Sensitivity Study of the RegCM4’s Surface Schemes in the Simulations of West Africa Climate

Two simulations of five years (2003-2007) were conducted with the Regional Climate models RegCM4, one coupled with Land surface models BATS and the other with CLM4.5 over West Africa, where simulated air temperature and precipitation were analyzed. The purpose of this study is to assess the performance of RegCM4 coupled with the new CLM4.5 Land surface scheme and the standard one named BATS in order to find the best configuration of RegCM4 over West African. This study could improve our understanding of the sensitivity of land surface model in West Africa climate simulation, and provide relevant information to RegCM4 users. The results show fairly realistic restitution of West Africa’s climatology and indicate correlations of 0.60 to 0.82 between the simulated fields (BATS and CLM4.5) for precipitation. The substitution of BATS surface scheme by CLM4.5 in the model configuration,


Introduction
The complexity and diversity of the dynamic and physical processes of the West African Monsoon (WAM) make its simulation a challenge for climate models.
Realistic climate reproduction requires accurate simulation. The most suitable tools to capture the characteristics of WAM are Regional Climate Models (RCM) [1] [2] [3]. Thus, optimizing RCMs configurations will allow for accurate climate simulations. To achieve this, several studies have been conducted, either by improving the current schemes in determining the best combination of existing physics schemes to use for a certain region, or implementing new physical parameterization schemes [4] [5] [6] [7] [8]. The performance of the models varies according to the region, season and physical configurations. In West Africa, many studies have attempted to improve the simulation of precipitation during the summer season, based on the sensitivity of the models to cumulus convection schemes [9]. Although the accurate simulation of precipitation is directly related to cumulus convection scheme and its interaction with other physical processes, the choice of land surface scheme can also contribute to improving the performance of models in simulating climate over West Africa. Chen et al. [10] demonstrated over eastern China, that the CLM3.5 surface scheme significantly improves precipitation and temperature simulations with respect to the NOAH land surface scheme when coupled with Weather Research and Forecasting (WRF). Steiner et al. [11] also showed the role of the land surface scheme in improving land-atmosphere moisture, energy exchange and associated surface climate feedback by replacing the standard BATS surface scheme with CLM in RegCM. Nevertheless, Halder et al. [12] demonstrated that the use of CLM can lead to poorer performance of RegCM than BATS over India. The new RegCM version 4 has been improved with substantial development of the software code and physical representations [2] and the introduction of CLM4.5 as a new option for describing land surface processes.
Several studies have shown that the choice of a land surface scheme has a substantial impact on climate model simulation [13]. Thus, the performance of RegCM4 coupled with the new CLM4.5 surface scheme needs to be investigated together with the standard BATS scheme to find the best configuration of RegCM4 for the West African, and provide relevant information to RegCM4 users.
In addition, this study could improve our understanding of the sensitivity of surface schemes in West Africa climate simulation. It will also contribute to estimating the advantages and disadvantages of the RegCM4 model by coupling it with the BATS and CLM4. 5  over West Africa. The analysis will focus on precipitation and temperature, a statistical analysis will be conducted to better highlight the model's performance.
The main conclusions are summarized in Section 4.

Model Description
The regional climate model, used in this study is named RegCM4. It is a limited area model, with finite-difference discretization using a terrain-following σ (sigma) pressure vertical coordinate system and an Arakawa B-grid finite differencing algorithm [2]. RegCM4 was developed at ICTP and aimed to study mesoscale processes in the atmosphere over a selected area of the Earth.
The release used in this study is RegCM4.7. The non-hydrostatic dynamical core of the mesoscale model version 5 (MM5) [14] has been ported to RegCM4 while maintaining the existing hydrostatic core. We used in this study the non-hydrostatic core of RegCM4. The radiation scheme is derived from the NCAR Community climate Model 3 (CCM3) described in Kiehl et al. [15], and includes a representation of aerosols following Zakey et al. [16]. It also includes many complex detailed processes such as, radiative transfer processes, grid-scale and subgrid-scale cloud processes, turbulence mixing processes and land-surface processes [2]. RegCM4 model shows an improvement performance compared to elder version especially in the formulation of surface scheme.

