Abstract

No abstract

Coupled climate models used for long-term future climate projections and seasonal or decadal predictions share a systematic and persistent warm sea surface temperature (SST) bias in the tropical Atlantic. This study attempts to better understand the physical mechanisms responsible for the development of systematic biases in the tropical Atlantic using the so-called Transpose-CMIP protocol in a multi-model context. Six global climate models have been used to perform seasonal forecasts starting both in May and February over the period 2000-2009. In all models, the growth of SST biases is rapid. Significant biases are seen in the first month of forecast and, by 6 months, the root-mean-square SST bias is 80% of the climatological bias. These control experiments show that the equatorial warm SST bias is not driven by surface heat flux biases in all models, whereas in the southeastern Atlantic the solar heat flux could explain the setup of an initial warm bias in the first few days. A set of sensitivity experiments with prescribed wind stress confirm the leading role of wind stress biases in driving the equatorial SST bias, even if the amplitude of the SST bias is model dependent. A reduced SST bias leads to a reduced precipitation bias locally, but there is no robust remote effect on West African Monsoon rainfall. Over the southeastern part of the basin, local wind biases tend to have an impact on the local SST bias (except in the high resolution model). However, there is also a non-local effect of equatorial wind correction in two models. This can be explained by sub-surface advection of water from the equator, which is colder when the bias in equatorial wind stress is corrected. In terms of variability, it is also shown that improving the mean state in the equatorial Atlantic leads to a beneficial intensification of the Bjerknes feedback loop. In conclusion, we show a robust effect of wind stress biases on tropical mean climate and variability in multiple climate models.

Land surface initial conditions have been recognized as a potential source of predictability in sub-seasonal to seasonal forecast systems, at least for near-surface air temperature prediction over the mid-latitude continents. Yet, few studies have systematically explored such an influence over a sufficient hindcast period and in a multi-model framework to produce a robust quantitative assessment. Here, a dedicated set of twin experiments has been carried out with boreal summer retrospective forecasts over the 1992–2010 period performed by five different global coupled ocean–atmosphere models. The impact of a realistic versus climatological soil moisture initialization is assessed in two regions with high potential previously identified as hotspots of land–atmosphere coupling, namely the North American Great Plains and South-Eastern Europe. Over the latter region, temperature predictions show a significant improvement, especially over the Balkans. Forecast systems better simulate the warmest summers if they follow pronounced dry initial anomalies. It is hypothesized that models manage to capture a positive feedback between high temperature and low soil moisture content prone to dominate over other processes during the warmest summers in this region. Over the Great Plains, however, improving the soil moisture initialization does not lead to any robust gain of forecast quality for near-surface temperature. It is suggested that models biases prevent the forecast systems from making the most of the improved initial conditions. © 2017 Springer-Verlag Berlin Heidelber