad, M., Omid, M., Sarmadian, F., \u0026amp; Hosseinpour, S. (2019). Predicting sugarcane yields in Khuzestan using a large time-series of remote sensing imagery region. \u003cem\u003eInt Arch Photogramm Remote Sens Spat Inf Sci XLII\u003c/em\u003e, \u003cem\u003e-4/W18\u003c/em\u003e, 645\u0026ndash;648.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003e\u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.5194/isprs-archives-xlii-4-w18-645-2019\u003c/span\u003e\u003cspan address=\"10.5194/isprs-archives-xlii-4-w18-645-2019\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eKhuimphukhieo, I., Bhandari, M., Enciso, J., \u0026amp; da Silva, J. A. (2025). Estimating sugarcane yield and its components using unoccupied aerial systems (UAS)-based high throughput phenotyping (HTP). Computers and Electronics in Agriculture, 237 (Part B), 110658. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1016/j.compag.2025.110658\u003c/span\u003e\u003cspan address=\"10.1016/j.compag.2025.110658\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eLi, D., Miao, Y., Ransom, C. J., Bean, G. M., Kitchen, N. R., Fern\u0026aacute;ndez, F. G., Sawyer, J. E., Camberato, J. J., Carter, P. R., Ferguson, R. B., et al. (2022). Corn nitrogen nutrition index prediction improved by integrating genetic, environmental, and management factors with active canopy sensing using machine learning. \u003cem\u003eRemote Sensing\u003c/em\u003e, \u003cem\u003e14\u003c/em\u003e, 394. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.3390/rs14020394\u003c/span\u003e\u003cspan address=\"10.3390/rs14020394\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eLi, D., Wang, C., Miao, Y., et al. (2026). Developing an innovative stacking ensemble machine learning and multi-source data fusion-based precision nitrogen management strategy for corn. \u003cem\u003ePrecision Agriculture\u003c/em\u003e, \u003cem\u003e27\u003c/em\u003e, 41. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1007/s11119-026-10344-7\u003c/span\u003e\u003cspan address=\"10.1007/s11119-026-10344-7\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eLi, J., Duan, T. T., Deng, K., Jing, Z. X., Lin, Z., Liang, Y. Q., Deng, C. N., Gao, Y., \u0026amp; Zheng, C. (2019). Responses of sugarcane varieties growth to nitrogen fertilizer application and related use efficiency. \u003cem\u003eChinese Journal of Tropical Crops\u003c/em\u003e, \u003cem\u003e40\u003c/em\u003e(4), 649\u0026ndash;656. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.3969/j.issn.1000-2561.2019.04.001\u003c/span\u003e\u003cspan address=\"10.3969/j.issn.1000-2561.2019.04.001\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eLi, Q., Song, X., Meng, X., Zhang, J., Zhang, M., Chen, L. Y., Li, J., \u0026amp; Wang, B. (2025). Shaping future sugarcane: ideal plant architecture and breeding strategies. \u003cem\u003eMolecular Plant\u003c/em\u003e, \u003cem\u003e18\u003c/em\u003e, 725\u0026ndash;728. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1016/j.molp.2025.04.004\u003c/span\u003e\u003cspan address=\"10.1016/j.molp.2025.04.004\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eLiu, Y., Feng, H., Yue, J., Fan, Y., Bian, M., Ma, Y., Jin, X., Song, X., \u0026amp; Yang, G. (2023a). Estimating potato above-ground biomass by using integrated unmanned aerial system-based optical, structural, and textural canopy measurements. \u003cem\u003eComputers and Electronics in Agriculture\u003c/em\u003e, \u003cem\u003e213\u003c/em\u003e, 108229. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1016/j.compag.2023.108229\u003c/span\u003e\u003cspan address=\"10.1016/j.compag.2023.108229\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eLi, Y. R. (2022). Growth and development of sugarcane (Saccharum spp. hybrid) and its relationship with environmental factors. In K. K. Verma, X. P. Song, V. D. Rajput, S. Solomon, Y. R. Li, \u0026amp; G. P. Rao (Eds.), \u003cem\u003eAgro-industrial Perspectives on Sugarcane Production under Environmental Stress\u003c/em\u003e. Springer. