Significantly reducing the impact of light: a novel framework for smartphone to predict water turbidity

Significantly reducing the impact of light: a novel framework for smartphone to predict water turbidity | 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 Significantly reducing the impact of light: a novel framework for smartphone to predict water turbidity Lingyan Qi, Kejia Zhang, Mingzhu Guo, Xinzhe Jiang, Weiwei Zhao, and 3 more This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-9317150/v1 This work is licensed under a CC BY 4.0 License Status: Under Review Version 1 posted 5 You are reading this latest preprint version Abstract Water turbidity is a key indicator of water quality, reflecting the concentration of suspended particles and affecting aquatic ecosystems. This study proposes a novel smartphone-based framework to predict water turbidity. The framework uses light correction to reduce the effect of varying illumination. It uses Mask R-CNN to extract a shadow-free water surface and a convolutional neural network (CNN) to predict turbidity. Light correction substantially improves prediction accuracy and stability, especially for turbidity values between 10 and 20 NTU. Evaluation shows that after correction, R² increased from 0.49 to 0.68, and RMSE decreased from 5.19 to 4.13. Overall, the framework provides a universal preprocessing approach that reduces the influence of light and enables more consistent and accurate turbidity prediction in the field. Turbidity Image Processing Smartphone Deep learning Figures Figure 1 Figure 2 Figure 3 Figure 4 Figure 5 Figure 6 Figure 7 Figure 8 1. Introduction Turbidity is a key indicator of suspended particulate and optical properties (Li et al., 2023b; Ding et al., 2021). It influences aquatic ecosystems and regulates key biogeochemical processes (Sahoo and Anandhi, 2023; Sehgal et al., 2022; Wu et al., 2025b). Global observations also show strong turbidity contrasts among lakes (Caroni et al., 2025; Chowdhury et al., 2025), Across 774 lakes worldwide, 63.4% show low turbidity (≤ 5 NTU), while 40% show significant decreases and 32% show significant increases over time (Wu et al., 2025a). In China, a 40-year Landsat analysis reported strong regional contrasts in river turbidity, with mean values of 32.5 NTU in the east and 22.8 NTU in the west, and long-term declines across 65.48% of river areas (Yan et al., 2025). In the Yangtze River Basin, mean turbidity is 60.1 NTU in river channels and 29.6 NTU in nearby lakes and reservoirs (Li and Xia, 2023), and long-term turbidity trends differ among water-body types with rates spanning − 1.3 to + 0.7 NTU yr⁻¹ (Lin et al., 2023). At the national lake scale, a China-wide assessment reported a wide turbidity range, from 0.01–98.63 NTU in 2015 to 0.01-122.09 NTU in 2020, together with a shift in mean turbidity across years (Li et al., 2023a). High-frequency and low-cost turbidity measurement remains challenging in such highly variable waters. Remote sensing is widely used to monitor surface-water quality (Palmer et al., 2015; Odermatt et al., 2012; Gholizadeh et al., 2016), and it is increasingly coupled with machine learning and deep learning (Pang et al., 2025; Deng et al., 2024; Chen et al., 2022). Random Forest has been applied to map turbidity dynamics in the Yangtze River Delta(Lin et al., 2023), to predict turbidity in Lake Taihu using meteorological drivers (Zhang et al., 2021), and to model sediment-related turbidity signals in the Yellow River (Qiu et al., 2025). XGBoost has been used to retrieve turbidity in Lake Taihu from Sentinel-2A/B and Landsat-8/9 (Yang et al., 2023), and to forecast turbidity in river networks using Sentinel-2 (Santos et al., 2025). Gradient boosting models have also been used for turbidity retrieval, including GBDT for lakes in Northeast China using Sentinel-2 (Ma et al., 2021), CatBoost for coastal-water quality inversion along the Fujian coast using Sentinel-3 OLCI (Chen et al., 2024).It is worth noting that convolutional neural networks offer distinct advantages, as CNNs can efficiently extract key features from remote sensing imagery and maintain strong model stability and generalization under optically complex or heterogeneous water conditions (Aimin et al., 2025; Pu et al., 2019). However, remote sensing remains unsuitable for many small inland waters. The limitations are evident and include, but are not limited to: (1) Satellite-based remote sensing is highly dependent on atmospheric conditions at the time of observation. Cloud cover, haze conditions, and terrain shadows often render images unusable or substantially reduce the signal to noise ratio, with cloud obstruction being one of the dominant atmospheric disturbances that markedly limits data availability and stability (Tan et al., 2022; Wu et al., 2025b; Overstreet and Legleiter, 2017); (2) Remote sensing is also constrained by insufficient spatial and temporal resolution. Coarse spatial resolution limits the number of valid water pixels in small inland waters, while low revisit frequency prevents the capture of rapid hydrological or optical variations, thereby increasing uncertainty in turbidity estimation (Coffer et al., 2020; Shi et al., 2019). In recent years, smartphone-based imaging has been increasingly applied in turbidity monitoring. The HydroColor smartphone application is used to measure above-water reflectance and support portable turbidity estimation, and has been validated across multiple aquatic environments (Leeuw and Boss, 2018). Images acquired by smartphones are used to develop Bayesian models for turbidity prediction and uncertainty quantification (Huang et al., 2021). Fixed surveillance cameras combined with deep learning models are used for long-term, high-frequency monitoring in rivers and glacial-lake systems by capturing temporal variations in water optical characteristic (Lu et al., 2025; Zhou et al., 2024). However, camera-based approaches are highly sensitive to illumination. Light variation and shadow interference can substantially alter the radiometric properties of images, reducing the stability of image-based turbidity retrieval (Liu et al., 2018; Zhou et al., 2024). To address this issue, we present a smartphone-image framework for turbidity estimation as a rapid and cost-effective alternative to conventional turbidimeters. The framework introduces an easy-to-place three-color background to support image correction under variable illumination. Compared with previous image-based turbidity studies, our approach reduces the sensitivity of turbidity estimates to changing lighting conditions. We collected smartphone images and paired turbidity measurements from diverse small inland waters across China to evaluate the framework’s performance across different water types and field settings. 2. Material and methods 2.1. Study area and datasets From March 6 to December 27, 2025, we implemented a nationwide sampling plan across multiple regions of China. The sampling sites spanned a broad geographic range and included both urban and rural settings. The sampling site locations and the turbidity range are shown in Fig. 1 . Samples were collected from diverse inland waters, including rivers, lakes, and ponds, yielding 653 paired records. Each record contains a smartphone image and the corresponding turbidity measurement. The following procedures were used to collect paired samples of images and turbidity measurements. (1) Pour each water sample into a bucket and place a three-color background next to the bucket. (2) Capture an RGB image of the water surface using a smartphone. (3) Turbidity was measured using a YSI EXO2 portable water-quality analyzer. According to the above sampling procedures, we adopted more detailed designs to ensure that the collected data were accurate and reliable for evaluate ng the framework. (1) Setting the background, angle, and depth for smartphone imaging The three-color background was placed next to the bucket to provide similar ambient light for both the background and the water surface. The smartphone was held stable at a consistent viewing angle for all samples. The water depth in the bucket was fixed at 15 cm to minimize depth-related effects on the images. (2) Diverse Sampling Settings for Model Generalization Water samples were collected from various small freshwater environments (e.g., rivers, lakes, and ponds) to capture a wide range of conditions. Images were taken using different smartphones (e.g., Huawei and iPhone) under various weather conditions (e.g., sunny and cloudy). These diverse sampling settings aimed to improve the model's generalizability. Detailed records of the sampling locations, dates, weather conditions, smartphone models, and waterbody types are provided in Fig. S1 of the Supplementary material. Based on the above procedures, two datasets were compiled. The water-quality dataset consisted of turbidity measurements from 653 water samples collected from various types of small waterbodies across different regions. The image dataset consisted of the corresponding raw smartphone images for these 653 samples (Fig S2 in the Supplementary material). To enrich image features and improve model generalization, the raw images were further augmented by cropping, rotation, and flipping before being used for model training and evaluation. 2.2. A novel framework for smartphone to predict water turbidity To address the impact of illumination variations on turbidity prediction, this study developed a novel framework that that incorporates light correction to adjust the RGB values of smartphone images. In the developed framework, a smartphone image is acquired with the water surface and the three-color background in the same view. For each image, we obtained the RGB values of the three channels and performed light correction based on these values. At the same time, the water region is segmented and shadow-affected areas are removed to obtain a shadow-free water surface image. The framework aims to minimize the influence of variable lighting conditions on smartphone turbidity prediction and provides a universal preprocessing method applicable to smartphone monitoring (Fig. 2 ). The framework consists of three main steps: (1) RGB identification and light correction, (2) Extraction of shadow-free water surface, and (3) Turbidity prediction. The critical procedures of the framework were described as follows: (1) RGB identification and light correction The background region was identified using a threshold segmentation approach to distinguish it from the surroundings. The region containing the three-color background was selected (Fig. 3 a) to reduce interference from the environment and improve color reference recognition accuracy. The image was then converted from RGB to HSV to separate chromatic and brightness information. Based on the average brightness value of the V channel, images were classified into good-light and low-light conditions. Different thresholds were applied for background recognition under each lighting condition (Table 1 ). For low-light images, wider HSV ranges were used for the blue and green regions to ensure stable background detection (Fig. 3 d). $$\:k=\frac{\stackrel{-}{{R}_{r}}}{R}\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:（1）$$ where k represents the factor. R r represents the average value of the red channel in the image, and R represents the standard value of the red channel (255). The same applies to the blue and green channels. Images captured under good-light and low-light conditions but with similar turbidity showed clear color inconsistency before correction (Fig. 4 a–b). After light correction, the images appeared more consistent across different illumination conditions. Correction performance was evaluated using the CIEDE2000 color difference metric (ΔE₀₀), a standard measure of objective color consistency (Sharma et al., 2005). A clear reduction in ΔE₀₀ after correction indicates improved color consistency and reduced lighting-induced bias (Fig. 4 c). In a few cases, ΔE₀₀ decreased less due to strong, spatially non-uniform shadows that cannot be fully compensated by a global linear correction. (2) Extraction of shadow-free water surface Shadows can change pixel brightness and introduce bias in turbidity prediction. Therefore, shadow-free water regions were extracted as a preprocessing step. A pretrained Mask R-CNN model was used as an instance-segmentation tool within the MMDetection framework to obtain water-surface masks (Fig. 5 ). A confidence threshold of 0.7 was applied. The mask with the highest confidence score was selected for each image. The maximum inscribed rectangle of this mask was then used as the final shadow-free water region for subsequent analysis. All masks were visually checked to avoid obvious errors. For more details on the Mask R-CNN architecture, please refer to the official MMDetection repository ( https://github.com/open-mmlab/mmdetection ). (3) Turbidity prediction Turbidity prediction was performed on corrected shadow-free water surface images using a convolutional neural network (CNN) (Qi et al., 2024). The network was implemented in TensorFlow. It consists of two convolutional layers (32 and 16 kernels), one max-pooling layer, and two fully connected layers (256 and 128 neurons). The network outputs a single turbidity value (Fig. 6 ). To achieve efficient data loading and standardization, a custom data generator was designed to load images and their corresponding turbidity values in batches. Normalization was applied to both the RGB pixels and the turbidity data. During training, 70% of the samples were used for training and 30% for validation. Early stopping and learning rate decay strategies were employed to ensure stable convergence and prevent overfitting. The development of CNN regression model was implemented in PyCharm Community Edition and TensorFlow ( https://www.TensorFlow.org ). 