Work

Remote sensing-driven machine learning models for SMI-based drought assessment and forecasting

Year: 2026
Type: article
Abstract: The complex and interconnected nature of climate systems makes drought forecasting highly challenging. This study integrates remote sensing and machine learning for Soil Moisture Index (SMI)-based drought prediction. In the first stage, Landsat 8 imagery and SRTM DEM data are used to estimate SMI in drought-prone regions of southern and western Turkey. A comparative analysis of four machine learning algorithms—Random Forests (RF), Gradient Boosting Machines (GBM), eXtreme Gradient Boosting (XGBoost), and LightGBM—is conducted to evaluate their ability to represent SMI from remotely sensed land surface factors. Results show that RF provides the best accuracy, with feature importance analysis highlighting Landsat 8 thermal bands as the most influential predictors. In the second stage, a multistep Convolutional Long Short-Term Memory (ConvLSTM) network is applied to a time series of Landsat 8 images to forecast SMI ~1 month (32 days) ahead. The forecasting framework is evaluated using both the original image sequence and a cloud-gap-filled sequence to assess the impact of cloud contamination on prediction performance. This dual-stage framework demonstrates the value of combining machine learning with remote sensing for both SMI-based drought modeling and forecasting. By capturing key spatial and temporal patterns, the proposed approach offers a robust tool for monitoring drought risk and supporting agricultural management in climate-sensitive regions.
Language: English
Open Access status: gold
Estimated APC paid: $3,020
OpenAlex ID: W7237179484
FWCI:
Cites: 9
Cited by:
Related to: 10