Scholarlink

<p>Agricultural drought poses a persistent challenge in semi-arid regions of India, where monsoon variability significantly affects crop productivity and rural livelihoods. This study integrates remote sensing indices and deep learning modelling to characterize and forecast drought conditions in the Papagni River Basin, Andhra Pradesh. The Vegetation Health Index (VHI), derived from the MODIS Normalized Difference Vegetation Index (NDVI) and land surface temperature (LST), was computed for the period 2001–2022 to assess historical drought dynamics. The main findings are (i) Spatial analysis revealed that mild drought (30 &lt; VHI &lt; 40) dominated, affecting 50–80% of the basin annually, while moderate drought (20 &lt; VHI &lt; 30) was episodic but severe in years such as 2005, 2006, and 2010. No drought conditions (40 &lt; VHI &lt; 60) fluctuated widely, and healthy vegetation (VHI &gt; 60) remained negligible. (ii) Drought hotspots were consistently observed in the northern and central basin. (iii) A comparative evaluation of SPI-12, SPEI-12, and VHI shows that although meteorological drought years generally correspond with vegetation stress, divergences arise due to evapotranspiration effects, irrigation buffering, and seasonal rainfall variability. This highlights that agricultural drought reflects the realized vegetation response rather than a direct linear response to precipitation anomalies, underscoring the complementary roles of meteorological and vegetation-based drought metrics. (iv) To predict future drought, a Long Short-Term Memory (LSTM) network was developed using multi-source predictors (NDVI, LST, Normalized Difference Built-up Index (NDBI), Modified Normalized Difference Water Index (MNDWI), Distance to Built-up area (DBU), Distance to Bare soil (DBS)) along with lagged VHI sequences, and the model achieved high accuracy (R² = 0.9473, MAE = 0.4949, MSE = 0.4246), with spatial errors largely within a 2–3% margin, confirming its robustness. (v) Projections for 2025–2040 indicate that mild drought will remain the chronic condition, covering 40–80% of the basin, while no drought conditions will fluctuate between 20 and 50%. Moderate drought is expected to be minimal, though the persistence of widespread mild drought underscores continued agricultural vulnerability. These findings highlight the potential of integrating VHI-based remote sensing with LSTM prediction for supporting strategic drought risk assessment and early warning frameworks. The study provides a transferable framework for climate-resilient agriculture and water resource management while also identifying future research opportunities to incorporate climate scenarios, soil moisture, and hybrid deep learning approaches.</p> Graphical Abstract <p></p> <p>This graphical abstract illustrates an integrated framework for agricultural drought assessment and forecasting in the semi-arid Papagni River Basin, India. The left panel shows the study area and the multi-source predictor variables derived from MODIS remote sensing datasets, including NDVI, LST, NDBI, MNDWI, and additional biophysical indices. Historical drought conditions (2001–2022) were evaluated using the VHI, revealing widespread mild drought across 50–80% of the basin, with recurring hotspots in the northern and central regions. The central panel visualizes these spatial patterns of mild, moderate, and hotspot drought classes. The right panel depicts the LSTM-based drought forecasting model, which incorporates multi-source predictors and lagged VHI sequences to predict future drought conditions. The model’s high predictive accuracy (R² = 0.9473) is reflected in the spatial drought projections for 2025–2040, which indicate the continued dominance of mild drought (40–80% areal coverage) and fluctuating no-drought conditions (20–50%). Moderate drought remains limited but persists locally. Overall, the graphical workflow highlights the integration of VHI-based remote sensing with deep learning for early-warning systems and climate-resilient agricultural planning in drought-prone regions.</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

Deep Learning-Based Agricultural Drought Monitoring and Prediction Using Vegetation Health Index in the Papagni River Basin, India

  • 期刊论文

  • 发表日期：

    2026年3月26日

查看全文
  • Chinthu Naresh,
  • Aneesh Mathew

摘要

Agricultural drought poses a persistent challenge in semi-arid regions of India, where monsoon variability significantly affects crop productivity and rural livelihoods. This study integrates remote sensing indices and deep learning modelling to characterize and forecast drought conditions in the Papagni River Basin, Andhra Pradesh. The Vegetation Health Index (VHI), derived from the MODIS Normalized Difference Vegetation Index (NDVI) and land surface temperature (LST), was computed for the period 2001–2022 to assess historical drought dynamics. The main findings are (i) Spatial analysis revealed that mild drought (30 < VHI < 40) dominated, affecting 50–80% of the basin annually, while moderate drought (20 < VHI < 30) was episodic but severe in years such as 2005, 2006, and 2010. No drought conditions (40 < VHI < 60) fluctuated widely, and healthy vegetation (VHI > 60) remained negligible. (ii) Drought hotspots were consistently observed in the northern and central basin. (iii) A comparative evaluation of SPI-12, SPEI-12, and VHI shows that although meteorological drought years generally correspond with vegetation stress, divergences arise due to evapotranspiration effects, irrigation buffering, and seasonal rainfall variability. This highlights that agricultural drought reflects the realized vegetation response rather than a direct linear response to precipitation anomalies, underscoring the complementary roles of meteorological and vegetation-based drought metrics. (iv) To predict future drought, a Long Short-Term Memory (LSTM) network was developed using multi-source predictors (NDVI, LST, Normalized Difference Built-up Index (NDBI), Modified Normalized Difference Water Index (MNDWI), Distance to Built-up area (DBU), Distance to Bare soil (DBS)) along with lagged VHI sequences, and the model achieved high accuracy (R² = 0.9473, MAE = 0.4949, MSE = 0.4246), with spatial errors largely within a 2–3% margin, confirming its robustness. (v) Projections for 2025–2040 indicate that mild drought will remain the chronic condition, covering 40–80% of the basin, while no drought conditions will fluctuate between 20 and 50%. Moderate drought is expected to be minimal, though the persistence of widespread mild drought underscores continued agricultural vulnerability. These findings highlight the potential of integrating VHI-based remote sensing with LSTM prediction for supporting strategic drought risk assessment and early warning frameworks. The study provides a transferable framework for climate-resilient agriculture and water resource management while also identifying future research opportunities to incorporate climate scenarios, soil moisture, and hybrid deep learning approaches.

Graphical Abstract

This graphical abstract illustrates an integrated framework for agricultural drought assessment and forecasting in the semi-arid Papagni River Basin, India. The left panel shows the study area and the multi-source predictor variables derived from MODIS remote sensing datasets, including NDVI, LST, NDBI, MNDWI, and additional biophysical indices. Historical drought conditions (2001–2022) were evaluated using the VHI, revealing widespread mild drought across 50–80% of the basin, with recurring hotspots in the northern and central regions. The central panel visualizes these spatial patterns of mild, moderate, and hotspot drought classes. The right panel depicts the LSTM-based drought forecasting model, which incorporates multi-source predictors and lagged VHI sequences to predict future drought conditions. The model’s high predictive accuracy (R² = 0.9473) is reflected in the spatial drought projections for 2025–2040, which indicate the continued dominance of mild drought (40–80% areal coverage) and fluctuating no-drought conditions (20–50%). Moderate drought remains limited but persists locally. Overall, the graphical workflow highlights the integration of VHI-based remote sensing with deep learning for early-warning systems and climate-resilient agricultural planning in drought-prone regions.