Scholarlink

<p>Climate change is intensifying hydroclimatic extremes worldwide, posing increasing challenges for ecologically sensitive regions. Kenya’s Tsavo Conservation Area (TCA) characterized by a bimodal rainfall regime is particularly vulnerable to recurrent droughts and prolonged wet conditions. This study develops a Random Forest (RF) classification framework to classify drought and wet conditions in the Tsavo Conservation Area. The framework integrates the Standardized Precipitation Index (SPI), the Vegetation Condition Index (VCI) and elevation to capture meteorological, ecological and topographic influences. SPI and VCI were derived from daily CHIRPS v2.0 precipitation data and MODIS products, representing hydrological and vegetation responses to moisture variability. Drought and wet extremes classes were generated using SPI- and VCI-based thresholds and used as target labels for RF classification. The RF framework was then applied to classify annual drought and wet conditions across the TCA for the period 2000-2024, achieving a macro-average F1-score of 0.998 and a Cohen’s Kappa coefficient of 0.996, demonstrating high reliability of drought and wet extremes classifications. The results reveal interannual spatial variability, highlighting pronounced drought in 2003, 2005, and 2007, as well as distinct wet conditions in 2006, 2018-2020 and 2023. Furthermore, the increasing temperature trend indicates rising moisture stress (0.04&#xa0;°C per year). Consecutive Dry Days (CDD) and Consecutive Wet Days (CWD) provide supporting evidence for the observed variability in drought and wet conditions. However, ground-based validation data for wet conditions remain limited. Integrating multi-source indicators within a machine-learning framework, this study demonstrates a practical approach for drought-wet extremes classification in conservation landscapes. The findings underscore the relevance of data-driven climate risk assessment tools for supporting early warning systems and climate-resilient land-use planning in protected areas.</p> Graphical Abstract <p></p> <p>Graphical abstract summarizes the overall study workflow. The Standardized Precipitation Index (SPI-12) and annual Vegetation Condition Index (VCI) were derived from CHIRPS v2.0 precipitation and MODIS vegetation indices respectively, with CHIRPS v2.0 validated against ground-based rainfall observations. Consecutive Dry Days (CDD) and Consecutive Wet Days (CWD) were computed across the Tsavo Conservation Area. Furthermore, precipitation and temperature trends were calculated using CHIRPS v2.0 and ERA5 reanalysis data. Random Forest classification framework was applied to classify drought and wet conditions using SPI, VCI and Digital Elevation Model (DEM) data as input variables. The results show major drought years (2003, 2005 and 2007) and wet years (2006, 2018, 2019, 2020 and 2023). Overall, the Random Forest classification framework provides a practical approach for characterizing the spatial and temporal variability of drought and wet extremes in the Tsavo Conservation Area, Kenya. These findings have important implications for land-use planning, water resource management, and human-wildlife conflict mitigation.</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

Machine Learning-Based Categorization of Drought and Wet Conditions Using SPI, VCI and Elevation in Tsavo Conservation Area, Kenya

  • 期刊论文

  • 发表日期：

    2026年7月3日

查看全文
  • Sudip Pandey,
  • Konduru Rakesh Teja,
  • Joseph Mukeka,
  • Shadrack Ngene,
  • William Dzekedzeke Jr,
  • Kabo Diraditsile,
  • Michael Gebreslasie,
  • Remi Chandran,
  • Fredrick Lala,
  • Ram Avtar

摘要

Climate change is intensifying hydroclimatic extremes worldwide, posing increasing challenges for ecologically sensitive regions. Kenya’s Tsavo Conservation Area (TCA) characterized by a bimodal rainfall regime is particularly vulnerable to recurrent droughts and prolonged wet conditions. This study develops a Random Forest (RF) classification framework to classify drought and wet conditions in the Tsavo Conservation Area. The framework integrates the Standardized Precipitation Index (SPI), the Vegetation Condition Index (VCI) and elevation to capture meteorological, ecological and topographic influences. SPI and VCI were derived from daily CHIRPS v2.0 precipitation data and MODIS products, representing hydrological and vegetation responses to moisture variability. Drought and wet extremes classes were generated using SPI- and VCI-based thresholds and used as target labels for RF classification. The RF framework was then applied to classify annual drought and wet conditions across the TCA for the period 2000-2024, achieving a macro-average F1-score of 0.998 and a Cohen’s Kappa coefficient of 0.996, demonstrating high reliability of drought and wet extremes classifications. The results reveal interannual spatial variability, highlighting pronounced drought in 2003, 2005, and 2007, as well as distinct wet conditions in 2006, 2018-2020 and 2023. Furthermore, the increasing temperature trend indicates rising moisture stress (0.04 °C per year). Consecutive Dry Days (CDD) and Consecutive Wet Days (CWD) provide supporting evidence for the observed variability in drought and wet conditions. However, ground-based validation data for wet conditions remain limited. Integrating multi-source indicators within a machine-learning framework, this study demonstrates a practical approach for drought-wet extremes classification in conservation landscapes. The findings underscore the relevance of data-driven climate risk assessment tools for supporting early warning systems and climate-resilient land-use planning in protected areas.

Graphical Abstract

Graphical abstract summarizes the overall study workflow. The Standardized Precipitation Index (SPI-12) and annual Vegetation Condition Index (VCI) were derived from CHIRPS v2.0 precipitation and MODIS vegetation indices respectively, with CHIRPS v2.0 validated against ground-based rainfall observations. Consecutive Dry Days (CDD) and Consecutive Wet Days (CWD) were computed across the Tsavo Conservation Area. Furthermore, precipitation and temperature trends were calculated using CHIRPS v2.0 and ERA5 reanalysis data. Random Forest classification framework was applied to classify drought and wet conditions using SPI, VCI and Digital Elevation Model (DEM) data as input variables. The results show major drought years (2003, 2005 and 2007) and wet years (2006, 2018, 2019, 2020 and 2023). Overall, the Random Forest classification framework provides a practical approach for characterizing the spatial and temporal variability of drought and wet extremes in the Tsavo Conservation Area, Kenya. These findings have important implications for land-use planning, water resource management, and human-wildlife conflict mitigation.