Naresh et al. (2026) Deep Learning-Based Agricultural Drought Monitoring and Prediction Using Vegetation Health Index in the Papagni River Basin, India

Identification

• Journal: Earth Systems and Environment
• Year: 2026
• Date: 2026-03-26
• Authors: Chinthu Naresh, Aneesh Mathew
• DOI: 10.1007/s41748-026-01097-4

Research Groups

• Department of Civil Engineering, National Institute of Technology, Tiruchirappalli, India

Short Summary

This study developed and validated a deep learning-based framework using the Vegetation Health Index (VHI) to monitor historical agricultural drought (2001–2022) and predict future conditions (2025–2040) in India's semi-arid Papagni River Basin, revealing chronic mild drought and achieving high predictive accuracy with an LSTM model.

Objective

• To assess historical agricultural drought variability (2001–2022) using MODIS-derived VHI, quantifying the area and percentage under different drought severity classes.
• To develop and validate a Long Short-Term Memory (LSTM) model incorporating multi-source remote sensing predictors to forecast future VHI and associated agricultural drought conditions.
• To evaluate projected agricultural drought risks and spatiotemporal trends for the near-future period (2025–2040) in the Papagni River Basin.

Study Configuration

• Spatial Scale: Papagni River Basin, Andhra Pradesh, India, covering approximately 8,500 square kilometers (km²). Data processed at 250 meter (m) resolution.
• Temporal Scale:
  • Historical assessment: 2001–2022 (22 years).
  • Prediction period: 2025–2040 (16 years).
  • Data temporal resolution: Annual mean values for VHI, SPI-12, and SPEI-12.

Methodology and Data

• Models used:
  • Vegetation Health Index (VHI) calculation (integrating VCI and TCI).
  • Standardized Precipitation Index (SPI-12).
  • Standardized Precipitation Evapotranspiration Index (SPEI-12).
  • Long Short-Term Memory (LSTM) deep learning network for VHI prediction.
• Data sources:
  • Satellite remote sensing: MODIS (MOD13A2 Normalized Difference Vegetation Index (NDVI), MOD11A2 Land Surface Temperature (LST)).
  • Derived indices: Normalized Difference Built-up Index (NDBI), Modified Normalized Difference Water Index (MNDWI), Dry Bare Soil (DBS) index, Dry Built-up (DBU) index.
  • Ground observations: Annual rainfall and temperature data (maximum and minimum) from the Indian Meteorological Department (IMD).

Main Results

• Historical Drought (2001–2022): Mild drought (30 < VHI < 40) was the dominant condition, affecting 50–80% of the basin annually. Moderate drought (20 < VHI < 30) was episodic but severe in years like 2005 (affecting 29% of the basin), 2006, and 2010. Healthy vegetation (VHI > 60) was negligible.
• Drought Hotspots: Consistently observed in the northern and central parts of the basin.
• Meteorological vs. Agricultural Drought: While SPI-12 and SPEI-12 generally corresponded with VHI-based vegetation stress, divergences were noted. For instance, 2018 had severe meteorological drought (SPEI-12 = -1.82) but limited VHI-based drought (17.53% mild, 0.03% moderate), whereas 2005 showed high moderate VHI drought (28.96%) despite positive SPI/SPEI values. This highlights the influence of evapotranspiration, irrigation buffering, and seasonal rainfall variability on agricultural drought.
• LSTM Model Performance: Achieved high predictive accuracy for VHI (R² = 0.9473, Mean Absolute Error (MAE) = 0.4949, Mean Squared Error (MSE) = 0.4246), with most spatial errors within a 2–3% margin.
• Future Projections (2025–2040): Mild drought is projected to remain the chronic condition, covering 40–80% of the basin. No drought conditions are expected to fluctuate between 20% and 55%. Moderate drought is anticipated to be minimal.

Contributions

• Introduced a novel approach to agricultural drought prediction by reframing it as a vegetation-response forecasting problem.
• Implemented pixel-wise deep sequence modeling using LSTM to capture spatial heterogeneity in agricultural drought, moving beyond basin-averaged or station-based approaches.
• Developed and demonstrated a transferable remote sensing–AI workflow for drought monitoring and prediction in semi-arid, data-scarce monsoon basins.
• Provided a robust framework for drought risk assessment, early warning, and climate-resilient agricultural and water resource management in semi-arid regions.

Funding

• No specific funding projects or programs were listed. Satellite data was made available by the U.S. Geological Survey (USGS).

Citation

@article{Naresh2026Deep,
  author = {Naresh, Chinthu and Mathew, Aneesh},
  title = {Deep Learning-Based Agricultural Drought Monitoring and Prediction Using Vegetation Health Index in the Papagni River Basin, India},
  journal = {Earth Systems and Environment},
  year = {2026},
  doi = {10.1007/s41748-026-01097-4},
  url = {https://doi.org/10.1007/s41748-026-01097-4}
}