An improved deep learning procedure for statistical downscaling of climate data.

Area of Science:

• Climate Science
• Artificial Intelligence
• Environmental Modeling

Background:

• Coupled Model Intercomparison Project Phase 6 (CMIP6) provides coarse-resolution climate change (CC) scenarios.
• Statistical downscaling using deep learning (DL) shows promise but requires further research, especially for arid regions.
• The suitability of DL for extrapolating future CC scenarios in arid environments remains underexplored.

Purpose of the Study:

• To analyze the suitability of DL for downscaling CMIP6 climate change scenarios in arid regions.
• To downscale maximum and minimum temperature over Egypt using a Convolutional Neural Network (CNN).
• To assess the performance of the DL model in reproducing historical and future climate data.

Main Methods:

• Employed a Convolutional Neural Network (CNN) for statistical downscaling, termed CNNSD.
• Utilized CMIP6 data from the CanESM5 General Circulation Model (GCM).
• Downscaled scenarios under two Shared Socioeconomic Pathways (SSPs): SSP4.5 and SSP8.5, for the Egyptian domain at 0.1° resolution.

Main Results:

• The CNNSD model accurately reproduced observed climate data and reduced biases in the original CMIP6 scenarios.
• Projected temperature increases by 2100 (compared to historical periods) under SSP4.5: 4.8°C (max) and 4.0°C (min).
• Projected temperature increases by 2100 under SSP8.5: 6.3°C (max) and 4.2°C (min).

Conclusions:

• CNN-based downscaling is effective for CMIP6 scenarios in arid regions like Egypt.
• The developed approach can support climate services, impact studies, and adaptation strategies in vulnerable developing countries.
• Further research incorporating multi-GCMs is recommended to quantify uncertainties and enhance climate change impact assessments.