Predicting daily maximum temperature over Andhra Pradesh using machine learning techniques

Surface Air Temperature (SAT) predictions, typically generated by Global Climate Models (GCMs), carry uncertainties, particularly across different greenhouse gas emission scenarios. Machine Learning (ML) techniques can be employed to forecast long-term temperature variations, although this is a challenging endeavour with few drawbacks, such as the influence of scenarios involving greenhouse gas emissions. Therefore, the present study utilized multiple ML approaches such as Artificial Neural Networks (ANN), multiple linear regression, support vector machine and random forest, along with various daily predicted results of GCMs from Coupled Model Intercomparison Project Phase 6 as predictors and the "India Meteorological Department's" Maximum SAT (MSAT) as the predictand, to predict daily MSAT in the months of March, April and May (MAM) over Andhra Pradesh (AP) for the period 1981-2022. The results show that ANN outperforms other ML techniques in predicting daily MSAT, with a root mean square error of 1.41, an index of agreement of 0.89 and a correlation coefficient of 0.81. The spatial distribution of hot and heat wave days indicates that the Multiple Model Mean (MMM) underestimates these occurrences, with a minimum bias of 9 and 6 days, respectively. In contrast, the ANN model exhibits much smaller biases, with a maximum underestimation of 3 hot and 2 heat wave days. These findings demonstrate that MMM does not capture the maximum temperatures well, resulting in poor predictability. Further, future temperature projections were analysed from 2023 to 2050, which display a gradual increase in mean MSAT during MAM over AP. This research demonstrates the potential of ML techniques to enhance temperature forecasting accuracy, offering valuable insights for climate modeling and adaptation. The results are crucial for stakeholders in agriculture, health, energy, water resources, socio-economic planning, and urban development, aiding in informed decision-making and improving resilience to climate change impacts.