Spatial and temporal variability of rainfall over Africa offers considerable challenges for assessing and understanding climate change over the continent. This is because of the complexity of African regional climates and the influence of regional geographic features, such as deserts, land cover variations, mountain chains, large lakes, land-sea contrasts and the sea surface temperatures (SSTs) of the surrounding Indian and Atlantic oceans. Such diversity influences the distribution and statistics of daily rainfall and extremes and the onset and evolution of the monsoon rainy season (Sylla et al., 2010a). This in turn may have a much greater impact on natural systems and human activities than mean precipitation (Parry et al., 2007). An assessment of climate change projections over the African continent thus requires a careful evaluation of a model's ability to simulate fine temporal scale hydroclimatic variables and the structure and daily evolution of the African monsoon systems.
Within this context, simulations of daily and extreme precipitation have been evaluated over many regions throughout the world in both global and regional climate models (Schmidli and Frei, 2005; Halenka et al., 2006; Buonomo et al., 2007; May, 2008; Tolika et al., 2008 and Torma et al.2011 for Europe; Park et al., 2001; Gao et al., 2002; Rahimzadeh et al., 2009; Liu et al., 2010 and Im et al., 2011 for Asia; Meehl et al., 2007; Marengo et al., 2009; Rusticucci et al., 2009; Walker and Diffenbaugh, 2009 and Wehner et al., 2009 for America). However, only few assessments of simulated daily-scale precipitation events over Africa (Sylla et al., 2010b, 2012) are available in the literature and they focus on relatively low resolution General Circulation Models (GCMs) and mainly over southern Africa (Rocha et al., 2008; Shongwe et al., 2009; Williams et al., 2010), where daily data are of better quality.
In fact, reliable observations of daily rainfall are crucial for assessing the model performance. The spatial distribution of rain gauges throughout Africa is not adequate enough to allow the description of local, regional, or even some large-scale hydrological and climatic phenomena (Love et al., 2004). Therefore, although rain gauge datasets are available for different African regions, these products by themselves are not sufficient to capture rainfall at both fine temporal and spatial scales. This lack of spatially and temporally continuous and accurate long-term precipitation datasets over Africa has prompted the generation of daily climatologies of gridded rainfall covering the entire continent based on satellite information.
Various merged satellite-gauge gridded rainfall datasets are now available at daily timescale and at relatively high spatial resolutions. These include the Global Precipitation Climatology Project (GPCP, 1° × 1° resolution; Huffman et al., 2001); Tropical Rainfall Measuring Mission (TRMM, 0.25° × 0.25° resolution; Kummerow et al., 2001; Huffman et al., 2007) and Famine Early Warning System (FEWS, 0.1° × 0.1° resolution; Herman et al., 1997; Love et al., 2004). However, the number of observations used in these products typically varies over the years and over different regions. Other differences may include retrieval, merging and interpolation techniques. These methodological differences may lead to notable discrepancies across the products, resulting in significant uncertainties in the estimates of observed precipitation. In particular, such differences may cancel out when looking at mean climatologies, while being evident in higher order daily statistics. Therefore, the evaluation and intercomparison of different rainfall products over the African continent is important in order to place their use in climate model assessment into a proper context.
Based on these premises, in this paper we first present an intercomparison across different merged satellite-gauge rainfall products in terms of daily characteristics and extreme events over the entire African continent in order to assess uncertainties across these products. These uncertainties are then evaluated within the context of the assessment of the International Centre for Theoretical Physics (ICTP) Regional Climate Model (RegCM3; Pal et al., 2007) in a simulation over an all-Africa domain driven at the lateral boundaries by the ERA-Interim reanalysis of observations (Uppala et al., 2008). In particular, we examine a range of daily precipitation statistics as well as the onset and evolution of the monsoon over different African sub-regions.
