We used the estimates of surface water extent (SWE) derived from GIEMS-2, which provides global coverage at a monthly time step of different water bodies, including wetlands, rivers, and lakes at 0.25∘ (∼27 km) spatial resolution at the Equator (on an equal-area grid, i.e. each pixel covering 773 km2; Prigent et al., 2007, 2020). The dataset was developed by merging observations from different sensors, as described in Prigent et al. (2007) and Papa et al. (2010). The last version used in this study spans over a long-term period from 1992 to 2015. For more details about the technique, we refer the reader to Prigent et al. (2007, 2020).

Several studies, such as Prigent et al. (2007, 2020), Papa et al. (2008, 2010, 2013), and Decharme et al. (2011), have been assessing the interannual and seasonal dynamics of the long-term SWE in different environments against several variables, such as the in situ river discharges, in situ and satellite-derived water level in rivers, lakes, wetlands, the total water storage from GRACE, and the satellite-derived rainfall. Recently, the characterization and evaluation of the 24-year SWE from GIEMS-2 have been conducted in the CRB against the in situ river discharge and water level and the performance gave satisfactory results (see Figs. 6 to 10 of Kitambo et al., 2022a, for details on the characterization and the assessment of GIEMS-2 over the CRB).

Satellite radar altimetry provides a systematic monitoring of the surface water height (SWH) of large rivers, lakes, wetlands, and floodplains at the virtual station (VS), defined as the intersection of a water body with the satellite theoretical ground track. The temporal variation of SWH is retrieved according to the repeat cycle of the satellite (Da Silva et al., 2010; Cretaux et al., 2017), a cycle that varies from 10 to 27 d for current operational satellites. Several studies, including Frappart et al. (2006), Da Silva et al. (2010), Papa et al. (2010, 2015), Kao et al. (2019), Kittel et al. (2021), Paris et al. (2022), and Kitambo et al. (2022a), to name a few, have been conducted in different river basins to validate SWH variations against in situ water levels. Results generally show a good capability of radar altimeter to retrieve SWH variability with uncertainties ranging from a few centimetres to tens of centimetres, depending on the acquisition mode of the satellite and the environmental characteristics (Bogning et al., 2018; Normandin et al., 2018; Jiang et al., 2020; Kittel et al., 2021; Kitambo et al., 2022a).

Over the CRB, Kitambo et al. (2022a) used ∼2300 VSs from different satellite missions and their pooling based on the principle of the nearest neighbour (located at a minimum distance of 2 km; see Da Silva et al., 2010; Cretaux et al., 2017) to generate long-term time series with record lengths ranging from 12 to 25 years (Fig. 2d of Kitambo et al., 2022a). A thorough assessment and validation of these long-term satellite-derived surface water height at nine in situ gauge stations provided a root-mean-squared error ranging from 10 (with Sentinel-3A) to 75 cm (with the European Remote Sensing satellite-2 – ERS-2) (see Table 2 of Kitambo et al., 2022a).

In the current study, the satellite-derived SWHs used are the ones spanning the record period 1995–2015 acquired from three satellite missions, (1) ERS-2, with observations spanning April 1995 to June 2003, (2) ENVISAT (hereafter named ENV), with observations spanning March 2002 to June 2012, and (3) the Satellite with ARgos and ALtiKa (SARAL/Altika, hereafter named SRL), from which we use observations from February 2013 to July 2016. All three satellite missions have a 35 d repeat cycle. These datasets were made available by the Centre de Topographie des Océans et de l'Hydrosphère (CTOH, http://ctoh.legos.obs-mip.fr, last access: 17 May 2023). They come from the geophysical data records made available by space agencies. For ERS-2, land reprocessing was used (Frappart et al., 2016). These datasets were processed using either the Multi-mission Altimetry Processing Software (MAPS, Frappart et al., 2015b) or the Altimetry Time Series Software (Frappart et al., 2021) to generate the time series of water levels. Therefore, the 160 generated VSs cover the entire CRB (see Fig. S1 in the Supplement) and a period of ∼21 years.

