Accuracy assessment of the CMIP6 precipitation projections
The evolution of precipitation distributions in the context of climate change is a sensitive issue and this presentation will focus on the CMIP6 (Eyring et al., 2016) database which is the main information source on projections from Global Circulation Models (GCM). However these models are primarily designed for global change studies and their accuracy on a regional scale has to be evaluated.
This model quality assessment encounters three areas of difficulties :
- Rainfall events display highly variabiity in space and time and thus all observation dataset gets its own bias. The GPM-IMERG (Huffman et al., 2019) will be used as reference data on the basis it is composed of the largest input database (ground and satellite) among global precipitation products.
- GCM do not pretend to supply a deterministic forecast but a distribution of atmospheric pardirecameters. Hence an observation cannot be directly compared to a projection.
- Actual data are used for GCM tuning via the parametrization process (Hourdin et al., 2017). Therefore models are considered only from the cutoff year (2015) onward.
The analysis has been carried out on available GCM output with a daily spatial resolution of 100 km or less, the experiments SSP585 and High-Res being selected. For each accumulation period some distribution parameters have been mapped. On a first step models projected value are compared GCM estimation on 2015-2020 period, then model dspersion has been assessed for medium term (2045-2050) and long term (2095-2100).
Our results match with the overall remark of a much better retrieval of temperature than precipitation. However some regional trends appear as noticed by Mansour et al. (2021) on South America. Models efficiency is highly region dependent and the ensemble process is not optimal. For regional studies, we would advocate the procedure proposed by X.Yang et al. (2021) where projections are based on a subset of models selected by their local adequacy with observation data.
Keywords: Precipitation|Global Circulation Model|Regional analysis|Error assessment
A104308JB