2. Description of model, experiment and observational dataset
2.1. The regional climate model
We analyse here a simulation with the version of the ICTP Regional Climate Model, RegCM3 (Giorgi et al., 1993a, 1993b; Pal et al., 2007). RegCM3 is a primitive equation, sigma vertical coordinate, regional climate model based on the hydrostatic version of the dynamical core of the NCAR/PSU's mesoscale meteorological model MM5 (Grell et al., 1994). Radiation is represented by the parameterization of Kiehl et al. (1996) and the planetary boundary scheme is from Holtslag et al. (1990) in the implementation of Giorgi et al. (1993a). Interactions between the land surface and the atmosphere are described using the Biosphere Atmosphere Transfer Scheme (BATS1E; Dickinson et al., 1993), while the scheme of Zeng et al. (1998) is used to represent fluxes from water surfaces. In RegCM3, convective precipitation can be represented by a number of schemes and here we use that of Grell et al. (1994) with the Fritsch and Chappell (1980) closure assumption. This choice was based on an analysis of the model performance in some preliminary tests (Sylla et al., 2010b). Resolvable scale precipitation processes are treated using the sub-grid explicit moisture scheme of Pal et al. (2000), which is a physically based parameterization including sub-grid scale clouds, cloud water accretion, and evaporation of falling raindrops.
The initial and lateral boundary conditions for the RegCM3 simulation are obtained from the new ERA-Interim 0.75° × 0.75° gridded reanalysis (Simmons et al., 2007; Uppala et al., 2008), which is the third generation ECMWF reanalysis product. The main advances in this reanalysis compared to ERA-40 are that ERA-Interim is carried out with a higher horizontal resolution with four-dimensional variational analysis, a new humidity analysis, improved model physics, bias correction of satellite radiance data and an improved fast radiative transfer code. ERA-Interim uses mostly the sets of observations acquired for ERA-40 with the addition of new altimeter wave-height data of more uniform quality reprocessed Meteosat data for wind and clear-sky radiance, and new ozone profile information from 1995 onwards. Several problems experienced in the ERA-40 reanalysis were eliminated or significantly improved in the ERA-Interim, particularly in the humidity and hydrologic cycle over the tropics (Uppala et al., 2008). The ERA-Interim boundary conditions are updated four times daily in RegCM3.
The regional model is integrated over the African domain of Figure 1(a) continuously for the entire 19-year period of January 1989 through December 2007, at a spatial resolution of 50 km. This domain follows the CORDEX Africa specifications (Giorgi et al., 2009). Note that the domain exhibits complex terrains especially in the southern and eastern regions. Some localized highlands are also present over West Africa around Cameroun (Cameroun Mountain), Central Nigeria (Jos Plateau) and Guinea (Guinea Highlands). Also shown in Figure 1(b) are three climate sub-regions selected for more detailed analysis of the daily distribution of precipitation (monsoon onset/retreat). Sylla et al. (2010c) provided an overall assessment of the mean seasonal climatology, annual cycles and internannual variability of this simulation.
2.3. Observation products and analysis measures
As mentioned, we employ here three observed daily precipitation gridded products: GPCP (1° × 1° resolution; Huffman et al., 2001); TRMM (0.25° × 0.25° resolution; Kummerow et al., 2001; Huffman et al., 2007) and FEWS (0.1° × 0.1° resolution; Herman et al., 1997; Love et al., 2004). The GPCP daily rainfall product (GPCP 1DD) is a satellite-derived dataset developed under the Global Precipitation Climatology Project and made available from late 1996 to present. Another satellite-derived daily rainfall, the 0.25° resolution TRMM (3B42 version 6), provides data for the entire tropics since November 1997. Finally, the high resolution FEWS daily dataset (RFE1.0 and RFE2.0), a new 0.10° gridded satellite-derived precipitations estimate covering the entire African continent, is produced and updated since 1995 at the Climate Prediction Center (CPC) for the Famine Early Warning System Network (FEWS NET) project. These rainfall datasets are mainly based on satellite observations; however, they have also been compared and merged with ground station rain gauges using different merging techniques and merging periods to create the final products. Only for mean precipitation and for limited illustrative purposes, we also include in the analysis the half-degree resolution dataset of the Climatic Research Unit (CRU) of the University of East Anglia (Mitchell et al., 2004). The CRU dataset includes rain gauge station-based monthly precipitation for the period 1901–2002 and it is widely used in model performance assessment.