The south-eastern portion of the basin, including the Lake Upemba and Lake Bangweulu regions, was not covered by the SWH VSs due to the simultaneous lack of data from the three aforementioned satellite missions. Over this region, missing data were replaced by the annual cycle computed using the altimetry-based water levels available during the study period.

We used four freely available global digital elevation models (DEMs) (Table 1): (1) ASTER version 3, (2) the Advanced Land Observing Satellite (ALOS), (3) Multi-Error-Removed Improved Terrain (MERIT), and (4) the Forest And Buildings removed Copernicus DEM (FABDEM).

DEMs are divided broadly into two categories based on the specific topographic surfaces they represent, which are digital surface model (DSM) and digital terrain model (DTM). DSM refers to the upper surface of natural and built or artificial features of the environment such as buildings, artificial features, and trees, while DTM represents the elevation of the Earth's surface with all natural and built features removed, i.e. the bare Earth surface (Guth et al., 2021; Hawker et al., 2022). Among the DEMs used, ASTER and ALOS are classified as DSMs. MERIT is closer to a DTM because of the removal of tree height bias, but it is not a complete DTM (Yamazaki et al., 2017; Hawker et al., 2022) due to other artifacts such as artificial features. In this study, only FABDEM can be considered a DTM (Hawker et al., 2022).

Therefore, in order to remove the presence of tree bias in DSMs, we subtract from them the forest canopy height from a global dataset (Potapov et al., 2020). For that, the global canopy height dataset, ASTER, and ALOS were all resampled to 90 m spatial resolution using the nearest-neighbour resampling method.

The global forest canopy height (available at https://glad.umd.edu/dataset/gedi/, last access: 17 May 2023) is a global dataset developed by combining the Global Ecosystem Dynamics Investigation (GEDI) lidar forest structure measurement and Landsat analysis-ready data time series (Potapov et al., 2020). GEDI is a new spaceborne lidar instrument on board the International Space Station collecting data on the vegetation structure since April 2019 (Dubayah et al., 2020). The spatial resolution of the dataset is 30 m, providing the global forest canopy height map for the year 2019 in the WGS84 reference system. The dataset covers zones between 54∘ N and 52∘ S globally, and then we use it over the CRB. For more details on the dataset, we refer the reader to Potapov et al. (2020).

Over the largest lakes of the CRB, time series of the monthly water storage anomaly for Lakes Bangweulu, Kivu, Mweru, Tanganyika, and Upemba (see Fig. 1 for their locations and Fig. S2 for their time series) are estimated using surface water extent and water level time series obtained from the HydroSat database (accessible at https://www.gis.uni-stuttgart.de/en/, last access: 17 May 2023; Tourian et al., 2022). After collecting the simultaneous lake water area and height measurements, the empirical relationship between lake surface water level and area is developed. This model represents the bathymetry of the lake for the part which is captured by remote-sensing observations. By assuming that the lake has a regular morphology and a pyramid shape between two consecutive measurement epochs, the lake water level area storage model is developed. Finally, time series of the lake water storage anomaly are calculated using the developed model and lake water level or surface extent measurements.

The multi-satellite approach (Fig. 2a) consists of the combination of the SWE and satellite-derived SWH over inland water bodies (rivers, lakes, reservoirs, wetlands, and floodplains), generally derived from radar altimetry over a common period of availability for both datasets (Frappart et al., 2008, 2011; Becker et al., 2018; Papa and Frappart, 2021). Therefore, this complementarity of multi-satellite observations offers the possibility of quantifying SWS changes and water volume variations in a watershed. SWE and SWH used in this study are respectively from GIEMS-2 and the family of spaceborne radar altimeters with a 35 d repeat cycle (hereafter ERS-2, ENV, and SRL). Their common period of availability is 1995–2015.

We summarize in the next sections the two-step methodology and, for more details, we refer the reader to Frappart et al. (2008, 2011, 2012, 2019).

In complement to the multi-satellite approach, we also used the hypsometric curve approach that consists of the combination of SWE and DEM-based topographic data. Following Papa et al. (2013), we summarize here the four-step process (Fig. 2b) to estimate SWS, using as an example the combination of GIEMS-2 SWE and FABDEM topography resampled at 90 m.