3. Framework application in small water bodies 3.1 Water turbidity measurement and prediction Small water bodies, such as ponds, rivers, and urban drainage channels, are vital components of local hydrological and ecological systems. They play an important role in regulating surface runoff, retaining nutrients, and supporting biodiversity (Pi et al., 2022; Zeng et al., 2019). In this study, the proposed light correction and turbidity prediction framework was applied to the Huajin River, a representative small water body located within Anhui Normal University in Wuhu, China. The river is a typical urban tributary with shallow water depth, limited flow, and strong temporal variations in water quality. It spans approximately 1.2 km in length, with a width ranging from 25 to 50 m, covering an area of 0.07 km² (Fig. 7a). The river is strongly influenced by human activities, such as drainage outfalls, waterfowl rearing, and boating events, which often result in high turbidity and frequent harmful algal blooms. Traditional remote sensing technology had difficulty in capturing this river for its small size. The Huajin River was therefore used as a field case to demonstrate the applicability of the proposed smartphone-based framework in a small water body. Four monitoring sites were arranged along the river. Smartphone images were collected at each site and processed using the full workflow, including three-color background calibration, shadow-free water extraction, and CNN-based turbidity prediction. Measured turbidity was obtained using a portable water quality analyzer, the YSI EXO2 designed by Xylem Analytics (USA), and used as the reference for evaluation. We captured 16 smartphone images along the river under different lighting conditions and fed them into the trained model for turbidity estimation. The results show that the corrected images have much higher accuracy than the uncorrected ones. This demonstrates the potential of the framework for turbidity monitoring in small waterbodies. Importantly, the framework helps improve monitoring consistency across lighting conditions, supporting better decision-making and efficient water management. 3.2 Performance evaluation The performance of turbidity prediction before and after light correction was evaluated using measured data from the Huajin River. The correlation between the measured and predicted turbidity improved significantly after applying the light correction. Before light correction, the prediction model achieved an R² of 0.49, root mean square error (RMSE) of 5.19, mean absolute error (MAE) of 4.04, and Nash–Sutcliffe efficiency (NS) of 0.49. After light correction, R² increased to 0.68, while RMSE and MAE decreased to 4.13 and 2.52, respectively, and NS improved to 0.68 (Fig. 8ab). This demonstrates that light correction effectively improves the stability and accuracy of turbidity estimation by eliminating the influence of lighting variations on RGB features. Figures 8 c and 8 d further illustrate the relationship between image brightness and turbidity prediction under varying light conditions. Before light correction, predicted turbidity values fluctuated strongly with image brightness. After light correction, predicted turbidity values were more consistent with measured data, and the influence of brightness variation was greatly reduced. These results confirm that the proposed light correction effectively minimizes illumination-induced bias, ensuring robust and reliable turbidity prediction in variable-light outdoor environments. The framework’s performance varied across turbidity ranges, with more accurate predictions observed in the 10–20 NTU range. This may be due to low light and high turbidity, which make images appear dark and lead to mispredictions. 4. Discussion 4.1 Comparison with existing methods Compared with previous smartphone-based turbidity estimation methods (Trejo-Zúñiga et al., 2024; Ceylan Koydemir et al., 2019; Zhou et al., 2024), this study achieved reliable results (Table S1 in the Supplementary material). This framework uses light correction for RGB adjustment and effectively reduces the impact of changing illumination. The accuracy of this framework is slightly lower than some fixed-camera monitoring approaches (Lu et al., 2025; Zhou et al., 2024), because our dataset includes images under diverse illumination, while those studies often exclude or do not evaluate low-light images. Turbidity monitoring has traditionally relied on satellite remote sensing. Table S1 shows that our performance is comparable to most previously published remote-sensing studies (Cui et al., 2022; Magrì et al., 2023; Kong et al., 2025). Some satellite-based estimates may still show slightly higher accuracy than our smartphone results (Feng et al., 2020; Hossain et al., 2021). This is expected because satellites capture richer spectral information, while smartphone images have fewer bands and contain unavoidable platform-related noise. Importantly, our approach provides much finer spatial resolution, down to the millimeter scale. This improvement is valuable but makes validation harder because turbidity can vary over very short distances. In addition, our field dataset for validation includes more samples than what is commonly used in many remote-sensing studies, which supports a robust comparison. 4.2 Framework Advantages The design and application of the developed framework in our case study demonstrated its high potential in estimating turbidity described as follows. 4.2.1 Well-suited to use in measuring turbidity in small waterbodies Remote sensing images face limitations when monitoring small water bodies, such as ponds, rivers, and urban drainage channels, where their effectiveness is often compromised. For example, the Huajin River in this study cannot be clearly observed in remote sensing images (Fig. 7c). Traditional remote sensing methods often fail to meet the real-time monitoring requirements for small water bodies and are susceptible to factors like spatial resolution and cloud cover. In contrast, the smartphone-based framework offers clear advantages, particularly for monitoring small water bodies. The framework provides an efficient and cost-effective method to measure turbidity with higher accuracy, overcoming the limitations of remote sensing in small water bodies. It is particularly suitable for complex environments, such as urban and rural areas, where real-time water quality monitoring is needed. 4.2.2 Significantly reduced the impact of different lighting conditions Different lighting conditions often change image colors, which can directly affect the accuracy of smartphone-based turbidity predictions. By applying RGB identification and light correction, the framework reduces color variations under different lighting conditions and improves the accuracy and stability of turbidity predictions. Furthermore, by extracting shadow-free water surfaces, the framework removes interference from shadowed areas and ensures accurate water surface information. As a result, the framework demonstrates stronger adaptability under varying lighting conditions, providing stable turbidity predictions with significant practical value. 4.2.3 A universal data preprocessing method The framework provides a universal data preprocessing method applicable to a wide range of image data for turbidity prediction. Through the image processing pipeline, the method provides stable predictions across different water bodies, environmental conditions, and imaging devices. The framework handles diverse input data and adapts to various monitoring scenarios, making it a valuable tool for water quality monitoring. This universal applicability allows seamless integration into various real-world applications, from large-scale environmental monitoring to localized assessments in resource-limited settings. 4.3 Potential applications in water management Integrating the framework into a smartphone application is a promising approach for water quality prediction. Given the widespread use of smartphones, they are ideal tools for water quality prediction (Ceylan Koydemir et al., 2019; Leeuw and Boss, 2018). Integrating the framework into a smartphone application provides an efficient and cost-effective alternative to conventional turbidity meters for predicting water turbidity. In addition to portable smartphone-based turbidity prediction, the framework can also be applied to monitoring cameras for long-term, real-time monitoring of the study area. Small waterbodies are highly susceptible to external influences due to their limited size, mobility, and self-purification capacity. They can undergo rapid changes over short periods. Therefore, real-time turbidity monitoring in small waterbodies enables timely assessment of water quality and supports pollution prevention and control. 4.4 Uncertainty in the framework Although the developed framework possesses several significant advantages as described in Section 4.1 , it is important to note that uncertainties remain. Some aspects require further improvement in future work. (1) Sensitivity of light correction to field illumination variability Light correction depends on stable illumination during imaging. Direct sunlight, glare, and strong shadows can affect the extracted RGB values. This may weaken global correction and introduce bias into turbidity prediction. This sensitivity can be reduced by preferring stable lighting conditions during imaging and by avoiding reflective surfaces. In future work, background localization could be improved by replacing threshold segmentation with deep learning–based object detection, followed by RGB extraction within the detected region. (2) Dependence of model robustness on training data representativeness Model robustness depends on the representativeness of the training data. More paired records of images and turbidity measurements are needed across different seasons, weather conditions, and small waterbodies. The dataset should also cover a wider range of turbidity levels and illumination conditions. Broader sampling would help reduce uncertainty and improve generalization. 