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1007/978-981-19-3955-6_1\u003c/span\u003e\u003cspan address=\"10.1007/978-981-19-3955-6_1\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eLiu, J., Zhu, Y., Song, L., Su, X., Li, J., Zheng, J., Zhu, X., Ren, L., Wang, W., \u0026amp; Li, X. (2023b). Optimizing window size and directional parameters of GLCM texture features for estimating rice AGB based on UAV multispectral imagery. \u003cem\u003eFrontiers in Plant Science\u003c/em\u003e, \u003cem\u003e14\u003c/em\u003e, 1284235. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.3389/fpls.2023.1284235\u003c/span\u003e\u003cspan address=\"10.3389/fpls.2023.1284235\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eLuciano, A. C. S., Picoli, M. C. A., Duft, D. G., Rocha, J. V., Leal, M. R. L. V., \u0026amp; le Maire, G. (2021). Empirical model for forecasting sugarcane yield on a local scale in Brazil using Landsat imagery and random forest algorithm. \u003cem\u003eComputers and Electronics in Agriculture\u003c/em\u003e, \u003cem\u003e184\u003c/em\u003e, 106063. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1016/j.compag.2021.106063\u003c/span\u003e\u003cspan address=\"10.1016/j.compag.2021.106063\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eLundberg, S. M., \u0026amp; Lee, S. I. (2017). A unified approach to interpreting model predictions. In: Proceedings of the 31st International Conference on Neural Information Processing Systems (NIPS\u0026rsquo;17), pp. 4768\u0026ndash;4777. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1016/j.compag.2021.106063\u003c/span\u003e\u003cspan address=\"10.1016/j.compag.2021.106063\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eMeier, E. A., \u0026amp; Thorburn, P. J. (2016). Long term sugarcane crop residue retention offers limited potential to reduce nitrogen fertilizer rates in Australian wet tropical environments. \u003cem\u003eFrontiers in Plant Science\u003c/em\u003e, \u003cem\u003e7\u003c/em\u003e, 1017. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.3389/fpls.2016.01017\u003c/span\u003e\u003cspan address=\"10.3389/fpls.2016.01017\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eOzkan, U. Y., \u0026amp; Demirel, T. (2021). The influence of window size on remote sensing-based prediction of forest structural variables. \u003cem\u003eEcological Processes\u003c/em\u003e, \u003cem\u003e10\u003c/em\u003e, 60. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1186/s13717-021-00330-4\u003c/span\u003e\u003cspan address=\"10.1186/s13717-021-00330-4\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003ePhang, S. K., Chiang, T. H. A., Happonen, A., \u0026amp; Chang, M. M. L. (2023). From satellite to UAV-based remote sensing: a review on precision agriculture. \u003cem\u003eIeee Access : Practical Innovations, Open Solutions\u003c/em\u003e, \u003cem\u003e11\u003c/em\u003e, 127057\u0026ndash;127076. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1109/ACCESS.2023.3330886\u003c/span\u003e\u003cspan address=\"10.1109/ACCESS.2023.3330886\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003ePaudel, D., Boogaard, H., de Wit, A., Janssen, S., Osinga, S., Pylianidis, C., \u0026amp; Athanasiadis, I. N. (2021). Machine learning for large-scale crop yield forecasting. \u003cem\u003eAgricultural Systems\u003c/em\u003e, \u003cem\u003e187\u003c/em\u003e, 103016. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1016/j.agsy.2020.103016\u003c/span\u003e\u003cspan address=\"10.1016/j.agsy.2020.103016\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eRasmussen, C. E., \u0026amp; Williams, C. K. I. (2006). \u003cem\u003eGaussian Processes for Machine Learning\u003c/em\u003e. The MIT Press. 978-0-262-18253-9.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eRahman, M. M., \u0026amp; Robson, A. (2020). Integrating Landsat-8 and Sentinel-2 time series data for yield prediction of sugarcane crops at the block level. \u003cem\u003eRemote Sensing\u003c/em\u003e, \u003cem\u003e12\u003c/em\u003e, 1313. