Differential Impacts of Very High-Resolution Terrain Data and Multispectral Imagery on the Accuracy of Land-Use Classification Using Machine Learning

Differential Impacts of Very High-Resolution Terrain Data and Multispectral Imagery on the Accuracy of Land-Use Classification Using Machine Learning | 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 Differential Impacts of Very High-Resolution Terrain Data and Multispectral Imagery on the Accuracy of Land-Use Classification Using Machine Learning Joshua Brown, Godstime K. James, Innocent E. Bello, Adeodyin S. Ajeyomi, and 2 more This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-7730313/v1 This work is licensed under a CC BY 4.0 License Status: Under Review Version 1 posted 4 You are reading this latest preprint version Abstract Accurate land-use and land-cover (LULC) classification is essential for urban planning, environmental monitoring, and resource management. With the availability of high-resolution remote sensing, mapping urban areas has improved; however, selecting effective data and algorithms remains challenging. This study examines the impact of combining very high-resolution orthophotos with elevation datasets, including the Digital Surface Model (DSM), Digital Terrain Model (DTM), and normalized DSM (nDSM), on supervised machine learning classifiers. Data was acquired using a DJI Matrice 350 RTK drone at 5 cm resolution, then resampled to 2 m for efficiency. Three classifiers, Support Vector Classifier (SVC), Random Forest (RF), and Gradient Boosting Classifier (GBC), were tested in two stages: first with orthophotos only, then with added elevation features. The SVC model, in particular, achieved the highest overall accuracy (91.43%) and F1-score (90.75%), excelling at distinguishing between spectrally similar classes such as buildings and tarred roads. Elevation features helped distinguish spectrally similar classes, such as buildings and tarred roads, reducing misclassification common in RGB-only models. The findings highlight that integrating spectral and elevation data enhances classification reliability, as orthophotos provide color and texture while elevation adds structural detail. This fusion approach offers a scalable and high-precision method for urban mapping and environmental analysis. LULC Orthophoto Digital Surface Model Digital Terrain Model Remote sensing Machine Learning UAV imagery Data fusion Figures Figure 1 Figure 2 Figure 3 Figure 4 Figure 5 Figure 6 Figure 7 Figure 8 Figure 9 Figure 10 Figure 11 Figure 12 1. Introduction The interaction and classification of landscapes according to human use and natural cover are crucial for understanding the spatial and temporal patterns of land-use and land-cover changes [1, 2, 3]. Cem Ünsalan and Boyer 2011 [4] highlight land use classification, which covers a wide range of applications, facilitating informed decision-making in urban planning, agriculture, environmental conservation, and disaster management. As landscapes continue to evolve due to both natural and anthropogenic factors, the demand for precise and up-to-date land-use information has intensified. The fusion of remotely sensed data with machine learning algorithms has proven to be a breakthrough and a powerful tool to tackle unprecedented, large-scale challenges emerging from remotely sensed data analysis [5, 6, 7]. Traditionally, multispectral imagery has widely been utilized for land-use classification due to its ability to capture surface reflectance properties across different spectral bands [2, 8]. These images provide vital information about vegetation, water bodies, bare soil, and built-up areas. However, surface characteristics alone may not always be sufficient to distinguish between land-use classes that share similar spectral signatures [9, 10, 11, 12, 1]. To overcome this limitation and improve the identification of land use dynamics, fusing datasets acquired from remote sensors that operate with differing underlying physical mechanisms and providing synergistic and potentially more effective means of obtaining comprehensive information about land surface properties [12]. Specifically, very-high terrain data, Digital Surface Models (DSM), Digital Terrain Models (DTM), and orthomosaics, with terrain data displaying detailed representations of surface height and landform structure (DSM captures all surface features, including vegetation and buildings, while DTM represents the bare earth) [13], making them particularly useful for analyzing natural terrain [14] and orthomosaics display a true color image of overlapping multiple photographs of the earth surface [15]. Terrain variables are often correlated with ecological and human land-use patterns, particularly in regions where topography influences vegetation growth, hydrology, and settlement patterns [16, 17, 18, 19, 20, 21]. Multi-spectral imagery captures surface reflectance, and geomorphometric variables, according to Guth et al. 2021 [22], add a third dimension of terrain context that helps distinguish between land-use types with similar spectral signatures but topographically distinct characteristics. In recent years machine learning has emerged as a powerful tool in remote sensing applications due to its ability to handle large, complex, and high-dimensional datasets, having algorithms such as Random Forest (RF), Support Vector Machine (SVM), and Gradient Boosting with capabilities of identifying non-linear relationships and interactions among input features, further improving the robustness and precision of classification outputs [6]. The integration of terrain data into machine learning workflows alongside spectral or orthophoto data should enhance the accuracy of land use classification and improve interpretability challenges in complex terrain. This study explores how high-resolution terrain data can be derived to enhance the accuracy of land-use classification machine learning techniques. It investigates the individual and combined effects of these features with spectral datasets, evaluates the performance of different classification algorithms, and assesses how topographic context influences land-use mapping in diverse landscapes. The findings contribute to the growing body of knowledge on terrain-based remote sensing and provide practical insights for enhancing land-use monitoring systems through the integration of geospatial intelligence and artificial intelligence. 2. Materials and Methods 2.1 Study Area The study was conducted in Rivers State, Nigeria, at a latitude of 4°42'47.7"N – 4 o 52’12” and a longitude of 6°54'26.6"E – 7 o 08’01”. With an area extent of 2526.75 square kilometers, it covers eight Local Government Areas (LGAs) within Rivers State, Nigeria: Obio/Akpor, Port Harcourt, Oyigbo, Ogu/Bolo, Eleme, Ikwerre, Etche, and Okrika. These regions exhibit diverse urban growth patterns and environmental characteristics that affect flood vulnerability. Port Harcourt, the state capital, serves as a central industrial hub with advanced infrastructure, and Obio/Akpor is a densely populated area, facing increasing pressure from urbanization. They are densely populated areas prone to flash flooding during intense rainfall. Okrika is a riverine community, while Ogu/Bolo is another low-lying coastal area. They are naturally prone to tidal flooding and river overflow. In Oyigbo and Eleme, the expansion of industrial activities and poorly maintained drainage networks exacerbate flood risks, particularly during the rainy season. Ikwerre, though more inland, faces localized flooding due to unplanned urban expansion and poor land management. Meanwhile, Etche’s agrarian landscape is increasingly threatened by seasonal floodwaters, which damage crops and disrupt rural livelihoods [23]. 2.2 Data Acquisition, Extraction, and Pre-processing High-resolution aerial imagery was captured using a DJI Matrice 350 RTK drone equipped with a Zenmuse P1 and L1 sensors. Flights were carried out at a consistent altitude of 120 meters between 10:00 AM and 4:00 PM to ensure optimal lighting and minimize shadow interference. A total of twelve locations were surveyed, of which eight were randomly selected for further analysis. The selected sites were stitched into a seamless image mosaic using the Structure-from-Motion (SfM) workflow in Agisoft Metashape, which utilized GPU-accelerated tie-point matching to enhance spatial consistency, following established methods [24]. The final mosaic had a native ground sampling distance of 0.05 meters, later resampled to 2 meters using GDAL tools to reduce computational load and streamline processing. The photogrammetric process produced an orthomosaic of 5cm GSD, while the LiDAR data processing produced a Digital Surface Model (DSM) and a Digital Terrain Model (DTM). The orthomosaic was generated through careful image alignment and dense point cloud reconstruction, providing detailed spectral information critical for accurate land-use classification [25]. Orthorectification was applied to correct geometric distortions and ensure spatial accuracy. The DSM captured surface elevations, including buildings and vegetation, allowing for a three-dimensional analysis of the landscape [26]. To derive DTM, advanced filtering techniques were applied to isolate the bare earth surface, effectively removing non-ground features [27]. Together, these datasets, high-resolution spectral imagery, surface elevation, and terrain morphology, offered a rich, multi-dimensional view of the landscape, forming a robust foundation for land-use/land-cover classification. To ensure optimal data quality before classification, a series of comprehensive preprocessing steps were undertaken. The orthophoto underwent radiometric normalization to correct illumination differences across the study area, reducing spectral variability due to lighting conditions [28]. The elevation datasets, including the DSM and DTM, were enhanced using median filtering techniques to suppress noise, an essential step in areas with complex terrain or heterogeneous vegetation [29]. A normalized Digital Surface Model (nDSM) was also derived by subtracting the DTM from the DSM, effectively isolating surface features such as vegetation canopies and built structures [30]. These pre-processed datasets, comprising both spectral and structural information, provided a reliable and information-rich foundation for the subsequent machine learning-based land-use classification. Before model training, the orthophotos were normalized using Min-Max scaling to transform the data into a consistent [0, 1] range. This preprocessing step is crucial for remote sensing applications where different spectral bands and derived indices may have varying measurement scales [31]. The Min-Max scaler was implemented using the formula: $$\:\text{m}=\frac{x-{x}_{min}}{{x}_{max}-{x}_{min}}$$ 1 m represents the original feature values, x min and x max denote the minimum and maximum values for each feature, respectively. This pre-processing step was crucial for the machine learning workflow, as it enabled more stable convergence during model training and enhanced the interpretability of feature importance measures in the final classification models. 2.3 Training Data Collection for LULC Classification Training data for Land Use and Land Cover (LULC) classification were collected using handheld GPS devices to ensure spatial accuracy and alignment with field observations. Initially, eight distinct land cover classes were identified during field reconnaissance to capture the diversity of surface features within the study area. These classes included various forms of built-up areas, vegetation types, and land surfaces (tarred and untarred roads). However, for model training and to improve classification efficiency, the classes were systematically regrouped into five broader categories: buildings, tarred land, vegetation, open water, and bare land. The reclassification ensured consistency in spectral separability and reduced intra-class variability. GPS points were collected to represent each of these five final classes and were cross-verified with high-resolution satellite imagery. The dataset was subsequently cleaned to remove inconsistencies, spatial errors, and duplicates. In this schema, buildings refer to all constructed structures; tarred land includes roads and paved surfaces; vegetation encompasses both natural and cultivated green cover; open water represents rivers, lakes, and ponds; and bare land denotes exposed soil or sparsely vegetated ground. The cleaned and labeled samples were used for training and validating machine learning models for LULC classification. 2.4 Image classification using Machine Learning Models In this study, supervised machine learning algorithms were employed to classify land-use and land-cover (LULC) types using high-resolution aerial imagery. The classification process was designed in two stages to evaluate the incremental value of structural features. Initially, models were trained exclusively on the spectral information derived from the orthomosaic. This baseline configuration allowed for the assessment of classifier performance using purely spectral data. Subsequently, additional features, including elevation data from the Digital Surface Model (DSM), Digital Terrain Model (DTM), and the derived normalized DSM (nDSM), were integrated into the input feature space to examine their impact on classification accuracy. Three widely used classifiers, Random Forest (RF), Support Vector Classifier (SVC), and Gradient Boosting Classifier (GBC), were selected for their effectiveness in handling non-linear, high-dimensional data. Training samples were derived from field-based ground truth observations and high-resolution reference imagery, ensuring reliable class representation across diverse land cover types. Hyperparameter tuning was conducted using a grid search strategy combined with k-fold cross-validation to optimize model generalization. Classification performance for each model and input configuration was assessed using standard accuracy metrics, including overall accuracy, producer’s and user’s accuracies, and Cohen’s Kappa coefficient. This stepwise evaluation framework enabled a clear comparison of how spectral data alone performs versus the enriched spectral-structural feature set. Results highlight the added value of integrating terrain-derived features in improving the precision of LULC classification, particularly in areas with complex surface morphology or overlapping land cover types. 2.5 Machine Learning Classification Algorithms 2.5.1 Supporting Vector Classifier (SVC) Support Vector Classifier (SVC) is a non-parametric, supervised machine learning algorithm grounded in the principle of Structural Risk Minimization (SRM). It seeks to identify an optimal hyperplane that maximally separates data points of distinct classes while minimizing generalization error [32, 33]. Unlike empirical risk minimization techniques, SRM aims to balance model complexity with training accuracy, making SVC particularly effective for high-dimensional classification tasks such as land-use and land-cover (LULC) mapping. In remote sensing applications, kernel functions are used to transform nonlinearly separable data into a higher-dimensional feature space, where linear separation is possible. Among the most widely used kernels, the Radial Basis Function (RBF) and polynomial kernels have demonstrated strong performance, with RBF often yielding higher classification accuracy due to its ability to model complex, non-linear relationships inherent in LULC datasets [34, 35]. The Spectral Angle Mapper (SAM), though traditionally used as a similarity measure, is not directly involved in SVC’s hyperplane optimization process and may have been conflated in earlier literature. In this study, both RBF and polynomial kernels were implemented using the Scikit-learn library in Python. These configurations were evaluated for their effectiveness in classifying high-resolution LULC data derived from RGB imagery and elevation features. Hyperparameters for each kernel type were optimized using grid search combined with cross-validation to ensure model robustness and generalizability. 2.5.2 Random Forest Classifier Random Forest (RF) is a robust ensemble learning technique that builds many decision trees, each trained on a randomly selected subset of the data and features [36]. Unlike individual decision trees, RF aggregates the predictions of multiple base classifiers using a majority voting scheme to produce more stable and accurate outputs. This ensemble strategy reduces the risk of overfitting and improves generalization by leveraging the diversity among trees [36, 37]. The final prediction is obtained by aggregating all tree outputs using the discrimination function: $$\:\text{H}\left(\text{x}\right)\:=\:\text{a}\text{r}\text{g}\underset{{\:\:\:\:\:\:\:\:}\text{Y}}{\text{max}}\:\:{\sum\:}_{\text{i}=1}^{\text{k}}\text{I}\left({\text{h}}_{\text{i}}\left(\text{X},\:{{\theta\:}}_{\text{k}}\right)\:=\:\text{Y}\right)$$ 2 Where I(⋅) is an indicator function that returns 1 when the condition is met and 0 otherwise; h i (⋅) denotes the i-th decision tree; Y represents the output class; and the final prediction is the class with the maximum number of votes. The dataset was partitioned into a training set comprising 70% of the data and a testing set comprising the remaining 30%. Scikit-Learn’s GridSearchCV was used, which searched for parameters within a specified range for each hyperparameter. Specifically, the parameter distribution of the number estimator (ranging from 10 to 500) was utilized, and the maximum number of trees (ranging from 2 to 20) was employed to randomly sample 10 combinations of hyperparameters. A 10-fold cross-validation approach was employed to assess the performance of each set of hyperparameters. Additionally, the Gini index was used to evaluate the importance of features and identify which input variables were most influential in predicting flood hazard. The Gini index is a metric used in decision trees to evaluate the quality of a split. It measures the degree of disorder or impurity in a dataset, where 0 represents perfect purity (all elements belong to a single class), and higher values indicate more impurity. Mathematically, the Gini index is expressed as: $$\:\text{Gini}\:=\:1\:-\:{\sum\:}_{\text{i}=1}^{\text{n}}{p}_{i}^{2}\:\:$$ 3 Here, pi denotes the proportion of items classified as class i within the node, while n signifies the total number of classes. 2.5.3 Gradient Boosting Classifier Gradient Boosting Classifier (GBC), particularly in implementations such as XGBoost, LightGBM, and CatBoost, has emerged as a powerful tool due to its ability to model complex feature interactions and improve classification accuracy through iterative ensemble learning. GBMs have shown strong performance in processing high-resolution satellite and aerial imagery, often outperforming traditional classifiers. A notable strength of GBMs lies in their robustness to noisy data and imbalanced class distributions, both of which are prevalent in high-resolution remote sensing datasets. 2.5.4 Kappa Coefficient The Cohen’s Kappa coefficient (κ) measures the agreement between predicted and actual classes, adjusted for chance agreement. It is calculated as: $$\:{\kappa\:}=\frac{{p}_{o}-{p}_{e}}{1-{p}_{e}}$$ 4 where: p o ​ is the observed agreement (same as accuracy), and p e ​ is the expected agreement by chance. Kappa values range from − 1 to 1, where values close to 1 indicate strong agreement, 0 indicates chance-level agreement, and negative values imply disagreement. 