5. Conclusion This study proposes a smartphone-based framework that reduces the influence of illumination on turbidity estimation. The framework combines light correction for RGB adjustment, shadow-free water extraction, and CNN regression. Field results based on 653 paired image–turbidity samples show that the method improves prediction stability under diverse outdoor light. The Huajin River case study further shows that light correction improves agreement with in situ turbidity compared with uncorrected images. Overall, the framework is low-cost, scalable, and suitable for small waterbodies that require dense spatial sampling. Future work will focus on expanding the dataset to cover more seasons, water types, and extreme lighting, and on improving background recognition to enhance robustness in complex field conditions. Declarations Funding The present study was supported by the National Natural Science Foundation of China (42001087), the Water Resources Science and Technology Program of Jiangxi, China (202426ZDKT09), and the Natural Science Research Project of Anhui Educational Committee (2023AH050151). Acknowledgments The authors would like to express their special thanks to Liuyi Dai, Zhengxin Wang, and Xudong Wang (Anhui Normal University, China) for their great support in water sampling and measurement. Author Contribution Jiacong Huang and Lingyan Qi provided the research concept and methodological guidance. Kejia Zhang wrote the main manuscript text. Jiacong Huang and Lingyan Qi reviewed and revised the manuscript. Kejia Zhang, Mingzhu Guo, Xinzhe Jiang, and Weiwei Zhao prepared Figures 1–8 and Table 1. Jiefa Cai, Yaqin Duan, and Jiacong Huang conducted field surveys and prepared Figures S1–S2 and Table S1. All authors reviewed and approved the manuscript. Acknowledgement Special thanks to Liuyi Dai, Zhengxin Wang, and Xudong Wang (Anhui Normal University, China) for their solid support in water sampling and measurement. Data Availability All raw and processed datasets, including 653 paired smartphone images and corresponding turbidity measurements, as well as supporting figures and tables (Figures S1–S2, Table S1), are available in the supplementary material. Additional data and scripts used for analysis are available from the corresponding author upon reasonable request. References Aimin, L. I., Zheng, Y., Zhenqiang, G. U. O., Ziyi, C., 2025. A Remote Sensing Retrieval Method For River Water Turbidity Based on a Convolutional Neural Network. Journal of Geo-information Science 27(6), 1305–1316. https://doi.org/10.12082/dqxxkx.2025.240613 Caroni, R., Greife, A. J., Bresciani, M., Giardino, C., Tellina, G., Carrea, L., Liu, X., Simis, S., Albergel, C., Pinardi, M., 2025. Examining global trends of satellite-derived water quality variables in shallow lakes. Remote Sensing Applications: Society and Environment 38. https://doi.org/10.1016/j.rsase.2025.101565 Ceylan Koydemir, H., Rajpal, S., Gumustekin, E., Karinca, D., Liang, K., Gorocs, Z., Tseng, D., Ozcan, A., 2019. Smartphone-based turbidity reader. Sci Rep 9(1), 19901. https://doi.org/10.1038/s41598-019-56474-z Chen, B., Chen, Y., Chen, H., 2024. An Interpretable CatBoost Model Guided by Spectral Morphological Features for the Inversion of Coastal Water Quality Parameters. Water 16(24). https://doi.org/10.3390/w16243615 Chen, J., Chen, S., Fu, R., Li, D., Jiang, H., Wang, C., Peng, Y., Jia, K., Hicks, B. J., 2022. Remote Sensing Big Data for Water Environment Monitoring: Current Status, Challenges, and Future Prospects. Earth's Future 10(2). https://doi.org/10.1029/2021ef002289 Chowdhury, M., de la Calle, I., Laiz, I., Ruescas, A. B., 2025. Near-Real-Time Turbidity Monitoring at Global Scale Using Sentinel-2 Data and Machine Learning Techniques. Remote Sensing 17(22). https://doi.org/10.3390/rs17223716 Coffer, M. M., Schaeffer, B. A., Darling, J. A., Urquhart, E. A., Salls, W. B., 2020. Quantifying national and regional cyanobacterial occurrence in US lakes using satellite remote sensing. Ecol Indic 111, 105976. https://doi.org/10.1016/j.ecolind.2019.105976 Cui, M., Sun, Y., Huang, C., Li, M., 2022. Water Turbidity Retrieval Based on UAV Hyperspectral Remote Sensing. Water 14(1). https://doi.org/10.3390/w14010128 Deng, Y., Zhang, Y., Pan, D., Yang, S. X., Gharabaghi, B., 2024. Review of Recent Advances in Remote Sensing and Machine Learning Methods for Lake Water Quality Management. Remote Sensing 16(22). https://doi.org/10.3390/rs16224196 Ding, W., Zhao, J., Qin, B., Wu, T., Zhu, S., Li, Y., Xu, S., Ruan, S., Wang, Y., 2021. Exploring and quantifying the relationship between instantaneous wind speed and turbidity in a large shallow lake: case study of Lake Taihu in China. Environmental Science and Pollution Research 28(13), 16616–16632. https://doi.org/10.1007/s11356-020-11544-y Feng, J., Chen, H., Zhang, H., Li, Z., Yu, Y., Zhang, Y., Bilal, M., Qiu, Z., 2020. Turbidity Estimation from GOCI Satellite Data in the Turbid Estuaries of China’s Coast. Remote Sensing 12(22). https://doi.org/10.3390/rs12223770 Gholizadeh, M. H., Melesse, A. M., Reddi, L., 2016. A Comprehensive Review on Water Quality Parameters Estimation Using Remote Sensing Techniques. Sensors (Basel) 16(8). https://doi.org/10.3390/s16081298 Hossain, A. K. M. A., Mathias, C., Blanton, R., 2021. Remote Sensing of Turbidity in the Tennessee River Using Landsat 8 Satellite. Remote Sensing 13(18). https://doi.org/10.3390/rs13183785 Huang, J., Qian, R., Gao, J., Bing, H., Huang, Q., Qi, L., Song, S., Huang, J., 2021. A novel framework to predict water turbidity using Bayesian modeling. Water Res 202, 117406. https://doi.org/10.1016/j.watres.2021.117406 Kong, Y., Jimenez, K., Lee, C. M., Winter, S., Summers-Evans, J., Cao, A., Menczer, M., Han, R., Mills, C., McCarthy, S., Blatzheim, K., Jay, J. A., 2025. Monitoring Coastal Water Turbidity Using Sentinel2—A Case Study in Los Angeles. Remote Sensing 17(2). https://doi.org/10.3390/rs17020201 Leeuw, T., Boss, E., 2018. The HydroColor App: Above Water Measurements of Remote Sensing Reflectance and Turbidity Using a Smartphone Camera. Sensors (Basel) 18(1). https://doi.org/10.3390/s18010256 Li, J., Xia, C., 2023. Drivers of Spatial and Temporal Dynamics in Water Turbidity of China Yangtze River Basin. Water 15(7). https://doi.org/10.3390/w15071264 Li, S., Kutser, T., Song, K., Liu, G., Li, Y., 2023a. Lake Turbidity Mapping Using an OWTs-bp Based Framework and Sentinel-2 Imagery. Remote Sensing 15(10). https://doi.org/10.3390/rs15102489 Li, Y., Li, S., Song, K., Liu, G., Wen, Z., Fang, C., Shang, Y., Lyu, L., Zhang, L., 2023b. Sentinel-3 OLCI observations of Chinese lake turbidity using machine learning algorithms. Journal of Hydrology 622. https://doi.org/10.1016/j.jhydrol.2023.129668 Lin, X., Wu, M., Shao, X., Li, G., Hong, Y., 2023. Water turbidity dynamics using random forest in the Yangtze River Delta Region, China. Sci Total Environ 903, 166511. https://doi.org/10.1016/j.scitotenv.2023.166511 Liu, Y., Chen, Y., Fang, X., 2018. A Review of Turbidity Detection Based on Computer Vision. IEEE Access 6, 60586–60604. https://doi.org/10.1109/access.2018.2875071 Lu, Z., Zhu, X., Li, J., Li, M., Wang, J., Wang, W., Zheng, Y., Zhang, Q., 2025. Detecting glacial lake water quality indicators from RGB surveillance images via deep learning. International Journal of Applied Earth Observation and Geoinformation 136. https://doi.org/10.1016/j.jag.2025.104392 Ma, Y., Song, K., Wen, Z., Liu, G., Shang, Y., Lyu, L., Du, J., Yang, Q., Li, S., Tao, H., Hou, J., 2021. Remote Sensing of Turbidity for Lakes in Northeast China Using Sentinel-2 Images With Machine Learning Algorithms. IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing 14, 9132–9146. https://doi.org/10.1109/jstars.2021.3109292 Magrì, S., Ottaviani, E., Prampolini, E., Besio, G., Fabiano, B., Federici, B., 2023. Application of machine learning techniques to derive sea water turbidity from Sentinel-2 imagery. Remote Sensing Applications: Society and Environment 30. https://doi.org/10.1016/j.rsase.2023.100951 Odermatt, D., Gitelson, A., Brando, V. E., Schaepman, M., 2012. Review of constituent retrieval in optically deep and complex waters from satellite imagery. Remote Sensing of Environment 118, 116–126. https://doi.org/10.1016/j.rse.2011.11.013 Overstreet, B. T., Legleiter, C. J., 2017. Removing sun glint from optical remote sensing images of shallow rivers. Earth Surface Processes and Landforms 42(2), 318–333. https://doi.org/https://doi.org/10.1002/esp.4063 Palmer, S. C. J., Kutser, T., Hunter, P. D., 2015. Remote sensing of inland waters: Challenges, progress and future directions. Remote Sensing of Environment 157, 1–8. https://doi.org/10.1016/j.rse.2014.09.021 Pang, Z., Zhou, Z., Fu, J. e., Jiang, W., Qin, X., Sun, M., 2025. Deep learning-based remote sensing retrieval of inland water quality: A review. Journal of Hydrology: Regional Studies 61. https://doi.org/10.1016/j.ejrh.2025.102759 Pi, X., Luo, Q., Feng, L., Xu, Y., Tang, J., Liang, X., Ma, E., Cheng, R., Fensholt, R., Brandt, M., Cai, X., Gibson, L., Liu, J., Zheng, C., Li, W., Bryan, B. A., 2022. Mapping global lake dynamics reveals the emerging roles of small lakes. Nature Communications 13(1). https://doi.org/10.1038/s41467-022-33239-3 Pu, F., Ding, C., Chao, Z., Yu, Y., Xu, X., 2019. Water-Quality Classification of Inland Lakes Using Landsat8 Images by Convolutional Neural Networks. Remote Sensing 11(14). https://doi.org/10.3390/rs11141674 Qi, L., Yin, H., Wang, Z., Ye, L., Zhang, S., Dai, L., Wu, F., Jiang, X., Huang, Q., Huang, J., 2024. Smartphone as an alternative to measure chlorophyll-a concentration in small waterbodies. J Environ Manage 368, 122135. https://doi.org/10.1016/j.jenvman.2024.122135 Qiu, Z., Liu, D., Yan, N., Yan, Y., Yang, C., Zhang, C., Duan, H., 2025. Landsat and dual random forest modelling reveal sediment fining in the Yellow River shaped by ecological restoration on China's loess plateau. Remote Sensing of Environment 330. https://doi.org/10.1016/j.rse.2025.114994 Sahoo, D., Anandhi, A., 2023. Conceptualizing turbidity for aquatic ecosystems in the context of sustainable development goals. Environmental Science: Advances 2(9), 1220–1234. https://doi.org/10.1039/d2va00327a Santos, V. O., Rocha, P. A. C., Thé, J. V. G., Gharabaghi, B., 2025. Evaluation of machine learning methods for forecasting turbidity in river networks using Sentinel-2 remote sensing data. Ecological Informatics 90. https://doi.org/10.1016/j.ecoinf.2025.103313 Sehgal, D., Martínez‐Carreras, N., Hissler, C., Bense, V. F., Hoitink, A. J. F., 2022. A Generic Relation Between Turbidity, Suspended Particulate Matter Concentration, and Sediment Characteristics. Journal of Geophysical Research: Earth Surface 127(12). https://doi.org/10.1029/2022jf006838 Sharma, G., Wu, W., Dalal, E. N., 2005. The CIEDE2000 color-difference formula: Implementation notes, supplementary test data, and mathematical observations. Color Research & Application 30(1), 21–30. https://doi.org/https://doi.org/10.1002/col.20070 Shi, K., Zhang, Y., Qin, B., Zhou, B., 2019. Remote sensing of cyanobacterial blooms in inland waters: present knowledge and future challenges. Sci Bull (Beijing) 64(20), 1540–1556. https://doi.org/10.1016/j.scib.2019.07.002 Tan, Z., Gao, M., Yuan, J., Jiang, L., Duan, H., 2022. A Robust Model for MODIS and Landsat Image Fusion Considering Input Noise. IEEE Transactions on Geoscience and Remote Sensing 60, 1–17. https://doi.org/10.1109/tgrs.2022.3145086 Trejo-Zúñiga, I., Moreno, M., Santana-Cruz, R. F., Meléndez-Vázquez, F., 2024. Deep-Learning-Driven Turbidity Level Classification. Big Data and Cognitive Computing 8(8). https://doi.org/10.3390/bdcc8080089 Wu, D., Tang, T., Odermatt, D., Liu, W., 2025a. Spatiotemporal variability in global lakes turbidity derived from satellite imageries. Environmental Research Communications 7(3). https://doi.org/10.1088/2515-7620/adb941 Wu, Z., Pang, J., Li, J., Wang, Y., Ruan, J., Zhang, X., Yang, L., Pang, Y., Gao, Y., 2025b. A review of remote sensing-based water quality monitoring in turbid coastal waters. Intelligent Marine Technology and Systems 3(1). https://doi.org/10.1007/s44295-025-00075-2 Yan, N., Qiu, Z., Zhang, C., Liu, J., Liu, D., 2025. Observing water turbidity in Chinese rivers using Landsat series data over the past 40 years. Journal of Cleaner Production 494. https://doi.org/10.1016/j.jclepro.2025.145001 