Diagnosing GCM errors over West Africa using relaxation experiments. Part I: summer monsoon climatology and interannual variability

Abstract

The CNRM atmospheric general circulation model Arpege-Climat is relaxed towards atmospheric reanalyses outside the 10°S–32°N 30°W–50°E domain in order to disentangle the regional versus large-scale sources of climatological biases and interannual variability of the West African monsoon (WAM). On the one hand, the main climatological features of the monsoon, including the spatial distribution of summer precipitation, are only weakly improved by the nudging, thereby suggesting the regional origin of the Arpege-Climat biases. On the other hand, the nudging technique is relatively efficient to control the interannual variability of the WAM dynamics, though the impact on rainfall variability is less clear. Additional sensitivity experiments focusing on the strong 1994 summer monsoon suggest that the weak sensitivity of the model biases is not an artifact of the nudging design, but the evidence that regional physical processes are the main limiting factors for a realistic simulation of monsoon circulation and precipitation in the Arpege-Climat model. Sensitivity experiments to soil moisture boundary conditions are also conducted and highlight the relevance of land–atmosphere coupling for the amplification of precipitation biases. Nevertheless, the land surface hydrology is not the main explanation for the model errors that are rather due to deficiencies in the atmospheric physics. The intraseasonal timescale and the model internal variability are discussed in a companion paper.

Appendix: Methodology for the computation of the moisture fluxes

Calculating the atmospheric water budget over a given area is not an easy task because of several methodology problems that can arise at different stages of the computation (Meynadier et al. 2010a):

• (1) reanalyses do not offer the possibility to obtain realistic closure terms for the atmospheric water vapor, because of data assimilation during the integration of numerical weather models.

• (2) tropical rainfall fields in atmospheric models are subject to huge uncertainties because of the perfectible parameterization of deep convective processes.

• (3) daily and even 6-hourly model outputs sub-sample the diurnal cycle, which may lead to wrong representations of, e.g. afternoon tropical storms.

• (4) vertical integration of, e.g. moisture fluxes, are somewhat uncertain because they are often based on a rather low number of pressure levels, interpolated from the model sigma levels.

When working with a GCM, one is still submitted to the errors linked to point (2). It is however possible to estimate to what extent points (3) and (4) may contribute to wrong estimates of the different terms of the equation for water vapor closure. To that end, we computed daily vertically integrated moisture fluxes based on daily outputs interpolated on similar pressure levels as those available for ERA40 reanalyses, and alternatively, moisture fluxes integrated on the model sigma levels at each time step (30 min here). In other words, the first method is that applied on reanalyzed data, but the second one offers better estimates of the moisture fluxes and theoretically allows for a perfect water vapor closure in each grid point. The results are shown in Figs. 14 (for the Sahel) and 15 (over Africa).

Fig. 14

a Scatter plot intersecting the pentad mean moisture divergence (mm day⁻¹) over the Sahel (10°N–20°N, 20°W–20°E) computed with daily pressure fields (abscissa) and with vertical integrations on the model levels at each model time step (ordinates), JJAS 1971–2000. The common variance between the two series is labeled on the figure. b Seasonality of moisture divergence according to daily pressure fields (blue curve) and 30-min-resolution fields available on sigma levels (red curve) for each pentad of the season. Bars difference between the two methods of vertical integration (c) As (b) but for seasonal means, period 1971–2000

Fig. 15

a Seasonal mean moisture fluxes integrated over the air column (vectors) and moisture divergence (shading), period JJAS 1971–2000. b Difference between the two methods of vertical integrations. Significance tested and shown as for Fig. 2

Figure 14 shows that, over the Sahel, the two methods lead to a difference of 2 mm day⁻¹. Daily fields over-estimate moisture divergence over the Sahel. The error is almost constant from early June to late September (Fig. 14b) and, to a lesser extent, from 1 year to another (Fig. 14c). For comparison purposes, the difference between ERA40 rainfall and GPCP estimates, shown on Fig. 3 to be rather marked over the region, is 0.5 mm day⁻¹. The methodology for the computation of moisture fluxes is thus of primary importance, and has a first order impact for the water vapor closure term.

Figure 15 shows the JJAS climatological integrated moisture fluxes and divergence for each grid point over Africa, and the error due to the computation method at each grid point. Note that the color scale is the same for the seasonal mean field and the differences between the two computation methods. The model climatology (Fig. 15a) shows the well-known features of large-scale atmospheric circulation over Africa, with inter-tropical moisture convergence located north of the equator and the very strong southerly moisture fluxes related to the Indian monsoon and found over the western Indian Ocean. Errors (Fig. 15b) confirm that moisture divergence is over-estimated over the Sahel when daily pressure fields are used. On the contrary, enhanced moisture convergence takes place over the Guinean belt. Daily pressure fields tend thus to exaggerate the model climatology.

For these reasons, all figures presented in this paper result of moisture fluxes calculated at the model time step and integrated on sigma levels. Errors nonetheless persist when comparisons with reanalyses are performed, because vertically integrated moisture fluxes are not available for these datasets.