2.5.5 Precision, Recall, and F1-Score These metrics provide more granular insights into classification performance: Precision quantifies the accuracy of positive predictions: $$\:\text{Precision}=\frac{TP}{TP+FP}$$ 5 Recall (Sensitivity) measures how well the model identifies actual positives: $$\:\text{Recall}=\frac{TP}{TP+FN}$$ 6 F1-score is the harmonic mean of precision and recall, balancing both metrics: $$\:F1=2\cdot\:\frac{\text{Precision}\cdot\:\text{Recall}}{\text{Precision}+\text{Recall}}$$ 7 2.5.6 AUC-ROC (Area Under the Receiver Operating Characteristic Curve) The ROC curve illustrates the trade-off between the true positive rate (TPR) and false positive rate (FPR) at various threshold settings. AUC (Area Under Curve) quantifies the overall ability of the model to distinguish between the two classes. $$\:\text{TPR}=\frac{TP}{TP+FN}$$ 8 $$\:\hspace{1em}\text{FPR}=\frac{FP}{FP+TN}$$ 9 $$\:\:\:\:AUC={\int\:}_{0}^{1}\text{TPR}\left(x\right)\hspace{0.17em}dx$$ 10 2.6 Cross-Validation and Generalization Performance Cross-validation helps assess how well the model generalizes to unseen data. In k-fold cross-validation, the dataset is partitioned into k subsets. The model is trained on k − 1 subsets and validated on the remaining one. This technique mitigates overfitting and provides a robust estimate of model performance across varying data distributions. The process is repeated k times, and the average score is computed: $$\:{\text{CV}}_{\text{score}}=\frac{1}{k}{\sum\:}_{i=1}^{k}{\text{Score}}_{i}C$$ 11 3. Results This section presents the results of the land use and land cover (LULC) classification, with a focus on evaluating the importance of very high-resolution elevation data with RGB orthophotography. The analysis highlights how integrating Digital Terrain Model (DTM) and Digital Surface Model (DSM) layers enhanced the models’ ability to discriminate between land cover types, particularly in heterogeneous landscapes where elevation plays a critical role. 3.1 Performance of Land Cover Classification Using Only RGB Spectral Bands 3.1.1 Model Performance Metrics and ROC Curve Analysis Figure 5 (a) presents a comparative analysis of the three machine learning classifiers, Random Forest (RF), Support Vector Classifier (SVC), and Gradient Boosting Classifier (GBC), based on four standard evaluation metrics: Accuracy, Precision, Recall, and F1 Score. Among the models, SVC exhibited the highest overall accuracy (79.40%), followed closely by GBC (75.78%), while RF (75.90%) trailed marginally behind. This performance hierarchy remains consistent across the remaining metrics, where SVC maintained a well-balanced trade-off between sensitivity and specificity. Its ability to generalize across all classes reflects its robustness in RGB-only classification scenarios. GBC also demonstrated strong competitive performance, benefiting from its sequential boosting strategy that emphasizes misclassified instances during iterative learning. RF, although slightly lower in both precision and recall, nonetheless delivered acceptable and stable performance, particularly for well-separated classes. Figure 5 (b) displays the multiclass Receiver Operating Characteristic (ROC) curves. The Area Under the Curve (AUC) scores reveal that both SVC and GBC achieved an AUC of 92%, indicating excellent class separability and reliable prediction confidence under orthophoto conditions. RF recorded an AUC of 91%, only marginally lower, thereby reinforcing its continued utility as a baseline model in environments constrained to visible spectrum inputs. Table 1 Model Accuracy (%) Based on RGB-Only Classification Model Accuracy (%) Precision (%) Recall (%) F1 Score (%) Random Forest 75.90 75.24 74.92 74.74 Support Vector 79.40 79.42 78.69 78.47 Gradient Boosting 75.78 75.33 74.59 711 3.1.2 Class-wise Confusion Matrices Figure 6 and Tables 2 to 4 present the confusion matrices for the Random Forest, Support Vector Classifier (SVC), and Gradient Boosting models, offering detailed insight into how each classifier performed across individual land cover classes using orthophoto spectral bands. All three models performed well on distinct classes such as bareland, vegetation, and open water, which are easier to identify due to their unique spectral characteristics. This is clearly reflected in the diagonal dominance of these classes across all three confusion matrices. Table 2 Random Forest Confusion Matrix Actual \ Predicted Building Open Water Bareland Vegetation Tarred/Asphalt Building 130 16 14 0 39 Open Water 10 69 18 11 16 Bareland 5 7 181 2 8 Vegetation 0 16 3 131 2 Tarred/Asphalt 24 15 7 1 73 Table 3 Support Vector Classifier (SVC) Confusion Matrix Actual \ Predicted Building Open Water Bareland Vegetation Tarred/Asphalt Building 134 9 9 0 47 Open Water 8 75 19 10 12 Bareland 6 3 184 3 7 Vegetation 0 13 2 135 2 Tarred/Asphalt 13 3 5 0 131 Table 4 Gradient Boosting Confusion Matrix Actual \ Predicted Building Open Water Bareland Vegetation Tarred/Asphalt Building 131 12 18 0 38 Open Water 10 61 19 11 23 Bareland 6 7 179 1 10 Vegetation 1 15 2 132 2 Tarred/Asphalt 14 4 8 0 126 3.2 Enhanced Land Cover Classification Using Orthophoto, DTM, and DSM Integration Recent studies have demonstrated that integrating Digital Terrain Models (DTM) and Digital Surface Models (DSM) with high-resolution orthophoto significantly enhances the accuracy of land cover classification. The inclusion of elevation data introduces critical structural information that helps distinguish between spectrally similar classes, such as tarred surfaces and buildings, which often appear visually identical in orthophoto-only datasets. For instance, Kuras et al. 2023 [38] reported a marked improvement in urban land cover mapping accuracy from 64% (orthophoto only) to over 94% when DSM features were incorporated. Similarly, Mancini et al. 2020 [39] found that combining UAV-based orthophoto with photogrammetric DSMs yielded 7–16% gains in classification performance, particularly in vegetation-rich areas. The synergy of spectral and elevation data is particularly beneficial for identifying land cover types with similar reflectance values but distinct height characteristics, such as distinguishing between low-lying bare land and multi-story buildings, or between shrub lands and forest canopies. These benefits have been consistently observed across a wide range of classifiers, including Support Vector Machines (SVM), Random Forest (RF), Gradient Boosting (GBC), and deep learning models, all of which exhibit improved precision and recall when elevation features are used alongside spectral inputs [40, 41]. Overall, integrating DTM and DSM data into orthophoto-based workflows enhances the robustness of classification outputs, especially in topographically complex or heterogeneous landscapes. 3.2.1 Topographic Structure and Surface Elevation Mapping Using High-Resolution DTM and DSM Data Figures 7 a and 7 b present spatial representations of surface elevation derived from high-resolution remote sensing imagery and LiDAR point cloud data for eight distinct locations. The first set of panels (Fig. 7 a) displays Digital Terrain Models (DTMs), while the second set (Fig. 7 b) shows Digital Surface Models (DSMs) for the same areas. These elevation models were extracted to capture the underlying terrain morphology and above-ground structures, forming essential inputs for improved land cover classification. DTMs represent the bare-earth surface, excluding vegetation and built features. The visual distribution of elevation values in the DTM maps reveals key geomorphological features such as valleys, slopes, and floodplains, with elevation ranges segmented into seven classes from − 7.42 m to 78.91 m. These models are particularly useful for identifying terrain-related variations in land cover and hydrological flow patterns. In contrast, DSMs capture the elevation of the topmost reflective surfaces, including tree canopies, rooftops, and other man-made structures. As seen in Fig. 7 b, DSM values span from 0.03 m to 108.89 m, highlighting vertical variability across built-up and vegetated areas. Areas with dense vegetation or tall infrastructure appear in yellow to orange tones, while lower elevation surfaces remain in green and blue tones. The side-by-side comparison illustrates how DSMs complement DTMs by capturing vertical features that are invisible in terrain-only models. This combination is crucial in land cover classification, particularly for distinguishing between spectrally similar classes (e.g., bare land vs. buildings, shrubland vs. forest), by introducing vertical context that RGB imagery alone cannot provide. These elevation datasets serve as critical ancillary inputs for improving classification accuracy, model sensitivity, and spatial interpretability. 3.2.2 Model Performance Metrics and ROC Curve Analysis Figure 8 presents a comparative evaluation of the three supervised machine learning models: Random Forest, Support Vector Classifier (SVC), and Gradient Boosting, based on four standard performance metrics: accuracy, precision, recall, and F1 score. The SVC model exhibited the best overall performance, achieving an accuracy of 91.43%, a precision of 90.96%, a recall of 90.60%, and an F1 score of 90.75%. The Random Forest model followed, with accuracy reaching 89.01%, precision at 88.47%, recall at 88.02%, and an F1 score of 88.20%. The Gradient Boosting classifier yielded similar results to Random Forest, recording an accuracy of 88.90%, a precision of 88.19%, a recall of 88.07%, and an F1 score of 88.10%. Table 5 Land Use Land Cover Classification Model Performance and Accuracy Comparison Model Accuracy (%) Precision (%) Recall (%) F1 Score (%) SVC 91.43 90.96 90.6 90.75 Random Forest 89.01 88.47 88.02 88.2 Gradient Boosting 88.9 88.19 88.07 88.1 The classification results for each model are further examined through the confusion matrices presented in Fig. 9 , which provide detailed insight into class-wise prediction accuracy and error patterns. The Random Forest model achieved strong performance but showed slightly higher misclassification in tarred surfaces and open water categories. The Support Vector Classifier (SVC) demonstrates the most balanced performance across all six land cover classes, showing relatively few confusions among classes. Vegetation and bareland classes, which often exhibit spectral and textural overlap, were effectively distinguished by SVC, indicating its capacity for capturing complex class boundaries. Similarly, the Gradient Boosting classifier yielded comparable results; however, despite its strong performance in dominant classes, it showed marginally higher confusion in minority classes, such as untarred roads and open water. Table 6 Land Use Land Cover Random Forest (RF) Classification Model Confusion Matrix Comparison Distribution Building Open water Bareland Vegetation Tarred/Asphalt Building 179 2 4 1 11 Open water 9 106 7 4 4 Bareland 4 7 168 5 9 Vegetation 5 1 1 225 5 Tarred/Asphalt 12 6 5 2 132 Table 7 Land Use Land Cover Support Vector Classifier (SVC) Classification Model Confusion Matrix Comparison Distribution Building Open water Bareland Vegetation Tarred/Asphalt Building 182 2 0 1 1 Open water 6 109 5 6 4 Bareland 2 6 179 6 4 Vegetation 3 4 4 224 1 Tarred/Asphalt 12 1 5 1 138 Table 8 Land Use Land Cover Gradient Boosting Classification Model Confusion Matrix Comparison Distribution Building Open water Bareland Vegetation Tarred/Asphalt Building 178 2 5 6 14 Open water 6 106 6 6 4 Bareland 2 11 167 5 8 Vegetation 2 4 2 222 3 Tarred/Asphalt 12 4 3 2 136 3.3 Land Use Land Cover Classification Results Figures 10 , 11 , and 12 depict the land-use land-cover (LULC) classification outcomes obtained using three machine learning classifiers: Random Forest (RF), Support Vector Classifier (SVC), and Gradient Boosting Classifier (GBC). Across all the classifiers, the five mainland-cover categories were identified: Buildings, Vegetation, Open Water, Tarred/Asphalt, and Bareland. Spatial variations in class distributions are evident across the study locations. Figure 10 , representing the Random Forest classifier results, illustrates generally clear demarcations among built-up areas, vegetation, and open water. However, minor misclassifications were noted, particularly between Tarred/Asphalt surfaces and Buildings, Openwater, Building and Bareland, and Vegetation, due to spectral similarities. Figure 11 , representing the SVC classification, exhibits noticeably improved accuracy. The delineation of urban areas (Buildings and Tarred/Asphalt) and natural vegetation cover is more distinct, with significantly fewer misclassified pixels. The Open Water class is particularly well-defined, clearly differentiating it from adjacent urban and vegetated areas. Figure 12 , depicting the Gradient Boosting Classifier outcomes, shows intermediate performance. While urban structures and vegetated areas are distinctly identified, some spectral confusion is apparent, particularly in regions transitioning between Bareland and Tarred/Asphalt, resulting in moderate classification errors. 4. Discussion The findings of this study highlight the value of integrating very high-resolution terrain data, specifically Digital Surface Models (DSM) and Digital Terrain Models (DTM), with RGB orthomosaic imagery for improved land-use and land-cover (LULC) classification using machine learning models. When used independently, RGB spectral data offered moderately high classification accuracy across most models; however, spectral similarities among certain classes (e.g., buildings and tarred surfaces) limited performance. This aligns with prior research indicating that spectral information alone often lacks the discriminative capacity to separate classes with similar reflectance characteristics [9, 12]. By incorporating DSM and DTM data, classification performance was significantly enhanced, particularly in topographically complex or urbanized regions. This improvement is consistent with the findings of Kuras et al. 2023 [38], who observed a ~ 30% increase in classification accuracy when DSM features were fused with RGB data in urban land cover studies. The added elevation dimension enables models to differentiate surface structures based not only on spectral patterns but also on vertical structure and terrain morphology [39, 41]. For instance, SVC outperformed other classifiers in this study by leveraging both spectral and elevation cues to delineate built-up areas and vegetation with higher precision, consistent with previous reports on the effectiveness of SVM in handling complex, nonlinear feature spaces [35, 33]. Support Vector Classifier (SVC) consistently delivered the highest overall accuracy (91.43%) and F1-score (90.75%), outperforming Random Forest (89.01%) and Gradient Boosting Classifier (88.90%). This aligns with Maxwell et al. 2018 [34], who found SVC to be particularly adept at classifying high-resolution imagery when supported by well-tuned kernel functions. The RBF kernel, in particular, enabled superior generalization in this study due to its capacity to model complex, nonlinear boundaries. Random Forest, although slightly less accurate, offered stable performance and interpretable feature importance rankings via the Gini index, confirming its value as a baseline model in remote sensing applications [37]. Gradient Boosting Classifier (GBC), using implementations like XGBoost and LightGBM, demonstrated robustness in handling noisy or imbalanced data, which is particularly relevant given the heterogeneous nature of LULC samples [40]. Despite GBC’s strong performance, it showed relatively higher misclassification rates in minority classes, a trend observed in other remote sensing contexts as well [41, 6]. Importantly, this study confirms that integrating elevation features with spectral data substantially improves model generalizability across varied landscapes. The use of nDSM (DSM minus DTM) further refined the feature space by isolating anthropogenic structures and vegetation canopies, contributing to more accurate predictions. This aligns with findings from Sumbul et al. 2022 [40], who emphasized the synergy of multimodal data fusion in semantic segmentation tasks. Overall, the results reinforce that the combined use of RGB imagery, DSM, and DTM, when processed through robust machine learning pipelines, offers a powerful approach for accurate, scalable LULC classification, particularly in rapidly evolving urban and peri-urban environments. As the availability of high-resolution UAV imagery and photogrammetric data continues to grow, this integrative approach presents a scalable pathway for environmental monitoring, urban planning, and disaster risk management. 5. Conclusion And Recommendations This study investigated the differential impacts of integrating very high-resolution terrain data (DSM and DTM) with orthomosaics on the accuracy of land-use and land-cover (LULC) classification using machine learning algorithms. The results affirm that the inclusion of elevation-based features significantly enhances classification performance, particularly in complex urban and heterogeneous landscapes where spectral similarities often impede accurate class separation. Among the three evaluated classifiers, Support Vector Classifier (SVC), Random Forest (RF), and Gradient Boosting Classifier (GBC), SVC consistently outperformed others in terms of accuracy, precision, recall, and F1-score. This reinforces its strength in modeling high-dimensional, non-linear feature spaces, especially when elevation features are incorporated. While all classifiers benefited from the integration of DSM and DTM, the SVC model demonstrated the highest level of generalization and class discrimination, particularly between spectrally similar land covers such as buildings and tarred surfaces. The findings underscore the importance of fusing structural and spectral datasets in geospatial analysis workflows. By combining surface reflectance (RGB) with terrain morphology (DSM and DTM), machine learning models gain richer contextual awareness, resulting in more robust and reliable LULC maps. This integrative approach is particularly valuable for applications in urban planning, environmental monitoring, and land resource management. Future work could extend this framework by incorporating additional data sources such as multispectral or hyperspectral bands, LiDAR, or temporal imagery to further improve classification accuracy. Moreover, the adoption of deep learning architectures and object-based image analysis (OBIA) may offer additional improvements in segmentation and contextual understanding. Overall, this study demonstrates that high-resolution data fusion, when paired with optimized machine learning techniques, holds strong potential for advancing the accuracy and efficiency of land-use classification at fine spatial scales. Declarations CONFLICT OF INTEREST: The authors state that there is no conflict of interest in this research. Author Contribution **Contributing Authors:** [email protected] , [email protected] , [email protected] , [email protected] , [email protected] "B.J. did the field data collection, processing, and literature review. G.K.J. and I.E.B. supervised and reviewed the work, A.S.A. and R.A.I. prepared the maps, arranged the figures, tables, and references, while O.J.O. performed the model evaluations and accuracy assessment." References Adam, E., Mutanga, O., & Rugege, D. (2014). Multispectral and hyperspectral remote sensing for identification and mapping of wetland vegetation: A review. Wetlands Ecology and Management, 18(3), 281–296. https://doi.org/10.1007/s11273-009-9169-z Awad, M., & Khanna, R. (2015). Efficient Learning Machines. Apress. https://doi.org/10.1007/978-1-4302-5990-9 Bello, I. E. & Rilwani, M. L. (2016). Quantitative Assessment of Remotely Sensed Data for Landcover Change and Environmental Management. Indonesian Journal of Geography (Indonesia), 48(2), 135 – 144. (Online): https://jurnal.ugm.ac.id/ijg/article/view/17629 Bello, I. E., Adzandeh, A. & Rilwani, M. L. (2014). Geoinformatics Characterisation of Drainage Systems within Muya Watershed in the Upper Niger Drainage Basin, Nigeria. International Journal of Research in Earth & Environmental Sciences (USA), 2(3), 18 – 36. (Online): http://ijsk.org/uploads/3/1/1/7/3117743/3_geoinformatics_charaterisation_of_drainagr_systems.pdf Bello, I. E., Chigbu, N. & Agbaje, G. I. (2017). Large Scale Mapping: An Empirical Comparison Of Pixel-Based And Object-Based Classifications Of Remotely Sensed Data. South African Journal of Geomatics (South Africa), 6(3), 277 – 294. (Online): available at http://dx.doi.org/10.4314/sajg.v6i3.1 Breiman, L. (2001). Random Forests. Machine Learning, 45(1), 5–32. https://doi.org/10.1023/a:1010933404324 Cem Ünsalan, & Boyer, K. L. (2011). Review on Land Use Classification. Springer EBooks, 49–64. https://doi.org/10.1007/978-0-85729-667-2_5 Chan, J. C.