Yang, Z., Gong, C., Lu, Z., Wu, E., Huai, H., Hu, Y., Li, L., Dong, L., 2023. Combined Retrievals of Turbidity from Sentinel-2A/B and Landsat-8/9 in the Taihu Lake through Machine Learning. Remote Sensing 15(17). https://doi.org/10.3390/rs15174333 Zeng, Z., Wang, D., Tan, W., Huang, J., 2019. Extracting aquaculture ponds from natural water surfaces around inland lakes on medium resolution multispectral images. International Journal of Applied Earth Observation and Geoinformation 80, 13–25. https://doi.org/10.1016/j.jag.2019.03.019 Zhang, Y., Yao, X., Wu, Q., Huang, Y., Zhou, Z., Yang, J., Liu, X., 2021. Turbidity prediction of lake-type raw water using random forest model based on meteorological data: A case study of Tai lake, China. J Environ Manage 290, 112657. https://doi.org/10.1016/j.jenvman.2021.112657 Zhou, N., Chen, H., Chen, N., Liu, B., 2024. An Approach Based on Video Image Intelligent Recognition for Water Turbidity Monitoring in River. ACS ES&T Water 4(2), 543–554. https://doi.org/10.1021/acsestwater.3c00596 Table Table 1. HSV threshold range condition color H range S range V range Good light Red [0–12], [58-180] 40–255 90–255 Green 40–85 40–255 90–255 Blue 100–132 45–255 80–255 Low light Red [0–12], [158–180] 35–255 60–255 Green 25–90 35–255 30–255 Blue 96–138 30–255 35–255 Additional Declarations No competing interests reported. Supplementary Files Supplementarymaterial.docx Cite Share Download PDF Status: Under Review Version 1 posted Reviewers agreed at journal 29 Apr, 2026 Reviewers invited by journal 29 Apr, 2026 Editor assigned by journal 21 Apr, 2026 Submission checks completed at journal 21 Apr, 2026 First submitted to journal 03 Apr, 2026 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-9317150","acceptedTermsAndConditions":true,"allowDirectSubmit":false,"archivedVersions":[],"articleType":"Research Article","associatedPublications":[],"authors":[{"id":634146261,"identity":"f83297fe-c71d-48af-9171-3d87ca7fc5ce","order_by":0,"name":"Lingyan Qi","email":"","orcid":"","institution":"Anhui Normal University","correspondingAuthor":false,"prefix":"","firstName":"Lingyan","middleName":"","lastName":"Qi","suffix":""},{"id":634146263,"identity":"00cdc295-5d97-4eaf-9f4d-e7dba68175e0","order_by":1,"name":"Kejia Zhang","email":"","orcid":"","institution":"Anhui Normal University","correspondingAuthor":false,"prefix":"","firstName":"Kejia","middleName":"","lastName":"Zhang","suffix":""},{"id":634146266,"identity":"bd6c38fd-f54a-4ab8-b466-b140a4afacfc","order_by":2,"name":"Mingzhu Guo","email":"","orcid":"","institution":"Anhui Normal University","correspondingAuthor":false,"prefix":"","firstName":"Mingzhu","middleName":"","lastName":"Guo","suffix":""},{"id":634146268,"identity":"eae78e8d-4688-4fa0-a954-d5ccceaa5867","order_by":3,"name":"Xinzhe Jiang","email":"","orcid":"","institution":"Anhui Normal University","correspondingAuthor":false,"prefix":"","firstName":"Xinzhe","middleName":"","lastName":"Jiang","suffix":""},{"id":634146270,"identity":"a0a5d868-5259-4fa1-b262-efbfc0fd49d0","order_by":4,"name":"Weiwei Zhao","email":"","orcid":"","institution":"Anhui Normal University","correspondingAuthor":false,"prefix":"","firstName":"Weiwei","middleName":"","lastName":"Zhao","suffix":""},{"id":634146283,"identity":"48ec6226-c1ab-447e-9df3-85d3ec925203","order_by":5,"name":"Jiefa Cai","email":"","orcid":"","institution":"Anhui Normal University","correspondingAuthor":false,"prefix":"","firstName":"Jiefa","middleName":"","lastName":"Cai","suffix":""},{"id":634146285,"identity":"c5e2b69d-81d5-4ec2-95d5-b15cd4f723c8","order_by":6,"name":"Yaqin Duan","email":"","orcid":"","institution":"Anhui Normal University","correspondingAuthor":false,"prefix":"","firstName":"Yaqin","middleName":"","lastName":"Duan","suffix":""},{"id":634146287,"identity":"7495fb76-718f-499b-9092-2bc8de8a02b1","order_by":7,"name":"Jiacong Huang","email":"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAZAAAAAyAQMAAABI0h/eAAAABlBMVEX///8AAABVwtN+AAAACXBIWXMAAA7EAAAOxAGVKw4bAAAA+klEQVRIie3RMUsDMRTA8YQHueW155jA4X2FB10civWjKEInB8cOglcCdVG6tjj4FTo55wiki9j1wEXwC6R7B3PtaM3h5pD/nB95L2EslfqnGT+5O82BV0bSEPOs6jgvGK8Xb26gHmBqzm7HhXo0nQRsbwZXq02mzcTbITUXcVG+PH2GWwRf2TBYQxtkDeN+e/M7IZdR2KUAFUi9oA/kzxWo5WuECMHaW0Q/ECsDgcII6EVIOROs3QVZS3b0jkJexglzByJP7P6RDWIXITduB3Ok9J5co8RaR3cptQMfvvJ+nq+/vNydj0ZrXfttbLAj8epv51OpVCr1o2+NX1knzXKmmgAAAABJRU5ErkJggg==","orcid":"","institution":"Nanjing Institute of Geography and Limnology","correspondingAuthor":true,"prefix":"","firstName":"Jiacong","middleName":"","lastName":"Huang","suffix":""}],"badges":[],"createdAt":"2026-04-04 03:23:19","currentVersionCode":1,"declarations":"","doi":"10.21203/rs.3.rs-9317150/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-9317150/v1","draftVersion":[],"editorialEvents":[],"editorialNote":"","failedWorkflow":false,"files":[{"id":108713646,"identity":"9e5d9306-4932-45de-8d17-f7adfd895d9c","added_by":"auto","created_at":"2026-05-07 14:41:59","extension":"png","order_by":1,"title":"Figure 1","display":"","copyAsset":false,"role":"figure","size":279194,"visible":true,"origin":"","legend":"\u003cp\u003eLocation of 653 sampling sites (a) and its distribution (b). Each record included an image (c) and measured NTU (d).\u003c/p\u003e","description":"","filename":"image1.png","url":"https://assets-eu.researchsquare.com/files/rs-9317150/v1/e6de6633eeb365b85239bf62.png"},{"id":108713689,"identity":"cd44298f-37cd-4090-bd57-96e6688ec260","added_by":"auto","created_at":"2026-05-07 14:42:09","extension":"png","order_by":2,"title":"Figure 2","display":"","copyAsset":false,"role":"figure","size":293260,"visible":true,"origin":"","legend":"\u003cp\u003eFramework schematic illustration including RGB identification and light correction (step I), Extraction of shadow-free water surface (step II), and Turbidity prediction (step III).\u003c/p\u003e","description":"","filename":"image2.png","url":"https://assets-eu.researchsquare.com/files/rs-9317150/v1/b11ce9491a5e37578d25e657.png"},{"id":108713688,"identity":"3833d817-7446-4c4a-82ca-5856a7d07da8","added_by":"auto","created_at":"2026-05-07 14:42:09","extension":"png","order_by":3,"title":"Figure 3","display":"","copyAsset":false,"role":"figure","size":501229,"visible":true,"origin":"","legend":"\u003cp\u003eRGB identification and light correction schematic diagram\u003c/p\u003e","description":"","filename":"image3.png","url":"https://assets-eu.researchsquare.com/files/rs-9317150/v1/cec8073a9f22472cf78c5369.png"},{"id":108713673,"identity":"ecc31b75-a054-491b-bdc8-a7cc30d4b147","added_by":"auto","created_at":"2026-05-07 14:42:02","extension":"jpeg","order_by":4,"title":"Figure 4","display":"","copyAsset":false,"role":"figure","size":437557,"visible":true,"origin":"","legend":"\u003cp\u003eRaw image and corrected image under different lighting conditions (a, b). Effectiveness of image correction under different lighting conditions (c).\u003c/p\u003e","description":"","filename":"image4.jpeg","url":"https://assets-eu.researchsquare.com/files/rs-9317150/v1/6453e95579bd6380dd4d73ae.jpeg"},{"id":108713723,"identity":"35924ab5-0cf3-4911-aa77-e6cfaaee5c20","added_by":"auto","created_at":"2026-05-07 14:42:17","extension":"jpeg","order_by":5,"title":"Figure 5","display":"","copyAsset":false,"role":"figure","size":339970,"visible":true,"origin":"","legend":"\u003cp\u003eExtraction of shadow-free water surface\u003c/p\u003e","description":"","filename":"image5.jpeg","url":"https://assets-eu.researchsquare.com/files/rs-9317150/v1/3213510f39979f2d97088cff.jpeg"},{"id":108713685,"identity":"a7fd55dd-8293-45fd-92f8-245ccfb6b70f","added_by":"auto","created_at":"2026-05-07 14:42:09","extension":"png","order_by":6,"title":"Figure 6","display":"","copyAsset":false,"role":"figure","size":196327,"visible":true,"origin":"","legend":"\u003cp\u003eStructure and parameters of the CNN regression model\u003c/p\u003e","description":"","filename":"image6.png","url":"https://assets-eu.researchsquare.com/files/rs-9317150/v1/792d59ac372dd78b2a0fa098.png"},{"id":108713692,"identity":"ca79e405-4559-4a54-aca0-237d19869cff","added_by":"auto","created_at":"2026-05-07 14:42:10","extension":"png","order_by":7,"title":"Figure 7","display":"","copyAsset":false,"role":"figure","size":343277,"visible":true,"origin":"","legend":"\u003cp\u003eLocation of 4 sampling sites in Huajin River (a), smartphone-based images (b), and satellite images (c) from MODIS, Landsat , and Sentinel-2A\u003c/p\u003e","description":"","filename":"image7.png","url":"https://assets-eu.researchsquare.com/files/rs-9317150/v1/502557132a3a1727247d48b4.png"},{"id":108713722,"identity":"8d89d376-0186-489a-add6-f8ddc7bcc4c9","added_by":"auto","created_at":"2026-05-07 14:42:17","extension":"png","order_by":8,"title":"Figure 8","display":"","copyAsset":false,"role":"figure","size":725459,"visible":true,"origin":"","legend":"\u003cp\u003eThe relative frequency (RF, %) of estimated and measured turbidity, and their model fit metrics (NS, R2, RMSE, and MAE) with 563 data records.\u003c/p\u003e","description":"","filename":"image8.png","url":"https://assets-eu.researchsquare.com/files/rs-9317150/v1/9f0ba89d30f020adc40a8a81.png"},{"id":108713742,"identity":"fa1b87cf-5866-44df-8d96-1edb347fde51","added_by":"auto","created_at":"2026-05-07 14:42:23","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":3382499,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-9317150/v1/a2959e5d-26dc-4b4f-96ed-55ccce5168c4.pdf"},{"id":108713672,"identity":"bfdcb1c6-f787-4383-8623-019f51f02f7d","added_by":"auto","created_at":"2026-05-07 14:42:02","extension":"docx","order_by":0,"title":"","display":"","copyAsset":false,"role":"supplement","size":7230504,"visible":true,"origin":"","legend":"","description":"","filename":"Supplementarymaterial.docx","url":"https://assets-eu.researchsquare.com/files/rs-9317150/v1/b528819ea65a5c7e0904abbb.docx"}],"financialInterests":"No competing interests reported.","formattedTitle":"Significantly reducing the impact of light: a novel framework for smartphone to predict water turbidity","fulltext":[{"header":"1. Introduction","content":"\u003cp\u003eTurbidity is a key indicator of suspended particulate and optical properties (Li et al., 2023b; Ding et al., 2021). It influences aquatic ecosystems and regulates key biogeochemical processes (Sahoo and Anandhi, 2023; Sehgal et al., 2022; Wu et al., 2025b). Global observations also show strong turbidity contrasts among lakes (Caroni et al., 2025; Chowdhury et al., 2025), Across 774 lakes worldwide, 63.4% show low turbidity (\u0026le;\u0026thinsp;5 NTU), while 40% show significant decreases and 32% show significant increases over time (Wu et al., 2025a). In China, a 40-year Landsat analysis reported strong regional contrasts in river turbidity, with mean values of 32.5 NTU in the east and 22.8 NTU in the west, and long-term declines across 65.48% of river areas (Yan et al., 2025). In the Yangtze River Basin, mean turbidity is 60.1 NTU in river channels and 29.6 NTU in nearby lakes and reservoirs (Li and Xia, 2023), and long-term turbidity trends differ among water-body types with rates spanning\u0026thinsp;\u0026minus;\u0026thinsp;1.3 to +\u0026thinsp;0.7 NTU yr⁻\u0026sup1; (Lin et al., 2023). At the national lake scale, a China-wide assessment reported a wide turbidity range, from 0.01\u0026ndash;98.63 NTU in 2015 to 0.01-122.09 NTU in 2020, together with a shift in mean turbidity across years (Li et al., 2023a). High-frequency and low-cost turbidity measurement remains challenging in such highly variable waters.