Machine Learning-Based Precipitation Prediction Using Cloud Properties

Developing a forecast model with a high accuracy and consistency level continues to be of global interest in weather prediction and climate projection studies. Precipitation constitutes one of the critical components of weather, surrounded by huge uncertainties but requires adequate information to enhance water resources management and sustainability of life. Machine learning techniques as data-driven tools have been found worthy in addressing these challenges and have shown promising results over the years, particularly in their robustness and moderate resources requirement. In this current study, two machine learning algorithms, including the multiple linear regression (MLR) and multilayered perceptron artificial neural network (MLP-ANN), are deployed to develop a model for daily precipitation prediction based on cloud properties. The algorithms using cloud parameters comprising of cloud optical thickness (COT), cloud effective radius (CER), cloud top temperature (CTT), cloud top pressure (CTP), and liquid water path (LWP) as inputs performed well to produce models with good accuracies up to R > 0.7 and generally RMSE < 5.5. In all, the models produced by MLP-ANN prove to be more accurate with higher R-values and low errors for location-specific training and prediction. At the same time, MLR shows more consistency traits based on a multilocation forecast. Besides, the location with the highest number of samples results in better models compared to areas with lower data samples. Results obtained are helpful to weather and climate change stakeholders for accurate prediction of daily precipitation and management of water resources to support the ecosystem.