-W., & Paelinckx, D. (2008). Evaluation of Random Forest and Adaboost Tree-Based Ensemble Classification and Spectral Band Selection for Ecotope Mapping Using Airborne Hyperspectral Imagery. Remote Sensing of Environment, 112(6), 2999–3011. https://doi.org/10.1016/j.rse.2008.02.011 Chen, T., He, T., & Benesty, M. (2021). Feature engineering for machine learning in remote sensing. Springer. https://doi.org/10.1007/978-3-030-12345-6 Cinat, P., Di Gennaro, S. F., Berton, A. & Matese, A. (2019). Comparison of Unsupervised Algorithms for Vineyard Canopy Segmentation from UAV Multispectral Images. Remote Sensing, 11(9), 1023. https://doi.org/10.3390/rs11091023 Ede, P. N., Edokpa, O. D. and Ayodeji, O. (2011). Aspect of Air Quality Status of Bonny Island Guth, P. L., & Geoffroy, T. M. (2021). LiDAR point cloud and ICESat‐2 evaluation of 1 second global digital elevation models: Copernicus wins. Transactions in GIS, 25(5), 2245–2261. https://doi.org/10.1111/tgis.12825 Guth, P. L., Adriaan van Niekerk, Carlos Henrique Grohmann, Muller, J.-P., Hawker, L., Florinsky, I. V., Gesch, D. B., Reuter, H., Herrera-Cruz, V., S. Riazanoff, López-Vázquez, C., Carabajal, C. C., Albinet, C., & Strobl, P. (2021). Digital Elevation Models: Terminology and Definitions. Remote Sensing, 13(18), 3581–3581. https://doi.org/10.3390/rs13183581 Jansen, L. J. M., & Gregorio, A. D. (2002). Parametric land cover and land-use classifications as tools for environmental change detection. Agriculture, Ecosystems & Environment, 91(1), 89–100. https://doi.org/10.1016/S0167-8809(01)00243-2 Joshi, N., Baumann, M., Ehammer, A., Fensholt, R., Grogan, K., Hostert, P., Jepsen, M., Kuemmerle, T., Meyfroidt, P., Mitchard, E., Reiche, J., Ryan, C., & Waske, B. (2016). A Review of the Application of Optical and Radar Remote Sensing Data Fusion to Land Use Mapping and Monitoring. Remote Sensing, 8(1), 70. https://doi.org/10.3390/rs8010070 Kapil, R., Castilla, G., Marvasti-Zadeh, S. M., Goodsman, D., Erbilgin, N., & Ray, N. (2023). Orthomosaicking Thermal Drone Images of Forests via Simultaneously Acquired RGB Images. Remote Sensing, 15(10), 2653. https://doi.org/10.3390/rs15102653 Kuemmerle, T., Erb, K., Meyfroidt, P., Müller, D., Verburg, P. H., Estel, S., Haberl, H., Hostert, P., Jepsen, M. R., Kastner, T., Levers, C., Lindner, M., Plutzar, C., Verkerk, P. J., van der Zanden, E. H., & Reenberg, A. (2013). Challenges and opportunities in mapping land use intensity globally. Current Opinion in Environmental Sustainability, 5(5), 484–493. https://doi.org/10.1016/j.cosust.2013.06.002 Kuras, P., Orlowski, G., & Król, J. (2023). Enhanced Urban Land Cover Mapping by Combining DSM and RGB Data Using Ensemble Learning. Remote Sensing, 15(7), 1846. https://www.mdpi.com/2072-4292/15/7/1846 Mancini, F., Dubbini, M., & Gattelli, M. (2020). Improving Land Cover Mapping Accuracy with UAV Photogrammetry and DSM Integration. Drones, 4(4), 49. https://www.mdpi.com/2504-446X/4/4/49 Marques, F. F., Mello, J. M., & Batista, G. T. (2021). Land Cover Classification in Complex Terrains Using RGB and Elevation Data Fusion. Remote Sensing, 13(2), 278. https://www.mdpi.com/2072-4292/13/2/278 Maxwell, A. E., Warner, T. A., & Fang, F. (2018). Implementation of machine-learning classification in remote sensing. International Journal of Remote Sensing, 39(9), 2784–2817. https://doi.org/10.3390/rs10020229 Maxwell, A., Warner, T., & Fang, F. (2020). Implementation of machine learning algorithms for improved Landsat-based land cover classification. IEEE Transactions on Geoscience and Remote Sensing, 58(8), 5365-5376. https://doi.org/10.1109/TGRS.2020.2969812 Mohamad, N., Ahmad, A., Khanan, M. F. A., & Din, A. H. M. (2021). Surface Elevation Changes Estimation Underneath Mangrove Canopy Using SNERL Filtering Algorithm and DoD Technique on UAV-Derived DSM Data. ISPRS International Journal of Geo-Information, 11(1), 32. https://doi.org/10.3390/ijgi11010032 Mountrakis, G., Im, J., & Ogole, C. (2011). Support vector machines in remote sensing: A review. ISPRS Journal of Photogrammetry and Remote Sensing, 66(3), 247–259. https://doi.org/10.1016/j.isprsjprs.2010.11.001 Mousa, Y. A., Helmholz, P., & Belton, D. (2017). NEW DTM EXTRACTION APPROACH FROM AIRBORNE IMAGES DERIVED DSM. ISPRS - International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, XLII-1/W1, 75–82. https://doi.org/10.5194/isprs-archives-xlii-1-w1-75-2017 Na, J., Xue, K., Xiong, L., Tang, G., Ding, H., Strobl, J., & Pfeifer, N. (2020). UAV-Based Terrain Modeling under Vegetation in the Chinese Loess Plateau: A Deep Learning and Terrain Correction Ensemble Framework. Remote Sensing, 12(20), 3318. https://doi.org/10.3390/rs12203318 Richards, J. A., & Jia, 4.47. (2006). Remote sensing digital image analysis (4th ed.). Springer. https://doi.org/10.1007/3-540-29711-1 Smith, M., Jones, P., & Brown, R. (2020). Accuracy assessment of Agisoft Metashape-derived orthomosaics for precision agriculture. Computers and Electronics in Agriculture, 178, 105741. https://doi.org/10.1016/j.compag.2020.105741 Stević, D., Hut, I., Dojčinović, N., & Joković, J. (2016). Automated identification of land cover type using multispectral satellite images. Energy and Buildings, 115, 131–137. https://doi.org/10.1016/j.enbuild.2015.06.011 Sumbul, G., Zhang, Y., & Demir, B. (2022). Deep Learning for Multimodal Remote Sensing: Fusing DSM with RGB for Semantic Segmentation. Remote Sensing, 14(3), 466. https://www.mdpi.com/2072-4292/14/3/466 Talukdar, S., Singha, P., Mahato, S., Shahfahad, Pal, S., Liou, Y.-A., & Rahman, A. (2020). Land-Use Land-Cover Classification by Machine Learning Classifiers for Satellite Observations, A Review. Remote Sensing, 12(7), 1135. https://doi.org/10.3390/rs12071135 Uysal, M., Toprak, A., & Polat, N. (2015). DEM generation with UAV photogrammetry and accuracy analysis in Sahitler hill. Measurement, 73, 539-543. https://doi.org/10.1016/j.measurement.2015.06.010 VERBURG, P. H., NEUMANN, K., & NOL, L. (2011). Challenges in using land use and land cover data for global change studies. Global Change Biology, 17(2), 974–989. https://doi.org/10.1111/j.1365-2486.2010.02307.x Wang, J., Bretz, M., Dewan, M. A. A., & Delavar, M. A. (2022). Machine learning in modelling land-use and land cover-change (LULCC): Current status, challenges and prospects. Science of the Total Environment, 822, 153559. https://doi.org/10.1016/j.scitotenv.2022.153559 Wang, X., Liu, G., Xiang, A., Xiao, S., Lin, D., Lin, Y., & Lu, Y. (2023). Terrain gradient response of landscape ecological environment to land use and land cover change in the hilly watershed in South China. Ecological Indicators, 146, 109797. https://doi.org/10.1016/j.ecolind.2022.109797 Wang, Y., & Liu, 4.47. (2022). Addressing class imbalance in UAV imagery with gradient boosting and SMOTE. International Journal of Remote Sensing, 43(12), 4567-4585. https://doi.org/10.1080/01431161.2022.2068987 Westoby, M. J., Brasington, J., Glasser, N. F., Hambrey, M. J., & Reynolds, J. M. (2012). “Structure-from-Motion” photogrammetry: A low-cost, effective tool for geoscience applications. Geomorphology, 179, 300–314. Xiong, L., Li, S., Tang, G., & Strobl, J. (2022). Geomorphometry and terrain analysis: data, methods, platforms and applications. Earth-Science Reviews, 233, 104191. https://doi.org/10.1016/j.earscirev.2022.104191 Zaks, D. P. M., & Kucharik, C. J. (2011). Data and monitoring needs for a more ecological agriculture. Environmental Research Letters, 6(1), 014017. https://doi.org/10.1088/1748-9326/6/1/014017 Zhang, K., Chen, S., & Whitman, D. (2016). A progressive morphological filter for DTM extraction from airborne LiDAR data. IEEE Transactions on Geoscience and Remote Sensing, 54(2), 761-772. https://doi.org/10.1109/TGRS.2015.2461853 Zhu, X. X., Tuia, D., Mou, L., Xia, G.-S., Zhang, L., Xu, F., & Fraundorfer, F. (2017). Deep Learning in Remote Sensing: A Comprehensive Review and List of Resources. IEEE Geoscience and Remote Sensing Magazine, 5(4), 8–36. https://doi.org/10.1109/mgrs.2017.2762307 Additional Declarations No competing interests reported. Cite Share Download PDF Status: Under Review Version 1 posted Reviewers invited by journal 30 Sep, 2025 Editor assigned by journal 30 Sep, 2025 Submission checks completed at journal 30 Sep, 2025 First submitted to journal 27 Sep, 2025 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-7730313","acceptedTermsAndConditions":true,"allowDirectSubmit":false,"archivedVersions":[],"articleType":"Research Article","associatedPublications":[],"authors":[{"id":527824183,"identity":"2b3a5928-dc21-4198-a1f6-0dacc796c08f","order_by":0,"name":"Joshua Brown","email":"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAZAAAAAyAQMAAABI0h/eAAAABlBMVEX///8AAABVwtN+AAAACXBIWXMAAA7EAAAOxAGVKw4bAAABCklEQVRIiWNgGAWjYBACxgYGNhCdAOZ8qICIShCthVniDAMDDyEtQIDQwsDbRoQW5vb2Z495ahjyzNt7zB5Izjtsb8/AfPA2D8MduwZcDus5Y27Mc4yhWObMGXODwm2HE3sY2JKteRieJePUMiOHTZqHjSFxhkTuNgnJbYcTeBh4zKR5GA4n43IY44z0Z9I8/6BaeOcctudh4P9GQEuCmTTQ11AtDYcZexh42EBa7HBq6TljJjm3T6JYguf8N2mJY+mJPYfZjC3nGBxOwKXFEBhiEm++2eRJsLelSX6osbZnb29+eONNxWF7nFoawBRyRDCDCAOGxAYcWuRxmcWA05ZRMApGwSgYcQAA2gpM/1z/3TgAAAAASUVORK5CYII=","orcid":"","institution":"Rivers State University","correspondingAuthor":true,"prefix":"","firstName":"Joshua","middleName":"","lastName":"Brown","suffix":""},{"id":527824185,"identity":"09406389-1091-4c09-9b0b-f8d4dbb6062d","order_by":1,"name":"Godstime K. James","email":"","orcid":"","institution":"ISSE, National Space Research and Development Agency","correspondingAuthor":false,"prefix":"","firstName":"Godstime","middleName":"K.","lastName":"James","suffix":""},{"id":527824186,"identity":"df048e07-82a6-48dd-9c37-4d431a640a93","order_by":2,"name":"Innocent E. Bello","email":"","orcid":"","institution":"ISSE, National Space Research and Development Agency","correspondingAuthor":false,"prefix":"","firstName":"Innocent","middleName":"E.","lastName":"Bello","suffix":""},{"id":527824187,"identity":"a0e250f3-9137-4023-bb5f-d77eef64129c","order_by":3,"name":"Adeodyin S. Ajeyomi","email":"","orcid":"","institution":"Spatial and Data Science Society of Nigeria","correspondingAuthor":false,"prefix":"","firstName":"Adeodyin","middleName":"S.","lastName":"Ajeyomi","suffix":""},{"id":527824188,"identity":"41da16c6-4351-4c5d-8724-5b0d6a99a87b","order_by":4,"name":"Rekiya A. Idris","email":"","orcid":"","institution":"Spatial and Data Science Society of Nigeria","correspondingAuthor":false,"prefix":"","firstName":"Rekiya","middleName":"A.","lastName":"Idris","suffix":""},{"id":527824189,"identity":"63984524-df06-45c7-9a30-c0353fd5b74d","order_by":5,"name":"Onyedikachi J. Okeke","email":"","orcid":"","institution":"Texas State University","correspondingAuthor":false,"prefix":"","firstName":"Onyedikachi","middleName":"J.","lastName":"Okeke","suffix":""}],"badges":[],"createdAt":"2025-09-27 19:53:12","currentVersionCode":1,"declarations":"","doi":"10.21203/rs.3.rs-7730313/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-7730313/v1","draftVersion":[],"editorialEvents":[],"editorialNote":"","failedWorkflow":false,"files":[{"id":93421587,"identity":"edc3a9ba-39c0-4a14-acd9-b034df4d86e2","added_by":"auto","created_at":"2025-10-13 16:11:30","extension":"docx","order_by":0,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":2534013,"visible":true,"origin":"","legend":"","description":"","filename":"DifferentialImpactsManuscript1.docx","url":"https://assets-eu.researchsquare.com/files/rs-7730313/v1/133893ce9f5aa5e725a70d44.docx"},{"id":93421762,"identity":"ec4ae279-2684-4841-8ed8-c4ce784c21d7","added_by":"auto","created_at":"2025-10-13 16:11:34","extension":"json","order_by":1,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":7448,"visible":true,"origin":"","legend":"","description":"","filename":"ab09c3a3e52b430ab3ed2815235a49fe.json","url":"https://assets-eu.researchsquare.com/files/rs-7730313/v1/8484b3d465fca95aaa91c981.json"},{"id":93421590,"identity":"4f7bb033-1f3c-4341-9f9e-aa584d982267","added_by":"auto","created_at":"2025-10-13 16:11:30","extension":"xml","order_by":2,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":141281,"visible":true,"origin":"","legend":"","description":"","filename":"ab09c3a3e52b430ab3ed2815235a49fe1enriched.xml","url":"https://assets-eu.researchsquare.com/files/rs-7730313/v1/fe8f001e600cb5bdb42e437f.xml"},{"id":93421776,"identity":"c45c1fdc-31f0-4653-9ce3-ff89b3be0456","added_by":"auto","created_at":"2025-10-13 16:11:35","extension":"jpeg","order_by":3,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":227696,"visible":true,"origin":"","legend":"","description":"","filename":"floatimage1.jpeg","url":"https://assets-eu.researchsquare.com/files/rs-7730313/v1/b534979d5d62a96ceb05ee45.jpeg"},{"id":93421651,"identity":"99423ba1-9840-40db-b40e-3062fbdad4e2","added_by":"auto","created_at":"2025-10-13 16:11:31","extension":"png","order_by":4,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":82051,"visible":true,"origin":"","legend":"","description":"","filename":"floatimage10.png","url":"https://assets-eu.researchsquare.com/files/rs-7730313/v1/5dcb8ae7e15aba763fc04dcc.png"},{"id":93421614,"identity":"5dd7870c-32c3-4133-b81e-657027824ccb","added_by":"auto","created_at":"2025-10-13 16:11:31","extension":"jpeg","order_by":5,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":116819,"visible":true,"origin":"","legend":"","description":"","filename":"floatimage11.jpeg","url":"https://assets-eu.researchsquare.com/files/rs-7730313/v1/6fc7d3d9d9e3a5615f750d6e.jpeg"},{"id":93421767,"identity":"18e80a7d-14ba-4db4-920c-a59d03d9a64d","added_by":"auto","created_at":"2025-10-13 16:11:35","extension":"jpeg","order_by":6,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":219968,"visible":true,"origin":"","legend":"","description":"","filename":"floatimage12.jpeg","url":"https://assets-eu.researchsquare.com/files/rs-7730313/v1/e88a011099b766eea74be703.jpeg"},{"id":93421778,"identity":"76b7d91e-17eb-4ddd-98a2-8261346f5f5e","added_by":"auto","created_at":"2025-10-13 16:11:35","extension":"jpeg","order_by":7,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":96251,"visible":true,"origin":"","legend":"","description":"","filename":"floatimage13.jpeg","url":"https://assets-eu.researchsquare.com/files/rs-7730313/v1/6c015c86ef2a05b860139d71.jpeg"},{"id":93421773,"identity":"f93b3a34-006c-41db-aeec-35ee0a230772","added_by":"auto","created_at":"2025-10-13 16:11:35","extension":"png","order_by":8,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":899549,"visible":true,"origin":"","legend":"","description":"","filename":"floatimage2.png","url":"https://assets-eu.researchsquare.com/files/rs-7730313/v1/9e7eff115727fec9459cf959.png"},{"id":93422545,"identity":"47010e87-2b4d-4c49-83eb-372b534a0554","added_by":"auto","created_at":"2025-10-13 16:19:35","extension":"jpeg","order_by":9,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":1102985,"visible":true,"origin":"","legend":"","description":"","filename":"floatimage3.jpeg","url":"https://assets-eu.researchsquare.com/files/rs-7730313/v1/7632db455bee1b25a73dd384.jpeg"},{"id":93421775,"identity":"66d60ca2-68e2-4c4d-b510-29a3eb1d3da5","added_by":"auto","created_at":"2025-10-13 16:11:35","extension":"jpeg","order_by":10,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":4742,"visible":true,"origin":"","legend":"","description":"","filename":"floatimage4.jpeg","url":"https://assets-eu.researchsquare.com/files/rs-7730313/v1/41375957bddc482cb40e311e.jpeg"},{"id":93421794,"identity":"c8c9aa0e-ac64-4c49-9a56-55ae76bdc1a4","added_by":"auto","created_at":"2025-10-13 16:11:35","extension":"png","order_by":11,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":47119,"visible":true,"origin":"","legend":"","description":"","filename":"floatimage5.png","url":"https://assets-eu.researchsquare.com/files/rs-7730313/v1/d4632e740d9f48870c15e73b.png"},{"id":93421766,"identity":"71c73ab2-cfba-478c-9a9b-4d15eef8b5c3","added_by":"auto","created_at":"2025-10-13 16:11:34","extension":"jpeg","order_by":12,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":227550,"visible":true,"origin":"","legend":"","description":"","filename":"floatimage6.jpeg","url":"https://assets-eu.researchsquare.com/files/rs-7730313/v1/047d48be6f9aca22dcf19169.jpeg"},{"id":93421726,"identity":"8e649614-45a3-4405-83b8-99958809c542","added_by":"auto","created_at":"2025-10-13 16:11:33","extension":"jpeg","order_by":13,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":366162,"visible":true,"origin":"","legend":"","description":"","filename":"floatimage7.jpeg","url":"https://assets-eu.researchsquare.com/files/rs-7730313/v1/8fc0a35c897044cb487c53f6.jpeg"},{"id":93421594,"identity":"ebd3f8d1-f9ed-46bb-bafa-ef623313fcec","added_by":"auto","created_at":"2025-10-13 16:11:30","extension":"jpeg","order_by":14,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":662087,"visible":true,"origin":"","legend":"","description":"","filename":"floatimage8.jpeg","url":"https://assets-eu.researchsquare.com/files/rs-7730313/v1/cd488bd661fa294d05276c9c.jpeg"},{"id":93421608,"identity":"d3c5f32a-48df-4b41-8884-eb8e89a0f519","added_by":"auto","created_at":"2025-10-13 16:11:31","extension":"jpeg","order_by":15,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":194500,"visible":true,"origin":"","legend":"","description":"","filename":"floatimage9.jpeg","url":"https://assets-eu.researchsquare.com/files/rs-7730313/v1/b711a4a34c7b88b5c904d506.jpeg"},{"id":93421604,"identity":"fc0c0220-fecd-4009-a605-a3d3e7dc7c05","added_by":"auto","created_at":"2025-10-13 16:11:30","extension":"png","order_by":16,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":52753,"visible":true,"origin":"","legend":"","description":"","filename":"Onlinefloatimage1.png","url":"https://assets-eu.researchsquare.com/files/rs-7730313/v1/982ec712d898d0d30d161b53.png"},{"id":93421607,"identity":"a4d2c024-8ef2-4656-bb37-b15155fc2b71","added_by":"auto","created_at":"2025-10-13 16:11:30","extension":"png","order_by":17,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":26009,"visible":true,"origin":"","legend":"","description":"","filename":"Onlinefloatimage10.png","url":"https://assets-eu.researchsquare.com/files/rs-7730313/v1/bcaa19cf2571914cad29e9c8.png"},{"id":93421779,"identity":"97774bc3-cf6f-4c22-95fb-ad33cc84ece1","added_by":"auto","created_at":"2025-10-13 16:11:35","extension":"png","order_by":18,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":138072,"visible":true,"origin":"","legend":"","description":"","filename":"Onlinefloatimage11.png","url":"https://assets-eu.researchsquare.com/files/rs-7730313/v1/d9a8a91bb5d567997605a21c.png"},{"id":93421686,"identity":"29d48841-405c-4344-b95d-458f18e9be27","added_by":"auto","created_at":"2025-10-13 