\u003c/p\u003e \u003cp\u003eRemote sensing is widely used to monitor surface-water quality (Palmer et al., 2015; Odermatt et al., 2012; Gholizadeh et al., 2016), and it is increasingly coupled with machine learning and deep learning (Pang et al., 2025; Deng et al., 2024; Chen et al., 2022). Random Forest has been applied to map turbidity dynamics in the Yangtze River Delta(Lin et al., 2023), to predict turbidity in Lake Taihu using meteorological drivers (Zhang et al., 2021), and to model sediment-related turbidity signals in the Yellow River (Qiu et al., 2025). XGBoost has been used to retrieve turbidity in Lake Taihu from Sentinel-2A/B and Landsat-8/9 (Yang et al., 2023), and to forecast turbidity in river networks using Sentinel-2 (Santos et al., 2025). Gradient boosting models have also been used for turbidity retrieval, including GBDT for lakes in Northeast China using Sentinel-2 (Ma et al., 2021), CatBoost for coastal-water quality inversion along the Fujian coast using Sentinel-3 OLCI (Chen et al., 2024).It is worth noting that convolutional neural networks offer distinct advantages, as CNNs can efficiently extract key features from remote sensing imagery and maintain strong model stability and generalization under optically complex or heterogeneous water conditions (Aimin et al., 2025; Pu et al., 2019). However, remote sensing remains unsuitable for many small inland waters. The limitations are evident and include, but are not limited to: (1) Satellite-based remote sensing is highly dependent on atmospheric conditions at the time of observation. Cloud cover, haze conditions, and terrain shadows often render images unusable or substantially reduce the signal to noise ratio, with cloud obstruction being one of the dominant atmospheric disturbances that markedly limits data availability and stability (Tan et al., 2022; Wu et al., 2025b; Overstreet and Legleiter, 2017); (2) Remote sensing is also constrained by insufficient spatial and temporal resolution. Coarse spatial resolution limits the number of valid water pixels in small inland waters, while low revisit frequency prevents the capture of rapid hydrological or optical variations, thereby increasing uncertainty in turbidity estimation (Coffer et al., 2020; Shi et al., 2019).\u003c/p\u003e \u003cp\u003eIn recent years, smartphone-based imaging has been increasingly applied in turbidity monitoring. The HydroColor smartphone application is used to measure above-water reflectance and support portable turbidity estimation, and has been validated across multiple aquatic environments (Leeuw and Boss, 2018). Images acquired by smartphones are used to develop Bayesian models for turbidity prediction and uncertainty quantification (Huang et al., 2021). Fixed surveillance cameras combined with deep learning models are used for long-term, high-frequency monitoring in rivers and glacial-lake systems by capturing temporal variations in water optical characteristic (Lu et al., 2025; Zhou et al., 2024). However, camera-based approaches are highly sensitive to illumination. Light variation and shadow interference can substantially alter the radiometric properties of images, reducing the stability of image-based turbidity retrieval (Liu et al., 2018; Zhou et al., 2024).\u003c/p\u003e \u003cp\u003eTo address this issue, we present a smartphone-image framework for turbidity estimation as a rapid and cost-effective alternative to conventional turbidimeters. The framework introduces an easy-to-place three-color background to support image correction under variable illumination. Compared with previous image-based turbidity studies, our approach reduces the sensitivity of turbidity estimates to changing lighting conditions. We collected smartphone images and paired turbidity measurements from diverse small inland waters across China to evaluate the framework\u0026rsquo;s performance across different water types and field settings.\u003c/p\u003e"},{"header":"2. Material and methods","content":"\u003cdiv id=\"Sec3\"\u003e\n \u003ch2\u003e2.1. Study area and datasets\u003c/h2\u003e\n \u003cp\u003eFrom March 6 to December 27, 2025, we implemented a nationwide sampling plan across multiple regions of China. The sampling sites spanned a broad geographic range and included both urban and rural settings. The sampling site locations and the turbidity range are shown in Fig.\u0026nbsp;\u003cspan refid=\"Fig1\"\u003e1\u003c/span\u003e. Samples were collected from diverse inland waters, including rivers, lakes, and ponds, yielding 653 paired records. Each record contains a smartphone image and the corresponding turbidity measurement. The following procedures were used to collect paired samples of images and turbidity measurements. (1) Pour each water sample into a bucket and place a three-color background next to the bucket. (2) Capture an RGB image of the water surface using a smartphone. (3) Turbidity was measured using a YSI EXO2 portable water-quality analyzer.\u003c/p\u003e\n \u003cp\u003eAccording to the above sampling procedures, we adopted more detailed designs to ensure that the collected data were accurate and reliable for evaluate ng the framework.\u003c/p\u003e\n \u003cp\u003e(1) Setting the background, angle, and depth for smartphone imaging\u003c/p\u003e\n \u003cp\u003eThe three-color background was placed next to the bucket to provide similar ambient light for both the background and the water surface. The smartphone was held stable at a consistent viewing angle for all samples. The water depth in the bucket was fixed at 15 cm to minimize depth-related effects on the images.\u003c/p\u003e\n \u003cp\u003e(2) Diverse Sampling Settings for Model Generalization\u003c/p\u003e\n \u003cp\u003eWater samples were collected from various small freshwater environments (e.g., rivers, lakes, and ponds) to capture a wide range of conditions. Images were taken using different smartphones (e.g., Huawei and iPhone) under various weather conditions (e.g., sunny and cloudy). These diverse sampling settings aimed to improve the model\u0026apos;s generalizability. Detailed records of the sampling locations, dates, weather conditions, smartphone models, and waterbody types are provided in Fig. \u003cspan refid=\"MOESM1\"\u003eS1\u003c/span\u003e of the Supplementary material.\u003c/p\u003e\n \u003cp\u003eBased on the above procedures, two datasets were compiled. The water-quality dataset consisted of turbidity measurements from 653 water samples collected from various types of small waterbodies across different regions. The image dataset consisted of the corresponding raw smartphone images for these 653 samples (Fig S2 in the Supplementary material). To enrich image features and improve model generalization, the raw images were further augmented by cropping, rotation, and flipping before being used for model training and evaluation.\u003c/p\u003e\n\u003c/div\u003e\n\u003cdiv id=\"Sec4\"\u003e\n \u003ch2\u003e2.2. A novel framework for smartphone to predict water turbidity\u003c/h2\u003e\n \u003cp\u003eTo address the impact of illumination variations on turbidity prediction, this study developed a novel framework that that incorporates light correction to adjust the RGB values of smartphone images. In the developed framework, a smartphone image is acquired with the water surface and the three-color background in the same view. For each image, we obtained the RGB values of the three channels and performed light correction based on these values. At the same time, the water region is segmented and shadow-affected areas are removed to obtain a shadow-free water surface image. The framework aims to minimize the influence of variable lighting conditions on smartphone turbidity prediction and provides a universal preprocessing method applicable to smartphone monitoring (Fig.\u0026nbsp;\u003cspan refid=\"Fig2\"\u003e2\u003c/span\u003e). The framework consists of three main steps: (1) RGB identification and light correction, (2) Extraction of shadow-free water surface, and (3) Turbidity prediction. The critical procedures of the framework were described as follows:\u003c/p\u003e\n \u003cp\u003e(1) RGB identification and light correction\u003c/p\u003e\n \u003cp\u003eThe background region was identified using a threshold segmentation approach to distinguish it from the surroundings. The region containing the three-color background was selected (Fig.\u0026nbsp;\u003cspan refid=\"Fig3\"\u003e3\u003c/span\u003ea) to reduce interference from the environment and improve color reference recognition accuracy. The image was then converted from RGB to HSV to separate chromatic and brightness information. Based on the average brightness value of the V channel, images were classified into good-light and low-light conditions. Different thresholds were applied for background recognition under each lighting condition (Table\u0026nbsp;\u003cspan refid=\"Tab1\"\u003e1\u003c/span\u003e). For low-light images, wider HSV ranges were used for the blue and green regions to ensure stable background detection (Fig.\u0026nbsp;\u003cspan refid=\"Fig3\"\u003e3\u003c/span\u003ed).\u003c/p\u003e\n \u003cdiv id=\"Equa\"\u003e\n \u003cdiv format=\"TEX\" id=\"FileID_Equa\" name=\"EquationSource\"\u003e$$\\:k=\\frac{\\stackrel{-}{{R}_{r}}}{R}\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:（1）$$\u003c/div\u003e\n \u003c/div\u003e\n \u003cp\u003ewhere k represents the factor. R\u003csub\u003er\u003c/sub\u003e represents the average value of the red channel in the image, and R represents the standard value of the red channel (255). The same applies to the blue and green channels.\u003c/p\u003e\n \u003cp\u003eImages captured under good-light and low-light conditions but with similar turbidity showed clear color inconsistency before correction (Fig. \u003cspan refid=\"Fig4\"\u003e4\u003c/span\u003ea\u0026ndash;b). After light correction, the images appeared more consistent across different illumination conditions. Correction performance was evaluated using the CIEDE2000 color difference metric (\u0026Delta;E₀₀), a standard measure of objective color consistency (Sharma et al., 2005). A clear reduction in \u0026Delta;E₀₀ after correction indicates improved color consistency and reduced lighting-induced bias (Fig. \u003cspan refid=\"Fig4\"\u003e4\u003c/span\u003ec). In a few cases, \u0026Delta;E₀₀ decreased less due to strong, spatially non-uniform shadows that cannot be fully compensated by a global linear correction.\u003c/p\u003e\n \u003cp\u003e(2) Extraction of shadow-free water surface\u003c/p\u003e\n \u003cp\u003eShadows can change pixel brightness and introduce bias in turbidity prediction. Therefore, shadow-free water regions were extracted as a preprocessing step. A pretrained Mask R-CNN model was used as an instance-segmentation tool within the MMDetection framework to obtain water-surface masks (Fig. \u003cspan refid=\"Fig5\"\u003e5\u003c/span\u003e). A confidence threshold of 0.7 was applied. The mask with the highest confidence score was selected for each image. The maximum inscribed rectangle of this mask was then used as the final shadow-free water region for subsequent analysis. All masks were visually checked to avoid obvious errors. For more details on the Mask R-CNN architecture, please refer to the official MMDetection repository (\u003cspan\u003e\u003cspan\u003ehttps://github.com/open-mmlab/mmdetection\u003c/span\u003e\u003c/span\u003e).\u003c/p\u003e\n \u003cp\u003e(3) Turbidity prediction\u003c/p\u003e\n \u003cp\u003eTurbidity prediction was performed on corrected shadow-free water surface images using a convolutional neural network (CNN) (Qi et al., 2024). The network was implemented in TensorFlow. It consists of two convolutional layers (32 and 16 kernels), one max-pooling layer, and two fully connected layers (256 and 128 neurons). The network outputs a single turbidity value (Fig. \u003cspan refid=\"Fig6\"\u003e6\u003c/span\u003e).\u003c/p\u003e\n \u003cp\u003eTo achieve efficient data loading and standardization, a custom data generator was designed to load images and their corresponding turbidity values in batches. Normalization was applied to both the RGB pixels and the turbidity data. During training, 70% of the samples were used for training and 30% for validation. Early stopping and learning rate decay strategies were employed to ensure stable convergence and prevent overfitting. The development of CNN regression model was implemented in PyCharm Community Edition and TensorFlow (\u003cspan\u003e\u003cspan\u003ehttps://www.TensorFlow.org\u003c/span\u003e\u003c/span\u003e).