The analysis is carried out by considering only the common period across the observations and the simulation (1998–2007). For the CRU dataset, which includes data only up to 2002, we use the long-term climatology for the period 1971–2000. After a brief assessment of mean precipitation, we focus on the time-latitude precipitation cross sections over East, South and West Africa to evaluate onset, peak and withdrawal of monsoon rainy seasons over the three sub-regions (Le Barbé et al., 2002; Hourdin et al., 2010). We then use different hydroclimatic indices (summarized in Table 1) to assess the spatial characteristics of daily rainfall: frequency of wet days and the simple daily index, extreme precipitation represented here as the 95th percentile, and maximum wet spell and dry spell length. These indices are calculated at the annual timescale and for both boreal and austral summer seasons. For more detailed analysis, we evaluate quantitative measures of precipitation uncertainty including the Mean Difference (or bias) over three key African sub-regions (Figure 1(b)) and the Pattern Correlation calculated over the entire African continent for the mean rainfall climatology. We intercompare the observation datasets across each other and with the model results.
Table 1. Selected indices and their definition
Number of wet days
Maximum number of days with precipitation > 1 mm
Simple daily index
Precipitation intensity due to the wet days only
Frequency of heavy rainfall events
Maximum number of days with precipitation > 20 mm
Extremes: 95th Percentile
Only 5% of the data are above this value
Mean maximum wet spell length
Maximum number of consecutive wet days
Mean Maximum dry spell length
Maximum number of consecutive dry days
3. Results and discussion
3.1. Mean precipitation
Before evaluating and intercomparing the characteristics of daily rainfall events from the different observation and modeling products, in this section we briefly examine the mean precipitation distribution over the African continent. Figure 2 thus shows the averaged precipitation from CRU (first column), FEWS (second column), TRMM (third column), GPCP (fourth column) and RegCM3 (fifth column) during DJF (upper panels), JJA (middle panels) and ANN (lower panels) over Continental Africa. Tables 2 and 3 report the seasonal precipitation mean differences and pattern correlation coefficients across the datasets for three key sub-regions identified in Figure 1: Sahel (June–August: JJA), northern East Africa (June–August: JJA) and Southern Africa (December–February: DJF).
Table 2. Differences in regional precipitation averages between the different products over the Sahel, Northern East Africa and Central Southern Africa, in descending order. The differences are presented in the table as ‘column minus row’. Units are expressed as percentage of the corresponding ‘row’ value. Values of regional differences with RegCM3 that are within the range of those across the observation products are in bold
Table 3. Pattern correlation coefficients across the different precipitation products calculated over the whole Africa continent for June–July–August, December–January–February and annual averages (in descending order). The coefficients are presented in the table as ‘column with respect to row’
During the boreal winter season, rainfall is confined to the southern hemisphere, while regions north of the equator are predominantly dry. In the observations (FEWS, TRMM, GPCP and CRU), rainfall maxima are located over the complex terrains of the eastern part of Central Southern Africa along the Zaire Air Boundary (ZAB) south of the Congo Basin. In JJA (boreal summer), the rain band moves to the northern hemisphere where the Intertropical Convergence Zone (ITCZ) is at its northernmost location. The observation products locate maximum rainfall along the monsoon belt and over the mountainous regions of Cameroun, Guinea and the Ethiopian highlands. Annual precipitation is therefore widely spread across the continent, except for the desert areas of the Sahara and Namibia.
A number of differences can be found across the observation datasets with regard to the magnitude and spatial extent of the mean rainfall. In general, FEWS shows lower mean precipitation values than the other datasets, especially CRU and GPCP, in almost all seasons and regions, particularly in orographic areas and in the convective regions of the ITCZ, while TRMM displays a better defined ITCZ.
An analysis of Tables 2 and 3 shows that, in general, high correlation coefficients, exceeding 0.9 are found between all pairs of observation datasets, indicating a good level of agreement in the observed precipitation patterns. However, significant differences are found in the magnitude of precipitation. FEWS is the dataset with the lowest precipitation amounts over all regions, by up to factors exceeding 30%. Conversely, GPCP shows ubiquitously the largest amounts. CRU and TRMM have intermediate values, with the CRU ones being larger than TRMM over almost all regions. Table 2 thus indicates substantial systematic differences across the datasets.
