Abstract

In the field of forestry studies, microwave remote sensing has broad applications due to the penetration into the semi-transparent media.This feature is used for the estimation of biophysical parameters and monitoring of deforestation.Therefore, the estimation of biophysical parameters is essential for assessing carbon stock management. Hence, the aboveground biomass (AGB) using synthetic aperture radar (SAR) data is recognized as typical approaches in forest application. However, the integrated use of polarimetric (PolSAR) and interferometric (PolInSAR) data might be more efficient tools for AGB mapping. Accordingly, in this study with the integrated data, the efficiency of machine learning techniques including random forest regression (RFR) and multiple linear regression (MLR) model were assessed and compared for the prediction of AGB. The analyses were performed using an image pair of fully polarimetric Radarsat-2 C-band dataset and the related field data of Malhan Forest Range, Dehradun Forest Division, which were collected using the systematic sampling technique. Particularly, the training and testing of the models were done using the field sample plots. The experimental results showed that the RFR algorithm provided a better prediction result of AGB than the MLR model. The correlation coefficient (R²) and root mean square error (RMSE)for the RFR algorithm was estimated to be around 0.65 and 24.33 Mg/ha, respectively; while for the MLR model, R2 and RMSE are estimated as 0.54 and 33.05 Mg/ha, respectively. Therefore, it was concluded that the prediction of AGB through the machine learning technique using PolSAR and PolInSAR data has a significant advantage for accurate estimation of the AGB.

摘要

在林业研究领域，微波遥感由于已渗透到半透明介质中而得到了广泛的应用，此功能用于生物物理参数的估计和森林砍伐的监测，因此，生物物理参数的估计对于评估碳存量管理。因此，使用合成孔径雷达（SAR）数据的地上生物量（AGB）被认为是森林应用中的典型方法。但是，极化（PolSAR）和干涉（PolInSAR）数据的综合使用可能是用于AGB映射的更有效工具。因此，在这项具有综合数据的研究中，评估并比较了包括随机森林回归（RFR）和多元线性回归（MLR）模型在内的机器学习技术的效率，以预测AGB。使用全极化Radarsat-2 C波段数据对的图像对和Dehradun森林分区Malhan Forest Range的相关野外数据进行了分析，这些数据是使用系统采样技术收集的。特别是，使用现场样本图对模型进行了训练和测试。实验结果表明，与MLR模型相比，RFR算法提供了更好的AGB预测结果。相关系数（R 实验结果表明，与MLR模型相比，RFR算法提供了更好的AGB预测结果。相关系数（R 实验结果表明，与MLR模型相比，RFR算法提供了更好的AGB预测结果。相关系数（R²）和RFR算法的均方根误差（RMSE）估计分别约为0.65和24.33 Mg / ha；而对于MLR模型，R2和RMSE分别估计为0.54和33.05 Mg / ha。因此，可以得出结论，通过使用PolSAR和PolInSAR数据的机器学习技术对AGB进行预测对于准确估计AGB具有明显的优势。