\u003c/p\u003e\n\u003c/div\u003e"},{"header":"3. Framework application in small water bodies","content":"\u003cdiv id=\"Sec6\" class=\"Section2\"\u003e\n \u003ch2\u003e3.1 Water turbidity measurement and prediction\u003c/h2\u003e\n \u003cp\u003eSmall water bodies, such as ponds, rivers, and urban drainage channels, are vital components of local hydrological and ecological systems. They play an important role in regulating surface runoff, retaining nutrients, and supporting biodiversity (Pi et al., 2022; Zeng et al., 2019). In this study, the proposed light correction and turbidity prediction framework was applied to the Huajin River, a representative small water body located within Anhui Normal University in Wuhu, China. The river is a typical urban tributary with shallow water depth, limited flow, and strong temporal variations in water quality. It spans approximately 1.2 km in length, with a width ranging from 25 to 50 m, covering an area of 0.07 km\u0026sup2; (Fig.\u0026nbsp;7a). The river is strongly influenced by human activities, such as drainage outfalls, waterfowl rearing, and boating events, which often result in high turbidity and frequent harmful algal blooms.\u003c/p\u003e\n \u003cp\u003eTraditional remote sensing technology had difficulty in capturing this river for its small size. The Huajin River was therefore used as a field case to demonstrate the applicability of the proposed smartphone-based framework in a small water body. Four monitoring sites were arranged along the river. Smartphone images were collected at each site and processed using the full workflow, including three-color background calibration, shadow-free water extraction, and CNN-based turbidity prediction. Measured turbidity was obtained using a portable water quality analyzer, the YSI EXO2 designed by Xylem Analytics (USA), and used as the reference for evaluation. We captured 16 smartphone images along the river under different lighting conditions and fed them into the trained model for turbidity estimation. The results show that the corrected images have much higher accuracy than the uncorrected ones. This demonstrates the potential of the framework for turbidity monitoring in small waterbodies. Importantly, the framework helps improve monitoring consistency across lighting conditions, supporting better decision-making and efficient water management.\u003c/p\u003e\n\u003c/div\u003e\n\u003cdiv id=\"Sec7\" class=\"Section2\"\u003e\n \u003ch2\u003e3.2 Performance evaluation\u003c/h2\u003e\n \u003cp\u003eThe performance of turbidity prediction before and after light correction was evaluated using measured data from the Huajin River. The correlation between the measured and predicted turbidity improved significantly after applying the light correction. Before light correction, the prediction model achieved an R\u0026sup2; of 0.49, root mean square error (RMSE) of 5.19, mean absolute error (MAE) of 4.04, and Nash\u0026ndash;Sutcliffe efficiency (NS) of 0.49. After light correction, R\u0026sup2; increased to 0.68, while RMSE and MAE decreased to 4.13 and 2.52, respectively, and NS improved to 0.68 (Fig.\u0026nbsp;8ab). This demonstrates that light correction effectively improves the stability and accuracy of turbidity estimation by eliminating the influence of lighting variations on RGB features.\u003c/p\u003e\n \u003cp\u003eFigures \u003cspan refid=\"Fig7\" class=\"InternalRef\"\u003e8\u003c/span\u003ec and \u003cspan refid=\"Fig7\" class=\"InternalRef\"\u003e8\u003c/span\u003ed further illustrate the relationship between image brightness and turbidity prediction under varying light conditions. Before light correction, predicted turbidity values fluctuated strongly with image brightness. After light correction, predicted turbidity values were more consistent with measured data, and the influence of brightness variation was greatly reduced. These results confirm that the proposed light correction effectively minimizes illumination-induced bias, ensuring robust and reliable turbidity prediction in variable-light outdoor environments.\u003c/p\u003e\n \u003cp\u003eThe framework\u0026rsquo;s performance varied across turbidity ranges, with more accurate predictions observed in the 10\u0026ndash;20 NTU range. This may be due to low light and high turbidity, which make images appear dark and lead to mispredictions.\u003c/p\u003e\n\u003c/div\u003e"},{"header":"4. Discussion","content":"\u003cdiv id=\"Sec9\" class=\"Section2\"\u003e \u003ch2\u003e4.1 Comparison with existing methods\u003c/h2\u003e \u003cp\u003eCompared with previous smartphone-based turbidity estimation methods (Trejo-Z\u0026uacute;\u0026ntilde;iga et al., 2024; Ceylan Koydemir et al., 2019; Zhou et al., 2024), this study achieved reliable results (Table \u003cspan refid=\"MOESM1\" class=\"InternalRef\"\u003eS1\u003c/span\u003e in the Supplementary material). This framework uses light correction for RGB adjustment and effectively reduces the impact of changing illumination. The accuracy of this framework is slightly lower than some fixed-camera monitoring approaches (Lu et al., 2025; Zhou et al., 2024), because our dataset includes images under diverse illumination, while those studies often exclude or do not evaluate low-light images.\u003c/p\u003e \u003cp\u003eTurbidity monitoring has traditionally relied on satellite remote sensing. Table \u003cspan refid=\"MOESM1\" class=\"InternalRef\"\u003eS1\u003c/span\u003e shows that our performance is comparable to most previously published remote-sensing studies (Cui et al., 2022; Magr\u0026igrave; et al., 2023; Kong et al., 2025). Some satellite-based estimates may still show slightly higher accuracy than our smartphone results (Feng et al., 2020; Hossain et al., 2021). This is expected because satellites capture richer spectral information, while smartphone images have fewer bands and contain unavoidable platform-related noise. Importantly, our approach provides much finer spatial resolution, down to the millimeter scale. This improvement is valuable but makes validation harder because turbidity can vary over very short distances. In addition, our field dataset for validation includes more samples than what is commonly used in many remote-sensing studies, which supports a robust comparison.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec10\" class=\"Section2\"\u003e \u003ch2\u003e4.2 Framework Advantages\u003c/h2\u003e \u003cp\u003eThe design and application of the developed framework in our case study demonstrated its high potential in estimating turbidity described as follows.\u003c/p\u003e \u003cdiv id=\"Sec11\" class=\"Section3\"\u003e \u003ch2\u003e\u003cb\u003e4.2.1 Well-suited to use in measuring turbidity in small waterbodies\u003c/b\u003e\u003c/h2\u003e \u003cp\u003eRemote sensing images face limitations when monitoring small water bodies, such as ponds, rivers, and urban drainage channels, where their effectiveness is often compromised. For example, the Huajin River in this study cannot be clearly observed in remote sensing images (Fig.\u0026nbsp;7c). Traditional remote sensing methods often fail to meet the real-time monitoring requirements for small water bodies and are susceptible to factors like spatial resolution and cloud cover. In contrast, the smartphone-based framework offers clear advantages, particularly for monitoring small water bodies. The framework provides an efficient and cost-effective method to measure turbidity with higher accuracy, overcoming the limitations of remote sensing in small water bodies. It is particularly suitable for complex environments, such as urban and rural areas, where real-time water quality monitoring is needed.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec12\" class=\"Section3\"\u003e \u003ch2\u003e4.2.2 Significantly reduced the impact of different lighting conditions\u003c/h2\u003e \u003cp\u003eDifferent lighting conditions often change image colors, which can directly affect the accuracy of smartphone-based turbidity predictions. By applying RGB identification and light correction, the framework reduces color variations under different lighting conditions and improves the accuracy and stability of turbidity predictions. Furthermore, by extracting shadow-free water surfaces, the framework removes interference from shadowed areas and ensures accurate water surface information. As a result, the framework demonstrates stronger adaptability under varying lighting conditions, providing stable turbidity predictions with significant practical value.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec13\" class=\"Section3\"\u003e \u003ch2\u003e4.2.3 A universal data preprocessing method\u003c/h2\u003e \u003cp\u003eThe framework provides a universal data preprocessing method applicable to a wide range of image data for turbidity prediction. Through the image processing pipeline, the method provides stable predictions across different water bodies, environmental conditions, and imaging devices. The framework handles diverse input data and adapts to various monitoring scenarios, making it a valuable tool for water quality monitoring. This universal applicability allows seamless integration into various real-world applications, from large-scale environmental monitoring to localized assessments in resource-limited settings.\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv id=\"Sec14\" class=\"Section2\"\u003e \u003ch2\u003e4.3 Potential applications in water management\u003c/h2\u003e \u003cp\u003eIntegrating the framework into a smartphone application is a promising approach for water quality prediction. Given the widespread use of smartphones, they are ideal tools for water quality prediction (Ceylan Koydemir et al., 2019; Leeuw and Boss, 2018). Integrating the framework into a smartphone application provides an efficient and cost-effective alternative to conventional turbidity meters for predicting water turbidity.\u003c/p\u003e \u003cp\u003eIn addition to portable smartphone-based turbidity prediction, the framework can also be applied to monitoring cameras for long-term, real-time monitoring of the study area. Small waterbodies are highly susceptible to external influences due to their limited size, mobility, and self-purification capacity. They can undergo rapid changes over short periods. Therefore, real-time turbidity monitoring in small waterbodies enables timely assessment of water quality and supports pollution prevention and control.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec15\" class=\"Section2\"\u003e \u003ch2\u003e4.4 Uncertainty in the framework\u003c/h2\u003e \u003cp\u003eAlthough the developed framework possesses several significant advantages as described in Section \u003cspan refid=\"Sec9\" class=\"InternalRef\"\u003e4.1\u003c/span\u003e, it is important to note that uncertainties remain. Some aspects require further improvement in future work.\u003c/p\u003e \u003cp\u003e(1) Sensitivity of light correction to field illumination variability\u003c/p\u003e \u003cp\u003eLight correction depends on stable illumination during imaging. Direct sunlight, glare, and strong shadows can affect the extracted RGB values. This may weaken global correction and introduce bias into turbidity prediction. This sensitivity can be reduced by preferring stable lighting conditions during imaging and by avoiding reflective surfaces. In future work, background localization could be improved by replacing threshold segmentation with deep learning\u0026ndash;based object detection, followed by RGB extraction within the detected region.\u003c/p\u003e \u003cp\u003e(2) Dependence of model robustness on training data representativeness\u003c/p\u003e \u003cp\u003eModel robustness depends on the representativeness of the training data. More paired records of images and turbidity measurements are needed across different seasons, weather conditions, and small waterbodies. The dataset should also cover a wider range of turbidity levels and illumination conditions. Broader sampling would help reduce uncertainty and improve generalization.