Homtong et al. (2026) Mapping spatiotemporal agricultural droughts from 2019 to 2024 in Northeast Thailand using multi-temporal and multiple sensor data together with random forest algorithm

Identification

• Journal: Agricultural Water Management
• Year: 2026
• Date: 2026-02-10
• Authors: Nudthawud Homtong, Savittri Ratanopad Suwanlee, Surasak Keawsomsee, Kemin Kasa, Jaturong Som-ard, Sarawut Ninsawat, NARISSARA NUTHAMMACHOT, Dario Spiller, Filippo Sarvia
• DOI: 10.1016/j.agwat.2026.110216

Research Groups

• Department of Geotechnology, Faculty of Technology, Khon Kaen University, Khon Kaen Province, Thailand
• Department of Geography, Faculty of Humanities and Social Sciences, Mahasarakham University, Maha Sarakham, Thailand
• Earth Observation Technologies for Land and Agricultural Development Research Unit, Mahasarakham University, Maha Sarakham, Thailand
• Remote Sensing and GIS, School of Engineering and Technology, Asian Institute of Technology, Klong Luang, Pathum Thani, Thailand
• Faculty of Environmental Management, Prince of Songkla University, Hat Yai, Songkhla, Thailand
• School of Aerospace Engineering, Sapienza University of Rome, Rome, Italy
• Department of Agricultural, Forest and Food Sciences, University of Turin, Grugliasco, Italy

Short Summary

This study mapped spatiotemporal agricultural droughts in Northeast Thailand from 2019 to 2024 using multi-temporal Sentinel-2 imagery and a Random Forest Regression algorithm, with a Soil Moisture Index (SMI) derived from Landsat 8 as reference data. The models achieved high accuracy (R > 0.65), revealing consistent severe drought events between March and May annually, with irrigated areas showing lower severity.

Objective

• To investigate the relationship between the Soil Moisture Index (SMI) and the Standardized Precipitation Evapotranspiration Index (SPEI) across wet and dry seasons in Northeast Thailand.
• To develop Random Forest Regression (RFR) models integrating multi-temporal Sentinel-2 (S2) imagery with several vegetation indices and SMI-derived reference data to map agricultural drought occurrence from 2019 to 2024.
• To assess agricultural drought trends for major crops, including rice, sugarcane, cassava, and rubber trees, within Northeast Thailand.
• To generate high-resolution (10 m x 10 m) monthly drought maps for the study period.

Study Configuration

• Spatial Scale: Northeast Thailand (approximately 168,854 km²), with agricultural drought maps generated at 10 m x 10 m spatial resolution.
• Temporal Scale: January 2019 to December 2024 (6 years), with monthly drought mapping and seasonal/interannual trend analysis.

Methodology and Data

• Models used:
  • Random Forest Regression (RFR) for drought mapping.
  • Guided Regularized Random Forest (GRRF) for feature selection.
  • Mann–Kendall (MK) test for detecting drought trends.
  • Sen’s slope estimator for quantifying drought trend magnitudes.
  • Savitzky–Golay smoothing algorithm for noise reduction in time series.
  • Pearson correlation coefficient (R) for assessing relationships.
• Data sources:
  • Satellite:
    • Landsat 8 (L8) Surface Reflectance Tier 1 product (30 m, 100 m thermal band) for Soil Moisture Index (SMI) derivation.
    • Sentinel-2 (S2) MultiSpectral Instrument (MSI) Level-2A (L2A) dataset (10 m, 20 m, 60 m bands) for vegetation indices and drought mapping.
  • Observation:
    • Monthly precipitation mean and monthly maximum-minimum temperature from 27 weather stations (Thai Meteorological Department) for Standardized Precipitation Evapotranspiration Index (SPEI) calculation.
  • Derived Indices:
    • Soil Moisture Index (SMI) derived from L8 Land Surface Temperature (LST) and Normalized Difference Vegetation Index (NDVI) feature space.
    • Standardized Precipitation Evapotranspiration Index (SPEI-3) calculated from meteorological data.
    • Seven vegetation indices (VIs) from S2: NDVI, Soil-Adjusted Vegetation Index (SAVI), Normalized Difference Water Index (NDWI), Normalized Difference Infrared Index (NDII), Enhanced Vegetation Index (EVI), Normalized Difference Moisture Index (NDMI), and Vegetation Condition Index (VCI).

Main Results

• The Soil Moisture Index (SMI) showed strong positive correlations with SPEI-3, with Pearson R values of 0.84 during the dry season and 0.73 during the wet season, confirming its suitability as a reference dataset for predictive modeling.
• The Random Forest Regression (RFR) models achieved moderate to high accuracy for drought mapping, with R values ranging from 0.36 to 0.66 and Root Mean Square Error (RMSE) values between 0.06 and 0.15. The optimal RFR configuration used 500 trees.
• Key features for the monthly predictive models included Sentinel-2 spectral bands (B12, B11, B9, B4, B5) and derived indices (NDII, NDVI, NDWI, EVI, SAVI, VCI), with their importance varying seasonally based on crop phenology and hydroclimatic conditions.
• Spatiotemporal analysis indicated that the most severe drought events consistently occurred between March and May annually, corresponding to the summer season with low precipitation.
• Forested and irrigated zones, particularly in the southern and eastern regions, exhibited little to no drought severity, highlighting the buffering effect of managed water supply systems.
• Drought conditions were generally minimal or absent in January and during the post-monsoon months (September to December).
• Crop-specific analysis revealed that rice, sugarcane, and cassava cultivation areas were particularly affected by drought during the dry season, while deep-rooted rubber plantations showed greater resistance to short-term drought stress.
• Drought trend analysis from 2019 to 2024 showed predominantly increasing positive trends (increasing drought severity) in the dry season, especially in the central region, partly due to limited irrigation. In the wet season, decreasing negative trends (persistent wet conditions) were generally observed, though parts of the western agricultural region exhibited drought trends, largely attributed to the 2023–2024 El Niño phenomenon.

Contributions

• Developed a novel approach for mapping agricultural drought at a 10 m x 10 m field scale in a cloud-prone tropical region (Northeast Thailand) by combining satellite-derived Soil Moisture Index (SMI) as a robust reference dataset with Random Forest Regression (RFR) and high-resolution Sentinel-2 (S2) time series.
• Addressed the critical challenge of limited ground observations for machine learning model training in data-scarce regions by demonstrating the effectiveness of SMI as a proxy reference.
• Provided the first high-resolution (10 m x 10 m) monthly drought maps for Northeast Thailand from 2019 to 2024, offering unprecedented detail for understanding spatiotemporal drought dynamics across major crop systems.
• Enabled crop-specific drought assessment and short-term drought trend analysis, which was previously unavailable for this region at such fine spatial and temporal resolutions.
• Offered practical tools and insights to support agricultural drought mitigation, sustainable agricultural management, and evidence-based policymaking in climate-sensitive regions.

Funding

• Mahasarakham University

Citation

@article{Homtong2026Mapping,
  author = {Homtong, Nudthawud and Suwanlee, Savittri Ratanopad and Keawsomsee, Surasak and Kasa, Kemin and Som-ard, Jaturong and Ninsawat, Sarawut and NUTHAMMACHOT, NARISSARA and Spiller, Dario and Sarvia, Filippo},
  title = {Mapping spatiotemporal agricultural droughts from 2019 to 2024 in Northeast Thailand using multi-temporal and multiple sensor data together with random forest algorithm},
  journal = {Agricultural Water Management},
  year = {2026},
  doi = {10.1016/j.agwat.2026.110216},
  url = {https://doi.org/10.1016/j.agwat.2026.110216}
}