Large biases and inconsistent climate change signals in ENSEMBLES regional projections

In this paper we analyze some caveats found in the state-of-the-art ENSEMBLES regional projections dataset focusing on precipitation over Spain, and highlight the need of a task-oriented validation of the GCM-driven control runs. In particular, we compare the performance of the GCM-driven control runs (20C3M scenario) with the ERA40-driven ones (“perfect” boundary conditions) in a common period (1961–2000). Large deviations between the results indicate a large uncertainty/bias for the particular RCM-GCM combinations and, hence, a small confidence for the corresponding transient simulations due to the potential nonlinear amplification of biases. Specifically, we found large biases for some RCM-GCM combinations attributable to RCM in-house problems with the particular GCM coupling. These biases are shown to distort the corresponding climate change signal, or “delta”, in the last decades of the 21st century, considering the A1B scenario. Moreover, we analyze how to best combine the available RCMs to obtain more reliable projections.     

Validation of present-day regional climate simulations over Europe: LAM simulations with observed boundary conditions

 Nested limited-area modelling is one method of down-scaling general circulation model (GCM) climate change simulations. To give credibility to this method the nested limited-area model (LAM) must be shown to simulate local present-day climate conditions fairly accurately. Here seven different European limited-area models driven by observed boundary conditions (operational weather forecast analyses) are validated against observations, and inter-compared for summer and winter months. Relatively large biases are found. In summer large positive surface air temperature biases are found over southeast Europe. The main reason is deficiencies in the surface hydrological schemes causing an unrealistic drying of the soil. In at least one of the models, most likely several of them, an additional factor is an overestimation of incoming solar radiation. Apart from excessive precipitation in mountainous areas in some models they generally show a negative bias due to the drying and decreased advection from the Atlantic. In winter most models have a positive precipitation bias which seems to be caused by an enhancement of advection from the Atlantic and enhanced cyclone activity. Surface air temperature biases are negative probably due to an underestimation of the incoming longwave radiation. Received: 11 December 1996 / Accepted: 17 March 1997     

The WAMME regional model intercomparison study

Results from five regional climate models (RCMs) participating in the West African Monsoon Modeling and Evaluation (WAMME) initiative are analyzed. The RCMs were driven by boundary conditions from National Center for Environmental Prediction reanalysis II data sets and observed sea-surface temperatures (SST) over four May–October seasons, (2000 and 2003–2005). In addition, the simulations were repeated with two of the RCMs, except that lateral boundary conditions were derived from a continuous global climate model (GCM) simulation forced with observed SST data. RCM and GCM simulations of precipitation, surface air temperature and circulation are compared to each other and to observational evidence. Results demonstrate a range of RCM skill in representing the mean summer climate and the timing of monsoon onset. Four of the five models generate positive precipitation biases and all simulate negative surface air temperature biases over broad areas. RCM spatial patterns of June–September mean precipitation over the Sahel achieve spatial correlations with observational analyses of about 0.90, but within two areas south of 10°N the correlations average only about 0.44. The mean spatial correlation coefficient between RCM and observed surface air temperature over West Africa is 0.88. RCMs show a range of skill in simulating seasonal mean zonal wind and meridional moisture advection and two RCMs overestimate moisture convergence over West Africa. The 0.5° computing grid enables three RCMs to detect local minima related to high topography in seasonal mean meridional moisture advection. Sensitivity to lateral boundary conditions differs between the two RCMs for which this was assessed. The benefits of dynamic downscaling the GCM seasonal climate prediction are analyzed and discussed.     

Bias corrections of global models for regional climate simulations of high-impact weather

All global circulation models (GCMs) suffer from some form of bias, which when used as boundary conditions for regional climate models may impact the simulations, perhaps severely. Here we present a bias correction method that corrects the mean error in the GCM, but retains the six-hourly weather, longer-period climate-variability and climate change from the GCM. We utilize six different bias correction experiments; each correcting different bias components. The impact of the full bias correction and the individual components are examined in relation to tropical cyclones, precipitation and temperature. We show that correcting of all boundary data provides the greatest improvement.     

