Scholarlink

<p>This study assesses the potential of satellite-based Leaf Area Index (LAI) estimation for smallholder wheat fields (&lt; 1 acre) in the semi-arid Bundelkhand region of India using machine learning approaches. LAI was estimated for the spring wheat seasons of 2020–2021 and 2021–2022 across six villages in Jhansi and Niwari districts using Sentinel-2 and Landsat-8 data. Three machine learning algorithms—Random Forest (RF), Support Vector Machine (SVM) with linear and radial basis function (RBF) kernels, and Extreme Gradient Boosting (XGBoost)—were evaluated. LAI values derived from satellite data were validated against field observations collected during three crop growth stages using performance metrics including Pearson’s correlation coefficient (R), Root Mean Square Error (RMSE), Mean Absolute Error (MAE), Mean Square Error (MSE), Coefficient of Determination (R<sup>2</sup>), Multiplicative Bias (MBias), and F test. RF and SVM-RBF consistently achieved the highest accuracy across both satellite datasets. For Sentinel-2, both RF and SVM-RBF yielded R = 0.94, RMSE = 0.40, MAE = 0.29 and 0.30, R<sup>2</sup> = 0.88, MSE = 0.16, MBias = 1.02 and 1.00, and F test values of 1692.21 and 1664.69, respectively. For Landsat-8, SVM-RBF recorded the lowest RMSE of 0.37, with R = 0.94, MAE = 0.29, MBias = 0.99, R<sup>2</sup> = 0.88, MSE = 0.14, and F test = 1508.75, while RF achieved RMSE = 0.38, MAE = 0.28, MBias = 1.00, R<sup>2</sup> = 0.88, MSE = 0.15, and F test = 1499.62. XGBoost also produced high R-values (0.93–0.94) but exhibited slightly higher error metrics with RMSE = 0.43 (Sentinel-2) and 0.40 (Landsat-8), and corresponding F test values of 1064.99 and 1196.50. The SVM linear model consistently underperformed across all datasets, showing the lowest R (0.84 for Sentinel-2 and 0.78 for Landsat-8) and highest RMSE (0.62 and 0.69), MAE (0.48 and 0.53), and MSE (0.39 and 0.48), with F test values of 584.81 and 351.73, respectively. Overall, the integration of red-edge, near-infrared, and shortwave infrared bands, along with solar and view zenith angles, enhanced LAI retrieval accuracy. The study demonstrates that RF and SVM-RBF are robust tools for scalable LAI monitoring in smallholder systems and supports their application in precision agriculture and crop modeling frameworks in similar agro-ecological contexts.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Remote Sensing-Based Leaf Area Index Estimation of Spring Wheat Through Machine-Learning Approaches

  • 期刊论文

  • 发表日期：

    2025年12月4日

查看全文
  • Pratibha Prakash,
  • Swadhina Koley,
  • Soora Naresh Kumar,
  • Ramesh Chand Harit,
  • Jitendra Kumar Gupta,
  • Ravi Kumar

摘要

This study assesses the potential of satellite-based Leaf Area Index (LAI) estimation for smallholder wheat fields (< 1 acre) in the semi-arid Bundelkhand region of India using machine learning approaches. LAI was estimated for the spring wheat seasons of 2020–2021 and 2021–2022 across six villages in Jhansi and Niwari districts using Sentinel-2 and Landsat-8 data. Three machine learning algorithms—Random Forest (RF), Support Vector Machine (SVM) with linear and radial basis function (RBF) kernels, and Extreme Gradient Boosting (XGBoost)—were evaluated. LAI values derived from satellite data were validated against field observations collected during three crop growth stages using performance metrics including Pearson’s correlation coefficient (R), Root Mean Square Error (RMSE), Mean Absolute Error (MAE), Mean Square Error (MSE), Coefficient of Determination (R2), Multiplicative Bias (MBias), and F test. RF and SVM-RBF consistently achieved the highest accuracy across both satellite datasets. For Sentinel-2, both RF and SVM-RBF yielded R = 0.94, RMSE = 0.40, MAE = 0.29 and 0.30, R2 = 0.88, MSE = 0.16, MBias = 1.02 and 1.00, and F test values of 1692.21 and 1664.69, respectively. For Landsat-8, SVM-RBF recorded the lowest RMSE of 0.37, with R = 0.94, MAE = 0.29, MBias = 0.99, R2 = 0.88, MSE = 0.14, and F test = 1508.75, while RF achieved RMSE = 0.38, MAE = 0.28, MBias = 1.00, R2 = 0.88, MSE = 0.15, and F test = 1499.62. XGBoost also produced high R-values (0.93–0.94) but exhibited slightly higher error metrics with RMSE = 0.43 (Sentinel-2) and 0.40 (Landsat-8), and corresponding F test values of 1064.99 and 1196.50. The SVM linear model consistently underperformed across all datasets, showing the lowest R (0.84 for Sentinel-2 and 0.78 for Landsat-8) and highest RMSE (0.62 and 0.69), MAE (0.48 and 0.53), and MSE (0.39 and 0.48), with F test values of 584.81 and 351.73, respectively. Overall, the integration of red-edge, near-infrared, and shortwave infrared bands, along with solar and view zenith angles, enhanced LAI retrieval accuracy. The study demonstrates that RF and SVM-RBF are robust tools for scalable LAI monitoring in smallholder systems and supports their application in precision agriculture and crop modeling frameworks in similar agro-ecological contexts.