Land Surface Scheme
BATS has been described in detail by Dickinson et al. [17], it's the standard land surface scheme used with RegCM, since many years. This scheme has been im-

Methodology
The RegCM4 model is used in this study with the two available land-surface schemes (BATS and CLM4.5). Two simulations that are identical in all respects, except for the land surface scheme, have been launched for West Africa. The first simulation uses the BATS scheme [17] while the second uses the CLM 4.5 scheme [21].      Figure 4 shows In the Guinea Coast area (Figure 4(c)), we note a fairly good results (BATS and CLM4.5) in terms of bias (3˚C or 8%). However, the temporal location of the simulated data is not correctly represented. The analysis of Figure 4(d)

Annual Cycle
(West Africa area) shows a constant bias between CLM4.5 and the observations.

Statistical Analysis
In order to analyze the performance of each surface scheme, a statistical study of the simulated and observed fields is carried out in this section. It will concern both the geographical location (grid point) and climatic parameter values (temperature or precipitation). After bringing the data (observations and simulations) to the same spatial resolution, we determine the biases, root mean square deviations (RMSE) and correlations at each grid point, over the period (2003)(2004)(2005)(2006)(2007). The RMSE values presented by the CLM4.5 experiment are two time than BATS, which suggests that compared to CLM, BATS still more efficient on this region.

Monthly Variation
Monthly average variations of daily precipitation from observed and simulated data, for the period 2003-2007 are shown in Figure 5. The simulated and observed data show similar patterns, reaching peaks at the same times. We note, however, that the RegCM4 model, underestimates in both simulations, precipitation over  The precipitation resulting from CLM4.5 is lower than that of BATS by about 0.5 mm•day −1 .

Spatial Distribution
The spatial distribution of daily precipitation in JJAS (2003-2007) over West Africa is shown in Figure 6.   Except these areas, BATS and CLM experiments produce fairly realistic precipitation fields with errors of less than 20% or biases of less than 2 mm•day −1 .
Comparing the simulations experiments each other, analysis shows an underestimation of the precipitation of the BATS experiment compared to the CLM experiment over the Atlantic. Otherwise CLM improves the precipitation over the ocean, Figure 7(c).

Annual Mean Cycle
The annual precipitation cycle for the sub regions are presented in Figure 8. The analysis of Central Sahel zone (Figure 8(a)) indicates that simulated precipitation  Furthermore, in the Central Sahelian zone, 70% of the precipitation is not due to local and convective phenomena, but to large-scale, non-convective precipitation [13]. The amplification of the biases in JJAS could not be attributed to the surface or convective scheme. However, the surface scheme BATS seems to better reproduce the observational data compared to CLM, the biases are smaller 20% for BATS and 50% for CLM. The analysis of West Sahel zone (Figure 8(b)), shows significant biases between observations (peaks at 9mm•day −1 ) and simulations (peaks at 3mm•day −1 ). The model underestimates the daily precipitation Analysis of the entire West Africa domain indicates that the CLM4.5 and BATS remain fairly close, the biases are less than 1 mm•day −1 , and moreover the simulated data sometimes coincides with the observations. The model underestimates the daily precipitation, over the area with mean bias of 2 mm•day −1 . Table 4 shows the correlation coefficients and root mean square deviations of JJAS precipitation for the simulated and observed data. On the Central Sahel BATS has a root mean square deviation (RMSE) of 11.13 compared to 11.10 for CLM4.5 and a non-significant correlation about 0.3, the RMSE of the two experiments are of the

Taylor Diagram
The Taylor diagram, Figure 9, allow a combined synthesized view of the pattern correlation coefficient and the JJAS standard deviation of precipitation from the different sensitivity studies with respect to TRMM over West Africa sub regions.  Similar results were also reported by Chung et al. [8] and Wang et al. [

Conclusions and Perspectives
In order to improve the simulations of the regional climate model RegCM4 over complete this work. We also thank the Computing Center of the Bingerville scientific pole which provided the regional model RegCM, hosted us for part of this work and provided the equipment to simulate the climate of West Africa.
Thanks to Arona DIEDHIOU for his frequent help.