Figure 6 presents the characteristics of the SWS temporal dynamics at the CRB scale (anomaly time series vs. its long-term mean, deseasonalized anomaly, and annual cycle for SWS aggregated for the entire CRB). It shows all SWS estimates computed with both the multi-satellite (for 1995–2015) and hypsometric curve (for 1992–2015) approaches from the use of the four DEMs (ALOS, ASTER, MERIT, and FABDEM).

Figure 6a shows, for the very first time, the long-term month-to-month variations of the CRB's SWS over a period of 24 years. It shows a strong seasonal cycle of SWS over the CRB with comparable behaviour in the peak-to-peak SWS variations from both approaches. The SWS amplitude ranges from ∼50 to ∼150 km3 over the years, showing a large year-to-year variability. The bimodal patterns that characterize the hydrological regime of the CRB (Kitambo et al., 2022a), linked to the variability of the Intertropical Convergence Zone (Kitambo et al., 2022a), are well depicted in the SWS estimates.

All SWS estimates from the different DEMs show fair agreement in their variations between them (Fig. 6a). However, we observed that SWS from ASTER (violet colour) tends to overestimate the SWS at the first peak (i.e. spanning over August–February) along the time series.

Figure 6b displays the deseasonalized anomaly (obtained by subtracting the mean monthly values over the considered periods 1992–2015 or 1995–2015 from individual months) of the CRB's SWS. A similar observation of the matching of the SWS anomaly from different approaches and DEM products is observed.

These SWS estimates also show a substantial interannual variability at the basin scale, especially in terms of annual maximum and minimum, pointing out extreme events in terms of droughts and floods that recently affected the CRB. Figure 6b reveals interesting and strong anomaly signals, such as the large positive peaks observed in 1997–1998 and 2006–2007. These can be related to the positive Indian Ocean Dipole (pIOD) events in combination with the El Niño events occurring in 1997–1998 and 2006–2007 that triggered floods in the western Indian Ocean and eastern Africa (Mcphaden, 2002; Ummenhofer et al., 2009; Becker et al., 2018; Kitambo et al., 2022a). Another major event is the severe drought that occurred in 2005–2006, which is clearly depicted in the CRB's SWS time series anomaly as the minima of the record (Fig. 6b). This is in agreement with Ndehedehe et al. (2019), Ndehedehe and Agutu (2022) and Tshimanga et al. (2022), who reported that, in the 1990s and 2000s, multi-year droughts in the CRB affected a significant part of the Congo basin. This interannual variability is also superimposed on variations at a larger timescale, from a few years to decades, such as a large increase in the SWS anomaly in 1996–1997 followed by a slight decrease until the minimum that occurred in 2005–2006, before SWS starts to slowly increase again after the 2007 peak until 2015.

Figure 6c shows the CRB's SWS annual cycle (computed over the 1992–2015 and 1995–2015 periods for the hypsometric and multi-satellite approaches respectively), revealing a strong seasonal variation. Both approaches present a mean annual amplitude of the same order of magnitude (Table 3), with estimates ranging around 80±17, 101±23, 80±20 km3 (from ALOS, FABDEM, and MERIT respectively based on the hypsometric curve approach), and 70±17 km3 from the multi-satellite approach. These estimates are of the same order of magnitude as previous findings over the CRB, i.e. 81±24 km3 as in Becker et al. (2018). As a matter of comparison, the mean annual amplitude of SWS from ASTER represents ∼11 % of the Amazon basin's SWS mean annual amplitude of ∼1200 km3, as reported in Papa et al. (2013), Papa and Frappart (2021).

As observed in Fig. 6a, ASTER's SWS provides larger estimates, with a mean annual amplitude of ∼124±25 km3 (Table 3). This can be explained by the fact that, among the four DEMs (ALOS, ASTER, FABDEM, and MERIT), ASTER has a greater vertical error (i.e. 17 m; see Table 1 of Hawker et al., 2019) compared to the others and, consequently, this can impact the elevation amplitude (derived at step 2 of Sect. 4.2 used to calculate the SWS variations (step 3 of Sect. 4.2).

