Diagnosing CMIP6 Model Biases in East African Rainfall
Abstract
The countries of the Greater Horn of Africa (GHA), the driest region of the tropics, depend on two climatologically distinct rainy seasons: the generally heavier long rains in the boreal spring and the weaker short rains in the boreal autumn. Recent studies have suggested trends or the influence of low-frequency variability in both seasons, leading to increased use of climate models to project mid- to long-term changes in the regional hydrological cycle in both academic and government analyses. However, to provide credible climate projections for climate impacts analyses, models must be able to convincingly replicate not only the rainy seasons, but also the dynamic processes that influence them, neither of which has broadly been achieved in previous generations of CMIP models. We conduct a process-based analysis of statistics of both the timing and strength of the rainy seasons in the bimodal region of the GHA in CMIP6 models. We find that significant biases present in CMIP5 models persist in CMIP6 models. On average, CMIP6 models begin the long rains too late and produce long rains that are too weak and short rains that are too strong. To find the source of model biases, we examine the relationship between the rainy seasons and two characteristics of regional dynamics previously connected with GHA precipitation: sea surface temperatures (SSTs) and atmospheric circulation over the Indian Ocean basin. Atmospheric models forced with historical SSTs do not fix issues with the rainy seasons, despite strong correlations between oceanic variables and the rainy seasons in models. Instead, representations of atmospheric circulation may be at fault; for example, models produce convection that is too deep during the short rains compared to observations. Simulations of the upper-level zonal circulation, and by extension the structure of the Indian Ocean Walker-like Circulation more broadly, seem to play a role as well. Our process-based correlation analysis further suggests a method for selecting models for impacts studies if models fail to replicate observed relationships associated with the rainy seasons, then confidence in their projections is decreased.
- Publication:
-
AGU Fall Meeting Abstracts
- Pub Date:
- December 2021
- Bibcode:
- 2021AGUFM.A52H..01S