Moving to the comparison with the regional model simulation, Figure 2 shows that the RegCM3 captures the basic seasonal large-scale patterns of precipitation. The analyses of Sylla et al. (2010a) and Sylla et al. (2011) indicate that this is due to a good representation of the mean and the variability of the main dynamical features of African climate such as the monsoon flows, African Easterly Waves, African Easterly Jet (AEJ), Tropical Easterly Jet (TEJ), subtropical jet streams and ZAB. The model, however, overestimates precipitation over some regions of Southern and especially Eastern Africa, while it shows an underestimate over the Congo Basin in JJA (Figure 2). In terms of area averaged magnitude, the model agrees best with the GPCP and CRU data and worst with the FEWS data, obviously in particular over the regions where precipitation is overestimated. The area of largest overestimation is the mountainous region of the East African highlands, where the model evidently overestimates the effects of the local orographic forcing. This appears to be a common problem in regional models over East Africa (Vizy and Cook, 2002; Song et al., 2004; Paeth et al., 2005; Anyah et al., 2006; Davis et al., 2009; Segele et al., 2009a; Paeth, 2010). In this region, the observation datasets exhibit large differences suggesting that this excess rainfall could also be related to an underestimate of observed precipitation over high elevation areas where gauge stations are sparse.
Tables 2 and 3 report the seasonal precipitation biases for the maximum precipitation seasons over the three regions of Figure 1 and the pattern correlations between the RegCM3 seasonal and annual precipitation and the different observation products. Overall, the tendency of the model to overestimate precipitation is revealed by the consistent predominance of positive biases. Consistently with the data of Figure 3, the maximum overestimate occurs over northern east Africa and the minimum over central southern Africa. The most striking feature in Table 2 is the range of biases when comparing the model to different observation datasets. For example, over northern east Africa the bias varies from 109% compared to TRMM to 57% compared to GPCP. The model-observation correlations of Table 3 show a much lower variability across products, with correlation coefficient over the whole Africa ranging from 0.76 to 0.86. This is consistent with the high correlations found across observational datasets (greater than 0.9). Overall, the high precipitation correlations quantitatively confirm that the model reproduces the observed spatial precipitation patterns (Figure 3).
Summarizing the results of this section, we first find that the precipitation products analysed here do exhibit substantial systematic differences in the mean precipitation amounts, with FEWS showing minimum precipitation amounts and GPCP maximum amounts. The regional model shows a tendency to overestimate rainfall compared to most observation datasets over the regions examined, but mostly simulates precipitation amounts within the observational range. The model also reproduces the spatial patterns of rainfall climatology, although some fine scale structures over mountainous regions show noticeable differences compared to the observation datasets.
3.2. Mean daily monsoon evolution
We now turn to the analysis of the observed and simulated evolution of the monsoon over the West, East and South Africa regions based on daily data. Figure 3 first shows time-latitude diagrams of daily rainfall averaged over the West Africa region between 10°E and 10°W (Figure 1) for the FEWS (Figure 3(a)), TRMM (Figure 3(b)), GPCP (Figure 3(c)) and RegCM3 (Figure 3(d)). The averages are taken for the period 1998–2007 and displayed throughout the year. Observations (FEWS, TRMM and GPCP) show the monsoon onset in March-April over the Gulf of Guinea, characterized by a northward extension of the rain belt up to about 6°N. An abrupt shift (Sultan and Janicot, 2000) occurs at the beginning of the summer season in early June, when the rain band core moves suddenly northward to about 10°N. This indicates the beginning of the high rain period with a peak reached in August around 12°N over the Sahel. A gradual retreat of the monsoon starts in early September, yielding a decrease in intensity and a southward migration of the rain band.
There are both similarities and differences across the three observation datasets. All three datasets agree in showing the heavy intensity rainfall around 4°N in June and the northward sudden jump. FEWS shows lower intensity of peak rainfall over the Sahel compared to TRMM and GPCP during the summer season. Some disagreement among the observations is found in the August–September monsoon precipitation, which is lower in FEWS compared to TRMM and GPCP.