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.3390/rs12081313\u003c/span\u003e\u003cspan address=\"10.3390/rs12081313\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eRosa, J. M., Fattori Junior, I. M., Melo, M. L. A., \u0026amp; Marin, F. R. (2026). Remote sensing-driven ensemble smoother assimilation of LAI for regional sugarcane yield estimation. \u003cem\u003eRemote Sensing Applications: Society and Environment\u003c/em\u003e, \u003cem\u003e41\u003c/em\u003e, 101952. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1016/j.rsase.2026.101952\u003c/span\u003e\u003cspan address=\"10.1016/j.rsase.2026.101952\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eShendryk, Y., Sofonia, J., Garrard, R., Rist, Y., Skocaj, D., \u0026amp; Thorburn, P. (2020). Fine-scale prediction of biomass and leaf nitrogen content in sugarcane using UAV LiDAR and multispectral imaging. \u003cem\u003eInternational Journal of Applied Earth Observation and Geoinformation\u003c/em\u003e, \u003cem\u003e92\u003c/em\u003e, 102177. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1016/j.jag.2020.102177\u003c/span\u003e\u003cspan address=\"10.1016/j.jag.2020.102177\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eSingla, S. K., Garg, R. D., \u0026amp; Dubey, O. P. (2018). Sugarcane ratoon discrimination using LANDSAT NDVI temporal data. \u003cem\u003eSpatial Information Research\u003c/em\u003e, \u003cem\u003e26\u003c/em\u003e, 415\u0026ndash;425. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1007/s41324-018-0184-0\u003c/span\u003e\u003cspan address=\"10.1007/s41324-018-0184-0\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eShi, Y., Gao, Y., Wang, Y., Luo, D., Chen, S., Ding, Z., \u0026amp; Fan, K. (2022). Using unmanned aerial vehicle-based multispectral image data to monitor the growth of intercropping crops in tea plantation. \u003cem\u003eFrontiers in Plant Science\u003c/em\u003e, \u003cem\u003e13\u003c/em\u003e, 820585. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.3389/fpls.2022.820585\u003c/span\u003e\u003cspan address=\"10.3389/fpls.2022.820585\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eSofonia, J., Shendryk, Y., Phinn, S., Roelfsema, C., Kendoul, F., \u0026amp; Skocaj, D. (2019). Monitoring sugarcane growth response to varying nitrogen application rates: a comparison of UAV SLAM LiDAR and photogrammetry. \u003cem\u003eInternational Journal of Applied Earth Observation and Geoinformation\u003c/em\u003e, \u003cem\u003e82\u003c/em\u003e, 101878. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1016/j.jag.2019.05.011\u003c/span\u003e\u003cspan address=\"10.1016/j.jag.2019.05.011\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eSom-ard, J., Atzberger, C., Izquierdo-Verdiguier, E., Vuolo, F., \u0026amp; Immitzer, M. (2021). Remote sensing applications in sugarcane cultivation: a review. \u003cem\u003eRemote Sensing\u003c/em\u003e, \u003cem\u003e13\u003c/em\u003e, 4040. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.3390/rs13204040\u003c/span\u003e\u003cspan address=\"10.3390/rs13204040\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eVasconcelos, J. C. S., Arantes, C. S., Speranza, E. A., Antunes, J. F. G., Barbosa, L. A. F., Can\u0026ccedil;ado, G., \u0026amp; M.d.A. (2025). Predicting sugarcane yield through temporal analysis of satellite imagery during the growth phase. \u003cem\u003eAgronomy\u003c/em\u003e, \u003cem\u003e15\u003c/em\u003e, 793. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.3390/agronomy15040793\u003c/span\u003e\u003cspan address=\"10.3390/agronomy15040793\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eWang, H., Dai, Y., Yao, Q., Ma, L., Zhang, Z., \u0026amp; Lv, X. (2025). Multi-task learning model driven by climate and remote sensing data collaboration for mid-season cotton yield prediction. \u003cem\u003eField Crops Research\u003c/em\u003e, \u003cem\u003e333\u003c/em\u003e, 110070. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1016/j.fcr.2025.110070\u003c/span\u003e\u003cspan address=\"10.1016/j.fcr.2025.110070\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eWu, N., Zhou, Y., Yin, S., Gong, H., \u0026amp; Zhang, C. (2025). Revealing the nonlinear impact of environmental regulation on ecological resilience using the XGBoost-SHAP model: evidence from the Yangtze River Delta region, China. \u003cem\u003eJournal of Cleaner Production\u003c/em\u003e, \u003cem\u003e514\u003c/em\u003e, 145700. