Comparative study of three stochastic future weather forecast approaches: a case study

Weather forecasting is an essential component of different hydrological studies. This article compares the weather prediction performance of various machine learning models like k-nearest neighbours (KNN), Soil and Water Assessment Tools (SWAT), and Representative Concentration Pathway (RCP). KNN is more resistant to noisy data set and provides more reliable performance than RCP and SWAT models. We simulate temperature, precipitation, and wind speed using KNN, SWAT and RCP weather generators, and we compare the results with observed data. The analyses compare WP-KNN with state-of-the-art classification and prediction models. We also suggest a systematic forecasting methodology that uses an updated version of the KNN classification. Our extensive experimental modelling findings show that the proposed technique is much more effective in a noisy dataset. • Generate synthetic weather sequences that could be used as inputs into hydrological models. • The predictions are performed by constructing relevant Machine Learning models based on stochastic process concepts. • Simulate weather data using three models: KNN, SWAT and, CCSM-RCP. • The generated weather data were then validated against observed data. • An RMSE criterion based on a confidence interval of 95% was used to determine which model was most efficient.