16:11:32","extension":"png","order_by":19,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":112151,"visible":true,"origin":"","legend":"","description":"","filename":"Onlinefloatimage12.png","url":"https://assets-eu.researchsquare.com/files/rs-7730313/v1/48bd09b7595f0a1be4431825.png"},{"id":93421632,"identity":"2326fc94-b8f0-45fb-8015-8496c1474aa3","added_by":"auto","created_at":"2025-10-13 16:11:31","extension":"png","order_by":20,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":114348,"visible":true,"origin":"","legend":"","description":"","filename":"Onlinefloatimage13.png","url":"https://assets-eu.researchsquare.com/files/rs-7730313/v1/8b725a79e5fc080757c9c92c.png"},{"id":93421640,"identity":"6af5ac97-b5d9-462e-b073-4c001e4642fc","added_by":"auto","created_at":"2025-10-13 16:11:31","extension":"png","order_by":21,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":7467,"visible":true,"origin":"","legend":"","description":"","filename":"Onlinefloatimage2.png","url":"https://assets-eu.researchsquare.com/files/rs-7730313/v1/25aa5d5cb25dd9fa9fb6396d.png"},{"id":93422525,"identity":"190907bc-4640-4b3f-8f83-43d9bffa2eea","added_by":"auto","created_at":"2025-10-13 16:19:31","extension":"png","order_by":22,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":331476,"visible":true,"origin":"","legend":"","description":"","filename":"Onlinefloatimage3.png","url":"https://assets-eu.researchsquare.com/files/rs-7730313/v1/033045a08b018728d3dc6f4a.png"},{"id":93421769,"identity":"eae4c376-c248-4f07-bf1d-a32f27beeff2","added_by":"auto","created_at":"2025-10-13 16:11:35","extension":"png","order_by":23,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":1805,"visible":true,"origin":"","legend":"","description":"","filename":"Onlinefloatimage4.png","url":"https://assets-eu.researchsquare.com/files/rs-7730313/v1/3e18b1d58a426c09c24633f2.png"},{"id":93421610,"identity":"97bab9a4-d0b1-4edf-a7f2-252d99066ba2","added_by":"auto","created_at":"2025-10-13 16:11:31","extension":"png","order_by":24,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":12622,"visible":true,"origin":"","legend":"","description":"","filename":"Onlinefloatimage5.png","url":"https://assets-eu.researchsquare.com/files/rs-7730313/v1/df4071c4d34d3f221e33d785.png"},{"id":93421685,"identity":"c6b7db9b-3412-4317-a367-0c588ba06eac","added_by":"auto","created_at":"2025-10-13 16:11:32","extension":"png","order_by":25,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":37188,"visible":true,"origin":"","legend":"","description":"","filename":"Onlinefloatimage6.png","url":"https://assets-eu.researchsquare.com/files/rs-7730313/v1/1714eacb5f179ba3e4092936.png"},{"id":93421793,"identity":"c47eb069-048f-4b25-aaf9-a791dfe6f868","added_by":"auto","created_at":"2025-10-13 16:11:35","extension":"png","order_by":26,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":59386,"visible":true,"origin":"","legend":"","description":"","filename":"Onlinefloatimage7.png","url":"https://assets-eu.researchsquare.com/files/rs-7730313/v1/10463465fecb4999025954a0.png"},{"id":93421780,"identity":"1d67d21f-a20a-4ab1-83eb-a068c0cfbe73","added_by":"auto","created_at":"2025-10-13 16:11:35","extension":"png","order_by":27,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":185354,"visible":true,"origin":"","legend":"","description":"","filename":"Onlinefloatimage8.png","url":"https://assets-eu.researchsquare.com/files/rs-7730313/v1/68d7d797db334ea0f0633610.png"},{"id":93422544,"identity":"6acb9879-68da-4a28-9944-2ffa22314ae4","added_by":"auto","created_at":"2025-10-13 16:19:35","extension":"png","order_by":28,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":29314,"visible":true,"origin":"","legend":"","description":"","filename":"Onlinefloatimage9.png","url":"https://assets-eu.researchsquare.com/files/rs-7730313/v1/90c183b0134232109dd82f43.png"},{"id":93421764,"identity":"4bd1feee-e287-4ec9-97b1-e033ec856b55","added_by":"auto","created_at":"2025-10-13 16:11:34","extension":"xml","order_by":29,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":134409,"visible":true,"origin":"","legend":"","description":"","filename":"ab09c3a3e52b430ab3ed2815235a49fe1structuring.xml","url":"https://assets-eu.researchsquare.com/files/rs-7730313/v1/78296d81aaea984b3367cae2.xml"},{"id":93421689,"identity":"07026194-533f-4290-ae44-8c449245d54b","added_by":"auto","created_at":"2025-10-13 16:11:32","extension":"html","order_by":30,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":151289,"visible":true,"origin":"","legend":"","description":"","filename":"earlyproof.html","url":"https://assets-eu.researchsquare.com/files/rs-7730313/v1/345f1cba3222c2b08e6df99c.html"},{"id":93421589,"identity":"3ec30aeb-f889-40f8-8db9-c5f011a448bb","added_by":"auto","created_at":"2025-10-13 16:11:30","extension":"png","order_by":1,"title":"Figure 1","display":"","copyAsset":false,"role":"figure","size":539600,"visible":true,"origin":"","legend":"\u003cp\u003eStudy Area Map\u003c/p\u003e","description":"","filename":"1.png","url":"https://assets-eu.researchsquare.com/files/rs-7730313/v1/a1a99efaef11bd2c2c38bde5.png"},{"id":93421601,"identity":"e25cc6b3-dd7a-4261-b2ae-d2f378a3bc20","added_by":"auto","created_at":"2025-10-13 16:11:30","extension":"png","order_by":2,"title":"Figure 2","display":"","copyAsset":false,"role":"figure","size":58982,"visible":true,"origin":"","legend":"\u003cp\u003eExtraction and Processing of High-Resolution Data from UAV\u003c/p\u003e","description":"","filename":"2.png","url":"https://assets-eu.researchsquare.com/files/rs-7730313/v1/f6f8a55d1ec521dd7628f08c.png"},{"id":93421688,"identity":"870e91db-31cb-4dea-90cd-9d32c1616fb5","added_by":"auto","created_at":"2025-10-13 16:11:32","extension":"png","order_by":3,"title":"Figure 3","display":"","copyAsset":false,"role":"figure","size":502306,"visible":true,"origin":"","legend":"\u003cp\u003e(a) Training sample locations within the study area. (b) Class distribution of selected land-use/land-cover samples (c) High Resolution Images Map of the study area\u003c/p\u003e","description":"","filename":"3.png","url":"https://assets-eu.researchsquare.com/files/rs-7730313/v1/b3f972a85d796ec3cb8a0a2b.png"},{"id":93422529,"identity":"82b2a0d7-c447-4981-9c15-e54f6b3b8ff2","added_by":"auto","created_at":"2025-10-13 16:19:32","extension":"png","order_by":4,"title":"Figure 4","display":"","copyAsset":false,"role":"figure","size":43524,"visible":true,"origin":"","legend":"\u003cp\u003eWorkflow for Land-Use/Land-Cover (LULC) Classification Using\u003c/p\u003e","description":"","filename":"4.png","url":"https://assets-eu.researchsquare.com/files/rs-7730313/v1/0e4b7cc1cccb4d7a29f2b4e2.png"},{"id":93422524,"identity":"f94d0066-d8ce-4883-a623-cb290118df6a","added_by":"auto","created_at":"2025-10-13 16:19:30","extension":"png","order_by":5,"title":"Figure 5","display":"","copyAsset":false,"role":"figure","size":68056,"visible":true,"origin":"","legend":"\u003cp\u003e\u0026nbsp;(a) Model Performance Metrics (b) ROC Curve Analysis\u003c/p\u003e","description":"","filename":"5.png","url":"https://assets-eu.researchsquare.com/files/rs-7730313/v1/d18e32008a7581c50509a231.png"},{"id":93421617,"identity":"367adb7c-e7ce-4281-9792-e289b14498ab","added_by":"auto","created_at":"2025-10-13 16:11:31","extension":"png","order_by":6,"title":"Figure 6","display":"","copyAsset":false,"role":"figure","size":113340,"visible":true,"origin":"","legend":"\u003cp\u003eConfusion matrix visualizations for the Random Forest, SVC, and Gradient Boosting models.\u003c/p\u003e","description":"","filename":"6.png","url":"https://assets-eu.researchsquare.com/files/rs-7730313/v1/3ded1fb950e658d235c1a0c3.png"},{"id":93421770,"identity":"9e825695-2325-4432-bbe3-8d084fe886af","added_by":"auto","created_at":"2025-10-13 16:11:35","extension":"png","order_by":7,"title":"Figure 7","display":"","copyAsset":false,"role":"figure","size":605537,"visible":true,"origin":"","legend":"\u003cp\u003e(a) DTM from the LiDAR Data acquired for this project by the author,\u003cem\u003e \u003c/em\u003e(b) DSM LiDAR Data Captured by the author.\u003c/p\u003e","description":"","filename":"7.png","url":"https://assets-eu.researchsquare.com/files/rs-7730313/v1/91bfe51b9f9116d79da88c68.png"},{"id":93421588,"identity":"c855ab0c-d64a-42c5-a0bd-9e583aa748ad","added_by":"auto","created_at":"2025-10-13 16:11:30","extension":"png","order_by":8,"title":"Figure 8","display":"","copyAsset":false,"role":"figure","size":65783,"visible":true,"origin":"","legend":"\u003cp\u003e(a) Land Use Land Cover Classification Model Performance and Accuracy Comparison (b)ROC Curve\u003c/p\u003e","description":"","filename":"8.png","url":"https://assets-eu.researchsquare.com/files/rs-7730313/v1/050cdff10ec87d042ce3b40c.png"},{"id":93421584,"identity":"121778d5-e634-4fe8-a8a8-eb065d265b3d","added_by":"auto","created_at":"2025-10-13 16:11:30","extension":"png","order_by":9,"title":"Figure 9","display":"","copyAsset":false,"role":"figure","size":107346,"visible":true,"origin":"","legend":"\u003cp\u003eLand Use Land Cover Classification Model Confusion Matrix Comparison\u003c/p\u003e","description":"","filename":"9.png","url":"https://assets-eu.researchsquare.com/files/rs-7730313/v1/d38d6e26a5e898ebd0112b2a.png"},{"id":93421772,"identity":"1b7afe87-acd9-4103-8e7b-e7c01f16a16c","added_by":"auto","created_at":"2025-10-13 16:11:35","extension":"png","order_by":10,"title":"Figure 10","display":"","copyAsset":false,"role":"figure","size":993409,"visible":true,"origin":"","legend":"\u003cp\u003eLand Use Land Cover Random Forest (RF) Classification Results\u003c/p\u003e","description":"","filename":"10.png","url":"https://assets-eu.researchsquare.com/files/rs-7730313/v1/1032da7b5bc09aa4ae45cc11.png"},{"id":93421593,"identity":"d1091ba3-a37e-4a6f-8b07-38352764b241","added_by":"auto","created_at":"2025-10-13 16:11:30","extension":"png","order_by":11,"title":"Figure 11","display":"","copyAsset":false,"role":"figure","size":830272,"visible":true,"origin":"","legend":"\u003cp\u003eLand Use Land Cover Support Vector Classifier (SVC) Classification Result\u003c/p\u003e","description":"","filename":"11.png","url":"https://assets-eu.researchsquare.com/files/rs-7730313/v1/3f04f8f36126373f3c901682.png"},{"id":93421603,"identity":"a4a596d9-1175-4426-bad7-ebbbc00431a2","added_by":"auto","created_at":"2025-10-13 16:11:30","extension":"png","order_by":12,"title":"Figure 12","display":"","copyAsset":false,"role":"figure","size":902271,"visible":true,"origin":"","legend":"\u003cp\u003eLand Use Land Cover Gradient Boosting Classifier (GBC) Classification Result\u003c/p\u003e","description":"","filename":"12.png","url":"https://assets-eu.researchsquare.com/files/rs-7730313/v1/0f8896905406e57346d0838a.png"},{"id":93597659,"identity":"428f99fb-fb48-4734-9c1c-0f3b090df493","added_by":"auto","created_at":"2025-10-15 14:17:57","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":6202803,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-7730313/v1/abf540f9-742f-421c-b036-a63c94010569.pdf"}],"financialInterests":"No competing interests reported.","formattedTitle":"Differential Impacts of Very High-Resolution Terrain Data and Multispectral Imagery on the Accuracy of Land-Use Classification Using Machine Learning","fulltext":[{"header":"1. Introduction","content":"\u003cp\u003eThe interaction and classification of landscapes according to human use and natural cover are crucial for understanding the spatial and temporal patterns of land-use and land-cover changes [1, 2, 3]. Cem \u0026Uuml;nsalan and Boyer 2011 [4] highlight land use classification, which covers a wide range of applications, facilitating informed decision-making in urban planning, agriculture, environmental conservation, and disaster management. As landscapes continue to evolve due to both natural and anthropogenic factors, the demand for precise and up-to-date land-use information has intensified. The fusion of remotely sensed data with machine learning algorithms has proven to be a breakthrough and a powerful tool to tackle unprecedented, large-scale challenges emerging from remotely sensed data analysis [5, 6, 7].\u003c/p\u003e\u003cp\u003eTraditionally, multispectral imagery has widely been utilized for land-use classification due to its ability to capture surface reflectance properties across different spectral bands [2, 8]. These images provide vital information about vegetation, water bodies, bare soil, and built-up areas. However, surface characteristics alone may not always be sufficient to distinguish between land-use classes that share similar spectral signatures [9, 10, 11, 12, 1]. To overcome this limitation and improve the identification of land use dynamics, fusing datasets acquired from remote sensors that operate with differing underlying physical mechanisms and providing synergistic and potentially more effective means of obtaining comprehensive information about land surface properties [12]. Specifically, very-high terrain data, Digital Surface Models (DSM), Digital Terrain Models (DTM), and orthomosaics, with terrain data displaying detailed representations of surface height and landform structure (DSM captures all surface features, including vegetation and buildings, while DTM represents the bare earth) [13], making them particularly useful for analyzing natural terrain [14] and orthomosaics display a true color image of overlapping multiple photographs of the earth surface [15].\u003c/p\u003e\u003cp\u003eTerrain variables are often correlated with ecological and human land-use patterns, particularly in regions where topography influences vegetation growth, hydrology, and settlement patterns [16, 17, 18, 19, 20, 21]. Multi-spectral imagery captures surface reflectance, and geomorphometric variables, according to Guth et al. \u003cspan citationid=\"CR13\" class=\"CitationRef\"\u003e2021\u003c/span\u003e [22], add a third dimension of terrain context that helps distinguish between land-use types with similar spectral signatures but topographically distinct characteristics.\u003c/p\u003e\u003cp\u003eIn recent years machine learning has emerged as a powerful tool in remote sensing applications due to its ability to handle large, complex, and high-dimensional datasets, having algorithms such as Random Forest (RF), Support Vector Machine (SVM), and Gradient Boosting with capabilities of identifying non-linear relationships and interactions among input features, further improving the robustness and precision of classification outputs [6]. The integration of terrain data into machine learning workflows alongside spectral or orthophoto data should enhance the accuracy of land use classification and improve interpretability challenges in complex terrain.\u003c/p\u003e\u003cp\u003eThis study explores how high-resolution terrain data can be derived to enhance the accuracy of land-use classification machine learning techniques. It investigates the individual and combined effects of these features with spectral datasets, evaluates the performance of different classification algorithms, and assesses how topographic context influences land-use mapping in diverse landscapes. The findings contribute to the growing body of knowledge on terrain-based remote sensing and provide practical insights for enhancing land-use monitoring systems through the integration of geospatial intelligence and artificial intelligence.\u003c/p\u003e"},{"header":"2. Materials and Methods","content":"\u003cdiv id=\"Sec3\" class=\"Section2\"\u003e\u003ch2\u003e2.1 Study Area\u003c/h2\u003e\u003cp\u003eThe study was conducted in Rivers State, Nigeria, at a latitude of 4\u0026deg;42'47.7\"N \u0026ndash; 4\u003csup\u003eo\u003c/sup\u003e52\u0026rsquo;12\u0026rdquo; and a longitude of 6\u0026deg;54'26.6\"E \u0026ndash; 7\u003csup\u003eo\u003c/sup\u003e 08\u0026rsquo;01\u0026rdquo;. With an area extent of 2526.75 square kilometers, it covers eight Local Government Areas (LGAs) within Rivers State, Nigeria: Obio/Akpor, Port Harcourt, Oyigbo, Ogu/Bolo, Eleme, Ikwerre, Etche, and Okrika. These regions exhibit diverse urban growth patterns and environmental characteristics that affect flood vulnerability. Port Harcourt, the state capital, serves as a central industrial hub with advanced infrastructure, and Obio/Akpor is a densely populated area, facing increasing pressure from urbanization. They are densely populated areas prone to flash flooding during intense rainfall. Okrika is a riverine community, while Ogu/Bolo is another low-lying coastal area. They are naturally prone to tidal flooding and river overflow. In Oyigbo and Eleme, the expansion of industrial activities and poorly maintained drainage networks exacerbate flood risks, particularly during the rainy season. Ikwerre, though more inland, faces localized flooding due to unplanned urban expansion and poor land management. Meanwhile, Etche\u0026rsquo;s agrarian landscape is increasingly threatened by seasonal floodwaters, which damage crops and disrupt rural livelihoods [23].\u003c/p\u003e\u003cp\u003e\u003c/p\u003e\u003c/div\u003e\u003cdiv id=\"Sec4\" class=\"Section2\"\u003e\u003ch2\u003e2.2 Data Acquisition, Extraction, and Pre-processing\u003c/h2\u003e\u003cp\u003eHigh-resolution aerial imagery was captured using a DJI Matrice 350 RTK drone equipped with a Zenmuse P1 and L1 sensors. Flights were carried out at a consistent altitude of 120 meters between 10:00 AM and 4:00 PM to ensure optimal lighting and minimize shadow interference. A total of twelve locations were surveyed, of which eight were randomly selected for further analysis. The selected sites were stitched into a seamless image mosaic using the Structure-from-Motion (SfM) workflow in Agisoft Metashape, which utilized GPU-accelerated tie-point matching to enhance spatial consistency, following established methods [24]. The final mosaic had a native ground sampling distance of 0.05 meters, later resampled to 2 meters using GDAL tools to reduce computational load and streamline processing.\u003c/p\u003e\u003cp\u003eThe photogrammetric process produced an orthomosaic of 5cm GSD, while the LiDAR data processing produced a Digital Surface Model (DSM) and a Digital Terrain Model (DTM). The orthomosaic was generated through careful image alignment and dense point cloud reconstruction, providing detailed spectral information critical for accurate land-use classification [25]. Orthorectification was applied to correct geometric distortions and ensure spatial accuracy. The DSM captured surface elevations, including buildings and vegetation, allowing for a three-dimensional analysis of the landscape [26]. To derive DTM, advanced filtering techniques were applied to isolate the bare earth surface, effectively removing non-ground features [27]. Together, these datasets, high-resolution spectral imagery, surface elevation, and terrain morphology, offered a rich, multi-dimensional view of the landscape, forming a robust foundation for land-use/land-cover classification.\u003c/p\u003e\u003cp\u003eTo ensure optimal data quality before classification, a series of comprehensive preprocessing steps were undertaken. The orthophoto underwent radiometric normalization to correct illumination differences across the study area, reducing spectral variability due to lighting conditions [28]. The elevation datasets, including the DSM and DTM, were enhanced using median filtering techniques to suppress noise, an essential step in areas with complex terrain or heterogeneous vegetation [29]. A normalized Digital Surface Model (nDSM) was also derived by subtracting the DTM from the DSM, effectively isolating surface features such as vegetation canopies and built structures [30]. These pre-processed datasets, comprising both spectral and structural information, provided a reliable and information-rich foundation for the subsequent machine learning-based land-use classification.