Congo Basin Hydrology, Climate, and Biogeochemistry. Группа авторов

the observed data were thereafter used for forward prediction based on the hold‐out method of cross‐validation (e.g., Haley, 2017). The stratified partitioning of the data using this approach ensures that each partition includes similar amount of observations from each group. The predicted and observed discharge were then compared using Pearson correlation.

5.3. RESULTS

      5.3.1. Characteristics of Extreme Events in the Congo Basin

Three leading modes of variabilities, accounting for a total of 34.5% were identified in the statistically decomposed SPEI patterns (1980–2015) over the Congo Basin (Figure 5.1a–f). From the joint interpretation of the spatiotemporal patterns of SPEI localized over the tropical south of the basin, the longest drought duration occurred during the 1992–1996 and 2003–2006 periods (Figure 5.1a,b). During the 1985–1991 and 2007–2008 periods, the SPEI time series associated with this southern section of the basin were significantly wet. In terms of the total variability accounted for in the three SPEI modes over the basin, the southern section of the basin with considerable changes in rainfall shows the highest (17%) SPEI variability (Figure 5.1a,b). We agree that the central regions and areas of the Congo basin below the equator are apparently and significantly wet in terms of rainfall amount and the presence of surface water and fluxes (Section 5.3.2.2). It is also true that the southern section experiences drought and drier conditions more frequently compared to other regions (Figure 5.1a,b). However, as with wet regions of the basin, the amplitudes of rainfall in the tropical wet‐dry southern section are strong and show monthly averages of about 250 mm between December and February, consistent with rainfall amounts during the September–November period in the central region. There is considerable evidence in the literature regarding the distribution of rainfall in the south between December and March (e.g., Alsdorf et al., 2016; Creese et al., 2019; Munzimi et al., 2015; Ndehedehe et al., 2019) and our spatiotemporal analysis of land water storage highlights this pattern in the wet‐dry and temperate regions of the southern region (Section 5.3.2.1), given that rainfall is the main input to hydrological systems.

Figure 5.1 Spatiotemporal SPEI patterns of the Congo basin using 12‐month gridded SPEI values (a–f). Localized spatial SPEI patterns (right) correspond to the temporal evolutions (left) and actual SPEI values to be used for drought classification (drought threshold is in red) are jointly derived from the spatial and temporal patterns. The SPEI time series (blue) are filtered to cushion the effect of residual short‐term seasonal signals and also for better representation.

Moreover, there are few drought episodes in the southwestern region of the basin (Figure 5.1c,d). The evolution of wet episodes in the southwestern region suggests it appears to be wet most of the times (Figure 5.1c,d). However, since 2010, the temporal patterns of SPEI in the southwestern region have been largely somewhat less than moderate (Figure 5.1c,d). The decomposition of SPEI over the basin also shows that drought conditions of the early 1980s affected the northern section in the Central Africa Republic (CAR) (Figure 5.1e,f). The frequent episodes of droughts in CAR obviously are in sharp contrast to the wet episodes observed in the southeast region of the basin (SPEI‐3, Figure 5.1). Some moderately wet periods between 1995 and 2002 (Figure 5.1e,f) are also noted in the CAR region. The latter is a humid tropical wet and dry savannah ecosystem largely characterized by considerable changes in annual and seasonal rainfall that is in opposite phase with the southern section. Arguably, the nourishment of the Congo basin hydrology and freshwater ecosystems also emanates from the southern end of the basin where extreme droughts tend to be more frequent (Figure 5.1a,b). As this region is also characterized by high rainfall amounts, which occur all through the year except during the June–August period, a shift in hydrological regime of the Congo Basin is more likely. Drought intensities over the Congo basin are conspicuously moderate or probably less. Droughts persisted between 1992 and 2001 with more than 40% coverage between 1994/1995 and early 2006 (Figure 5.2a,b). The observed extreme drought between 2004 and 2006, one of the post‐2000 period with widely acknowledged hydrological drought period in the basin reported in the literature also persisted, fluctuating between 25% in 2004 and more than 40% in 2006 (Figure 5.2b). During the last few decades (between 1984 and 2011), recurring severe and extreme drought episodes have affected on the average at least 30% of the Congo Basin (Figure 5.2). While the 1994 extreme droughts reached 50%, in other periods (e.g., 1992, 1999, 2004, 2005/2006), only about 30% of the basin on average has been affected by extreme drought during the 1991–2011 period (Figure 5.2a,b). This increasing intensity in extreme drought episodes is consistent with the results indicated in Figure 5.1. However, between 2012 and 2016, drought episodes and their intensities have diminished over the Congo Basin (Figure 5.2), consistent with the temporal SPEI patterns shown in Figure 5.1. But the intensity of the well‐known large‐scale extreme droughts of the 1980s, which affected Africa, are less and not as widespread in the Congo Basin compared to other African sub‐regions where drought‐affected areas ranged from 70 to more than 90% (e.g., Agutu et al., 2017; Masih et al., 2014; Ndehedehe et al., 2019, 2020a).

Figure 5.2 Estimated areas affected by various drought intensities (extreme, severe, and moderate) over the Congo basin during 1980–2000 (a) and 2001–2015 (b) periods. The SPEI‐derived drought affected areas (%) are characterized based on the classification thresholds defined in McKee et al. (1993) and Ndehedehe et al. (2019). Note this SPEI is based on a 12‐month aggregation.

      5.3.2. Surface Water Hydrology of the Congo Basin

       Spatial and Temporal Patterns of Land Water Storage

Linear trends from three GRACE‐mascon solutions are summarized in Figure 5.3a–c for the period between Jan 2003 and Dec 2015. Despite having a spatial resolution of 0.5° × 0.5° (CSR and JPL), the mass changes show the overall structures of the 3.0° × 3.0° native resolution and its suitability to capture hydrological