Climate Downscaling and Prediction Using GIS-Based Machine Learning

Forecasters are finding it increasingly difficult to precisely anticipate the appearance of any weather factor due to fast climate change. The goal of this paper is to predict monthly temperature in the United States using different machine learning algorithms by performing time-series analysis. Climate model data for monthly temperature forecasts across the United States were obtained. The data was analysed, and the relationship between the predictors (GCM Variables) and the target variable (Temperature in months) was examined using several GIS based machine learning models such as GLR, GWR, and Random Forest (RF). The best predictor model was used to estimate five-year average monthly temperatures, meaning temperature forecasts for 60 time periods. A time-series mosaic was used to mix 60 rasters for improved visualisation. R-ArcGIS Bridge was used to do analyse the functional data as well as deconstruct temporal signals into analytical fundamental functions. Thus, climate data is downscaled statistically and temperature profiles are predicted over the United States.