Understanding and simulating the link between African easterly waves and Atlantic tropical cyclones using a regional climate model: the role of domain size and lateral boundary conditions

Using a suite of lateral boundary conditions, we investigate the impact of domain size and boundary conditions on the Atlantic tropical cyclone and african easterly Wave activity simulated by a regional climate model. Irrespective of boundary conditions, simulations closest to observed climatology are obtained using a domain covering both the entire tropical Atlantic and northern African region. There is a clear degradation when the high-resolution model domain is diminished to cover only part of the African continent or only the tropical Atlantic. This is found to be the result of biases in the boundary data, which for the smaller domains, have a large impact on TC activity. In this series of simulations, the large-scale Atlantic atmospheric environment appears to be the primary control on simulated TC activity. Weaker wave activity is usually accompanied by a shift in cyclogenesis location, from the MDR to the subtropics. All ERA40-driven integrations manage to capture the observed interannual variability and to reproduce most of the upward trend in tropical cyclone activity observed during that period. When driven by low-resolution global climate model (GCM) integrations, the regional climate model captures interannual variability (albeit with lower correlation coefficients) only if tropical cyclones form in sufficient numbers in the main development region. However, all GCM-driven integrations fail to capture the upward trend in Atlantic tropical cyclone activity. In most integrations, variations in Atlantic tropical cyclone activity appear uncorrelated with variations in African easterly wave activity.     

Impacts of bias correction of lateral boundary conditions on regional climate projections in West Africa

Biases existing in the lateral boundary conditions (LBCs) influence climate simulations in regional climate models (RCMs). Correcting the biases in global climate model (GCM)-produced LBCs before running RCMs was proposed in previous studies as a possible way to reduce the GCM-related model dependence of future climate projections using RCMs. In this study the ICTP Regional Climate Model Version 4 (RegCM4) is used to investigate the impact of LBC bias correction on projected future changes of regional climate in West Africa. To accomplish this, two types of present versus future simulations are conducted using RegCM4: a control type where both the present and future LBCs are derived directly from the GCM output (as is done in most regional climate downscaling studies); an experiment type where the present-day LBCs are from reanalysis data and future LBCs are derived by combining the reanalysis data and the GCM-projected LBC changes. For each type of simulations, three different sets of LBCs are experimented on: 6-hourly synoptic forcing directly from the reanalysis or GCM, 6-hourly data interpolated from monthly climatology (without diurnal cycle), and 6-hourly data interpolated from the month-specific climatology of diurnal cycles. It is found that the simulations using different LBCs produce similar present-day summer rainfall patterns, but the predicted future changes differ significantly depending on how the LBC bias correction is treated. Specifically, both the bias correction applied at the synoptic scale and the bias correction applied to the monthly interpolated LBCs without diurnal cycle produce a spurious drying signal caused by physical inconsistency in the corrected future LBCs. Interpolated monthly LBCs with diurnal cycle alleviate the problem to a large extent. These results suggest that using bias-corrected LBCs to drive regional climate models may not guarantee reliable future projections although reasonable present climate can be simulated. Physical inconsistencies may be contained in the bias-corrected LBCs, increasing the uncertainties of RCM-produced future projections.     

Examining climate variability over West Africa during the 1979–1993 period: observations and CCM3 comparisons

The Community Climate Model Version 3.6 is used to simulate the mean climate of West Africa during the Northern Hemisphere summer season (June-August). The climate model uses prescribed climatological sea surface temperatures (SSTs) and observed SSTs during the 1979-1993 period. Two important circulation features, the African Easterly Jet (AEJ) and the Tropical Easterly Jet (TEJ), are found in the simulations but a westerly wind bias is found with respect to 700 hPa winds. Consequently, easterly waves and rain rates are poorly simulated. The primary cause of the poorly simulated AEJ is the advection of cold air from Europe producing a cold bias over northern Africa and a weaker than observed meridional temperature gradient. The cold bias is caused by an eastward displacement of the simulated Azores surface high into Western Europe creating a stronger than observed meridional sea level pressure gradient over northern Africa. This bias systematically occurs in simulations using both climatological and observed SSTs. The biases in sea level pressure, temperature and zonal winds have the potential to produce poor regional climate model results for West Africa if the meteorological output from the CCM3 is used as lateral boundaries. Moreover, these biases introduce uncertainties to West African GCM sensitivity studies associated with interannual variability, land-use change and elevated anthropogenic greenhouse gases.     

