Multi-Index Assessment and Machine Learning Integration for Drought Monitoring Using Google Earth Engine
Open Access
- 17 June 2025
- journal article
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing
- Vol. 18 (19391404) , 16497-16516
Abstract
This study advances multi-sensor remote sensing data fusion integrating optical (Sentinel-2, MODIS), thermal (LST), and hydrological (SMAP) sensors with climate datasets to evaluate soil moisture dynamics at five depths (0–50 cm) across nine agricultural zones (October 2021–September 2023), leveraging AI and machine learning to address data quality challenges in heterogeneous sensor inputs. Our novel AI-driven approach improved data fusion accuracy by 27% compared to conventional methods, enabling more reliable drought detection even in areas with persistent cloud cover and sensor inconsistencies. Using Google Earth Engine for spatiotemporal fusion and Random Forest classification for feature optimization, we demonstrate the superiority of fused multi-sensor indices, with VCI (Vegetation Condition Index) and NVSWI (Normalized Vegetation-Soil Water Index) achieving the highest accuracy (r = 0.56–0.59) at 10–40 cm depths, while single-sensor optical indices underperformed at 50 cm (r < 0.30). Temporal analysis revealed cumulative drought signals in SPEI (Standardized Precipitation Evapotranspiration Index) (3–6-month lag) and rapid surface responses in TCI (Temperature Condition Index), emphasizing the need for adaptive sensor fusion. Land cover changes included an 18.5% reduction in surface water and a 3,714.6 km2 expansion of bare ground in 2023, correlating with SPI (Standardized Precipitation Index)-based drought severity (29.8% to 47.3% affected area). The framework's AI-driven error correction and multi-sensor synergy provide a scalable model for coastal applications, such as mangrove ecosystem resilience monitoring (integrating thermal, optical, and radar data for salinity intrusion analysis) and hydrological forecasting (fusing soil moisture, precipitation, and tidal datasets). By resolving sensor inconsistencies and enhancing reliability in complex environments, this work underscores the broader relevance of multi-sensor fusion for coastal vulnerability assessments, where land-sea interactions demand integrated sensor networks and machine learning to mitigate ecological and climatic risks.Keywords
Funding Information
- RUDN University Scientific Projects
- Northern Border University, Saudi Arabia
This publication has 69 references indexed in Scilit:
- Characterizing uncertainty in process-based hydraulic modeling, exemplified in a semiarid Inner Mongolia steppeGeoderma, 2023
- Assessing soil water balance to optimize irrigation schedules of flood-irrigated maize fields with different cultivation histories in the arid regionAgricultural Water Management, 2022
- Vegetation dynamics and responses to climate change and anthropogenic activities in the Three-River Headwaters Region, ChinaEcological Indicators, 2021
- Biochar aging increased microbial carbon use efficiency but decreased biomass turnover timeGeoderma, 2021
- Vegetation greening intensified soil drying in some semi-arid and arid areas of the worldAgricultural and Forest Meteorology, 2020
- A Machine Learning approach to the assessment of the vulnerability of Posidonia oceanica meadowsEcological Indicators, 2020
- Comparison of three crop water stress index models with sap flow measurements in maizeAgricultural Water Management, 2018
- The rise in global atmospheric CO2, surface temperature, and sea level from emissions traced to major carbon producersClimatic Change, 2017
- Prediction of kharif rice yield at Kharagpur using disaggregated extended range rainfall forecastsTheoretical and Applied Climatology, 2017
- Random forest in remote sensing: A review of applications and future directionsISPRS Journal of Photogrammetry and Remote Sensing, 2016