RegCM3 (Figure 3(d)) reproduces the seasonal evolution of the monsoon rain band and the three distinct phases of the rainy season, however some differences with respect to the observed data are found. The general timing of the onset monsoon jump is simulated but the associated rain maximum around 4°N over the Guinea coast is underestimated and it occurs somewhat earlier than observed. Conversely, precipitation is overestimated during the early stage of the retreat phase in September.
Over the East Africa region, observations (Figure 4(a)–(c) for FEWS, TRMM and GPCP respectively) show a number of features. First, the band of high intensity rainfall, confined in the southern hemisphere during the beginning of the year, undergoes a sudden jump northward from April to May, when the peak associated with the long rainy season occurs. This feature is followed by a small break in June and a precipitation maximum in the northern regions in July-September. This maximum corresponds to the East African short rainy season and it is associated with large-scale atmospheric and SST anomaly patterns discussed, for example, by Segele et al. (2009b). Another break occurs in September and October, while a second rainy season appears from late October in the East African equatorial regions. Observations agree in the timing and amplitude of the zonal migration of rainfall features, although the peak rainfall over the northern regions is more intense and somewhat more extended in GPCP than FEWS and TRMM.
The regional climate model (Figure 4(d)) captures the three phases of the seasonal evolution of the East Africa monsoon, but also shows significant differences compared to observations. First, the rainfall amounts during the East Africa long rainy season (July–September) are overestimated, mostly as a result of the overestimation of precipitation over the Ethiopian Highlands previously discussed (Figure 2). In addition, RegCM3 shows overly wet conditions during the retreat phase in August–September compared to observations. Part of this overestimate is associated with persistent rain over the Lake Victoria region associated with the lake evaporation source. In fact, Anyah and Semazzi (2007) coupled RegCM3 with a 1D Lake Model and found this precipitation overestimation significantly decreased.
Over Southern Africa, the daily rainfall variability exhibits three distinct features. Observations (FEWS, TRMM and GPCP, respectively; Figure 5(a)–(c)) show that during the austral winter, the South African continental regions are predominantly dry. The rainfall distribution displays two axes of intense convection: a northern axis following the ITCZ and a southern axis associated with the easterly subtropical trough over Zimbabwe-South Africa (Jury, 1999). Starting from July, high intensity rainfall moving along these two axes confluences around a frontal boundary between 10°S and 20°S by the end of October: the ZAB. This is the onset and inception phase of the rainy season over those regions. The wet conditions around the ZAB and complex terrains of Southern Africa persist until late April and proceed to a retreat causing the sudden termination of heavy precipitation and the inception of the dry season over the region. Also in this region, as in West and East Africa, GPCP shows higher intensity of rainfall during the peak of the rainy season (DJF) compared to the other observed rainfall products. The regional model reproduces well the onset, peak and northward and southward retreat of the high intensity rainfall axes associated to the ITCZ and the easterly subtropical trough, respectively. As in the other regions, however, RegCM3 overestimates the precipitation intensity along the ITCZ during its southward displacement and northward retreat.
Summarizing the results of this section, the three observations datasets show a general agreement in the representation of the monsoon rain band dynamics over the regions considered, while, as discussed in the previous section, they exhibit differences mostly in the precipitation intensity estimates. Overall, the RegCM3 reproduces the observed seasonal and intra-seasonal evolution of the different monsoon rain bands, but shows systematic biases, most noticeably an overestimation of precipitation over the Eastern and Southern Africa mature monsoon phases. These appear related to specific features, such as the precipitation overestimate over the Ethiopian highlands and the Lake Victoria areas.
3.3. Mean frequency, intensity and extremes of daily rainfall
We now turn our attention to the analysis of daily precipitation characteristics using the measures described in Table 1. Figure 6 shows the number of wet days (i.e. days with precipitation in excess of 1 mm) for DJF (upper panels), JJA (middle panels) and ANN (lower panels) in the FEWS (first column), TRMM (second column), GPCP(third column) and RegCM3 (fourth column) data. As can be expected, for all three observational datasets the spatial distribution of the number of wet days essentially follows the corresponding mean precipitation in each season. Key differences among the observed rainfall products are however observed. The most evident is the substantially larger number of rainy days in GPCP and FEWS compared to TRMM (more than 10–20%). This is consistent with the higher GPCP precipitation amounts, which thus appear to be related to a higher frequency of precipitation. On the other hand, larger numbers of precipitation days are found in FEWS compared to TRMM, even though the total precipitation amounts in FEWS are lower (Figure 2). This has important implications on the rainfall intensity, as discussed later.
