Quantum Versus Classical Recurrent Neural Networks for Solar Radiation Forecasting over Mali

Forecasting solar radiation is crucial for the deployment of photovoltaic systems and for supporting renewable energy and agricultural planning in Sub-Saharan Africa. This manuscript compares the performance of two machine learning algorithms for this purpose: a classical Recurrent Neural Network (RNN) and a hybrid Quantum Neural Network (QNN), applied to monthly SWGDN prediction across 20 cities in Mali using MERRA-2 data from 1980 to 2025. The QNN architecture, comprising an LSTM feature extractor followed by an 8-qubit, 3-layer variational quantum circuit with Pauli-Z readout, outperforms the RNN in 9 of 20 Malian cities (45%). Notably, the quantum advantage is most easily noticeable in cities characterized by strong seasonal variability, whereas the RNN remains competitive in more arid, stable regions. These findings underline the potential of quantum machine learning for regional climate forecasting.