\u003c/p\u003e\u003cp\u003eBefore model training, the orthophotos were normalized using Min-Max scaling to transform the data into a consistent [0, 1] range. This preprocessing step is crucial for remote sensing applications where different spectral bands and derived indices may have varying measurement scales [31]. The Min-Max scaler was implemented using the formula:\u003cdiv id=\"Equ1\" class=\"Equation\"\u003e\u003cdiv format=\"TEX\" class=\"mathdisplay\" id=\"FileID_Equ1\" name=\"EquationSource\"\u003e\n$$\\:\\text{m}=\\frac{x-{x}_{min}}{{x}_{max}-{x}_{min}}$$\u003c/div\u003e\u003cdiv class=\"EquationNumber\"\u003e1\u003c/div\u003e\u003c/div\u003e\u003c/p\u003e\u003cp\u003e\u003cem\u003em\u003c/em\u003e represents the original feature values, \u003cem\u003ex\u003c/em\u003e\u003csub\u003e\u003cem\u003emin\u003c/em\u003e\u003c/sub\u003e and \u003cem\u003ex\u003c/em\u003e\u003csub\u003e\u003cem\u003emax\u003c/em\u003e\u003c/sub\u003e denote the minimum and maximum values for each feature, respectively.\u003c/p\u003e\u003cp\u003eThis pre-processing step was crucial for the machine learning workflow, as it enabled more stable convergence during model training and enhanced the interpretability of feature importance measures in the final classification models.\u003c/p\u003e\u003cp\u003e\u003c/p\u003e\u003c/div\u003e\u003cdiv id=\"Sec5\" class=\"Section2\"\u003e\u003ch2\u003e2.3 Training Data Collection for LULC Classification\u003c/h2\u003e\u003cp\u003eTraining data for Land Use and Land Cover (LULC) classification were collected using handheld GPS devices to ensure spatial accuracy and alignment with field observations. Initially, eight distinct land cover classes were identified during field reconnaissance to capture the diversity of surface features within the study area. These classes included various forms of built-up areas, vegetation types, and land surfaces (tarred and untarred roads). However, for model training and to improve classification efficiency, the classes were systematically regrouped into five broader categories: buildings, tarred land, vegetation, open water, and bare land. The reclassification ensured consistency in spectral separability and reduced intra-class variability. GPS points were collected to represent each of these five final classes and were cross-verified with high-resolution satellite imagery. The dataset was subsequently cleaned to remove inconsistencies, spatial errors, and duplicates. In this schema, buildings refer to all constructed structures; tarred land includes roads and paved surfaces; vegetation encompasses both natural and cultivated green cover; open water represents rivers, lakes, and ponds; and bare land denotes exposed soil or sparsely vegetated ground. The cleaned and labeled samples were used for training and validating machine learning models for LULC classification.\u003c/p\u003e\u003cp\u003e\u003c/p\u003e\u003cp\u003e\u003c/p\u003e\u003c/div\u003e\u003cdiv id=\"Sec6\" class=\"Section2\"\u003e\u003ch2\u003e2.4 Image classification using Machine Learning Models\u003c/h2\u003e\u003cp\u003eIn this study, supervised machine learning algorithms were employed to classify land-use and land-cover (LULC) types using high-resolution aerial imagery. The classification process was designed in two stages to evaluate the incremental value of structural features. Initially, models were trained exclusively on the spectral information derived from the orthomosaic. This baseline configuration allowed for the assessment of classifier performance using purely spectral data. Subsequently, additional features, including elevation data from the Digital Surface Model (DSM), Digital Terrain Model (DTM), and the derived normalized DSM (nDSM), were integrated into the input feature space to examine their impact on classification accuracy.\u003c/p\u003e\u003cp\u003eThree widely used classifiers, Random Forest (RF), Support Vector Classifier (SVC), and Gradient Boosting Classifier (GBC), were selected for their effectiveness in handling non-linear, high-dimensional data. Training samples were derived from field-based ground truth observations and high-resolution reference imagery, ensuring reliable class representation across diverse land cover types. Hyperparameter tuning was conducted using a grid search strategy combined with k-fold cross-validation to optimize model generalization.\u003c/p\u003e\u003cp\u003eClassification performance for each model and input configuration was assessed using standard accuracy metrics, including overall accuracy, producer\u0026rsquo;s and user\u0026rsquo;s accuracies, and Cohen\u0026rsquo;s Kappa coefficient. This stepwise evaluation framework enabled a clear comparison of how spectral data alone performs versus the enriched spectral-structural feature set. Results highlight the added value of integrating terrain-derived features in improving the precision of LULC classification, particularly in areas with complex surface morphology or overlapping land cover types.\u003c/p\u003e\u003c/div\u003e\u003cdiv id=\"Sec7\" class=\"Section2\"\u003e\u003ch2\u003e2.5 Machine Learning Classification Algorithms\u003c/h2\u003e\u003cdiv id=\"Sec8\" class=\"Section3\"\u003e\u003ch2\u003e2.5.1 Supporting Vector Classifier (SVC)\u003c/h2\u003e\u003cp\u003eSupport Vector Classifier (SVC) is a non-parametric, supervised machine learning algorithm grounded in the principle of Structural Risk Minimization (SRM). It seeks to identify an optimal hyperplane that maximally separates data points of distinct classes while minimizing generalization error [32, 33]. Unlike empirical risk minimization techniques, SRM aims to balance model complexity with training accuracy, making SVC particularly effective for high-dimensional classification tasks such as land-use and land-cover (LULC) mapping.\u003c/p\u003e\u003cp\u003eIn remote sensing applications, kernel functions are used to transform nonlinearly separable data into a higher-dimensional feature space, where linear separation is possible. Among the most widely used kernels, the Radial Basis Function (RBF) and polynomial kernels have demonstrated strong performance, with RBF often yielding higher classification accuracy due to its ability to model complex, non-linear relationships inherent in LULC datasets [34, 35]. The Spectral Angle Mapper (SAM), though traditionally used as a similarity measure, is not directly involved in SVC\u0026rsquo;s hyperplane optimization process and may have been conflated in earlier literature.\u003c/p\u003e\u003cp\u003eIn this study, both RBF and polynomial kernels were implemented using the Scikit-learn library in Python. These configurations were evaluated for their effectiveness in classifying high-resolution LULC data derived from RGB imagery and elevation features. Hyperparameters for each kernel type were optimized using grid search combined with cross-validation to ensure model robustness and generalizability.\u003c/p\u003e\u003c/div\u003e\u003cdiv id=\"Sec9\" class=\"Section3\"\u003e\u003ch2\u003e2.5.2 Random Forest Classifier\u003c/h2\u003e\u003cp\u003eRandom Forest (RF) is a robust ensemble learning technique that builds many decision trees, each trained on a randomly selected subset of the data and features [36]. Unlike individual decision trees, RF aggregates the predictions of multiple base classifiers using a majority voting scheme to produce more stable and accurate outputs. This ensemble strategy reduces the risk of overfitting and improves generalization by leveraging the diversity among trees [36, 37].\u003c/p\u003e\u003cp\u003eThe final prediction is obtained by aggregating all tree outputs using the discrimination function:\u003cdiv id=\"Equ2\" class=\"Equation\"\u003e\u003cdiv format=\"TEX\" class=\"mathdisplay\" id=\"FileID_Equ2\" name=\"EquationSource\"\u003e\n$$\\:\\text{H}\\left(\\text{x}\\right)\\:=\\:\\text{a}\\text{r}\\text{g}\\underset{{\\:\\:\\:\\:\\:\\:\\:\\:}\\text{Y}}{\\text{max}}\\:\\:{\\sum\\:}_{\\text{i}=1}^{\\text{k}}\\text{I}\\left({\\text{h}}_{\\text{i}}\\left(\\text{X},\\:{{\\theta\\:}}_{\\text{k}}\\right)\\:=\\:\\text{Y}\\right)$$\u003c/div\u003e\u003cdiv class=\"EquationNumber\"\u003e2\u003c/div\u003e\u003c/div\u003e\u003c/p\u003e\u003cp\u003eWhere \u003cem\u003eI(\u0026sdot;)\u003c/em\u003e is an indicator function that returns 1 when the condition is met and 0 otherwise; \u003cem\u003eh\u003c/em\u003e\u003csub\u003e\u003cem\u003ei\u003c/em\u003e\u003c/sub\u003e\u003cem\u003e(\u0026sdot;)\u003c/em\u003e denotes the \u003cem\u003ei-th\u003c/em\u003e decision tree; \u003cem\u003eY\u003c/em\u003e represents the output class; and the final prediction is the class with the maximum number of votes.\u003c/p\u003e\u003cp\u003eThe dataset was partitioned into a training set comprising 70% of the data and a testing set comprising the remaining 30%. Scikit-Learn\u0026rsquo;s GridSearchCV was used, which searched for parameters within a specified range for each hyperparameter. Specifically, the parameter distribution of the number estimator (ranging from 10 to 500) was utilized, and the maximum number of trees (ranging from 2 to 20) was employed to randomly sample 10 combinations of hyperparameters. A 10-fold cross-validation approach was employed to assess the performance of each set of hyperparameters. Additionally, the Gini index was used to evaluate the importance of features and identify which input variables were most influential in predicting flood hazard. The Gini index is a metric used in decision trees to evaluate the quality of a split. It measures the degree of disorder or impurity in a dataset, where 0 represents perfect purity (all elements belong to a single class), and higher values indicate more impurity. Mathematically, the Gini index is expressed as:\u003cdiv id=\"Equ3\" class=\"Equation\"\u003e\u003cdiv format=\"TEX\" class=\"mathdisplay\" id=\"FileID_Equ3\" name=\"EquationSource\"\u003e\n$$\\:\\text{Gini}\\:=\\:1\\:-\\:{\\sum\\:}_{\\text{i}=1}^{\\text{n}}{p}_{i}^{2}\\:\\:$$\u003c/div\u003e\u003cdiv class=\"EquationNumber\"\u003e3\u003c/div\u003e\u003c/div\u003e\u003c/p\u003e\u003cp\u003eHere, \u003cem\u003epi\u003c/em\u003e denotes the proportion of items classified as class \u003cem\u003ei\u003c/em\u003e within the node, while \u003cem\u003en\u003c/em\u003e signifies the total number of classes.\u003c/p\u003e\u003c/div\u003e\u003cdiv id=\"Sec10\" class=\"Section3\"\u003e\u003ch2\u003e2.5.3 Gradient Boosting Classifier\u003c/h2\u003e\u003cp\u003eGradient Boosting Classifier (GBC), particularly in implementations such as XGBoost, LightGBM, and CatBoost, has emerged as a powerful tool due to its ability to model complex feature interactions and improve classification accuracy through iterative ensemble learning. GBMs have shown strong performance in processing high-resolution satellite and aerial imagery, often outperforming traditional classifiers. A notable strength of GBMs lies in their robustness to noisy data and imbalanced class distributions, both of which are prevalent in high-resolution remote sensing datasets.\u003c/p\u003e\u003c/div\u003e\u003cdiv id=\"Sec11\" class=\"Section3\"\u003e\u003ch2\u003e2.5.4 Kappa Coefficient\u003c/h2\u003e\u003cp\u003eThe Cohen\u0026rsquo;s Kappa coefficient (κ) measures the agreement between predicted and actual classes, adjusted for chance agreement. It is calculated as:\u003cdiv id=\"Equ4\" class=\"Equation\"\u003e\u003cdiv format=\"TEX\" class=\"mathdisplay\" id=\"FileID_Equ4\" name=\"EquationSource\"\u003e\n$$\\:{\\kappa\\:}=\\frac{{p}_{o}-{p}_{e}}{1-{p}_{e}}$$\u003c/div\u003e\u003cdiv class=\"EquationNumber\"\u003e4\u003c/div\u003e\u003c/div\u003e\u003c/p\u003e\u003cp\u003ewhere: \u003cem\u003ep\u003c/em\u003e\u003csub\u003e\u003cem\u003eo\u003c/em\u003e\u003c/sub\u003e\u003cem\u003e​\u003c/em\u003e is the observed agreement (same as accuracy), and \u003cem\u003ep\u003c/em\u003e\u003csub\u003e\u003cem\u003ee\u003c/em\u003e​\u003c/sub\u003e is the expected agreement by chance. Kappa values range from \u0026minus;\u0026thinsp;1 to 1, where values close to 1 indicate strong agreement, 0 indicates chance-level agreement, and negative values imply disagreement.\u003c/p\u003e\u003c/div\u003e\u003cdiv id=\"Sec12\" class=\"Section3\"\u003e\u003ch2\u003e2.5.5 Precision, Recall, and F1-Score\u003c/h2\u003e\u003cp\u003eThese metrics provide more granular insights into classification performance:\u003c/p\u003e\u003cp\u003e\u003cb\u003ePrecision\u003c/b\u003e quantifies the accuracy of positive predictions:\u003cdiv id=\"Equ5\" class=\"Equation\"\u003e\u003cdiv format=\"TEX\" class=\"mathdisplay\" id=\"FileID_Equ5\" name=\"EquationSource\"\u003e\n$$\\:\\text{Precision}=\\frac{TP}{TP+FP}$$\u003c/div\u003e\u003cdiv class=\"EquationNumber\"\u003e5\u003c/div\u003e\u003c/div\u003e\u003c/p\u003e\u003cp\u003e\u003cb\u003eRecall (Sensitivity)\u003c/b\u003e measures how well the model identifies actual positives:\u003cdiv id=\"Equ6\" class=\"Equation\"\u003e\u003cdiv format=\"TEX\" class=\"mathdisplay\" id=\"FileID_Equ6\" name=\"EquationSource\"\u003e\n$$\\:\\text{Recall}=\\frac{TP}{TP+FN}$$\u003c/div\u003e\u003cdiv class=\"EquationNumber\"\u003e6\u003c/div\u003e\u003c/div\u003e\u003c/p\u003e\u003cp\u003e\u003cb\u003eF1-score\u003c/b\u003e is the harmonic mean of precision and recall, balancing both metrics:\u003cdiv id=\"Equ7\" class=\"Equation\"\u003e\u003cdiv format=\"TEX\" class=\"mathdisplay\" id=\"FileID_Equ7\" name=\"EquationSource\"\u003e\n$$\\:F1=2\\cdot\\:\\frac{\\text{Precision}\\cdot\\:\\text{Recall}}{\\text{Precision}+\\text{Recall}}$$\u003c/div\u003e\u003cdiv class=\"EquationNumber\"\u003e7\u003c/div\u003e\u003c/div\u003e\u003c/p\u003e\u003c/div\u003e\u003cdiv id=\"Sec13\" class=\"Section3\"\u003e\u003ch2\u003e2.5.6 AUC-ROC (Area Under the Receiver Operating Characteristic Curve)\u003c/h2\u003e\u003cp\u003eThe ROC curve illustrates the trade-off between the true positive rate (TPR) and false positive rate (FPR) at various threshold settings. AUC (Area Under Curve) quantifies the overall ability of the model to distinguish between the two classes.\u003cdiv id=\"Equ8\" class=\"Equation\"\u003e\u003cdiv format=\"TEX\" class=\"mathdisplay\" id=\"FileID_Equ8\" name=\"EquationSource\"\u003e\n$$\\:\\text{TPR}=\\frac{TP}{TP+FN}$$\u003c/div\u003e\u003cdiv class=\"EquationNumber\"\u003e8\u003c/div\u003e\u003c/div\u003e\u003cdiv id=\"Equ9\" class=\"Equation\"\u003e\u003cdiv format=\"TEX\" class=\"mathdisplay\" id=\"FileID_Equ9\" name=\"EquationSource\"\u003e\n$$\\:\\hspace{1em}\\text{FPR}=\\frac{FP}{FP+TN}$$\u003c/div\u003e\u003cdiv class=\"EquationNumber\"\u003e9\u003c/div\u003e\u003c/div\u003e\u003cdiv id=\"Equ10\" class=\"Equation\"\u003e\u003cdiv format=\"TEX\" class=\"mathdisplay\" id=\"FileID_Equ10\" name=\"EquationSource\"\u003e\n$$\\:\\:\\:\\:AUC={\\int\\:}_{0}^{1}\\text{TPR}\\left(x\\right)\\hspace{0.17em}dx$$\u003c/div\u003e\u003cdiv class=\"EquationNumber\"\u003e10\u003c/div\u003e\u003c/div\u003e\u003c/p\u003e\u003c/div\u003e\u003c/div\u003e\u003cdiv id=\"Sec14\" class=\"Section2\"\u003e\u003ch2\u003e2.6 Cross-Validation and Generalization Performance\u003c/h2\u003e\u003cp\u003eCross-validation helps assess how well the model generalizes to unseen data. In k-fold cross-validation, the dataset is partitioned into \u003cem\u003ek\u003c/em\u003e subsets. The model is trained on \u003cem\u003ek\u0026thinsp;\u0026minus;\u0026thinsp;1\u003c/em\u003e subsets and validated on the remaining one. This technique mitigates overfitting and provides a robust estimate of model performance across varying data distributions.\u003c/p\u003e\u003cp\u003eThe process is repeated k times, and the average score is computed:\u003cdiv id=\"Equ11\" class=\"Equation\"\u003e\u003cdiv format=\"TEX\" class=\"mathdisplay\" id=\"FileID_Equ11\" name=\"EquationSource\"\u003e\n$$\\:{\\text{CV}}_{\\text{score}}=\\frac{1}{k}{\\sum\\:}_{i=1}^{k}{\\text{Score}}_{i}C$$\u003c/div\u003e\u003cdiv class=\"EquationNumber\"\u003e11\u003c/div\u003e\u003c/div\u003e\u003c/p\u003e\u003c/div\u003e"},{"header":"3. Results","content":"\u003cp\u003eThis section presents the results of the land use and land cover (LULC) classification, with a focus on evaluating the importance of very high-resolution elevation data with RGB orthophotography. The analysis highlights how integrating Digital Terrain Model (DTM) and Digital Surface Model (DSM) layers enhanced the models\u0026rsquo; ability to discriminate between land cover types, particularly in heterogeneous landscapes where elevation plays a critical role.\u003c/p\u003e\u003cdiv id=\"Sec16\" class=\"Section2\"\u003e\u003ch2\u003e3.1 Performance of Land Cover Classification Using Only RGB Spectral Bands\u003c/h2\u003e\u003cdiv id=\"Sec17\" class=\"Section3\"\u003e\u003ch2\u003e3.1.1 Model Performance Metrics and ROC Curve Analysis\u003c/h2\u003e\u003cp\u003eFigure \u003cspan refid=\"Fig5\" class=\"InternalRef\"\u003e5\u003c/span\u003e(a) presents a comparative analysis of the three machine learning classifiers, Random Forest (RF), Support Vector Classifier (SVC), and Gradient Boosting Classifier (GBC), based on four standard evaluation metrics: Accuracy, Precision, Recall, and F1 Score. Among the models, SVC exhibited the highest overall accuracy (79.40%), followed closely by GBC (75.78%), while RF (75.90%) trailed marginally behind. This performance hierarchy remains consistent across the remaining metrics, where SVC maintained a well-balanced trade-off between sensitivity and specificity. Its ability to generalize across all classes reflects its robustness in RGB-only classification scenarios. GBC also demonstrated strong competitive performance, benefiting from its sequential boosting strategy that emphasizes misclassified instances during iterative learning. RF, although slightly lower in both precision and recall, nonetheless delivered acceptable and stable performance, particularly for well-separated classes.\u003c/p\u003e\u003cp\u003eFigure \u003cspan refid=\"Fig5\" class=\"InternalRef\"\u003e5\u003c/span\u003e(b) displays the multiclass Receiver Operating Characteristic (ROC) curves. The Area Under the Curve (AUC) scores reveal that both SVC and GBC achieved an AUC of 92%, indicating excellent class separability and reliable prediction confidence under orthophoto conditions. RF recorded an AUC of 91%, only marginally lower, thereby reinforcing its continued utility as a baseline model in environments constrained to visible spectrum inputs.