An inter-comparison of regional climate models for Europe: model performance in present-day climate

The analysis of possible regional climate changes over Europe as simulated by 10 regional climate models within the context of PRUDENCE requires a careful investigation of possible systematic biases in the models. The purpose of this paper is to identify how the main model systematic biases vary across the different models. Two fundamental aspects of model validation are addressed here: the ability to simulate (1) the long-term (30 or 40 years) mean climate and (2) the inter-annual variability. The analysis concentrates on near-surface air temperature and precipitation over land and focuses mainly on winter and summer. In general, there is a warm bias with respect to the CRU data set in these extreme seasons and a tendency to cold biases in the transition seasons. In winter the typical spread (standard deviation) between the models is 1 K. During summer there is generally a better agreement between observed and simulated values of inter-annual variability although there is a relatively clear signal that the modeled temperature variability is larger than suggested by observations, while precipitation variability is closer to observations. The areas with warm (cold) bias in winter generally exhibit wet (dry) biases, whereas the relationship is the reverse during summer (though much less clear, coupling warm (cold) biases with dry (wet) ones). When comparing the RCMs with their driving GCM, they generally reproduce the large-scale circulation of the GCM though in some cases there are substantial differences between regional biases in surface temperature and precipitation.     

RegCM3 regional climatologies for South America using reanalysis and ECHAM global model driving fields

To enable downscaling of seasonal prediction and climate change scenarios, long-term baseline regional climatologies which employ global model forcing are needed for South America. As a first step in this process, this work examines climatological integrations with a regional climate model using a continental scale domain nested in both reanalysis data and multiple realizations of an atmospheric general circulation model (GCM). The analysis presents an evaluation of the nested model simulated large scale circulation, mean annual cycle and interannual variability which is compared against observational estimates and also with the driving GCM for the Northeast, Amazon, Monsoon and Southeast regions of South America. Results indicate that the regional climate model simulates the annual cycle of precipitation well in the Northeast region and Monsoon regions; it exhibits a dry bias during winter (July–September) in the Southeast, and simulates a semi-annual cycle with a dry bias in summer (December–February) in the Amazon region. There is little difference in the annual cycle between the GCM and renalyses driven simulations, however, substantial differences are seen in the interannual variability. Despite the biases in the annual cycle, the regional model captures much of the interannual variability observed in the Northeast, Southeast and Amazon regions. In the Monsoon region, where remote influences are weak, the regional model improves upon the GCM, though neither show substantial predictability. We conclude that in regions where remote influences are strong and the global model performs well it is difficult for the regional model to improve the large scale climatological features, indeed the regional model may degrade the simulation. Where remote forcing is weak and local processes dominate, there is some potential for the regional model to add value. This, however, will require improvments in physical parameterizations for high resolution tropical simulations.     

Atmospheric circulation in regional climate models over Central Europe: links to surface air temperature and the influence of driving data

The study examines simulation of atmospheric circulation, represented by circulation indices (flow direction, strength and vorticity), and links between circulation and daily surface air temperatures in regional climate models (RCMs) over Central Europe. We explore control simulations of five high-resolution RCMs from the ENSEMBLES project driven by re-analysis (ERA-40) and the same global climate model (ECHAM5 GCM) plus of one RCM (RCA) driven by different GCMs. The aims are to (1) identify errors in RCM-simulated distributions of circulation indices in individual seasons, (2) identify errors in simulated temperatures under particular circulation indices, and (3) compare performance of individual RCMs with respect to the driving data. Although most of the RCMs qualitatively reflect observed distributions of the airflow indices, each produces distributions significantly different from the observations. General biases include overestimation of the frequency of strong flow days and of strong cyclonic vorticity. Some circulation biases obviously propagate from the driving data. ECHAM5 and all simulations driven by ECHAM5 underestimate frequency of easterly flow, mainly in summer. Except for HIRHAM, however, all RCMs driven by ECHAM5 improve on the driving GCM in simulating atmospheric circulation. The influence on circulation characteristics in the nested RCM differs between GCMs, as demonstrated in a set of RCA simulations with different driving data. The driving data control on circulation in RCA is particularly weak for the BCM GCM, in which case RCA substantially modifies (but does not improve) the circulation from the driving data in both winter and summer. Those RCMs with the most distorted atmospheric circulation are HIRHAM driven by ECHAM5 and RCA driven by BCM. Relatively strong relationships between circulation indices and surface air temperatures were found in the observed data for Central Europe. The links differ by season and are usually stronger for daily maxima than minima. RCMs qualitatively reproduce these relationships. Effects of the driving model biases were found on RCMs’ performance in reproducing not only atmospheric circulation but also the links to surface temperature. However, the RCM formulation appears to be more important than the driving data in representing the latter. Differences of the circulation-to-temperature links among the RCA simulations are smaller and the links tend to be more realistic compared to the driving GCMs.     