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1016/j.jclepro.2025.145700\u003c/span\u003e\u003cspan address=\"10.1016/j.jclepro.2025.145700\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eXie, C., \u0026amp; Yang, C. (2020). A review on plant high-throughput phenotyping traits using UAV-based sensors. \u003cem\u003eComputers and Electronics in Agriculture\u003c/em\u003e, \u003cem\u003e178\u003c/em\u003e, 105731. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1016/j.compag.2020.105731\u003c/span\u003e\u003cspan address=\"10.1016/j.compag.2020.105731\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eXu, F., Wang, Z., Lu, G., Zeng, R., \u0026amp; Que, Y. (2021). Sugarcane ratooning ability: research status, shortcomings, and prospects. \u003cem\u003eBiology\u003c/em\u003e, \u003cem\u003e10\u003c/em\u003e(10), 1052. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.3390/biology10101052\u003c/span\u003e\u003cspan address=\"10.3390/biology10101052\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eXu, J., Du, X., Dong, T., Li, Q., Zhang, Y., Wang, H., Liu, M., Zhu, J., \u0026amp; Yang, J. (2025). Estimation of sugarcane biomass from Sentinel-2 leaf area index using an improved SAFY model (SAFY-Sugar). International Journal of Applied Earth Observation and Geoinformation, 140, 104570. https://doi.org/xxxxx.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eXuan, F., Su, W., Chen, Z., Huang, X., Zhai, W., Li, X., Zeng, Y., Li, Z., Li, J., \u0026amp; Huang, J. (2025). Performance of stacking machine learning and volume model for improving corn above ground biomass prediction. \u003cem\u003ePlant Phenomics\u003c/em\u003e, \u003cem\u003e7\u003c/em\u003e(3), 100068.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eYang, Y., Wu, Z., Wang, D., Wang, C., Yang, X., Wang, Y., Wang, J., Huang, Q., Hou, L., Wang, Z., et al. (2025). Sugarcane phenology retrieval in heterogeneous agricultural landscapes based on spatiotemporal fusion remote sensing data. \u003cem\u003eAgriculture\u003c/em\u003e, \u003cem\u003e15\u003c/em\u003e, 1578. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.3390/agriculture15151578\u003c/span\u003e\u003cspan address=\"10.3390/agriculture15151578\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eYu, J., Zhang, Y., Song, Z., Jiang, D., Guo, Y., Liu, Y., \u0026amp; Chang, Q. (2024). Estimating leaf area index in apple orchard by UAV multispectral images with spectral and texture information. \u003cem\u003eRemote Sensing\u003c/em\u003e, \u003cem\u003e16\u003c/em\u003e(17). \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e3237.https://doi.org/10.3390/rs16173237\u003c/span\u003e\u003cspan address=\"3237.10.3390/rs16173237\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eYu, Z., Chen, Q., Shan, Q., Lv, Y., Chen, Z., Zhou, Y., Zhang, P., Jiang, H., \u0026amp; Cao, W. (2025). Developing soybean yield prediction model based on multicolor space and texture features of UAV images. \u003cem\u003eComputers and Electronics in Agriculture\u003c/em\u003e, \u003cem\u003e235\u003c/em\u003e, 110345. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1016/j.compag.2025.110345\u003c/span\u003e\u003cspan address=\"10.1016/j.compag.2025.110345\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eYuan, W., Meng, Y., Li, Y., Ji, Z., Kong, Q., Gao, R., \u0026amp; Su, Z. (2023). Research on rice leaf area index estimation based on fusion of texture and spectral information. \u003cem\u003eComputers and Electronics in Agriculture\u003c/em\u003e, \u003cem\u003e211\u003c/em\u003e, 108016. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1016/j.compag.2023.108016\u003c/span\u003e\u003cspan address=\"10.1016/j.compag.2023.108016\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eYue, J., Yang, G., Tian, Q., Feng, H., Xu, K., \u0026amp; Zhou, C. (2019). Estimate of winter-wheat above-ground biomass based on UAV ultrahigh-ground-resolution image textures and vegetation indices. \u003cem\u003eISPRS Journal of Photogrammetry and Remote Sensing\u003c/em\u003e, \u003cem\u003e150\u003c/em\u003e, 226\u0026ndash;244. