Uncertainty quantification of deep learning based statistical downscaling of climatic parameters

Abstract

This study investigated the uncertainty involved in statistically downscaling of hydroclimatic time series obtained by Artificial Neural Networks (ANNs). The Coupled Model Intercomparison Project 6 (CMIP6) General Circulation Model (GCM) CanESM5 was used as large-scale predictor data for downscaling temperature and precipitation parameters. Two ANN, feed-forward and long short-term memory (LSTM) were utilized for statistical downscaling. To quantify the uncertainty of downscaling, prediction intervals (PIs) were estimated via the lower upper bound estimation (LUBE) method. To assess performance of proposed models in different climate regimes, data from Tabriz and Rasht stations were employed. The calibrated models via historical GCM data were used for future projections via the high-forcing and fossil fuel-driven development scenario (SSP5-8.5). Projections were compared with the Can-RCM4 projections via same scenario. Results indicated that both LSTM-based point predictions and PIs are more accurate than the FFNN-based predictions with an average of 55% higher Nash-Sutcliffe efficiency (NSE) for point predictions and 25% lower coverage width criterion (CWC) for PIs. Projections suggested that Tabriz is going to experience warmer climate by an increase in average temperature by 2 °C and 5 °C for near and far futures, respectively, and drier climate by a 20% decrease in precipitation until 2100. Future projections for the Rasht station however suggested a more uniform climate with less seasonal variability. Average precipitation will increase up to 25% and 70% until near and far future periods, respectively. Ultimately, point predictions show that the average temperature in Rasht will increase by 1 °C until near future and then a constant average temperature until far future.