Table 3 also presents the statistical errors comparing FABDEM's SWS dataset to other estimates and reinforces the difference highlighted in SWS magnitude between different approaches and DEMs. Note that trees and urban-area biases are only removed from FABDEM, and thus it seems to be the most adequate DEM in representing hydrology processes (Hawker et al., 2022). ALOS and MERIT have provided small errors, as reflected in the mean absolute error (MAE) (9 and 5 km3) and the root-mean-squared-error (RMSE) values (11 and 7 km3) compared to values greater than 15 km3 of MAE and above 20 km3 of the RMSE from the multi-satellite approach and the ASTER DEM.

At the basin scale, Fig. 6c clearly depicts a double peak, with a SWS maximum reached in November for the first peak and in April for the second peak. In comparison with the annual cycle of the GIEMS-2 SWE (Fig. 7c of Kitambo et al., 2022a), the first peak of the SWS maximum is in phase with the maximum SWE, while for the second peak there is a 1-month delay, with the maximum SWE occurring in March. This can be explained by the non-linear relationship between SWE and SWS through the hypsometric curve approach.

It is important to note that the SWS from FABDEM and MERIT shows a very good fit in all the seasons, whereas ALOS slightly underestimates the storage at the second peak. In contrast to the others, ASTER shows peculiar behaviour, with its SWS largely underestimating and overestimating the storage at the second and first peaks respectively of the hydrological regime.

In agreement with the results from the SWS hypsometric curve approach, the SWS from the multi-satellite approach also points out the maximum SWS in November and April for the first and second peaks. However, the dynamics of the SWS from the multi-satellite approach differ from the others over the period from February to May. Over that period of time, the hydrological regime of the basin is more controlled by the south-eastern region, especially by the Lake Bangweulu and Lake Upemba area (Kitambo et al., 2022a). In this region, the spatial distribution of VSs is not sufficient enough (Fig. S1) to obtain a very accurate representation of the temporal variations of the SWS even if the mean annual cycle variations of some VSs from ENV and SRL were used to account for the water level over the entire 1995–2015 period. This might explain in part the different dynamics observed in the SWS variations over February–May in the south-eastern region.

In parallel to Fig. 6, Fig. 7 presents the spatial distribution of SWS dynamics (mean annual, SD, mean annual maximum, and the average month of the maximum) for the entire CRB. In the following parts of the paper, the estimates obtained with the FABDEM DEM will become our reference, and we will use them to display the results. This is justified by FABDEM DEM topographic characteristics and properties, which makes it the closest to a DTM. The estimates obtained with the other DEM will be displayed in the Supplement (Fig. S6).

In agreement with the spatial distribution of SWE across the CRB (Fig. 6 of Kitambo et al., 2022a), SWS (Fig. 7a and b) shows realistic spatial patterns along the Congo River and the Cuvette Centrale and depicts quite well the other main structures of the basin, for instance the main tributaries (e.g. the Sangha, Ubangui, Luvua, Luapula, and Lualaba rivers) along with their large wetlands and floodplains. These are characterized by a strong variability in SWS (Fig. 7c and d).

Higher values of SWS ranging from 0.3 to 0.6 km3 in a 773 km2 pixel dominate the extensive wetlands and floodplains such as the Cuvette Centrale and in the south-eastern part of the basin (the Upemba and Bangweulu region). These regions display a large variability as well (in terms of the SD, Fig. 7c and d) and are characterized by maximum values of surface water storage generally greater than 0.6 km3 per 773 km2 pixel (Fig. 7e and f).