Ability of an ensemble of regional climate models to reproduce weather regimes over Europe-Atlantic during the period 1961–2000

One of the main concerns in regional climate modeling is to which extent limited-area regional climate models (RCM) reproduce the large-scale atmospheric conditions of their driving general circulation model (GCM). In this work we investigate the ability of a multi-model ensemble of regional climate simulations to reproduce the large-scale weather regimes of the driving conditions. The ensemble consists of a set of 13 RCMs on a European domain, driven at their lateral boundaries by the ERA40 reanalysis for the time period 1961–2000. Two sets of experiments have been completed with horizontal resolutions of 50 and 25 km, respectively. The spectral nudging technique has been applied to one of the models within the ensemble. The RCMs reproduce the weather regimes behavior in terms of composite pattern, mean frequency of occurrence and persistence reasonably well. The models also simulate well the long-term trends and the inter-annual variability of the frequency of occurrence. However, there is a non-negligible spread among the models which is stronger in summer than in winter. This spread is due to two reasons: (1) we are dealing with different models and (2) each RCM produces an internal variability. As far as the day-to-day weather regime history is concerned, the ensemble shows large discrepancies. At daily time scale, the model spread has also a seasonal dependence, being stronger in summer than in winter. Results also show that the spectral nudging technique improves the model performance in reproducing the large-scale of the driving field. In addition, the impact of increasing the number of grid points has been addressed by comparing the 25 and 50 km experiments. We show that the horizontal resolution does not affect significantly the model performance for large-scale circulation.     

Sources of systematic errors in the climatology of a regional climate model over Europe

 Two ten-year simulations made with a European regional climate model (RCM) are compared. They are driven by the same observed sea surface temperatures but use different lateral boundary forcing. For one simulation, RCM AMIP, this forcing is obtained from a standard integration of a global general circulation model (GCM AMIP), whereas for the other simulation, RCM ASSIM, it is derived from a time series of operational analyses. The archive of analysis fields (surface pressure plus winds and temperatures on various pressure levels) is not sufficiently comprehensive to provide directly the full set of driving fields required for the RCM (in particular, no moisture fields are present), so these are obtained via a GCM integration, GCM ASSIM, in which the model is continuously relaxed towards the analysis fields using a data assimilation technique. Errors in RCM AMIP can arise either from the internal RCM physics or from errors in the lateral boundary forcing inherited from GCM AMIP. Errors in RCM ASSIM can arise from the internal RCM physics or the boundary moisture forcing but not from the driving circulation. Although previous studies have considered RCM integrations driven either by output from standard GCM integrations or operational analyses, our study is the first to compare parallel integrations of each type. This allows the total systematic error in an RCM integration driven by standard GCM output to be partitioned into components arising from the driving circulation and the internal RCM physics. These components indicate the scope for reducing regional simulation biases by improving either the driving GCM or the RCM itself. The results relate mainly to elements of surface climate likely to be influenced by both the driving circulation and regional physical processes operating in the RCM. For cloud cover, errors are found to arise largely from the internal RCM physics. Values are too low despite a positive relative humidity bias, indicating shortcomings in the parametrisation scheme used to calculate cloud cover. In summer, surface temperature and precipitation errors are also explained principally by regional processes. For example excessive solar heating leads to anomalously high surface temperatures over southern Europe and excessive drying of the soil reduces precipitation in the south eastern sector of the domain. The lateral boundary forcing reduces precipitation in the south eastern sector of the domain. The lateral boundary forcing also exerts some influence, mainly via a tropospheric cold bias which partially offsets the warming over southern Europe and also increases precipitation. In other seasons the lateral boundary forcing and the regional physics both contribute significantly to the errors in surface temperature and precipitation. In winter the boundary forcing (apart from moisture) is responsible for about 60% of the total error variance for temperature and about 40% for precipitation, due to the cold bias and circulation errors such as a southward shift in the storm track. The remaining errors arise from the regional physics, although for precipitation an excessive supply of moisture from the lateral boundaries also contributes. The skill of the mesoscale component of the surface temperature and precipitation distributions exceeds previous estimates, due to more realistic observed climatology. The mesoscale patterns are very similar in the two RCM simulations indicating that errors in the simulation of fine scale detail arise mainly from inadequate representations of local forcings rather than errors in the large-scale circulation. Circulation errors in RCM AMIP (e.g. cold bias, southward shift of storm track) are also present in GCM AMIP, but are largely absent in RCM ASSIM except in summer. This confirms evidence from previous work that the key to reducing most circulation errors in the RCM lies in improving the driving GCM. Regional processes only make a major contribution to circulation errors in summer, when reduced advection from the boundaries allows errors in surface temperature to be transmitted more effectively into the troposphere. Finally, we find evidence of error balances in the GCM which act to minimise biases in important climatological variables. This reflects tuning of the model physics at GCM resolution. In order to achieve simultaneous optimisation of the RCM and GCM at widely differing resolutions it may be necessary to introduce explicit scale dependences into some aspects of the physics. Received: 17 September 1997/Accepted: 10 March 1998     