The RegCM3 simulation shows frequency values that are generally intermediate between those in the observations, in fact mostly in line with FEWS and GPCP in East and West Africa and with TRMM over South Africa and the Congo Basin. Some regional discrepancies with observations are observed: (1) an underestimate of rainfall events in DJF over most of the Congo basin; and (2) an excessive number of rainfall events in JJA over orographic zones of West Africa. In areas where RegCM3 shows lower frequencies of rainfall events with respect to FEWS and GPCP, simulated values are consistent with TRMM observations.
Figure 7 shows the mean intensity of precipitation in the three observation products and the RegCM3 simulation. For this variable, striking differences are found across the observations datasets. Specifically, TRMM and GPCP produce intensity values that are consistently and substantially larger (more than 10–12 mm d−1 and 4–6 mm d−1, respectively) than in FEWS, with discrepancies being most pronounced in JJA and ANN. In particular, the largest precipitation intensities are found in TRMM. Similarly to the case of wet day number, the RegCM3 shows values generally intermediate across the three datasets except over East Africa, where the precipitation intensity is consistently overestimated. In general, the regional model shows intensities that lie towards the upper end of the observation ensemble.
Observed and simulated 95th precipitation percentiles are shown in Figure 8. This measure of extreme essentially follows the same pattern as the daily precipitation intensity: TRMM and GPCP exhibit consistently and substantially larger observed estimates than FEWS (more than 10–15 mm d−1 and 5–10 mm d−1, respectively). The regional model produces values towards the upper end of the observed range, but generally in line with both TRMM and GPCP. Only over the mountainous regions of East Africa the regional model consistently overestimates the 95th precipitation percentile compared to all datasets.
It is thus clear from Figures 6, 7 and 8 that the three observation datasets provide a substantially different description of the daily characteristics of rainfall. TRMM produces the most intense and less frequent daily rainfall events. FEWS produces more frequent but lower intensity events, which contributes to lower mean precipitation amounts, while GPCP exhibits the largest frequency and high intensity of events, contributing to the largest total precipitation amounts (Figure 2). It should be noted that, although the TRMM, GPCP and FEWS merged products use similar procedures, the satellite and merged station input data are different. Differences may also arise from the merging techniques and the way the gauge data enter the satellite retrieval algorithms during their initial formulation (Huffman et al., 2001, 2007; Love et al., 2004). This should strongly impact the frequency, intensity and duration of the daily rainfall events across the different products. For example, the lack of microwave remote sensing techniques in the FEWS might have led to a failure to capture locally heavy precipitation events (Love et al., 2004).
An overall intercomparison of the observations and RegCM3 data of number of wet days, intensity and extremes shows mixed results. It indicates that the model has more intense but less frequent events over areas of East and South Africa than in FEWS and GPCP, and more frequent but less intense events than in TRMM over West Africa. Therefore, the large differences across the three observational datasets make it difficult to unambiguously assess the model performance in these measures. The model results appear generally within the bounds of the uncertainties in observations except over specific sub-regions, most noticeably the mountainous regions of East Africa.
3.4. Maximum duration of wet and dry spells
Two important aspects of daily precipitation characteristics for assessing possible climate impacts are the maximum lengths of consecutive dry and wet spells within the monsoon seasons.
Figure 9 shows the average maximum wet spell length (i.e. the maximum number of consecutive wet days in a season or year) for observations (FEWS, TRMM and GPCP) and model simulation at the seasonal (DJF and JJA) and annual (ANN) timescales. GPCP shows generally longer maximum wet day spells (more than 4–8 d) compared to FEWS and especially TRMM, especially over the complex terrains of Southern and Eastern Africa (including the Sudanese highlands), along the ITCZ, and in the Congo basin. This is broadly consistent with the number of wet days shown in Figure 6. Although FEWS and GPCP display similar magnitudes of wet spell over many regions, the prominent difference between them includes shorter maximum wet spells (within 2–5 d) in the Gulf of Guinea and along the eastern coastlines of Africa in FEWS.
