Abstract

Three Sentinel-1A datasets in vertical transmitted and horizontal received (VH) and vertical transmitted and vertical received (VV) polarisations, and the linear combination of VH and VV (VHVV) are evaluated for rice green leaf area index (LAI) estimation using four machine learning regression models [Support Vector Machine (SVM), k-Nearest Neighbour (k-NN), Random Forest (RF) and Gradient Boosting Decision Tree (GBDT)]. Results showed that for the entire growing season, VV outperformed VH, recording an R² of 0.68 and an RMSE of 0.98 m²/m² with the k-NN model. However, VHVV produced the most accurate estimates with GBDT (R² of 0.82 and RMSE of 0.68 m²/m²), followed by that of VHVV with RF (R² of 0.78 and RMSE of 0.90 m²/m²). Our findings have further confirmed that combining VH and VV data can achieve improved rice growth modelling, and that tree-based algorithms can better handle data dimensionality.

摘要

三个 Sentinel-1A 数据集在垂直传输和水平接收 (VH) 以及垂直传输和垂直接收 (VV) 极化，以及 VH 和 VV (VHVV) 的线性组合用于水稻绿叶面积指数 (LAI) 估计使用四个机器学习回归模型 [支持向量机 (SVM)、k-最近邻 (k-NN)、随机森林 (RF) 和梯度提升决策树 (GBDT)]。结果表明，在整个生长季节，VV 的表现优于 VH，使用 k-NN 模型记录的 R ²为 0.68，RMSE 为 0.98 m ² /m ^{2 。}然而，VHVV 使用 GBDT 产生了最准确的估计值（R ²为 0.82，RMSE 为 0.68 m ² /m ²)，其次是具有 RF 的 VHVV（R ²为 0.78，RMSE 为 0.90 m ² /m ²）。我们的研究结果进一步证实，结合 VH 和 VV 数据可以实现改进的水稻生长建模，并且基于树的算法可以更好地处理数据维度。