The simulation of cutoff lows in a regional climate model: reliability and future trends

Cutoff lows are an important source of rainfall in the mid-latitudes that climate models need to simulate accurately to give confidence in climate projections for rainfall. Coarse-scale general circulation models used for climate studies show some notable biases and deficiencies in the simulation of cutoff lows in the Australian region and important aspects of the broader circulation such as atmospheric blocking and the split jet structure observed over Australia. The regional climate model conformal cubic atmospheric model or CCAM gives an improvement in some aspects of the simulation of cutoffs in the Australian region, including a reduction in the underestimate of the frequency of cutoff days by more than 15 % compared to a typical GCM. This improvement is due at least in part to substantially higher resolution. However, biases in the simulation of the broader circulation, blocking and the split jet structure are still present. In particular, a northward bias in the central latitude of cutoff lows creates a substantial underestimate of the associated rainfall over Tasmania in April to October. Also, the regional climate model produces a significant north–south distortion of the vertical profile of cutoff lows, with the largest distortion occurring in the cooler months that was not apparent in GCM simulations. The remaining biases and presence of new biases demonstrates that increased horizontal resolution is not the only requirement in the reliable simulation of cutoff lows in climate models. Notwithstanding the biases in their simulation, the regional climate model projections show some responses to climate warming that are noteworthy. The projections indicate a marked closing of the split jet in winter. This change is associated with changes to atmospheric blocking in the Tasman Sea, which decreases in June to November (by up to 7.9 m s^?1), and increases in December to May. The projections also show a reduction in the number of annual cutoff days by 67 % over the century, together with an increase in their intensity, and these changes are strongest in spring and summer.     

Evaluation of the regional climate model ALADIN to simulate the climate over North America in the CORDEX framework

In this study, an ensemble of four multi-year climate simulations is performed with the regional climate model ALADIN to evaluate its ability to simulate the climate over North America in the CORDEX framework. The simulations differ in their driving fields (ERA-40 or ERA-Interim) and the nudging technique (with or without large-scale nudging). The validation of the simulated 2-m temperature and precipitation with observationally-based gridded data sets shows that ALADIN performs similarly to other regional climate models that are commonly used over North America. Large-scale nudging improves the temporal correlation of the atmospheric circulation between ALADIN and its driving field, and also reduces the warm and dry summer biases in central North America. The differences between the simulations driven with different reanalyses are small and are likely related to the regional climate model’s induced internal variability. In general, the impact of different driving fields on ALADIN is smaller than that of large-scale nudging. The analysis of the multi-year simulations over the prairie and the east taiga indicates that the ALADIN 2-m temperature and precipitation interannual variability is similar or larger than that observed. Finally, a comparison of the simulations with observations for the summer 1993 shows that ALADIN underestimates the flood in central North America mainly due to its systematic dry bias in this region. Overall, the results indicate that ALADIN can produce a valuable contribution to CORDEX over North America.     