The RegCM3 shows values that are mostly within the observed ones, particularly in line with the GPCP values, except for the occurrence of large localized maxima of maximum wet spell length in correspondence of the mountainous areas of Ethiopia, Cameroun and the Guinea coast. This is an indication of the persistent orographic forcing in the model which tends to sustain prolonged precipitation over mountainous systems once convection is activated, particularly because of the occurrence of medium to light precipitation events.
The spatial patterns of the maximum dry spell length (Figure 10) show a good general agreement (with 5–10% of uncertainties) across the different observation datasets and with the model simulation. Some regional differences across the observed dataset are found. For example, TRMM displays longer dry spell lengths in the northern Kalahari Desert, and FEWS and GPCP exhibit lower but similar minima in both magnitude and spatial expansion over sub-equatorial southern and northern Africa in DJF and JJA, respectively. Similarly, regional differences between simulations and observations occur, such as an overestimate of maximum dry spell length over the Congo Basin. In particular, the model appears to produce longer maximum dry spell lengths during the local dry season.
In summary, the observational and simulated datasets are broadly consistent in terms of maximum dry spell length, while greater differences are found for maximum wet spell length, most noticeably the presence of generally shorter maximum wet spells in TRMM and excessively persistent wet spells in the regional model in correspondence to topographical relief. Overall, the analysis shows the presence of substantial uncertainties among the various observational products that affect significantly the characteristics of their daily rainfall distributions over Africa. These differences give rise to complex issues regarding the analysis and improvements of regional climate models performance over the monsoon regions.
4. Summary and conclusion
Rainfall distribution is critical over Africa for many applications such as water resources, agriculture, and drought and flood forecasts. Simulating and understanding the spatial and temporal variability of precipitation at the daily timescale is a challenging problem as it requires high quality observed datasets for model performance evaluation. A number of such datasets are today available for the African continent through the blending of satellite and in situ station observations, but these data suffer from likely uncertainty due to limitations in density and quality of available stations and techniques for data blending.
In this paper, we presented an intercomparison of different observed precipitation datasets (TRMM, GPCP, FEWS and, for seasonal and annual average, CRU) and a validation exercise of a regional climate model simulation (with the RegCM3 model) viewed within the context of the observation uncertainties. We focused not only on mean precipitation but also on the characteristics of rainfall at the intraseasonal and daily scale. A range of model performance measures was investigated, including monsoon evolution, frequency, intensity and extremes of precipitation, along with maximum length of dry and wet spells.
Our analysis first indicates that substantial discrepancies exist among the different observational datasets in some of the metrics evaluated. In particular, the observation products can differ substantially in terms of mean precipitation amount (with FEWS exhibiting the lowest amounts and GPCP the largest). More specifically, these differences are associated with a wide range of estimates of higher order statistics, such as frequency, intensity and duration of rain events. For example, the FEWS dataset produces more frequent but less intense daily events than TRMM. In other words, the observation datasets analysed here provide quite different descriptions of daily rainfall characteristics and this adds a strong element of uncertainty in the model evaluation.
This uncertainty is indeed reflected in the quantitative assessment of the model performance. In fact, while the RegCM3 shows a tendency to overestimate precipitation, particularly in mountainous areas of East and Southern Africa, the model bias is strongly dependent on the observations used to calculate it. The model generally captures the spatial patterns of precipitation as well as the evolution of different monsoon systems at the intra-seasonal scale. At the daily scale, the model statistics are mostly within the relatively wide range found in the three observation datasets examined.
Although the comparison with observations indicates a realistic representation of daily precipitation statistics by the model, it is recognized that the large uncertainties in the observations make a rigorous and unambiguous model evaluation rather difficult. It is thus imperative that the quality and consistency of available high temporal and spatial resolution observation datasets is improved towards a better understanding of the response of the hydroclimatology of Africa to global warming.