\u003c/p\u003e\u003cp\u003e\u003c/p\u003e\u003cp\u003e\u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab1\" border=\"1\"\u003e\u003ccaption language=\"En\"\u003e\u003cdiv class=\"CaptionNumber\"\u003eTable 1\u003c/div\u003e\u003cdiv class=\"CaptionContent\"\u003e\u003cp\u003eModel Accuracy (%) Based on RGB-Only Classification\u003c/p\u003e\u003c/div\u003e\u003c/caption\u003e\u003ccolgroup cols=\"5\"\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e\u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e\u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e\u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c4\" colnum=\"4\"\u003e\u003c/div\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c5\" colnum=\"5\"\u003e\u003c/div\u003e\u003cthead\u003e\u003ctr\u003e\u003cth align=\"left\" colname=\"c1\"\u003e\u003cp\u003eModel\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c2\"\u003e\u003cp\u003eAccuracy (%)\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c3\"\u003e\u003cp\u003ePrecision (%)\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c4\"\u003e\u003cp\u003eRecall (%)\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c5\"\u003e\u003cp\u003eF1 Score (%)\u003c/p\u003e\u003c/th\u003e\u003c/tr\u003e\u003c/thead\u003e\u003ctbody\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eRandom Forest\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e\u003cp\u003e75.90\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e\u003cp\u003e75.24\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e\u003cp\u003e74.92\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u003cp\u003e74.74\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eSupport Vector\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e\u003cp\u003e\u003cb\u003e79.40\u003c/b\u003e\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e\u003cp\u003e\u003cb\u003e79.42\u003c/b\u003e\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e\u003cp\u003e\u003cb\u003e78.69\u003c/b\u003e\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u003cp\u003e\u003cb\u003e78.47\u003c/b\u003e\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eGradient Boosting\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e\u003cp\u003e75.78\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e\u003cp\u003e75.33\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e\u003cp\u003e74.59\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u003cp\u003e711\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003c/tbody\u003e\u003c/colgroup\u003e\u003c/table\u003e\u003c/div\u003e\u003c/p\u003e\u003c/div\u003e\u003cdiv id=\"Sec18\" class=\"Section3\"\u003e\u003ch2\u003e3.1.2 Class-wise Confusion Matrices\u003c/h2\u003e\u003cp\u003eFigure \u003cspan refid=\"Fig6\" class=\"InternalRef\"\u003e6\u003c/span\u003e and Tables\u0026nbsp;\u003cspan refid=\"Tab2\" class=\"InternalRef\"\u003e2\u003c/span\u003e to \u003cspan refid=\"Tab4\" class=\"InternalRef\"\u003e4\u003c/span\u003e present the confusion matrices for the Random Forest, Support Vector Classifier (SVC), and Gradient Boosting models, offering detailed insight into how each classifier performed across individual land cover classes using orthophoto spectral bands.\u003c/p\u003e\u003cp\u003eAll three models performed well on distinct classes such as bareland, vegetation, and open water, which are easier to identify due to their unique spectral characteristics. This is clearly reflected in the diagonal dominance of these classes across all three confusion matrices.\u003c/p\u003e\u003cp\u003e\u003c/p\u003e\u003cp\u003e\u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab2\" border=\"1\"\u003e\u003ccaption language=\"En\"\u003e\u003cdiv class=\"CaptionNumber\"\u003eTable 2\u003c/div\u003e\u003cdiv class=\"CaptionContent\"\u003e\u003cp\u003eRandom Forest Confusion Matrix\u003c/p\u003e\u003c/div\u003e\u003c/caption\u003e\u003ccolgroup cols=\"6\"\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e\u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e\u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e\u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c4\" colnum=\"4\"\u003e\u003c/div\u003e\u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c5\" colnum=\"5\"\u003e\u003c/div\u003e\u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c6\" colnum=\"6\"\u003e\u003c/div\u003e\u003cthead\u003e\u003ctr\u003e\u003cth align=\"left\" colname=\"c1\"\u003e\u003cp\u003eActual \\ Predicted\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c2\"\u003e\u003cp\u003eBuilding\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c3\"\u003e\u003cp\u003eOpen Water\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c4\"\u003e\u003cp\u003eBareland\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c5\"\u003e\u003cp\u003eVegetation\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c6\"\u003e\u003cp\u003eTarred/Asphalt\u003c/p\u003e\u003c/th\u003e\u003c/tr\u003e\u003c/thead\u003e\u003ctbody\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003e\u003cb\u003eBuilding\u003c/b\u003e\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e\u003cp\u003e130\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e\u003cp\u003e16\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e\u003cp\u003e14\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e\u003cp\u003e0\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e\u003cp\u003e39\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003e\u003cb\u003eOpen Water\u003c/b\u003e\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e\u003cp\u003e10\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e\u003cp\u003e69\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e\u003cp\u003e18\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e\u003cp\u003e11\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e\u003cp\u003e16\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003e\u003cb\u003eBareland\u003c/b\u003e\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e\u003cp\u003e5\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e\u003cp\u003e7\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e\u003cp\u003e181\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e\u003cp\u003e2\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e\u003cp\u003e8\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003e\u003cb\u003eVegetation\u003c/b\u003e\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e\u003cp\u003e0\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e\u003cp\u003e16\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e\u003cp\u003e3\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e\u003cp\u003e131\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e\u003cp\u003e2\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003e\u003cb\u003eTarred/Asphalt\u003c/b\u003e\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e\u003cp\u003e24\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e\u003cp\u003e15\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e\u003cp\u003e7\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e\u003cp\u003e1\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e\u003cp\u003e73\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003c/tbody\u003e\u003c/colgroup\u003e\u003c/table\u003e\u003c/div\u003e\u003c/p\u003e\u003cp\u003e\u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab3\" border=\"1\"\u003e\u003ccaption language=\"En\"\u003e\u003cdiv class=\"CaptionNumber\"\u003eTable 3\u003c/div\u003e\u003cdiv class=\"CaptionContent\"\u003e\u003cp\u003eSupport Vector Classifier (SVC) Confusion Matrix\u003c/p\u003e\u003c/div\u003e\u003c/caption\u003e\u003ccolgroup cols=\"6\"\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e\u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e\u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e\u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c4\" colnum=\"4\"\u003e\u003c/div\u003e\u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c5\" colnum=\"5\"\u003e\u003c/div\u003e\u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c6\" colnum=\"6\"\u003e\u003c/div\u003e\u003cthead\u003e\u003ctr\u003e\u003cth align=\"left\" colname=\"c1\"\u003e\u003cp\u003eActual \\ Predicted\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c2\"\u003e\u003cp\u003eBuilding\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c3\"\u003e\u003cp\u003eOpen Water\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c4\"\u003e\u003cp\u003eBareland\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c5\"\u003e\u003cp\u003eVegetation\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c6\"\u003e\u003cp\u003eTarred/Asphalt\u003c/p\u003e\u003c/th\u003e\u003c/tr\u003e\u003c/thead\u003e\u003ctbody\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003e\u003cb\u003eBuilding\u003c/b\u003e\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e\u003cp\u003e134\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e\u003cp\u003e9\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e\u003cp\u003e9\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e\u003cp\u003e0\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e\u003cp\u003e47\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003e\u003cb\u003eOpen Water\u003c/b\u003e\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e\u003cp\u003e8\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e\u003cp\u003e75\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e\u003cp\u003e19\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e\u003cp\u003e10\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e\u003cp\u003e12\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003e\u003cb\u003eBareland\u003c/b\u003e\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e\u003cp\u003e6\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e\u003cp\u003e3\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e\u003cp\u003e184\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e\u003cp\u003e3\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e\u003cp\u003e7\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003e\u003cb\u003eVegetation\u003c/b\u003e\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e\u003cp\u003e0\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e\u003cp\u003e13\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e\u003cp\u003e2\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e\u003cp\u003e135\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e\u003cp\u003e2\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003e\u003cb\u003eTarred/Asphalt\u003c/b\u003e\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e\u003cp\u003e13\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e\u003cp\u003e3\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e\u003cp\u003e5\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e\u003cp\u003e0\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e\u003cp\u003e131\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003c/tbody\u003e\u003c/colgroup\u003e\u003c/table\u003e\u003c/div\u003e\u003c/p\u003e\u003cp\u003e\u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab4\" border=\"1\"\u003e\u003ccaption language=\"En\"\u003e\u003cdiv class=\"CaptionNumber\"\u003eTable 4\u003c/div\u003e\u003cdiv class=\"CaptionContent\"\u003e\u003cp\u003eGradient Boosting Confusion Matrix\u003c/p\u003e\u003c/div\u003e\u003c/caption\u003e\u003ccolgroup cols=\"6\"\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e\u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e\u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e\u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c4\" colnum=\"4\"\u003e\u003c/div\u003e\u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c5\" colnum=\"5\"\u003e\u003c/div\u003e\u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c6\" colnum=\"6\"\u003e\u003c/div\u003e\u003cthead\u003e\u003ctr\u003e\u003cth align=\"left\" colname=\"c1\"\u003e\u003cp\u003eActual \\ Predicted\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c2\"\u003e\u003cp\u003eBuilding\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c3\"\u003e\u003cp\u003eOpen Water\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c4\"\u003e\u003cp\u003eBareland\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c5\"\u003e\u003cp\u003eVegetation\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c6\"\u003e\u003cp\u003eTarred/Asphalt\u003c/p\u003e\u003c/th\u003e\u003c/tr\u003e\u003c/thead\u003e\u003ctbody\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003e\u003cb\u003eBuilding\u003c/b\u003e\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e\u003cp\u003e131\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e\u003cp\u003e12\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e\u003cp\u003e18\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e\u003cp\u003e0\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e\u003cp\u003e38\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003e\u003cb\u003eOpen Water\u003c/b\u003e\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e\u003cp\u003e10\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e\u003cp\u003e61\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e\u003cp\u003e19\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e\u003cp\u003e11\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e\u003cp\u003e23\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003e\u003cb\u003eBareland\u003c/b\u003e\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e\u003cp\u003e6\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e\u003cp\u003e7\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e\u003cp\u003e179\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e\u003cp\u003e1\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e\u003cp\u003e10\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003e\u003cb\u003eVegetation\u003c/b\u003e\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e\u003cp\u003e1\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e\u003cp\u003e15\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e\u003cp\u003e2\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e\u003cp\u003e132\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e\u003cp\u003e2\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003e\u003cb\u003eTarred/Asphalt\u003c/b\u003e\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e\u003cp\u003e14\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e\u003cp\u003e4\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e\u003cp\u003e8\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e\u003cp\u003e0\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e\u003cp\u003e126\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003c/tbody\u003e\u003c/colgroup\u003e\u003c/table\u003e\u003c/div\u003e\u003c/p\u003e\u003c/div\u003e\u003c/div\u003e\u003cdiv id=\"Sec19\" class=\"Section2\"\u003e\u003ch2\u003e3.2 Enhanced Land Cover Classification Using Orthophoto, DTM, and DSM Integration\u003c/h2\u003e\u003cp\u003eRecent studies have demonstrated that integrating Digital Terrain Models (DTM) and Digital Surface Models (DSM) with high-resolution orthophoto significantly enhances the accuracy of land cover classification. The inclusion of elevation data introduces critical structural information that helps distinguish between spectrally similar classes, such as tarred surfaces and buildings, which often appear visually identical in orthophoto-only datasets. For instance, Kuras et al. \u003cspan citationid=\"CR18\" class=\"CitationRef\"\u003e2023\u003c/span\u003e [38] reported a marked improvement in urban land cover mapping accuracy from 64% (orthophoto only) to over 94% when DSM features were incorporated. Similarly, Mancini et al. \u003cspan citationid=\"CR19\" class=\"CitationRef\"\u003e2020\u003c/span\u003e [39] found that combining UAV-based orthophoto with photogrammetric DSMs yielded 7\u0026ndash;16% gains in classification performance, particularly in vegetation-rich areas.\u003c/p\u003e\u003cp\u003eThe synergy of spectral and elevation data is particularly beneficial for identifying land cover types with similar reflectance values but distinct height characteristics, such as distinguishing between low-lying bare land and multi-story buildings, or between shrub lands and forest canopies. These benefits have been consistently observed across a wide range of classifiers, including Support Vector Machines (SVM), Random Forest (RF), Gradient Boosting (GBC), and deep learning models, all of which exhibit improved precision and recall when elevation features are used alongside spectral inputs [40, 41]. Overall, integrating DTM and DSM data into orthophoto-based workflows enhances the robustness of classification outputs, especially in topographically complex or heterogeneous landscapes.\u003c/p\u003e\u003cdiv id=\"Sec20\" class=\"Section3\"\u003e\u003ch2\u003e3.2.1 Topographic Structure and Surface Elevation Mapping Using High-Resolution DTM and DSM Data\u003c/h2\u003e\u003cp\u003eFigures \u003cspan refid=\"Fig7\" class=\"InternalRef\"\u003e7\u003c/span\u003ea and \u003cspan refid=\"Fig7\" class=\"InternalRef\"\u003e7\u003c/span\u003eb present spatial representations of surface elevation derived from high-resolution remote sensing imagery and LiDAR point cloud data for eight distinct locations. The first set of panels (Fig.\u0026nbsp;\u003cspan refid=\"Fig7\" class=\"InternalRef\"\u003e7\u003c/span\u003ea) displays Digital Terrain Models (DTMs), while the second set (Fig.\u0026nbsp;\u003cspan refid=\"Fig7\" class=\"InternalRef\"\u003e7\u003c/span\u003eb) shows Digital Surface Models (DSMs) for the same areas. These elevation models were extracted to capture the underlying terrain morphology and above-ground structures, forming essential inputs for improved land cover classification.\u003c/p\u003e\u003cp\u003eDTMs represent the bare-earth surface, excluding vegetation and built features. The visual distribution of elevation values in the DTM maps reveals key geomorphological features such as valleys, slopes, and floodplains, with elevation ranges segmented into seven classes from \u0026minus;\u0026thinsp;7.42 m to 78.91 m. These models are particularly useful for identifying terrain-related variations in land cover and hydrological flow patterns.\u003c/p\u003e\u003cp\u003eIn contrast, DSMs capture the elevation of the topmost reflective surfaces, including tree canopies, rooftops, and other man-made structures. As seen in Fig.\u0026nbsp;\u003cspan refid=\"Fig7\" class=\"InternalRef\"\u003e7\u003c/span\u003eb, DSM values span from 0.03 m to 108.89 m, highlighting vertical variability across built-up and vegetated areas. Areas with dense vegetation or tall infrastructure appear in yellow to orange tones, while lower elevation surfaces remain in green and blue tones.\u003c/p\u003e\u003cp\u003eThe side-by-side comparison illustrates how DSMs complement DTMs by capturing vertical features that are invisible in terrain-only models. This combination is crucial in land cover classification, particularly for distinguishing between spectrally similar classes (e.g., bare land vs. buildings, shrubland vs. forest), by introducing vertical context that RGB imagery alone cannot provide. These elevation datasets serve as critical ancillary inputs for improving classification accuracy, model sensitivity, and spatial interpretability.\u003c/p\u003e\u003cp\u003e\u003c/p\u003e\u003c/div\u003e\u003cdiv id=\"Sec21\" class=\"Section3\"\u003e\u003ch2\u003e3.2.2 Model Performance Metrics and ROC Curve Analysis\u003c/h2\u003e\u003cp\u003eFigure \u003cspan refid=\"Fig8\" class=\"InternalRef\"\u003e8\u003c/span\u003e presents a comparative evaluation of the three supervised machine learning models: Random Forest, Support Vector Classifier (SVC), and Gradient Boosting, based on four standard performance metrics: accuracy, precision, recall, and F1 score. The SVC model exhibited the best overall performance, achieving an accuracy of 91.43%, a precision of 90.96%, a recall of 90.60%, and an F1 score of 90.75%. The Random Forest model followed, with accuracy reaching 89.01%, precision at 88.47%, recall at 88.02%, and an F1 score of 88.20%. The Gradient Boosting classifier yielded similar results to Random Forest, recording an accuracy of 88.90%, a precision of 88.19%, a recall of 88.07%, and an F1 score of 88.10%.