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1016/j.isprsjprs.2019.02.022\u003c/span\u003e\u003cspan address=\"10.1016/j.isprsjprs.2019.02.022\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eZhang, J., Qiu, X., Wu, Y., Zhu, Y., Cao, Q., Liu, X., \u0026amp; Cao, W. (2021). Combining texture, color, and vegetation indices from fixed-wing UAS imagery to estimate wheat growth parameters using multivariate regression methods. \u003cem\u003eComputers and Electronics in Agriculture\u003c/em\u003e, \u003cem\u003e185\u003c/em\u003e, 106138. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1016/j.compag.2021.106138\u003c/span\u003e\u003cspan address=\"10.1016/j.compag.2021.106138\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eZhang, Y., Zhao, Z., \u0026amp; Zheng, J. (2020). CatBoost: a new approach for estimating daily reference crop evapotranspiration in arid and semi-arid regions of Northern China. \u003cem\u003eJournal of Hydrology\u003c/em\u003e, \u003cem\u003e588\u003c/em\u003e, 125087. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1016/j.jhydrol.2020.125087\u003c/span\u003e\u003cspan address=\"10.1016/j.jhydrol.2020.125087\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eZhao, J., Li, J., Liu, Q., \u0026amp; Yang, L. (2012). A preliminary study on mechanism of LAI inversion saturation. The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, XXXIX-B1, 77\u0026ndash;81. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.5194/isprsarchives-XXXIX-B1-77-2012\u003c/span\u003e\u003cspan address=\"10.5194/isprsarchives-XXXIX-B1-77-2012\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eZheng, H., Cheng, T., Zhou, M., Li, D., Yao, X., Tian, Y., Cao, W., \u0026amp; Zhu, Y. (2019). Improved estimation of rice aboveground biomass combining textural and spectral analysis of UAV imagery. \u003cem\u003ePrecision Agriculture\u003c/em\u003e, \u003cem\u003e20\u003c/em\u003e, 611\u0026ndash;629. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1007/s11119-018-9600-7\u003c/span\u003e\u003cspan address=\"10.1007/s11119-018-9600-7\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eZhu, H., Dong, Z., Wei, L., Qin, S., Qin, X., \u0026amp; He, Y. (2026). From canopy segmentation to accurate prediction: an UAV-based multi-feature fusion framework for plot-scale ratoon sugarcane seedling counting. \u003cem\u003eInternational Journal of Applied Earth Observation and Geoinformation\u003c/em\u003e, \u003cem\u003e147\u003c/em\u003e, 105183. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1016/j.jag.2026.105183\u003c/span\u003e\u003cspan address=\"10.1016/j.jag.2026.105183\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003c/ol\u003e"}],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":true,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":false,"hideJournal":true,"highlight":"","institution":"","isAcceptedByJournal":false,"isAuthorSuppliedPdf":false,"isDeskRejected":"","isHiddenFromSearch":false,"isInQc":false,"isInWorkflow":false,"isPdf":false,"isPdfUpToDate":true,"isWithdrawnOrRetracted":false,"journal":{"display":true,"email":"[email protected]","identity":"researchsquare","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":true,"externalIdentity":"","sideBox":"","snPcode":"","submissionUrl":"/submission","title":"Research Square","twitterHandle":"researchsquare","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"","reportingPortfolio":"","inReviewEnabled":false,"inReviewRevisionsEnabled":true},"keywords":"Sugarcane phenotypic traits, Jilin-1, Texture indices, Machine learning, SHAP interpretability","lastPublishedDoi":"10.21203/rs.3.rs-9371174/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-9371174/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003ch2\u003ePurpose\u003c/h2\u003e \u003cp\u003eThe primary objective of this study was to evaluate the potential of high spatial resolution Jilin-1 satellite imagery for plot-level prediction of sugarcane yield, millable stalk population, and Brix, and to assess whether integrating spectral, texture, soil, and fertilization information could improve prediction performance for precision sugarcane management.