Abstract

This study explored the utilization of rotation forests (RTF) and extreme gradient boosting (XGBoost) machine learning algorithms (MLAs) to classify drought damage in commercial forests in KwaZulu-Natal (KZN). These algorithms were trained using information obtained from the Terra Moderate Resolution Imaging Spectroradiometer (MODIS) derived vegetation and conditional drought indices. The results demonstrated that both algorithms were capable of accurately detecting trees that exhibit drought damage and those that did not, this was more apparent when classifying based on information derived from conditional drought indices that yielded an overall accuracy of 82% and 76% for XGBoost and RTF, respectively. However, the accuracy decreased when using vegetation indices data, resulting in an accuracy of 69% and 72% for XGBoost and RTF, respectively. Overall, the results demonstrated that MLAs could be utilized for classifying drought damage on forest vegetation. Additionally, the study showed that MODIS imagery could be used for MLA classification.

摘要

本研究探讨了利用旋转森林 (RTF) 和极端梯度提升 (XGBoost) 机器学习算法 (MLA) 对夸祖鲁-纳塔尔 (KZN) 商业林的干旱损害进行分类。这些算法使用从 Terra 中分辨率成像光谱仪 (MODIS) 导出的植被和条件干旱指数获得的信息进行训练。结果表明，这两种算法都能够准确地检测出表现出干旱损害的树木和没有表现出干旱损害的树木，这在基于条件干旱指数的信息进行分类时更为明显，XGBoost 和RTF，分别。然而，当使用植被指数数据时，准确度下降，导致 XGBoost 和 RTF 的准确度分别为 69% 和 72%。总体而言，结果表明 MLA 可用于对森林植被的干旱损害进行分类。此外，研究表明 MODIS 图像可用于 MLA 分类。