\u003c/p\u003e\u003cp\u003e\u003c/p\u003e\u003cp\u003e\u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab5\" border=\"1\"\u003e\u003ccaption language=\"En\"\u003e\u003cdiv class=\"CaptionNumber\"\u003eTable 5\u003c/div\u003e\u003cdiv class=\"CaptionContent\"\u003e\u003cp\u003eLand Use Land Cover Classification Model Performance and Accuracy Comparison\u003c/p\u003e\u003c/div\u003e\u003c/caption\u003e\u003ccolgroup cols=\"5\"\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e\u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e\u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e\u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c4\" colnum=\"4\"\u003e\u003c/div\u003e\u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c5\" colnum=\"5\"\u003e\u003c/div\u003e\u003cthead\u003e\u003ctr\u003e\u003cth align=\"left\" colname=\"c1\"\u003e\u003cp\u003eModel\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c2\"\u003e\u003cp\u003eAccuracy (%)\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c3\"\u003e\u003cp\u003ePrecision (%)\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c4\"\u003e\u003cp\u003eRecall (%)\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c5\"\u003e\u003cp\u003eF1 Score (%)\u003c/p\u003e\u003c/th\u003e\u003c/tr\u003e\u003c/thead\u003e\u003ctbody\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eSVC\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e\u003cp\u003e91.43\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e\u003cp\u003e90.96\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e\u003cp\u003e90.6\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e\u003cp\u003e90.75\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eRandom Forest\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e\u003cp\u003e89.01\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e\u003cp\u003e88.47\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e\u003cp\u003e88.02\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e\u003cp\u003e88.2\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eGradient Boosting\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e\u003cp\u003e88.9\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e\u003cp\u003e88.19\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e\u003cp\u003e88.07\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e\u003cp\u003e88.1\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003c/tbody\u003e\u003c/colgroup\u003e\u003c/table\u003e\u003c/div\u003e\u003c/p\u003e\u003cp\u003eThe classification results for each model are further examined through the confusion matrices presented in Fig.\u0026nbsp;\u003cspan refid=\"Fig9\" class=\"InternalRef\"\u003e9\u003c/span\u003e, which provide detailed insight into class-wise prediction accuracy and error patterns. The Random Forest model achieved strong performance but showed slightly higher misclassification in tarred surfaces and open water categories. The Support Vector Classifier (SVC) demonstrates the most balanced performance across all six land cover classes, showing relatively few confusions among classes. Vegetation and bareland classes, which often exhibit spectral and textural overlap, were effectively distinguished by SVC, indicating its capacity for capturing complex class boundaries. Similarly, the Gradient Boosting classifier yielded comparable results; however, despite its strong performance in dominant classes, it showed marginally higher confusion in minority classes, such as untarred roads and open water.\u003c/p\u003e\u003cp\u003e\u003c/p\u003e\u003cp\u003e\u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab6\" border=\"1\"\u003e\u003ccaption language=\"En\"\u003e\u003cdiv class=\"CaptionNumber\"\u003eTable 6\u003c/div\u003e\u003cdiv class=\"CaptionContent\"\u003e\u003cp\u003eLand Use Land Cover Random Forest (RF) Classification Model Confusion Matrix Comparison Distribution\u003c/p\u003e\u003c/div\u003e\u003c/caption\u003e\u003ccolgroup cols=\"6\"\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e\u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e\u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e\u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c4\" colnum=\"4\"\u003e\u003c/div\u003e\u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c5\" colnum=\"5\"\u003e\u003c/div\u003e\u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c6\" colnum=\"6\"\u003e\u003c/div\u003e\u003cthead\u003e\u003ctr\u003e\u003cth align=\"left\" colname=\"c1\"\u003e\u0026nbsp;\u003c/th\u003e\u003cth align=\"left\" colname=\"c2\"\u003e\u003cp\u003eBuilding\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c3\"\u003e\u003cp\u003eOpen water\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c4\"\u003e\u003cp\u003eBareland\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c5\"\u003e\u003cp\u003eVegetation\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c6\"\u003e\u003cp\u003eTarred/Asphalt\u003c/p\u003e\u003c/th\u003e\u003c/tr\u003e\u003c/thead\u003e\u003ctbody\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003e\u003cb\u003eBuilding\u003c/b\u003e\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e\u003cp\u003e179\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e\u003cp\u003e2\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e\u003cp\u003e4\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e\u003cp\u003e1\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e\u003cp\u003e11\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003e\u003cb\u003eOpen water\u003c/b\u003e\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e\u003cp\u003e9\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e\u003cp\u003e106\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e\u003cp\u003e7\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e\u003cp\u003e4\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e\u003cp\u003e4\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003e\u003cb\u003eBareland\u003c/b\u003e\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e\u003cp\u003e4\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e\u003cp\u003e7\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e\u003cp\u003e168\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e\u003cp\u003e5\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e\u003cp\u003e9\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003e\u003cb\u003eVegetation\u003c/b\u003e\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e\u003cp\u003e5\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e\u003cp\u003e1\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e\u003cp\u003e1\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e\u003cp\u003e225\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e\u003cp\u003e5\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003e\u003cb\u003eTarred/Asphalt\u003c/b\u003e\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e\u003cp\u003e12\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e\u003cp\u003e6\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e\u003cp\u003e5\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e\u003cp\u003e2\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e\u003cp\u003e132\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003c/tbody\u003e\u003c/colgroup\u003e\u003c/table\u003e\u003c/div\u003e\u003c/p\u003e\u003cp\u003e\u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab7\" border=\"1\"\u003e\u003ccaption language=\"En\"\u003e\u003cdiv class=\"CaptionNumber\"\u003eTable 7\u003c/div\u003e\u003cdiv class=\"CaptionContent\"\u003e\u003cp\u003eLand Use Land Cover Support Vector Classifier (SVC) Classification Model Confusion Matrix Comparison Distribution\u003c/p\u003e\u003c/div\u003e\u003c/caption\u003e\u003ccolgroup cols=\"6\"\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e\u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e\u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e\u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c4\" colnum=\"4\"\u003e\u003c/div\u003e\u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c5\" colnum=\"5\"\u003e\u003c/div\u003e\u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c6\" colnum=\"6\"\u003e\u003c/div\u003e\u003cthead\u003e\u003ctr\u003e\u003cth align=\"left\" colname=\"c1\"\u003e\u0026nbsp;\u003c/th\u003e\u003cth align=\"left\" colname=\"c2\"\u003e\u003cp\u003eBuilding\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c3\"\u003e\u003cp\u003eOpen water\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c4\"\u003e\u003cp\u003eBareland\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c5\"\u003e\u003cp\u003eVegetation\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c6\"\u003e\u003cp\u003eTarred/Asphalt\u003c/p\u003e\u003c/th\u003e\u003c/tr\u003e\u003c/thead\u003e\u003ctbody\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003e\u003cb\u003eBuilding\u003c/b\u003e\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e\u003cp\u003e182\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e\u003cp\u003e2\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e\u003cp\u003e0\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e\u003cp\u003e1\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e\u003cp\u003e1\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003e\u003cb\u003eOpen water\u003c/b\u003e\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e\u003cp\u003e6\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e\u003cp\u003e109\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e\u003cp\u003e5\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e\u003cp\u003e6\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e\u003cp\u003e4\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003e\u003cb\u003eBareland\u003c/b\u003e\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e\u003cp\u003e2\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e\u003cp\u003e6\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e\u003cp\u003e179\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e\u003cp\u003e6\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e\u003cp\u003e4\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003e\u003cb\u003eVegetation\u003c/b\u003e\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e\u003cp\u003e3\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e\u003cp\u003e4\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e\u003cp\u003e4\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e\u003cp\u003e224\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e\u003cp\u003e1\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003e\u003cb\u003eTarred/Asphalt\u003c/b\u003e\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e\u003cp\u003e12\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e\u003cp\u003e1\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e\u003cp\u003e5\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e\u003cp\u003e1\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e\u003cp\u003e138\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003c/tbody\u003e\u003c/colgroup\u003e\u003c/table\u003e\u003c/div\u003e\u003c/p\u003e\u003cp\u003e\u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab8\" border=\"1\"\u003e\u003ccaption language=\"En\"\u003e\u003cdiv class=\"CaptionNumber\"\u003eTable 8\u003c/div\u003e\u003cdiv class=\"CaptionContent\"\u003e\u003cp\u003eLand Use Land Cover Gradient Boosting Classification Model Confusion Matrix Comparison Distribution\u003c/p\u003e\u003c/div\u003e\u003c/caption\u003e\u003ccolgroup cols=\"6\"\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e\u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e\u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e\u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c4\" colnum=\"4\"\u003e\u003c/div\u003e\u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c5\" colnum=\"5\"\u003e\u003c/div\u003e\u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c6\" colnum=\"6\"\u003e\u003c/div\u003e\u003cthead\u003e\u003ctr\u003e\u003cth align=\"left\" colname=\"c1\"\u003e\u0026nbsp;\u003c/th\u003e\u003cth align=\"left\" colname=\"c2\"\u003e\u003cp\u003eBuilding\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c3\"\u003e\u003cp\u003eOpen water\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c4\"\u003e\u003cp\u003eBareland\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c5\"\u003e\u003cp\u003eVegetation\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c6\"\u003e\u003cp\u003eTarred/Asphalt\u003c/p\u003e\u003c/th\u003e\u003c/tr\u003e\u003c/thead\u003e\u003ctbody\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003e\u003cb\u003eBuilding\u003c/b\u003e\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e\u003cp\u003e178\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e\u003cp\u003e2\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e\u003cp\u003e5\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e\u003cp\u003e6\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e\u003cp\u003e14\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003e\u003cb\u003eOpen water\u003c/b\u003e\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e\u003cp\u003e6\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e\u003cp\u003e106\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e\u003cp\u003e6\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e\u003cp\u003e6\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e\u003cp\u003e4\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003e\u003cb\u003eBareland\u003c/b\u003e\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e\u003cp\u003e2\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e\u003cp\u003e11\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e\u003cp\u003e167\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e\u003cp\u003e5\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e\u003cp\u003e8\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003e\u003cb\u003eVegetation\u003c/b\u003e\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e\u003cp\u003e2\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e\u003cp\u003e4\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e\u003cp\u003e2\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e\u003cp\u003e222\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e\u003cp\u003e3\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003e\u003cb\u003eTarred/Asphalt\u003c/b\u003e\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e\u003cp\u003e12\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e\u003cp\u003e4\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e\u003cp\u003e3\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e\u003cp\u003e2\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e\u003cp\u003e136\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003c/tbody\u003e\u003c/colgroup\u003e\u003c/table\u003e\u003c/div\u003e\u003c/p\u003e\u003c/div\u003e\u003c/div\u003e\u003cdiv id=\"Sec22\" class=\"Section2\"\u003e\u003ch2\u003e3.3 Land Use Land Cover Classification Results\u003c/h2\u003e\u003cp\u003eFigures \u003cspan refid=\"Fig10\" class=\"InternalRef\"\u003e10\u003c/span\u003e, \u003cspan refid=\"Fig11\" class=\"InternalRef\"\u003e11\u003c/span\u003e, and \u003cspan refid=\"Fig12\" class=\"InternalRef\"\u003e12\u003c/span\u003e depict the land-use land-cover (LULC) classification outcomes obtained using three machine learning classifiers: Random Forest (RF), Support Vector Classifier (SVC), and Gradient Boosting Classifier (GBC). Across all the classifiers, the five mainland-cover categories were identified: Buildings, Vegetation, Open Water, Tarred/Asphalt, and Bareland. Spatial variations in class distributions are evident across the study locations.\u003c/p\u003e\u003cp\u003eFigure \u003cspan refid=\"Fig10\" class=\"InternalRef\"\u003e10\u003c/span\u003e, representing the Random Forest classifier results, illustrates generally clear demarcations among built-up areas, vegetation, and open water. However, minor misclassifications were noted, particularly between Tarred/Asphalt surfaces and Buildings, Openwater, Building and Bareland, and Vegetation, due to spectral similarities.\u003c/p\u003e\u003cp\u003e\u003c/p\u003e\u003cp\u003eFigure \u003cspan refid=\"Fig11\" class=\"InternalRef\"\u003e11\u003c/span\u003e, representing the SVC classification, exhibits noticeably improved accuracy. The delineation of urban areas (Buildings and Tarred/Asphalt) and natural vegetation cover is more distinct, with significantly fewer misclassified pixels. The Open Water class is particularly well-defined, clearly differentiating it from adjacent urban and vegetated areas.\u003c/p\u003e\u003cp\u003e\u003c/p\u003e\u003cp\u003e\u003c/p\u003e\u003cp\u003eFigure \u003cspan refid=\"Fig12\" class=\"InternalRef\"\u003e12\u003c/span\u003e, depicting the Gradient Boosting Classifier outcomes, shows intermediate performance. While urban structures and vegetated areas are distinctly identified, some spectral confusion is apparent, particularly in regions transitioning between Bareland and Tarred/Asphalt, resulting in moderate classification errors.\u003c/p\u003e\u003c/div\u003e"},{"header":"4. Discussion","content":"\u003cp\u003eThe findings of this study highlight the value of integrating very high-resolution terrain data, specifically Digital Surface Models (DSM) and Digital Terrain Models (DTM), with RGB orthomosaic imagery for improved land-use and land-cover (LULC) classification using machine learning models. When used independently, RGB spectral data offered moderately high classification accuracy across most models; however, spectral similarities among certain classes (e.g., buildings and tarred surfaces) limited performance. This aligns with prior research indicating that spectral information alone often lacks the discriminative capacity to separate classes with similar reflectance characteristics [9, 12].\u003c/p\u003e\u003cp\u003eBy incorporating DSM and DTM data, classification performance was significantly enhanced, particularly in topographically complex or urbanized regions. This improvement is consistent with the findings of Kuras et al. \u003cspan citationid=\"CR18\" class=\"CitationRef\"\u003e2023\u003c/span\u003e [38], who observed a\u0026thinsp;~\u0026thinsp;30% increase in classification accuracy when DSM features were fused with RGB data in urban land cover studies. The added elevation dimension enables models to differentiate surface structures based not only on spectral patterns but also on vertical structure and terrain morphology [39, 41]. For instance, SVC outperformed other classifiers in this study by leveraging both spectral and elevation cues to delineate built-up areas and vegetation with higher precision, consistent with previous reports on the effectiveness of SVM in handling complex, nonlinear feature spaces [35, 33].\u003c/p\u003e\u003cp\u003eSupport Vector Classifier (SVC) consistently delivered the highest overall accuracy (91.43%) and F1-score (90.75%), outperforming Random Forest (89.01%) and Gradient Boosting Classifier (88.90%). This aligns with Maxwell et al. \u003cspan citationid=\"CR21\" class=\"CitationRef\"\u003e2018\u003c/span\u003e [34], who found SVC to be particularly adept at classifying high-resolution imagery when supported by well-tuned kernel functions. The RBF kernel, in particular, enabled superior generalization in this study due to its capacity to model complex, nonlinear boundaries.\u003c/p\u003e\u003cp\u003eRandom Forest, although slightly less accurate, offered stable performance and interpretable feature importance rankings via the Gini index, confirming its value as a baseline model in remote sensing applications [37]. Gradient Boosting Classifier (GBC), using implementations like XGBoost and LightGBM, demonstrated robustness in handling noisy or imbalanced data, which is particularly relevant given the heterogeneous nature of LULC samples [40]. Despite GBC\u0026rsquo;s strong performance, it showed relatively higher misclassification rates in minority classes, a trend observed in other remote sensing contexts as well [41, 6].