The heart of the Cuvette Centrale, Lake Mai-Ndombe (in the Kasaï sub-basin), and a large part of the main channel of the Congo (up to the Lomami River) presents the maximum values of the SWS change in the basin. The same observation is made for the lakes in the Upemba depression (e.g. Lake Upemba), Mweru, and Bangweulu. These maximum values are reached in September–October in the upper part of the Cuvette Centrale and November–December in its lower part (Fig. 7g and h). In the Lualaba sub-basin, the average month of the maximum of SWS is January–February, while in the other parts, such as Lake Mweru and the eastern part of Lake Bangweulu, with its large grassy swamps and floodplains, it recorded in March–April. Conversely to the general trend in the Cuvette Centrale, the regions of Lake Tumba and Lake Mai-Ndombe reached their maximum SWS in July–August.

In general, the results from the hypsometric curve (Fig. 7, left column) and multi-satellite (Fig. 7, right column) approaches are quite similar in terms of spatial distribution for both the magnitude and variability of the changes. Nevertheless, as expected, the multi-satellite dataset approach shows a limitation in terms of spatial distribution caused by the reduced availability of the combined long-term VSs in some regions. For instance, there is a lack of observations of the eastern part of Lake Tanganyika and in the Bangweulu region, where the spatial distribution of SWS from the multi-satellite approach is smaller than that of the FABDEM hypsometric curve approach. This is mainly explained by the sparse availability of long-term satellite-derived (ERS-2–ENV–SRL) time series in that region, leading to less distributed SWS estimates in these regions.

On the other hand, SWS variations from the hypsometric curve approach also present limitations, mainly on small lakes and around large lakes, where there are almost no variations in elevation from the DEMs, leading to flat hypsometric curves and therefore to the computed storage having zero changes. This is for instance the case around Lake Kivu in the Lualaba sub-basin. In this case, the SWS change in the lake is then added to the dataset using the lake water storage anomaly computed independently (see the data description in Sect. 3.5).

Finally, for comparison purposes, Fig. S6 shows the same analysis in terms of spatial distribution and variability of SWS across the basin for the other estimates based on the hypsometric curve approach (i.e. ALOS; see Fig. S6, left column; ASTER; see Fig. S6, middle column; MERIT; see Fig. S6, right column). In general terms, results are consistent between one another. The pattern of SWS in terms of the distribution and order of magnitude of the mean annual (Fig. S6a, b, and c), variability in terms of the SD (Fig. S6d, e, and f), mean annual maximum (Fig. S6g, h, and i), and average month of the maximum (Fig. S6j, k, and l) between the three DEMs and FABDEM are generally similar, although results from the ASTER DEM (Fig. S6b, e, and h) underestimate the storage (i.e. values ranging between 0.15 and 0.45 km3) compared to other DEMs (i.e. values greater than 0.6 km3) in the Bangweulu–Upemba region (e.g. Lake Bangweulu, Lake Upemba).

At the sub-basin scale, the mean annual amplitude for the five sub-basins is provided as follows. The Middle Congo is the sub-basin with a large amplitude (71±15 km3) associated with the strong variations of the SWS anomaly observed in the Cuvette Centrale. It is followed by the Lualaba sub-basin with 59±15 km3 due to the presence of major lakes (Kivu, Tanganyika, Mweru, Bangweulu) and large wetlands that show large variability and that are characterized by maximum values of surface water storage generally greater than 0.6 km3 per 773 km2 pixel. Sangha and Kasaï show quite similar annual amplitudes (i.e. 24±5 and 24±6 km3 respectively). Both sub-basins are overlapped at their mouth (i.e. the downstream part) by the Cuvette Centrale. Among the five sub-basins, Ubangui is the one with the smallest mean annual amplitude (13±4 km3), although it is among the two northern sub-basins (i.e. Ubangui and Sangha) with higher satellite-derived SWH mean maximum amplitude (see Fig. 5a of Kitambo et al., 2022a). As observed for the Sangha and Kasaï sub-basins, the Ubangui sub-basin is also occupied in its downstream part by the Cuvette Centrale (Fig. 1).

In order to evaluate the monthly estimates of the large-scale SWS over the CRB, we compare, at the basin and sub-basin levels, their seasonal and interannual variability against other independent hydrological variables such as precipitation data from MSWEP V2.8 and in situ discharges. For clarity purposes, only the SWS results from the hypsometric curve approach with FABDEM and from the multi-satellite approach are displayed and discussed here.