\u003c/p\u003e\u003ch2\u003eMethods\u003c/h2\u003e \u003cp\u003eJilin-1 satellite imagery acquired at four growth stages, from seedling to maturity, was used to derive vegetation indices (VIs) and texture indices (TIs), including the normalized difference texture index (NDTI), enhanced vegetation texture index (EVTI), and double-difference ratio texture index (DDRTI). Soil chemical properties (SCPs) and fertilization information (FI) were further incorporated with the remotely sensed variables. Machine learning models were developed for plot-level prediction of sugarcane traits across plant cane and first ratoon cane, and texture window size was optimized to improve TI extraction and model performance.\u003c/p\u003e\u003ch2\u003eResults\u003c/h2\u003e \u003cp\u003eFor yield, the combination of VIs and TIs outperformed VIs alone at the tillering stage (R\u003csup\u003e2\u003c/sup\u003e\u003csub\u003eCV\u003c/sub\u003e\u0026thinsp;=\u0026thinsp;0.65, RMSECV\u0026thinsp;=\u0026thinsp;15.06 t/ha, RPDCV\u0026thinsp;=\u0026thinsp;1.68). Adding SCPs and FI further improved yield prediction across plant cane and first ratoon cane (R\u003csup\u003e2\u003c/sup\u003e\u003csub\u003eCV\u003c/sub\u003e\u0026thinsp;=\u0026thinsp;0.70, RMSECV\u0026thinsp;=\u0026thinsp;13.84 t/ha, RPDCV\u0026thinsp;=\u0026thinsp;1.83). Millable stalk population was best predicted at the maturation stage by VIs and Tis, achieving the best performance (R\u003csup\u003e2\u003c/sup\u003e\u003csub\u003eCV\u003c/sub\u003e\u0026thinsp;=\u0026thinsp;0.63, RMSECV\u0026thinsp;=\u0026thinsp;6602 stalks/ha, RPDCV\u0026thinsp;=\u0026thinsp;1.66). The best Brix model integrated VIs, TIs, SCPs, and FI at the maturation stage (R\u003csup\u003e2\u003c/sup\u003e\u003csub\u003eCV\u003c/sub\u003e\u0026thinsp;=\u0026thinsp;0.44, RMSECV\u0026thinsp;=\u0026thinsp;0.53 \u0026deg;Bx, RPDCV\u0026thinsp;=\u0026thinsp;1.33). SHAP analysis identified VIs as the dominant features for sugarcane traits prediction. And, DDRTI contributed more than NDTI and EVTI in yield and Brix prediction.\u003c/p\u003e\u003ch2\u003eConclusion\u003c/h2\u003e \u003cp\u003eIt is concluded that integrating spectral, texture, soil, and fertilization information from high spatial resolution Jilin-1 imagery is a promising approach for improving plot-level prediction of key sugarcane traits.\u003c/p\u003e","manuscriptTitle":"Integrating spectral, texture, soil and fertilization information for plot-level prediction of sugarcane yield, millable stalk population and Brix from Jilin-1 imagery ","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2026-04-27 11:31:43","doi":"10.21203/rs.3.rs-9371174/v1","editorialEvents":[{"type":"communityComments","content":0}],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"researchsquare","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":true,"externalIdentity":"","sideBox":"","snPcode":"","submissionUrl":"/submission","title":"Research Square","twitterHandle":"researchsquare","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"","reportingPortfolio":"","inReviewEnabled":false,"inReviewRevisionsEnabled":true}}],"origin":"","ownerIdentity":"eee8b887-3e19-48da-b1e9-4105bb637a36","owner":[],"postedDate":"April 27th, 2026","published":true,"recentEditorialEvents":[{"type":"reviewerAgreed","content":"40123682213319747813217232662868514372","date":"2026-05-15T05:40:50+00:00","index":29,"fulltext":""},{"type":"reviewerAgreed","content":"178954124206794403423055689887972261398","date":"2026-05-13T06:43:47+00:00","index":28,"fulltext":""},{"type":"editorInvitedReview","content":"","date":"2026-05-08T22:02:44+00:00","index":21,"fulltext":""}],"rejectedJournal":[],"revision":"","amendment":"","status":"posted","subjectAreas":[],"tags":[],"updatedAt":"2026-04-27T11:31:43+00:00","versionOfRecord":[],"versionCreatedAt":"2026-04-27 11:31:43","video":"","vorDoi":"","vorDoiUrl":"","workflowStages":[]},"version":"v1","identity":"rs-9371174","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-9371174","identity":"rs-9371174","version":["v1"]},"buildId":"XKTyCvWXoU3ODBz1xrDgd","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}