Enhancing Ready-to-Implementation subseasonal crop growth predictions in central Southwestern Asia: A machine learning-climate dynamical hybrid strategy

Abstract

Responding to the urgent need for precise, one-month-ahead crop growth predictions in Central Southwestern Asia (CSWA), this study introduces a fully operational convolutional neural network (CNN)-climate dynamical hybrid model designed for real-time agricultural planning and management. It is engineered to accurately forecast the Normalized Difference Vegetation Index (NDVI), a vital indicator of crop health, with a one-month lead time. The model integrates multi-temporal data, including soil moisture and temperature from the preceding months, and historical NDVI, enhancing its predictive accuracy with 500hPa geopotential heights and 2-meter surface temperatures refined through a U-Net-based CNN. These meteorological inputs are sourced from the Flexible Global Ocean–Atmosphere–Land System Model finite volume version 2 (FGOALS-f2), an advanced global dynamical prediction system. Empirical validation across CSWA demonstrates the model’s robust performance, with pattern correlation coefficients of 0.60, 0.70, and 0.58, root mean squared errors of 0.036, 0.029, and 0.022, and sign consistency rates of 74.8 %, 77.1 %, and 73.3 % for April, May, and June, respectively. Seamlessly integrated into the operational framework of FGOALS-f2, this model enables real-time, one-month advance predictions of NDVI. This pioneering approach not only enhances the accuracy of subseasonal crop growth forecasts in CSWA but also sets a new standard for subseasonal climate services.