\u003c/p\u003e \u003c/div\u003e"},{"header":"5. Conclusion","content":"\u003cp\u003eThis study proposes a smartphone-based framework that reduces the influence of illumination on turbidity estimation. The framework combines light correction for RGB adjustment, shadow-free water extraction, and CNN regression. Field results based on 653 paired image\u0026ndash;turbidity samples show that the method improves prediction stability under diverse outdoor light. The Huajin River case study further shows that light correction improves agreement with in situ turbidity compared with uncorrected images. Overall, the framework is low-cost, scalable, and suitable for small waterbodies that require dense spatial sampling. Future work will focus on expanding the dataset to cover more seasons, water types, and extreme lighting, and on improving background recognition to enhance robustness in complex field conditions.\u003c/p\u003e"},{"header":"Declarations","content":"\u003ch2\u003eFunding\u003c/h2\u003e \u003cp\u003eThe present study was supported by the National Natural Science Foundation of China (42001087), the Water Resources Science and Technology Program of Jiangxi, China (202426ZDKT09), and the Natural Science Research Project of Anhui Educational Committee (2023AH050151).\u003c/p\u003e \u003cp\u003eAcknowledgments\u003c/p\u003e \u003cp\u003eThe authors would like to express their special thanks to Liuyi Dai, Zhengxin Wang, and Xudong Wang (Anhui Normal University, China) for their great support in water sampling and measurement.\u003c/p\u003e\u003ch2\u003eAuthor Contribution\u003c/h2\u003e\u003cp\u003eJiacong Huang and Lingyan Qi provided the research concept and methodological guidance. Kejia Zhang wrote the main manuscript text. Jiacong Huang and Lingyan Qi reviewed and revised the manuscript. Kejia Zhang, Mingzhu Guo, Xinzhe Jiang, and Weiwei Zhao prepared Figures 1\u0026ndash;8 and Table 1. Jiefa Cai, Yaqin Duan, and Jiacong Huang conducted field surveys and prepared Figures S1\u0026ndash;S2 and Table S1. All authors reviewed and approved the manuscript.\u003c/p\u003e\u003ch2\u003eAcknowledgement\u003c/h2\u003e\u003cp\u003eSpecial thanks to Liuyi Dai, Zhengxin Wang, and Xudong Wang (Anhui Normal University, China) for their solid support in water sampling and measurement.\u003c/p\u003e\u003ch2\u003eData Availability\u003c/h2\u003e\u003cp\u003eAll raw and processed datasets, including 653 paired smartphone images and corresponding turbidity measurements, as well as supporting figures and tables (Figures S1\u0026ndash;S2, Table S1), are available in the supplementary material. Additional data and scripts used for analysis are available from the corresponding author upon reasonable request.\u003c/p\u003e"},{"header":"References","content":"\u003col\u003e\n\u003cli\u003eAimin, L. I., Zheng, Y., Zhenqiang, G. U. O., Ziyi, C., 2025. A Remote Sensing Retrieval Method For River Water Turbidity Based on a Convolutional Neural Network. \u003cem\u003eJournal of Geo-information Science\u003c/em\u003e 27(6), 1305\u0026ndash;1316. https://doi.org/10.12082/dqxxkx.2025.240613 \u003c/li\u003e\n\u003cli\u003eCaroni, R., Greife, A. J., Bresciani, M., Giardino, C., Tellina, G., Carrea, L., Liu, X., Simis, S., Albergel, C., Pinardi, M., 2025. Examining global trends of satellite-derived water quality variables in shallow lakes. \u003cem\u003eRemote Sensing Applications: Society and Environment\u003c/em\u003e 38. https://doi.org/10.1016/j.rsase.2025.101565 \u003c/li\u003e\n\u003cli\u003eCeylan Koydemir, H., Rajpal, S., Gumustekin, E., Karinca, D., Liang, K., Gorocs, Z., Tseng, D., Ozcan, A., 2019. Smartphone-based turbidity reader. \u003cem\u003eSci Rep\u003c/em\u003e 9(1), 19901. https://doi.org/10.1038/s41598-019-56474-z \u003c/li\u003e\n\u003cli\u003eChen, B., Chen, Y., Chen, H., 2024. An Interpretable CatBoost Model Guided by Spectral Morphological Features for the Inversion of Coastal Water Quality Parameters. \u003cem\u003eWater\u003c/em\u003e 16(24). https://doi.org/10.3390/w16243615 \u003c/li\u003e\n\u003cli\u003eChen, J., Chen, S., Fu, R., Li, D., Jiang, H., Wang, C., Peng, Y., Jia, K., Hicks, B. J., 2022. Remote Sensing Big Data for Water Environment Monitoring: Current Status, Challenges, and Future Prospects. \u003cem\u003eEarth\u0026apos;s Future\u003c/em\u003e 10(2). https://doi.org/10.1029/2021ef002289 \u003c/li\u003e\n\u003cli\u003eChowdhury, M., de la Calle, I., Laiz, I., Ruescas, A. B., 2025. Near-Real-Time Turbidity Monitoring at Global Scale Using Sentinel-2 Data and Machine Learning Techniques. \u003cem\u003eRemote Sensing\u003c/em\u003e 17(22). https://doi.org/10.3390/rs17223716 \u003c/li\u003e\n\u003cli\u003eCoffer, M. M., Schaeffer, B. A., Darling, J. A., Urquhart, E. A., Salls, W. B., 2020. Quantifying national and regional cyanobacterial occurrence in US lakes using satellite remote sensing. \u003cem\u003eEcol Indic\u003c/em\u003e 111, 105976. https://doi.org/10.1016/j.ecolind.2019.105976 \u003c/li\u003e\n\u003cli\u003eCui, M., Sun, Y., Huang, C., Li, M., 2022. Water Turbidity Retrieval Based on UAV Hyperspectral Remote Sensing. \u003cem\u003eWater\u003c/em\u003e 14(1). https://doi.org/10.3390/w14010128 \u003c/li\u003e\n\u003cli\u003eDeng, Y., Zhang, Y., Pan, D., Yang, S. X., Gharabaghi, B., 2024. Review of Recent Advances in Remote Sensing and Machine Learning Methods for Lake Water Quality Management. \u003cem\u003eRemote Sensing\u003c/em\u003e 16(22). https://doi.org/10.3390/rs16224196 \u003c/li\u003e\n\u003cli\u003eDing, W., Zhao, J., Qin, B., Wu, T., Zhu, S., Li, Y., Xu, S., Ruan, S., Wang, Y., 2021. Exploring and quantifying the relationship between instantaneous wind speed and turbidity in a large shallow lake: case study of Lake Taihu in China. \u003cem\u003eEnvironmental Science and Pollution Research\u003c/em\u003e 28(13), 16616\u0026ndash;16632. https://doi.org/10.1007/s11356-020-11544-y \u003c/li\u003e\n\u003cli\u003eFeng, J., Chen, H., Zhang, H., Li, Z., Yu, Y., Zhang, Y., Bilal, M., Qiu, Z., 2020. Turbidity Estimation from GOCI Satellite Data in the Turbid Estuaries of China\u0026rsquo;s Coast. \u003cem\u003eRemote Sensing\u003c/em\u003e 12(22). https://doi.org/10.3390/rs12223770 \u003c/li\u003e\n\u003cli\u003eGholizadeh, M. H., Melesse, A. M., Reddi, L., 2016. A Comprehensive Review on Water Quality Parameters Estimation Using Remote Sensing Techniques. \u003cem\u003eSensors (Basel)\u003c/em\u003e 16(8). https://doi.org/10.3390/s16081298 \u003c/li\u003e\n\u003cli\u003eHossain, A. K. M. A., Mathias, C., Blanton, R., 2021. Remote Sensing of Turbidity in the Tennessee River Using Landsat 8 Satellite. \u003cem\u003eRemote Sensing\u003c/em\u003e 13(18). https://doi.org/10.3390/rs13183785 \u003c/li\u003e\n\u003cli\u003eHuang, J., Qian, R., Gao, J., Bing, H., Huang, Q., Qi, L., Song, S., Huang, J., 2021. A novel framework to predict water turbidity using Bayesian modeling. \u003cem\u003eWater Res\u003c/em\u003e 202, 117406. https://doi.org/10.1016/j.watres.2021.117406 \u003c/li\u003e\n\u003cli\u003eKong, Y., Jimenez, K., Lee, C. M., Winter, S., Summers-Evans, J., Cao, A., Menczer, M., Han, R., Mills, C., McCarthy, S., Blatzheim, K., Jay, J. A., 2025. Monitoring Coastal Water Turbidity Using Sentinel2\u0026mdash;A Case Study in Los Angeles. \u003cem\u003eRemote Sensing\u003c/em\u003e 17(2). https://doi.org/10.3390/rs17020201 \u003c/li\u003e\n\u003cli\u003eLeeuw, T., Boss, E., 2018. The HydroColor App: Above Water Measurements of Remote Sensing Reflectance and Turbidity Using a Smartphone Camera. \u003cem\u003eSensors (Basel)\u003c/em\u003e 18(1). https://doi.org/10.3390/s18010256 \u003c/li\u003e\n\u003cli\u003eLi, J., Xia, C., 2023. Drivers of Spatial and Temporal Dynamics in Water Turbidity of China Yangtze River Basin. \u003cem\u003eWater\u003c/em\u003e 15(7). https://doi.org/10.3390/w15071264 \u003c/li\u003e\n\u003cli\u003eLi, S., Kutser, T., Song, K., Liu, G., Li, Y., 2023a. Lake Turbidity Mapping Using an OWTs-bp Based Framework and Sentinel-2 Imagery. \u003cem\u003eRemote Sensing\u003c/em\u003e 15(10). https://doi.org/10.3390/rs15102489 \u003c/li\u003e\n\u003cli\u003eLi, Y., Li, S., Song, K., Liu, G., Wen, Z., Fang, C., Shang, Y., Lyu, L., Zhang, L., 2023b. Sentinel-3 OLCI observations of Chinese lake turbidity using machine learning algorithms. \u003cem\u003eJournal of Hydrology\u003c/em\u003e 622. https://doi.org/10.1016/j.jhydrol.2023.129668 \u003c/li\u003e\n\u003cli\u003eLin, X., Wu, M., Shao, X., Li, G., Hong, Y., 2023. Water turbidity dynamics using random forest in the Yangtze River Delta Region, China. \u003cem\u003eSci Total Environ\u003c/em\u003e 903, 166511. https://doi.org/10.1016/j.scitotenv.2023.166511 \u003c/li\u003e\n\u003cli\u003eLiu, Y., Chen, Y., Fang, X., 2018. A Review of Turbidity Detection Based on Computer Vision. \u003cem\u003eIEEE Access\u003c/em\u003e 6, 60586\u0026ndash;60604. https://doi.org/10.1109/access.2018.2875071 \u003c/li\u003e\n\u003cli\u003eLu, Z., Zhu, X., Li, J., Li, M., Wang, J., Wang, W., Zheng, Y., Zhang, Q., 2025. Detecting glacial lake water quality indicators from RGB surveillance images via deep learning. \u003cem\u003eInternational Journal of Applied Earth Observation and Geoinformation\u003c/em\u003e 136. https://doi.org/10.1016/j.jag.2025.104392 \u003c/li\u003e\n\u003cli\u003eMa, Y., Song, K., Wen, Z., Liu, G., Shang, Y., Lyu, L., Du, J., Yang, Q., Li, S., Tao, H., Hou, J., 2021. Remote Sensing of Turbidity for Lakes in Northeast China Using Sentinel-2 Images With Machine Learning Algorithms. \u003cem\u003eIEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing\u003c/em\u003e 14, 9132\u0026ndash;9146. https://doi.org/10.1109/jstars.2021.3109292 \u003c/li\u003e\n\u003cli\u003eMagr\u0026igrave;, S., Ottaviani, E., Prampolini, E., Besio, G., Fabiano, B., Federici, B., 2023. Application of machine learning techniques to derive sea water turbidity from Sentinel-2 imagery. \u003cem\u003eRemote Sensing Applications: Society and Environment\u003c/em\u003e 30. https://doi.org/10.1016/j.rsase.2023.100951 \u003c/li\u003e\n\u003cli\u003eOdermatt, D., Gitelson, A., Brando, V. E., Schaepman, M., 2012. Review of constituent retrieval in optically deep and complex waters from satellite imagery. \u003cem\u003eRemote Sensing of Environment\u003c/em\u003e 118, 116\u0026ndash;126. https://doi.org/10.1016/j.rse.2011.11.013 \u003c/li\u003e\n\u003cli\u003eOverstreet, B. T., Legleiter, C. J., 2017. Removing sun glint from optical remote sensing images of shallow rivers. \u003cem\u003eEarth Surface Processes and Landforms\u003c/em\u003e 42(2), 318\u0026ndash;333. https://doi.org/https://doi.org/10.1002/esp.4063 \u003c/li\u003e\n\u003cli\u003ePalmer, S. C. J., Kutser, T., Hunter, P. D., 2015. Remote sensing of inland waters: Challenges, progress and future directions. \u003cem\u003eRemote Sensing of Environment\u003c/em\u003e 157, 1\u0026ndash;8. https://doi.org/10.1016/j.rse.2014.09.021 \u003c/li\u003e\n\u003cli\u003ePang, Z., Zhou, Z., Fu, J. e., Jiang, W., Qin, X., Sun, M., 2025. Deep learning-based remote sensing retrieval of inland water quality: A review. \u003cem\u003eJournal of Hydrology: Regional Studies\u003c/em\u003e 61. https://doi.org/10.1016/j.ejrh.2025.102759 \u003c/li\u003e\n\u003cli\u003ePi, X., Luo, Q., Feng, L., Xu, Y., Tang, J., Liang, X., Ma, E., Cheng, R., Fensholt, R., Brandt, M., Cai, X., Gibson, L., Liu, J., Zheng, C., Li, W., Bryan, B. A., 2022. Mapping global lake dynamics reveals the emerging roles of small lakes. \u003cem\u003eNature Communications\u003c/em\u003e 13(1). https://doi.org/10.1038/s41467-022-33239-3 \u003c/li\u003e\n\u003cli\u003ePu, F., Ding, C., Chao, Z., Yu, Y., Xu, X., 2019. Water-Quality Classification of Inland Lakes Using Landsat8 Images by Convolutional Neural Networks. \u003cem\u003eRemote Sensing\u003c/em\u003e 11(14). https://doi.org/10.3390/rs11141674 \u003c/li\u003e\n\u003cli\u003eQi, L., Yin, H., Wang, Z., Ye, L., Zhang, S., Dai, L., Wu, F., Jiang, X., Huang, Q., Huang, J., 2024. Smartphone as an alternative to measure chlorophyll-a concentration in small waterbodies. \u003cem\u003eJ Environ Manage\u003c/em\u003e 368, 122135. https://doi.org/10.1016/j.jenvman.2024.122135 \u003c/li\u003e\n\u003cli\u003eQiu, Z., Liu, D., Yan, N., Yan, Y., Yang, C., Zhang, C., Duan, H., 2025. Landsat and dual random forest modelling reveal sediment fining in the Yellow River shaped by ecological restoration on China\u0026apos;s loess plateau. \u003cem\u003eRemote Sensing of Environment\u003c/em\u003e 330. https://doi.org/10.1016/j.rse.2025.114994 \u003c/li\u003e\n\u003cli\u003eSahoo, D., Anandhi, A., 