Downscaling & super-resolution
Assessing Three Perfect Prognosis Methods for Statistical Downscaling of Climate Change Precipitation Scenarios
In one sentence. Compares GLMs, CNNs and a posteriori random forests for perfect-prognosis precipitation downscaling of EC-Earth; APRFs best reproduce the GCM's future climate change signal.
Abstract
Abstract Under the perfect prognosis approach, statistical downscaling methods learn the relationships between large‐scale variables from reanalysis and local observational records. These relationships are subsequently applied to downscale future global climate model (GCM) simulations in order to obtain projections for the local region and variables of interest. However, the capability of such methods to produce future climate change signals consistent with those from the GCM, often referred to as transferability, is an important issue that remains to be carefully analyzed. Using the EC‐Earth GCM and focusing on precipitation, we assess the transferability of generalized linear models, convolutional neural networks and a posteriori random forests (APRFs). We conclude that APRFs present the best overall performance for the historical period, and future local climate change signals consistent with those projected by EC‐Earth. Moreover, we show how a slight modification of APRFs can greatly improve the temporal consistency of the downscaled series.
Labels
- Topic
- Downscaling & super-resolution
- Type
- Evaluation or intercomparison of existing models
- Method
- CNN / U-Net, Classical ML (trees, GP, linear, kernels)
- Time scale
- Climate projections / centennial
- Domain
- Regional, Local / station
- Data
- CMIP / climate model output, ERA5 / reanalysis, Station / in-situ
- Status
- Research
- Identifiers
- DOI:10.1029/2022gl102525 OpenAlex:W4376601314
- Added
- 2026-10-22
Labels and summary are generated by Claude and reviewed weekly. Wrong? Edit this entry.