What Control the Dry Bias over Amazonian by CMIP5 Models and Its Implication for Future Projections?

Underestimate rainfall, especially during the dry season, over Amazonia is a common problem of Coupled Model Intercomparison Project phase 3 (CMIP3) models. Such a dry bias is an important source of uncertainty in project global carbon-climate feedbacks. We investigate whether the dry bias still exists in the CMIP phase 5 (CMIP5) models. If so, what are responsible for the bias? Our evaluation of the CMIP5 historical simulations shows that some of the CMIP5 models still tend to underestimate rainfall over Amazonia. During the dry season, GFDL-ESM2M and IPSL show notably more pentads with no rain. In the dry and transition seasons, models with more realistic moisture convergence and surface evapotranspiration generally have more realistic rainfall amounts. In some models, overestimates of rainfall all associated with the adjacent tropical and eastern Pacific ITCZs, whereas in other models, too much surface net radiation and a resultant high Bowen ratio, appears to cause underestimates of rainfall. During the transition season, low pre-seasonal latent heat and high sensible flux and a weaker influence of cold air incursions contribute to the dry bias. About half the models can capture, but overestimate, the influences of teleconnection. Based on a simple metric, HadGEM2-ES outperforms other models especially for surface conditions and atmospheric circulation, whereas GFDL-ESM2M has the strongest dry bias presumably due to its overestimated moisture divergence induced by overestimated ITCZs in adjacent oceans, reinforced by positive feedbacks between reduced cloudiness, high Bowen ratio and suppression of rainfall during dry season, and weak incursions of extratropical disturbances during the transition season. These biases of processes are key factors for understanding the uncertainties of the future projections.