Simulation of South American wintertime climate with a nesting system

A numerical nesting system is developed to simulate wintertime climate of the eastern South Pacific-South America-western South Atlantic region, and preliminary results are presented. The nesting system consists of a large-scale global atmospheric general circulation model (GCM) and a regional climate model (RCM). The latter is driven at its boundaries by the GCM. The particularity of this nesting system is that the GCM itself has a variable horizontal resolution (stretched grid). Our main purpose is to assess the plausibility of such a technique to improve climate representation over South America. In order to evaluate how this nesting system represents the main features of the regional circulation, several mean fields have been analyzed. The global model, despite its relatively low resolution, could simulate reasonably well the more significant large-scale circulation patterns. The use of the regional model often results in improvements, but not universally. Many of the systematic errors of the global model are also present in the regional model, although the biases tend to be rectified. Our preliminary results suggest that nesting technique is a computationally low-cost alternative for simulating regional climate features. However, additional simulations, parametrizations tuning and further diagnosis are clearly needed to represent local patterns more precisely. Received: 18 February 1999 / Accepted: 31 May 2000     

A study on combining global and regional climate model results for generating climate scenarios of temperature and precipitation for the Netherlands

Climate scenarios for the Netherlands are constructed by combining information from global and regional climate models employing a simplified, conceptual framework of three sources (levels) of uncertainty impacting on predictions of the local climate. In this framework, the first level of uncertainty is determined by the global radiation balance, resulting in a range of the projected changes in the global mean temperature. On the regional (1,000–5,000 km) scale, the response of the atmospheric circulation determines the second important level of uncertainty. The third level of uncertainty, acting mainly on a local scale of 10 (and less) to 1,000 km, is related to the small-scale processes, like for example those acting in atmospheric convection, clouds and atmospheric meso-scale circulations—processes that play an important role in extreme events which are highly relevant for society. Global climate models (GCMs) are the main tools to quantify the first two levels of uncertainty, while high resolution regional climate models (RCMs) are more suitable to quantify the third level. Along these lines, results of an ensemble of RCMs, driven by only two GCM boundaries and therefore spanning only a rather narrow range in future climate predictions, are rescaled to obtain a broader uncertainty range. The rescaling is done by first disentangling the climate change response in the RCM simulations into a part related to the circulation, and a residual part which is related to the global temperature rise. Second, these responses are rescaled using the range of the predictions of global temperature change and circulation change from five GCMs. These GCMs have been selected on their ability to simulate the present-day circulation, in particular over Europe. For the seasonal means, the rescaled RCM results obey the range in the GCM ensemble using a high and low emission scenario. Thus, the rescaled RCM results are consistent with the GCM results for the means, while adding information on the small scales and the extremes. The method can be interpreted as a combined statistical–dynamical downscaling approach, with the statistical relations based on regional model output.     

Biases in the diurnal temperature range in Central Europe in an ensemble of regional climate models and their possible causes

The study examines how regional climate models (RCMs) reproduce the diurnal temperature range (DTR) in their control simulations over Central Europe. We evaluate 30-year runs driven by perfect boundary conditions (the ERA40 reanalysis, 1961–1990) and a global climate model (ECHAM5) of an ensemble of RCMs with 25-km resolution from the ENSEMBLES project. The RCMs’ performance is compared against the dataset gridded from a high-density stations network. We find that all RCMs underestimate DTR in all seasons, notwithstanding whether driven by ERA40 or ECHAM5. Underestimation is largest in summer and smallest in winter in most RCMs. The relationship of the models’ errors to indices of atmospheric circulation and cloud cover is discussed to reveal possible causes of the biases. In all seasons and all simulations driven by ERA40 and ECHAM5, underestimation of DTR is larger under anticyclonic circulation and becomes smaller or negligible for cyclonic circulation. In summer and transition seasons, underestimation tends to be largest for the southeast to south flow associated with warm advection, while in winter it does not depend on flow direction. We show that the biases in DTR, which seem common to all examined RCMs, are also related to cloud cover simulation. However, there is no general tendency to overestimate total cloud amount under anticyclonic conditions in the RCMs, which suggests the large negative bias in DTR for anticyclonic circulation cannot be explained by a bias in cloudiness. Errors in simulating heat and moisture fluxes between land surface and atmosphere probably contribute to the biases in DTR as well.     

