What question did this study set out to answer?

The aim is to assess drought conditions in Iran and develop a forecasting model for future scenarios.

March 22, 2026Open Access

Medium -term monitoring and machine learning-based forecasting of drought dynamics in Iran

Key Points

The aim is to assess drought conditions in Iran and develop a forecasting model for future scenarios.
Analyzed 58 years of climate data from 34 meteorological stations.
Applied the Reconnaissance Drought Index at various time scales.
Conducted benchmark analysis for optimization algorithms in a Support Vector Machine framework.
Developed and tested a model combining SVM with Pelican Optimization Algorithm and BiD-LSTM.
Identified SVM-POA as the most accurate forecasting model based on multiple performance metrics.
Forecasts suggest a shift towards N and Moderately Dry RDI classes over the coming decade.
Projected increase in stations showing significant drying trends, with 88.23% expected at the 1-month scale.

Abstract

Facing Iran’s acute water scarcity and the compounding effects of drought, this research undertakes a comprehensive assessment of past drought conditions, decadal forecast developments, and analyzes the progression of trends over time. The study draws on 58 years (1967–2024) of climate observations from 34 meteorological stations, applying the Reconnaissance Drought Index (RDI) at 1, 3, and 12-month time scales. To develop a robust forecasting model, we first conducted a benchmark analysis to select an optimal optimization algorithm for a Support Vector Machine (SVM). The Pelican Optimization Algorithm (POA) demonstrated superior performance compared to the Genetic Algorithm (GA) and Particle Swarm Optimization (PSO), leading to the selection of the SVM-POA framework. This validated model and a Bidirectional Long Short-Term Memory (BiD-LSTM) network were then tested for forecasting performance. A comprehensive evaluation using multiple performance metrics, including R², Mean Absolute Error (MAE), Normalized Root Mean Squared Error (NRMSE), and the Kappa index, identified SVM-POA as the most accurate model, which was subsequently employed to decadal forecast drought conditions for 2025–2036. Historically, the Normal (N) RDI class was the most frequent. Forecasts indicate that N and Moderately Dry (M-D) classes will be the most frequent in the coming decade, with extreme wet conditions expected to disappear. Importantly, trend diagnostics detected a prevailing shift toward drier conditions in all stations, whether statistically significant or not. The share of stations with significant drying is projected to rise sharply, from 26.47 to 88.23% for the 1-month scale and from 2.9 to 35.29% for the 3-month scale, between the historical and extended periods. For the 12-month scale, the proportion of stations with significant drying trends remains unchanged. These findings underscore the likelihood of increasingly severe droughts in Iran, demanding urgent, forward-looking strategies for sustainable water resource management.

AI से पूछें

Bookmark

View Full Paper

Cite This Study

Abdol Rassoul Zarei (Fri,) studied this question.

synapsesocial.com/papers/69bf8641f665edcd009e8d8c https://doi.org/https://doi.org/10.1038/s41598-026-45031-0

AI से पूछें

Bookmark

View Full Paper