At the basin level, Fig. 8 presents the comparison of the aggregated normalized SWS anomaly variation over the entire basin against the normalized precipitation anomaly and the in situ normalized discharge anomaly at the outlet of the basin (Brazzaville/Kinshasa station; see Table 2). The normalized anomalies here are estimated by subtracting the mean value of the time series from individual months and by dividing the obtained series by the SD of the original time series. As a complement, we report in Table 4 the maximum linear Pearson correlation coefficient along with its lag, calculated between the aggregated normalized SWS anomaly variation, the normalized precipitation anomaly, and the in situ normalized discharge anomaly at basin and sub-basin scales.

Figure 8a presents a fair agreement between the SWS and the other two hydrological variables, with a maximum correlation coefficient r of 0.56 (lag = 0; p value <0.01) between SWS and precipitation variations. A similar correlation coefficient (r=0.57 with a 0-month lag; p value <0.01) is found between the normalized SWS anomaly and the in situ normalized discharge anomaly. The deseasonalized normalized anomalies (acquired by subtracting the mean monthly values over the considered periods 1992–2015 or 1995–2015 from individual months and dividing by the SD of the raw series) (Fig. 8b) show correlation coefficients of r=0.52 (lag = 0; p value <0.01) and r=0.12 (lag = 0; p value <0.05) respectively between SWS vs. in situ discharge and SWS vs. precipitation.

Additionally, we also perform a comparison with previous estimates of SWS over the Congo basin from Becker et al. (2018), estimated using the multi-satellite approach and available over the period 2003–2007. The assessment with the SWS anomaly from FABDEM shows good agreement (Fig. 8a) with similar amplitude and a maximum correlation coefficient of r=0.73 (lag = 0; p value <0.01).

At the seasonal timescale, Fig. 8c reveals for the first peak (i.e. August–February) that the SWS anomaly reaches its maximum in November, 1 month before the maximum of the river's normalized discharge anomaly (December) and after the maximum of normalized precipitation anomaly data (October). The same observation is made in terms of the temporal shift for the second peak (i.e. March–July), where the maximum of the SWS anomaly occurs in April, 1 month later for the normalized discharge anomaly, and 1 month before for the normalized precipitation anomaly respectively in May and March.