2023. Conceptualizing turbidity for aquatic ecosystems in the context of sustainable development goals. \u003cem\u003eEnvironmental Science: Advances\u003c/em\u003e 2(9), 1220\u0026ndash;1234. https://doi.org/10.1039/d2va00327a \u003c/li\u003e\n\u003cli\u003eSantos, V. O., Rocha, P. A. C., Th\u0026eacute;, J. V. G., Gharabaghi, B., 2025. Evaluation of machine learning methods for forecasting turbidity in river networks using Sentinel-2 remote sensing data. \u003cem\u003eEcological Informatics\u003c/em\u003e 90. https://doi.org/10.1016/j.ecoinf.2025.103313 \u003c/li\u003e\n\u003cli\u003eSehgal, D., Mart\u0026iacute;nez‐Carreras, N., Hissler, C., Bense, V. F., Hoitink, A. J. F., 2022. A Generic Relation Between Turbidity, Suspended Particulate Matter Concentration, and Sediment Characteristics. \u003cem\u003eJournal of Geophysical Research: Earth Surface\u003c/em\u003e 127(12). https://doi.org/10.1029/2022jf006838 \u003c/li\u003e\n\u003cli\u003eSharma, G., Wu, W., Dalal, E. N., 2005. The CIEDE2000 color-difference formula: Implementation notes, supplementary test data, and mathematical observations. \u003cem\u003eColor Research \u0026amp; Application\u003c/em\u003e 30(1), 21\u0026ndash;30. https://doi.org/https://doi.org/10.1002/col.20070 \u003c/li\u003e\n\u003cli\u003eShi, K., Zhang, Y., Qin, B., Zhou, B., 2019. Remote sensing of cyanobacterial blooms in inland waters: present knowledge and future challenges. \u003cem\u003eSci Bull (Beijing)\u003c/em\u003e 64(20), 1540\u0026ndash;1556. https://doi.org/10.1016/j.scib.2019.07.002 \u003c/li\u003e\n\u003cli\u003eTan, Z., Gao, M., Yuan, J., Jiang, L., Duan, H., 2022. A Robust Model for MODIS and Landsat Image Fusion Considering Input Noise. \u003cem\u003eIEEE Transactions on Geoscience and Remote Sensing\u003c/em\u003e 60, 1\u0026ndash;17. https://doi.org/10.1109/tgrs.2022.3145086 \u003c/li\u003e\n\u003cli\u003eTrejo-Z\u0026uacute;\u0026ntilde;iga, I., Moreno, M., Santana-Cruz, R. F., Mel\u0026eacute;ndez-V\u0026aacute;zquez, F., 2024. Deep-Learning-Driven Turbidity Level Classification. \u003cem\u003eBig Data and Cognitive Computing\u003c/em\u003e 8(8). https://doi.org/10.3390/bdcc8080089 \u003c/li\u003e\n\u003cli\u003eWu, D., Tang, T., Odermatt, D., Liu, W., 2025a. Spatiotemporal variability in global lakes turbidity derived from satellite imageries. \u003cem\u003eEnvironmental Research Communications\u003c/em\u003e 7(3). https://doi.org/10.1088/2515-7620/adb941 \u003c/li\u003e\n\u003cli\u003eWu, Z., Pang, J., Li, J., Wang, Y., Ruan, J., Zhang, X., Yang, L., Pang, Y., Gao, Y., 2025b. A review of remote sensing-based water quality monitoring in turbid coastal waters. \u003cem\u003eIntelligent Marine Technology and Systems\u003c/em\u003e 3(1). https://doi.org/10.1007/s44295-025-00075-2 \u003c/li\u003e\n\u003cli\u003eYan, N., Qiu, Z., Zhang, C., Liu, J., Liu, D., 2025. Observing water turbidity in Chinese rivers using Landsat series data over the past 40 years. \u003cem\u003eJournal of Cleaner Production\u003c/em\u003e 494. https://doi.org/10.1016/j.jclepro.2025.145001 \u003c/li\u003e\n\u003cli\u003eYang, Z., Gong, C., Lu, Z., Wu, E., Huai, H., Hu, Y., Li, L., Dong, L., 2023. Combined Retrievals of Turbidity from Sentinel-2A/B and Landsat-8/9 in the Taihu Lake through Machine Learning. \u003cem\u003eRemote Sensing\u003c/em\u003e 15(17). https://doi.org/10.3390/rs15174333 \u003c/li\u003e\n\u003cli\u003eZeng, Z., Wang, D., Tan, W., Huang, J., 2019. Extracting aquaculture ponds from natural water surfaces around inland lakes on medium resolution multispectral images. \u003cem\u003eInternational Journal of Applied Earth Observation and Geoinformation\u003c/em\u003e 80, 13\u0026ndash;25. https://doi.org/10.1016/j.jag.2019.03.019 \u003c/li\u003e\n\u003cli\u003eZhang, Y., Yao, X., Wu, Q., Huang, Y., Zhou, Z., Yang, J., Liu, X., 2021. Turbidity prediction of lake-type raw water using random forest model based on meteorological data: A case study of Tai lake, China. \u003cem\u003eJ Environ Manage\u003c/em\u003e 290, 112657. https://doi.org/10.1016/j.jenvman.2021.112657 \u003c/li\u003e\n\u003cli\u003eZhou, N., Chen, H., Chen, N., Liu, B., 2024. An Approach Based on Video Image Intelligent Recognition for Water Turbidity Monitoring in River. \u003cem\u003eACS ES\u0026amp;T Water\u003c/em\u003e 4(2), 543\u0026ndash;554. https://doi.org/10.1021/acsestwater.3c00596 \u003c/li\u003e\n\u003c/ol\u003e"},{"header":"Table","content":"\u003cp\u003eTable 1.\u0026nbsp;HSV threshold range\u003c/p\u003e\n \u003ctable border=\"0\" cellspacing=\"0\" cellpadding=\"0\"\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 132px;\"\u003e\n \u003cp\u003e\u003cstrong\u003econdition\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 76px;\"\u003e\n \u003cp\u003e\u003cstrong\u003ecolor\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 132px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eH range\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 110px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eS range\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 88px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eV range\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 132px;\"\u003e\n \u003cp\u003eGood light\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 76px;\"\u003e\n \u003cp\u003eRed\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 132px;\"\u003e\n \u003cp\u003e[0\u0026ndash;12], [58-180]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 110px;\"\u003e\n \u003cp\u003e40\u0026ndash;255\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 88px;\"\u003e\n \u003cp\u003e90\u0026ndash;255\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 132px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 76px;\"\u003e\n \u003cp\u003eGreen\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 132px;\"\u003e\n \u003cp\u003e40\u0026ndash;85\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 110px;\"\u003e\n \u003cp\u003e40\u0026ndash;255\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 88px;\"\u003e\n \u003cp\u003e90\u0026ndash;255\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 132px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 76px;\"\u003e\n \u003cp\u003eBlue\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 132px;\"\u003e\n \u003cp\u003e100\u0026ndash;132\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 110px;\"\u003e\n \u003cp\u003e45\u0026ndash;255\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 88px;\"\u003e\n \u003cp\u003e80\u0026ndash;255\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 132px;\"\u003e\n \u003cp\u003eLow light\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 76px;\"\u003e\n \u003cp\u003eRed\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 132px;\"\u003e\n \u003cp\u003e[0\u0026ndash;12], [158\u0026ndash;180]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 110px;\"\u003e\n \u003cp\u003e35\u0026ndash;255\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 88px;\"\u003e\n \u003cp\u003e60\u0026ndash;255\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 132px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 76px;\"\u003e\n \u003cp\u003eGreen\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 132px;\"\u003e\n \u003cp\u003e25\u0026ndash;90\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 110px;\"\u003e\n \u003cp\u003e35\u0026ndash;255\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 88px;\"\u003e\n \u003cp\u003e30\u0026ndash;255\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 132px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 76px;\"\u003e\n \u003cp\u003eBlue\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 132px;\"\u003e\n \u003cp\u003e96\u0026ndash;138\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 110px;\"\u003e\n \u003cp\u003e30\u0026ndash;255\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 88px;\"\u003e\n \u003cp\u003e35\u0026ndash;255\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n \u003c/table\u003e"}],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":true,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":false,"hideJournal":false,"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":"environmental-monitoring-and-assessment","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"emas","sideBox":"Learn more about [Environmental Monitoring and Assessment](http://link.springer.com/journal/10661)","snPcode":"10661","submissionUrl":"https://submission.nature.com/new-submission/10661/3","title":"Environmental Monitoring and Assessment","twitterHandle":"","acdcEnabled":true,"dfaEnabled":true,"editorialSystem":"em","reportingPortfolio":"Springer Hybrid","inReviewEnabled":true,"inReviewRevisionsEnabled":false},"keywords":"Turbidity, Image Processing, Smartphone, Deep learning","lastPublishedDoi":"10.21203/rs.3.rs-9317150/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-9317150/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003cp\u003eWater turbidity is a key indicator of water quality, reflecting the concentration of suspended particles and affecting aquatic ecosystems. This study proposes a novel smartphone-based framework to predict water turbidity. The framework uses light correction to reduce the effect of varying illumination. It uses Mask R-CNN to extract a shadow-free water surface and a convolutional neural network (CNN) to predict turbidity. Light correction substantially improves prediction accuracy and stability, especially for turbidity values between 10 and 20 NTU. Evaluation shows that after correction, R\u0026sup2; increased from 0.49 to 0.68, and RMSE decreased from 5.19 to 4.13. Overall, the framework provides a universal preprocessing approach that reduces the influence of light and enables more consistent and accurate turbidity prediction in the field.\u003c/p\u003e","manuscriptTitle":"Significantly reducing the impact of light: a novel framework for smartphone to predict water turbidity","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2026-05-07 14:40:42","doi":"10.21203/rs.3.rs-9317150/v1","editorialEvents":[{"type":"communityComments","content":0},{"type":"reviewerAgreed","content":"236756511299057778215090875590821416482","date":"2026-04-29T21:10:07+00:00","index":"hide","fulltext":""},{"type":"reviewersInvited","content":"","date":"2026-04-29T19:21:32+00:00","index":"","fulltext":""},{"type":"editorAssigned","content":"","date":"2026-04-21T07:54:02+00:00","index":"","fulltext":""},{"type":"checksComplete","content":"","date":"2026-04-21T07:53:07+00:00","index":"","fulltext":""},{"type":"submitted","content":"Environmental Monitoring and Assessment","date":"2026-04-04T03:08:37+00:00","index":"","fulltext":""}],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"environmental-monitoring-and-assessment","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"emas","sideBox":"Learn more about [Environmental Monitoring and Assessment](http://link.springer.com/journal/10661)","snPcode":"10661","submissionUrl":"https://submission.nature.com/new-submission/10661/3","title":"Environmental Monitoring and Assessment","twitterHandle":"","acdcEnabled":true,"dfaEnabled":true,"editorialSystem":"em","reportingPortfolio":"Springer Hybrid","inReviewEnabled":true,"inReviewRevisionsEnabled":false}}],"origin":"","ownerIdentity":"4b282b37-bd53-4c75-a9ba-355a6a70e742","owner":[],"postedDate":"May 7th, 2026","published":true,"recentEditorialEvents":[{"type":"reviewerAgreed","content":"236756511299057778215090875590821416482","date":"2026-04-29T21:10:07+00:00","index":16,"fulltext":""},{"type":"reviewersInvited","content":"11","date":"2026-04-29T19:21:32+00:00","index":"","fulltext":""}],"rejectedJournal":[],"revision":"","amendment":"","status":"under-review","subjectAreas":[],"tags":[],"updatedAt":"2026-05-07T14:40:42+00:00","versionOfRecord":[],"versionCreatedAt":"2026-05-07 14:40:42","video":"","vorDoi":"","vorDoiUrl":"","workflowStages":[]},"version":"v1","identity":"rs-9317150","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-9317150","identity":"rs-9317150","version":["v1"]},"buildId":"XKTyCvWXoU3ODBz1xrDgd","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}