.png){kind=icon}
.png){kind=icon}
Return
Ensemble machine learning for meteorological drought assessment and forecasting with satellite and climate data (Urmia Lake basin, Iran)
.png){kind=icon}
PRE
AI
C
M
A
M
J
DOI：10.1016/j.watcyc.2026.02.002.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
This study assesses the ability of satellite, meteorological, and climate teleconnection indices to forecast drought at 3-, 6-, 9-, and 12-month intervals in the Urmia Lake basin (ULB) using machine learning models. Data sources include TRMM precipitation, MODIS-derived NDVI and daytime land surface temperature (LSTday), gridded potential evapotranspiration (PET), and large-scale indices (MEI, SOI, AMO, NAO) from 2001 to 2019. The methods compared feature Decision Tree (DT), Random Forest (RF), and Extremely Randomized Trees (ERT). Model performance was evaluated through cross-validated R-squared, root mean square error (RMSE), and mean absolute error (MAE), along with variable importance, SHAP (SHapley Additive exPlanations) interpretation, and wavelet coherence analysis. A key finding is that ERT consistently outperformed other algorithms across different accumulation periods, showing superior predictive accuracy (higher R-squared and lower RMSE/MAE) and more consistent cross-validation results; SHAP and wavelet coherence analyses indicate a systematic shift from local drivers (precipitation, NDVI, LSTday) at shorter timescales to teleconnection indices (MEI, AMO, SOI, NAO) at longer ones. Practical implications suggest that the ERT-based framework can support operational drought monitoring and water management in the Basin by integrating satellite data with climate indices to enhance early-warning systems and resource planning. The novelty of this approach lies in integrating multiscale SHAP interpretation and wavelet coherence to elucidate the shift from local to teleconnected drought drivers, offering a transferable framework for basin-scale drought assessment.
Keywords:
Drought forecasting
Machine learning
MODIS
Teleconnections
Urmia lake
Journal
W
IF:
8.7
Papers:
33
Citations:
630
Organization
Cited Papers