Figure 9 displays the comparison at the basin level between the aggregated normalized SWS anomaly and TWSA from GRACE. Both variables show a similar interannual variability during the common period of availability of data (i.e. 2002 to 2015), presenting a fair correlation of r=0.84 (lag = 1; p value <0.01; Fig. 9a). It is also worth mentioning that both datasets capture the bimodal patterns. Figure 9b presents the deseasonalized normalized anomaly for the two variables (r=0.4; lag = 0; p value <0.01), showing quite similar variations, especially in the long-term variability. We also notice the higher magnitude of the normalized SWS anomaly as compared to the normalized TWSA. At the seasonal timescale, Fig. 9c reveals a similar behaviour, with the two peaks depicted in the two variables, one in November–December and one in April. The lowest level of the SWS occurs in July, which is 1 month ahead of the TWSA minimum. Figure 10 presents the same comparison as done in Fig. 8 but at the sub-basin level (considering the Ubangui, Sangha, Middle Congo, Kasaï, and Lualaba sub-basins; see Fig. 1). The seasonal variations of all the sub-basins are provided in Fig. 10 (right column). The outlet of the Kasaï and Lualaba sub-basins provides historical observations (i.e. data before the 1990s), and thus the comparison with its in situ normalized discharge anomaly time series was integrated only at the annual cycle. For the Ubangui, Sangha, and Lualaba sub-basins, the maximum linear correlation coefficient is not significant between normalized SWS anomaly and normalized precipitation anomaly variations at the interannual level and their associated anomaly (Fig. 10a, b, d, e, m, and n, Table 4). This could be associated with the bimodal dynamics observed in the precipitation data, whereas the SWS variations do not show that behaviour. Becker et al. (2018), using similar datasets (GIEMS-1 for SWE and VS from ENVISAT), reported the same observation between precipitation data with bimodal patterns and SWE with unimodal patterns (Fig. 4 of Becker et al., 2018). Another reason could be that, for Ubangui and Sangha, SWS is mainly a function of discharge variations, while for the Lualaba sub-basin, which encompasses many lakes and floodplains, the various processes and their link for instance with evaporation lead to an insignificant precipitation–SWS correlation. Additionally, one can observe a negative lag between the normalized SWS anomaly and the normalized precipitation anomaly for the Ubangui and Lualaba sub-basins, which is not physically acceptable, since we expect a positive temporal shift between SWS and precipitation. This is probably due to the fact that SWS shows a unimodal pattern, while precipitation shows a bimodal pattern. Nevertheless, the comparison between the normalized SWS anomaly vs. the normalized discharge anomaly for Ubangui and Sangha, except for the Lualaba sub-basin, shows a fair correlation coefficient (r>0.7; p value <0.01) at the interannual timescale (Fig. 10a and d). Their related deseasonalized normalized anomalies (Fig. 10b and e) present lower values of correlation coefficients (r<0.4; p value <0.01). Regarding the Middle Congo and Kasaï sub-basins (Fig. 10g and j), the maximum linear correlation coefficients are r=0.32 and r=0.69 (lag = 0; p value <0.01) respectively between the normalized SWS anomaly and the normalized precipitation anomaly. Their associated deseasonalized normalized anomaly (Fig. 10h and k) does not show a significant correlation coefficient. In contrast to this, for the Middle Congo basin, the comparison between SWS and discharge provides a moderate correlation coefficient for both the interannual and deseasonalized anomaly variations (r>0.5 with a delay lag of 1 month; p value <0.01). In accordance with other results (see Fig. 10h of Kitambo et al., 2022a), the Middle Congo basin appears to be the main sub-basin for which the variability of the normalized discharge anomaly at the outlet Brazzaville/Kinshasa station is fairly related (∼35 %) to the variations of the normalized SWS anomaly in the Cuvette Centrale region due to its significant correlation coefficient of r=0.58 between the deseasonalized normalized anomaly of both SWS and discharge. The northern and Middle Congo sub-basins reach their SWS anomaly maximum in November (for Sangha and the Middle Congo, Fig. 9f and i) and October (for Ubangui, Fig. 10c), and this is in phase with the maximum of the normalized discharge anomaly and a backward temporal shift of 1 month with the normalized precipitation anomaly. Compared to northern sub-basins, southern sub-basins (Kasaï and Lualaba) for the period January to May reach their SWS anomaly maximum in March (Fig. 10l and o), which is in phase with the occurrence of the maximum of the normalized precipitation anomaly. The maximum of the normalized discharge anomaly occurs 2 months later in May for Lualaba and 1 month later for Kasaï. In contrast to the other sub-basins, Kasaï and the Middle Congo have depicted the bimodal patterns in SWS anomaly variations. For the Middle Congo, the first peak is reached in November and the second in May, while for Kasaï, the first peak occurs in December, and the second peak with a steady evolution occurs in March and May. Similar results were observed in the Cuvette Centrale by Frappart et al. (2021).

The temporal patterns in Figs. 8 and 10 follow alternatively wet and dry events associated with large-scale climatic phenomena for all the datasets (SWS, precipitation, and discharge). A focus on the Lualaba and Kasaï deseasonalized normalized anomalies of SWS reveals that there are two main sub-basins significantly impacted (large positive anomaly in Fig. 10k and n) by the major flood event triggered by the positive Indian Ocean Dipole in combination with the El Niño event that characterized the period 1997–1998. Conversely, recent studies using hydrometeorological datasets have shown that some parts of the CRB are subject to a long-term drying trend over the past decades (Hua et al., 2016; Ndehedehe et al., 2018). The droughts that affected large areas of the CRB in recent years are amongst the most severe ones in the past decades (Ndehedehe and Agutu, 2022), including the large anomalous event of the 2005–2006 drought (Fig. 8). This is further investigated below.