Scholarlink

<p>Despite widespread use of conventional drought indices, accurately capturing the spatial–temporal dynamics of drought remains challenging, particularly in semi-arid regions where vegetation–climate interactions are highly complex. Therefore, this study aimed to conduct spatial modeling and time-series analysis of drought, with a comparative application of Machine Learning Algorithms (MLAs) and object-based methods in the Taleqan watershed, Iran. At first, drought conditioning indices, including Modified Chlorophyll Absorption Ratio Index Improved (MCARI2), Moisture Stress Index (MSI), Normalized-Difference Snow Index (NDSI), Normalized Difference Vegetation Index (NDVI), Radar Vegetation Index (RVI), Soil and Atmospherically Resistant Vegetation Index (SARVI), Soil Adjusted Vegetation Index (SAVI), Vegetation Condition Index (VCI), and Water Index (WI), were selected. After calculating the indices, ML algorithms, including Support Vector Regression (SVR), Random Forest (RF), K-Nearest Neighbors (KNN), Simple Linear Regression (SLR), and Support Vector Machine (SVM), were implemented in Python. The object-based method used vegetation cover indices, including Temperature Condition Index (TCI), Vegetation Health Index (VHI), and VCI, for spatial and temporal drought investigation. The results showed that among the MLAs, the SVR algorithm was selected as optimal, with MAE, MSE, RMSE, and AUC values of 0.29, 0.15, 0.30, and 0.93, respectively. Based on the SVR algorithm, the spatial change analysis showed that about 19.84% of the studied area was in the no-drought class. Meanwhile, classes with low and moderate drought were included at 16.97 and 22.35%, respectively. Finally, the highest value in the drought map corresponded to the class with the highest drought, at about 40.85%. Monitoring the temporal drought changes using the object-based method also showed that all three indices, TCI, VHI, and VCI, had significant trends in 2001–2022. The comparison between machine learning algorithms (MLAs) and object-based methods indicates that MLAs provide a clearer and more interpretable representation of spatial drought patterns. This enhanced interpretability facilitates the identification of critical drought-prone areas and supports more informed, evidence-based decision-making by water resource managers and policymakers.</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

Advanced spatial–temporal drought mapping using machine learning and object-based methods

  • 期刊论文

  • 发表日期：

    2026年6月1日

查看全文
  • Ali Nasiri Khiavi,
  • Mohammad Tavoosi,
  • Alban Kuriqi

摘要

Despite widespread use of conventional drought indices, accurately capturing the spatial–temporal dynamics of drought remains challenging, particularly in semi-arid regions where vegetation–climate interactions are highly complex. Therefore, this study aimed to conduct spatial modeling and time-series analysis of drought, with a comparative application of Machine Learning Algorithms (MLAs) and object-based methods in the Taleqan watershed, Iran. At first, drought conditioning indices, including Modified Chlorophyll Absorption Ratio Index Improved (MCARI2), Moisture Stress Index (MSI), Normalized-Difference Snow Index (NDSI), Normalized Difference Vegetation Index (NDVI), Radar Vegetation Index (RVI), Soil and Atmospherically Resistant Vegetation Index (SARVI), Soil Adjusted Vegetation Index (SAVI), Vegetation Condition Index (VCI), and Water Index (WI), were selected. After calculating the indices, ML algorithms, including Support Vector Regression (SVR), Random Forest (RF), K-Nearest Neighbors (KNN), Simple Linear Regression (SLR), and Support Vector Machine (SVM), were implemented in Python. The object-based method used vegetation cover indices, including Temperature Condition Index (TCI), Vegetation Health Index (VHI), and VCI, for spatial and temporal drought investigation. The results showed that among the MLAs, the SVR algorithm was selected as optimal, with MAE, MSE, RMSE, and AUC values of 0.29, 0.15, 0.30, and 0.93, respectively. Based on the SVR algorithm, the spatial change analysis showed that about 19.84% of the studied area was in the no-drought class. Meanwhile, classes with low and moderate drought were included at 16.97 and 22.35%, respectively. Finally, the highest value in the drought map corresponded to the class with the highest drought, at about 40.85%. Monitoring the temporal drought changes using the object-based method also showed that all three indices, TCI, VHI, and VCI, had significant trends in 2001–2022. The comparison between machine learning algorithms (MLAs) and object-based methods indicates that MLAs provide a clearer and more interpretable representation of spatial drought patterns. This enhanced interpretability facilitates the identification of critical drought-prone areas and supports more informed, evidence-based decision-making by water resource managers and policymakers.