Diagnosing southeast tropical Atlantic SST and ocean circulation biases in the CMIP5 ensemble

Warm sea-surface temperature (SST) biases in the southeastern tropical Atlantic (SETA), which is defined by a region from 5°E to the west coast of southern Africa and from 10°S to 30°S, are a common problem in many current and previous generation climate models. The Coupled Model Intercomparison Project Phase 5 (CMIP5) ensemble provides a useful framework to tackle the complex issues concerning causes of the SST bias. In this study, we tested a number of previously proposed mechanisms responsible for the SETA SST bias and found the following results. First, the multi-model ensemble mean shows a positive shortwave radiation bias of ~20 W m^?2, consistent with models’ deficiency in simulating low-level clouds. This shortwave radiation error, however, is overwhelmed by larger errors in the simulated surface turbulent heat and longwave radiation fluxes, resulting in excessive heat loss from the ocean. The result holds for atmosphere-only model simulations from the same multi-model ensemble, where the effect of SST biases on surface heat fluxes is removed, and is not sensitive to whether the analysis region is chosen to coincide with the maximum warm SST bias along the coast or with the main SETA stratocumulus deck away from the coast. This combined with the fact that there is no statistically significant relationship between simulated SST biases and surface heat flux biases among CMIP5 models suggests that the shortwave radiation bias caused by poorly simulated low-level clouds is not the leading cause of the warm SST bias. Second, the majority of CMIP5 models underestimate upwelling strength along the Benguela coast, which is linked to the unrealistically weak alongshore wind stress simulated by the models. However, a correlation analysis between the model simulated vertical velocities and SST biases does not reveal a statistically significant relationship between the two, suggesting that the deficient coastal upwelling in the models is not simply related to the warm SST bias via vertical heat advection. Third, SETA SST biases in CMIP5 models are correlated with surface and subsurface ocean temperature biases in the equatorial region, suggesting that the equatorial temperature bias remotely contributes to the SETA SST bias. Finally, we found that all CMIP5 models simulate a southward displaced Angola–Benguela front (ABF), which in many models is more than 10° south of its observed location. Furthermore, SETA SST biases are most significantly correlated with ABF latitude, which suggests that the inability of CMIP5 models to accurately simulate the ABF is a leading cause of the SETA SST bias. This is supported by simulations with the oceanic component of one of the CMIP5 models, which is forced with observationally derived surface fluxes. The results show that even with the observationally derived surface atmospheric forcing, the ocean model generates a significant warm SST bias near the ABF, underlining the important role of ocean dynamics in SETA SST bias problem. Further model simulations were conducted to address the impact of the SETA SST biases. The results indicate a significant remote influence of the SETA SST bias on global model simulations of tropical climate, underscoring the importance and urgency to reduce the SETA SST bias in global climate models.     

Present and future climatologies in the phase I CREMA experiment

We provide an overall assessment of the surface air temperature and precipitation present day (1976–2005) and future (2070–2099) ensemble climatologies in the Phase I CREMA experiment. This consists of simulations performed with different configurations (physics schemes) of the ICTP regional model RegCM4 over five CORDEX domains (Africa, Mediterranean, Central America, South America, South Asia), driven by different combinations of three global climate models (GCMs) and two greenhouse gas (GHG) representative concentration pathways (RCP8.5 and RCP4.5). The biases (1976–2005) in the driving and nested model ensembles compared to observations show a high degree of spatial variability and, when comparing GCMs and RegCM4, similar magnitudes and more similarity for precipitation than for temperature. The large scale patterns of change (2070–2099 minus 1976–2005) are broadly consistent across the GCM and RegCM4 ensembles and with previous analyses of GCM projections, indicating that the GCMs selected in the CREMA experiment are representative of the more general behavior of current GCMs. The RegCM4, however, shows a lower climate sensitivity (reduced warming) than the driving GCMs, especially when using the CLM land surface scheme. While the broad patterns of precipitation change are consistent across the GCM and RegCM4 ensembles, greater differences are found at sub-regional scales over the various domains, evidently tied to the representation of local processes. This paper serves to provide a reference view of the behavior of the CREMA ensemble, while more detailed and process-based analysis of individual domains is left to companion papers of this special issue.