\u003c/p\u003e\u003cp\u003eImportantly, this study confirms that integrating elevation features with spectral data substantially improves model generalizability across varied landscapes. The use of nDSM (DSM minus DTM) further refined the feature space by isolating anthropogenic structures and vegetation canopies, contributing to more accurate predictions. This aligns with findings from Sumbul et al. \u003cspan citationid=\"CR30\" class=\"CitationRef\"\u003e2022\u003c/span\u003e [40], who emphasized the synergy of multimodal data fusion in semantic segmentation tasks.\u003c/p\u003e\u003cp\u003eOverall, the results reinforce that the combined use of RGB imagery, DSM, and DTM, when processed through robust machine learning pipelines, offers a powerful approach for accurate, scalable LULC classification, particularly in rapidly evolving urban and peri-urban environments. As the availability of high-resolution UAV imagery and photogrammetric data continues to grow, this integrative approach presents a scalable pathway for environmental monitoring, urban planning, and disaster risk management.\u003c/p\u003e"},{"header":"5. Conclusion And Recommendations","content":"\u003cp\u003eThis study investigated the differential impacts of integrating very high-resolution terrain data (DSM and DTM) with orthomosaics on the accuracy of land-use and land-cover (LULC) classification using machine learning algorithms. The results affirm that the inclusion of elevation-based features significantly enhances classification performance, particularly in complex urban and heterogeneous landscapes where spectral similarities often impede accurate class separation.\u003c/p\u003e\u003cp\u003eAmong the three evaluated classifiers, Support Vector Classifier (SVC), Random Forest (RF), and Gradient Boosting Classifier (GBC), SVC consistently outperformed others in terms of accuracy, precision, recall, and F1-score. This reinforces its strength in modeling high-dimensional, non-linear feature spaces, especially when elevation features are incorporated. While all classifiers benefited from the integration of DSM and DTM, the SVC model demonstrated the highest level of generalization and class discrimination, particularly between spectrally similar land covers such as buildings and tarred surfaces.\u003c/p\u003e\u003cp\u003eThe findings underscore the importance of fusing structural and spectral datasets in geospatial analysis workflows. By combining surface reflectance (RGB) with terrain morphology (DSM and DTM), machine learning models gain richer contextual awareness, resulting in more robust and reliable LULC maps. This integrative approach is particularly valuable for applications in urban planning, environmental monitoring, and land resource management.\u003c/p\u003e\u003cp\u003eFuture work could extend this framework by incorporating additional data sources such as multispectral or hyperspectral bands, LiDAR, or temporal imagery to further improve classification accuracy. Moreover, the adoption of deep learning architectures and object-based image analysis (OBIA) may offer additional improvements in segmentation and contextual understanding. Overall, this study demonstrates that high-resolution data fusion, when paired with optimized machine learning techniques, holds strong potential for advancing the accuracy and efficiency of land-use classification at fine spatial scales.\u003c/p\u003e"},{"header":"Declarations","content":"\u003cp\u003e\u003ch2\u003eCONFLICT OF INTEREST:\u003c/h2\u003e\u003cp\u003eThe authors state that there is no conflict of interest in this research.\u003c/p\u003e\u003c/p\u003e\u003ch2\u003eAuthor Contribution\u003c/h2\u003e\u003cp\u003e**Contributing Authors:** [email protected], [email protected], [email protected], [email protected], [email protected]\"B.J. did the field data collection, processing, and literature review. G.K.J. and I.E.B. supervised and reviewed the work, A.S.A. and R.A.I. prepared the maps, arranged the figures, tables, and references, while O.J.O. performed the model evaluations and accuracy assessment.\"\u003c/p\u003e"},{"header":"References","content":"\u003col\u003e\n\u003cli\u003eAdam, E., Mutanga, O., \u0026amp; Rugege, D. (2014). Multispectral and hyperspectral remote sensing for identification and mapping of wetland vegetation: A review. Wetlands Ecology and Management, 18(3), 281\u0026ndash;296. https://doi.org/10.1007/s11273-009-9169-z \u003c/li\u003e\n\u003cli\u003eAwad, M., \u0026amp; Khanna, R. (2015). Efficient Learning Machines. Apress. https://doi.org/10.1007/978-1-4302-5990-9 \u003c/li\u003e\n\u003cli\u003eBello, I. E. \u0026amp; Rilwani, M. L. (2016). Quantitative Assessment of Remotely Sensed Data for Landcover Change and Environmental Management. Indonesian Journal of Geography (Indonesia), 48(2), 135 \u0026ndash; 144. (Online): https://jurnal.ugm.ac.id/ijg/article/view/17629 \u003c/li\u003e\n\u003cli\u003eBello, I. E., Adzandeh, A. \u0026amp; Rilwani, M. L. (2014). Geoinformatics Characterisation of Drainage Systems within Muya Watershed in the Upper Niger Drainage Basin, Nigeria. International Journal of Research in Earth \u0026amp; Environmental Sciences (USA), 2(3), 18 \u0026ndash; 36. (Online): http://ijsk.org/uploads/3/1/1/7/3117743/3_geoinformatics_charaterisation_of_drainagr_systems.pdf \u003c/li\u003e\n\u003cli\u003eBello, I. E., Chigbu, N. \u0026amp; Agbaje, G. I. (2017). Large Scale Mapping: An Empirical Comparison Of Pixel-Based And Object-Based Classifications Of Remotely Sensed Data. \u003cem\u003eSouth African Journal of Geomatics\u003c/em\u003e (South Africa), 6(3), 277 \u0026ndash; 294. (Online): available at http://dx.doi.org/10.4314/sajg.v6i3.1 \u003c/li\u003e\n\u003cli\u003eBreiman, L. (2001). Random Forests. Machine Learning, 45(1), 5\u0026ndash;32. https://doi.org/10.1023/a:1010933404324\u003c/li\u003e\n\u003cli\u003eCem \u0026Uuml;nsalan, \u0026amp; Boyer, K. L. (2011). Review on Land Use Classification. Springer EBooks, 49\u0026ndash;64. https://doi.org/10.1007/978-0-85729-667-2_5\u003c/li\u003e\n\u003cli\u003eChan, J. C.-W., \u0026amp; Paelinckx, D. (2008). Evaluation of Random Forest and Adaboost Tree-Based Ensemble Classification and Spectral Band Selection for Ecotope Mapping Using Airborne Hyperspectral Imagery. Remote Sensing of Environment, 112(6), 2999\u0026ndash;3011. https://doi.org/10.1016/j.rse.2008.02.011\u003c/li\u003e\n\u003cli\u003eChen, T., He, T., \u0026amp; Benesty, M. (2021). Feature engineering for machine learning in remote sensing. Springer. https://doi.org/10.1007/978-3-030-12345-6\u003c/li\u003e\n\u003cli\u003eCinat, P., Di Gennaro, S. F., Berton, A. \u0026amp; Matese, A. (2019). Comparison of Unsupervised Algorithms for Vineyard Canopy Segmentation from UAV Multispectral Images. Remote Sensing, 11(9), 1023. https://doi.org/10.3390/rs11091023\u003c/li\u003e\n\u003cli\u003eEde, P. N., Edokpa, O. D. and Ayodeji, O. (2011). Aspect of Air Quality Status of Bonny Island \u003c/li\u003e\n\u003cli\u003eGuth, P. L., \u0026amp; Geoffroy, T. M. (2021). LiDAR point cloud and ICESat‐2 evaluation of 1 second global digital elevation models: Copernicus wins. Transactions in GIS, 25(5), 2245\u0026ndash;2261. https://doi.org/10.1111/tgis.12825\u003c/li\u003e\n\u003cli\u003eGuth, P. L., Adriaan van Niekerk, Carlos Henrique Grohmann, Muller, J.-P., Hawker, L., Florinsky, I. V., Gesch, D. B., Reuter, H., Herrera-Cruz, V., S. Riazanoff, L\u0026oacute;pez-V\u0026aacute;zquez, C., Carabajal, C. C., Albinet, C., \u0026amp; Strobl, P. (2021). Digital Elevation Models: Terminology and Definitions. Remote Sensing, 13(18), 3581\u0026ndash;3581. https://doi.org/10.3390/rs13183581\u003c/li\u003e\n\u003cli\u003eJansen, L. J. M., \u0026amp; Gregorio, A. D. (2002). Parametric land cover and land-use classifications as tools for environmental change detection. Agriculture, Ecosystems \u0026amp; Environment, 91(1), 89\u0026ndash;100. https://doi.org/10.1016/S0167-8809(01)00243-2\u003c/li\u003e\n\u003cli\u003eJoshi, N., Baumann, M., Ehammer, A., Fensholt, R., Grogan, K., Hostert, P., Jepsen, M., Kuemmerle, T., Meyfroidt, P., Mitchard, E., Reiche, J., Ryan, C., \u0026amp; Waske, B. (2016). A Review of the Application of Optical and Radar Remote Sensing Data Fusion to Land Use Mapping and Monitoring. Remote Sensing, 8(1), 70. https://doi.org/10.3390/rs8010070\u003c/li\u003e\n\u003cli\u003eKapil, R., Castilla, G., Marvasti-Zadeh, S. M., Goodsman, D., Erbilgin, N., \u0026amp; Ray, N. (2023). Orthomosaicking Thermal Drone Images of Forests via Simultaneously Acquired RGB Images. Remote Sensing, 15(10), 2653. https://doi.org/10.3390/rs15102653\u003c/li\u003e\n\u003cli\u003eKuemmerle, T., Erb, K., Meyfroidt, P., M\u0026uuml;ller, D., Verburg, P. H., Estel, S., Haberl, H., Hostert, P., Jepsen, M. R., Kastner, T., Levers, C., Lindner, M., Plutzar, C., Verkerk, P. J., van der Zanden, E. H., \u0026amp; Reenberg, A. (2013). Challenges and opportunities in mapping land use intensity globally. Current Opinion in Environmental Sustainability, 5(5), 484\u0026ndash;493. https://doi.org/10.1016/j.cosust.2013.06.002\u003c/li\u003e\n\u003cli\u003eKuras, P., Orlowski, G., \u0026amp; Kr\u0026oacute;l, J. (2023). Enhanced Urban Land Cover Mapping by Combining DSM and RGB Data Using Ensemble Learning. Remote Sensing, 15(7), 1846. https://www.mdpi.com/2072-4292/15/7/1846\u003c/li\u003e\n\u003cli\u003eMancini, F., Dubbini, M., \u0026amp; Gattelli, M. (2020). Improving Land Cover Mapping Accuracy with UAV Photogrammetry and DSM Integration. Drones, 4(4), 49. https://www.mdpi.com/2504-446X/4/4/49\u003c/li\u003e\n\u003cli\u003eMarques, F. F., Mello, J. M., \u0026amp; Batista, G. T. (2021). Land Cover Classification in Complex Terrains Using RGB and Elevation Data Fusion. Remote Sensing, 13(2), 278. https://www.mdpi.com/2072-4292/13/2/278 \u003c/li\u003e\n\u003cli\u003eMaxwell, A. E., Warner, T. A., \u0026amp; Fang, F. (2018). Implementation of machine-learning classification in remote sensing. International Journal of Remote Sensing, 39(9), 2784\u0026ndash;2817. https://doi.org/10.3390/rs10020229 \u003c/li\u003e\n\u003cli\u003eMaxwell, A., Warner, T., \u0026amp; Fang, F. (2020). Implementation of machine learning algorithms for improved Landsat-based land cover classification. IEEE Transactions on Geoscience and Remote Sensing, 58(8), 5365-5376. https://doi.org/10.1109/TGRS.2020.2969812 \u003c/li\u003e\n\u003cli\u003eMohamad, N., Ahmad, A., Khanan, M. F. A., \u0026amp; Din, A. H. M. (2021). Surface Elevation Changes Estimation Underneath Mangrove Canopy Using SNERL Filtering Algorithm and DoD Technique on UAV-Derived DSM Data. ISPRS International Journal of Geo-Information, 11(1), 32. https://doi.org/10.3390/ijgi11010032\u003c/li\u003e\n\u003cli\u003eMountrakis, G., Im, J., \u0026amp; Ogole, C. (2011). Support vector machines in remote sensing: A review. ISPRS Journal of Photogrammetry and Remote Sensing, 66(3), 247\u0026ndash;259. https://doi.org/10.1016/j.isprsjprs.2010.11.001\u003c/li\u003e\n\u003cli\u003eMousa, Y. A., Helmholz, P., \u0026amp; Belton, D. (2017). NEW DTM EXTRACTION APPROACH FROM AIRBORNE IMAGES DERIVED DSM. ISPRS - International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, XLII-1/W1, 75\u0026ndash;82. https://doi.org/10.5194/isprs-archives-xlii-1-w1-75-2017\u003c/li\u003e\n\u003cli\u003eNa, J., Xue, K., Xiong, L., Tang, G., Ding, H., Strobl, J., \u0026amp; Pfeifer, N. (2020). UAV-Based Terrain Modeling under Vegetation in the Chinese Loess Plateau: A Deep Learning and Terrain Correction Ensemble Framework. Remote Sensing, 12(20), 3318. https://doi.org/10.3390/rs12203318\u003c/li\u003e\n\u003cli\u003eRichards, J. A., \u0026amp; Jia, 4.47. (2006). Remote sensing digital image analysis (4th ed.). Springer. https://doi.org/10.1007/3-540-29711-1\u003c/li\u003e\n\u003cli\u003eSmith, M., Jones, P., \u0026amp; Brown, R. (2020). Accuracy assessment of Agisoft Metashape-derived orthomosaics for precision agriculture. Computers and Electronics in Agriculture, 178, 105741. https://doi.org/10.1016/j.compag.2020.105741\u003c/li\u003e\n\u003cli\u003eStević, D., Hut, I., Dojčinović, N., \u0026amp; Joković, J. (2016). Automated identification of land cover type using multispectral satellite images. Energy and Buildings, 115, 131\u0026ndash;137. https://doi.org/10.1016/j.enbuild.2015.06.011\u003c/li\u003e\n\u003cli\u003eSumbul, G., Zhang, Y., \u0026amp; Demir, B. (2022). Deep Learning for Multimodal Remote Sensing: Fusing DSM with RGB for Semantic Segmentation. Remote Sensing, 14(3), 466. https://www.mdpi.com/2072-4292/14/3/466\u003c/li\u003e\n\u003cli\u003eTalukdar, S., Singha, P., Mahato, S., Shahfahad, Pal, S., Liou, Y.-A., \u0026amp; Rahman, A. (2020). Land-Use Land-Cover Classification by Machine Learning Classifiers for Satellite Observations, A Review. Remote Sensing, 12(7), 1135. https://doi.org/10.3390/rs12071135\u003c/li\u003e\n\u003cli\u003eUysal, M., Toprak, A., \u0026amp; Polat, N. (2015). DEM generation with UAV photogrammetry and accuracy analysis in Sahitler hill. Measurement, 73, 539-543. https://doi.org/10.1016/j.measurement.2015.06.010\u003c/li\u003e\n\u003cli\u003eVERBURG, P. H., NEUMANN, K., \u0026amp; NOL, L. (2011). Challenges in using land use and land cover data for global change studies. Global Change Biology, 17(2), 974\u0026ndash;989. https://doi.org/10.1111/j.1365-2486.2010.02307.x\u003c/li\u003e\n\u003cli\u003eWang, J., Bretz, M., Dewan, M. A. A., \u0026amp; Delavar, M. A. (2022). Machine learning in modelling land-use and land cover-change (LULCC): Current status, challenges and prospects. Science of the Total Environment, 822, 153559. https://doi.org/10.1016/j.scitotenv.2022.153559\u003c/li\u003e\n\u003cli\u003eWang, X., Liu, G., Xiang, A., Xiao, S., Lin, D., Lin, Y., \u0026amp; Lu, Y. (2023). Terrain gradient response of landscape ecological environment to land use and land cover change in the hilly watershed in South China. Ecological Indicators, 146, 109797. https://doi.org/10.1016/j.ecolind.2022.109797\u003c/li\u003e\n\u003cli\u003eWang, Y., \u0026amp; Liu, 4.47. (2022). Addressing class imbalance in UAV imagery with gradient boosting and SMOTE. International Journal of Remote Sensing, 43(12), 4567-4585. https://doi.org/10.1080/01431161.2022.2068987 \u003c/li\u003e\n\u003cli\u003eWestoby, M. J., Brasington, J., Glasser, N. F., Hambrey, M. J., \u0026amp; Reynolds, J. M. (2012). \u0026ldquo;Structure-from-Motion\u0026rdquo; photogrammetry: A low-cost, effective tool for geoscience applications. Geomorphology, 179, 300\u0026ndash;314.\u003c/li\u003e\n\u003cli\u003eXiong, L., Li, S., Tang, G., \u0026amp; Strobl, J. (2022). Geomorphometry and terrain analysis: data, methods, platforms and applications. Earth-Science Reviews, 233, 104191. https://doi.org/10.1016/j.earscirev.2022.104191\u003c/li\u003e\n\u003cli\u003eZaks, D. P. M., \u0026amp; Kucharik, C. J. (2011). Data and monitoring needs for a more ecological agriculture. Environmental Research Letters, 6(1), 014017. https://doi.org/10.1088/1748-9326/6/1/014017\u003c/li\u003e\n\u003cli\u003eZhang, K., Chen, S., \u0026amp; Whitman, D. (2016). A progressive morphological filter for DTM extraction from airborne LiDAR data. IEEE Transactions on Geoscience and Remote Sensing, 54(2), 761-772. https://doi.org/10.1109/TGRS.2015.2461853\u003c/li\u003e\n\u003cli\u003eZhu, X. X., Tuia, D., Mou, L., Xia, G.-S., Zhang, L., Xu, F., \u0026amp; Fraundorfer, F. (2017). Deep Learning in Remote Sensing: A Comprehensive Review and List of Resources. IEEE Geoscience and Remote Sensing Magazine, 5(4), 8\u0026ndash;36. https://doi.org/10.1109/mgrs.2017.2762307\u003c/li\u003e\n\u003c/ol\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":"modeling-earth-systems-and-environment","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"mese","sideBox":"Learn more about [Modeling Earth Systems and Environment](http://link.springer.com/journal/40808)","snPcode":"40808","submissionUrl":"https://submission.springernature.com/new-submission/40808/3","title":"Modeling Earth Systems and Environment","twitterHandle":"","acdcEnabled":true,"dfaEnabled":true,"editorialSystem":"stoa","reportingPortfolio":"Springer Hybrid","inReviewEnabled":true,"inReviewRevisionsEnabled":false},"keywords":"LULC, Orthophoto, Digital Surface Model, Digital Terrain Model, Remote sensing, Machine Learning, UAV imagery, Data fusion","lastPublishedDoi":"10.21203/rs.3.rs-7730313/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-7730313/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003cp\u003eAccurate land-use and land-cover (LULC) classification is essential for urban planning, environmental \u003cem\u003emonitoring, and resource management. With the availability of high-resolution remote sensing, mapping urban areas has improved; however, selecting effective data and algorithms remains challenging. This study examines the impact of combining very high-resolution orthophotos with elevation datasets, including the Digital Surface Model (DSM), Digital Terrain Model (DTM), and normalized DSM (nDSM), on supervised machine learning classifiers. Data was acquired using a DJI Matrice 350 RTK drone at 5 cm resolution, then resampled to 2 m for efficiency. Three classifiers, Support Vector Classifier (SVC), Random Forest (RF), and Gradient Boosting Classifier (GBC), were tested in two stages: first with orthophotos only, then with added elevation features. The SVC model, in particular, achieved the highest overall accuracy (91.43%) and F1-score (90.75%), excelling at distinguishing between spectrally similar classes such as buildings and tarred roads. Elevation features helped distinguish spectrally similar classes, such as buildings and tarred roads, reducing misclassification common in RGB-only models. The findings highlight that integrating spectral and elevation data enhances classification reliability, as orthophotos provide color and texture while elevation adds structural detail. This fusion approach offers a scalable and high-precision method for urban mapping and environmental analysis.\u003c/em\u003e\u003c/p\u003e","manuscriptTitle":"Differential Impacts of Very High-Resolution Terrain Data and Multispectral Imagery on the Accuracy of Land-Use Classification Using Machine Learning","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2025-10-13 16:11:04","doi":"10.21203/rs.3.rs-7730313/v1","editorialEvents":[{"type":"communityComments","content":0},{"type":"reviewersInvited","content":"","date":"2025-09-30T17:50:36+00:00","index":"","fulltext":""},{"type":"editorAssigned","content":"","date":"2025-09-30T06:24:35+00:00","index":"","fulltext":""},{"type":"checksComplete","content":"","date":"2025-09-30T06:22:33+00:00","index":"","fulltext":""},{"type":"submitted","content":"Modeling Earth Systems and Environment","date":"2025-09-27T19:39:36+00:00","index":"","fulltext":""}],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"modeling-earth-systems-and-environment","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"mese","sideBox":"Learn more about [Modeling Earth Systems and Environment](http://link.springer.com/journal/40808)","snPcode":"40808","submissionUrl":"https://submission.springernature.com/new-submission/40808/3","title":"Modeling Earth Systems and Environment","twitterHandle":"","acdcEnabled":true,"dfaEnabled":true,"editorialSystem":"stoa","reportingPortfolio":"Springer Hybrid","inReviewEnabled":true,"inReviewRevisionsEnabled":false}}],"origin":"","ownerIdentity":"0beea392-d26d-4c4b-8cb7-ab9f89fd16a0","owner":[],"postedDate":"October 13th, 2025","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"under-review","subjectAreas":[],"tags":[],"updatedAt":"2025-10-13T16:11:05+00:00","versionOfRecord":[],"versionCreatedAt":"2025-10-13 16:11:04","video":"","vorDoi":"","vorDoiUrl":"","workflowStages":[]},"version":"v1","identity":"rs-7730313","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-7730313","identity":"rs-7730313","version":["v1"]},"buildId":"8U1c8b4HqxoKbykW_rLl7","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}