Surface water dynamics of Lake Chad Basin (Sahelian Africa) based on daily temporal resolution Earth observation time series

BGR

Federal Institute for Geosciences and Natural Resources

DOY

Day of year

EO

Earth observation

FAO

Food and Agriculture Organization of the United Nations

GIZ

The German Agency for International Cooperation (Deutsche Gesellschaft für Internationale Zusammenarbeit)

GWP

Global WaterPack

GWSP IPO

Global Water System Project International Project Office

HVIP

Hadejia Valley Irrigation Project

JRC

Joint Research Centre

KRIP

Kano River Irrigation Project

LCB

Lake Chad Basin

LCBC

Lake Chad Basin Commission

MK/MMK

Mann–Kendall/modified Mann–Kendall

UNESCO

United Nations Educational, Scientific and Cultural Organization

The Southern pool is further segmented into zones of open water and archipelagoes. While the Northern pool contains mostly seasonal water, the Southern pool remains quasi-stable but fluctuates between wet and dry episodes. Despite recent increases in precipitation over the basin, the lake does not completely recover due to groundwater level decline, population pressure, and reduced discharges (Geerken et al. 2012). Whilst freshwater resources remain vital for sustenance in semi-arid and arid areas, water resources management is already hard hit by climate change. Therefore, global monitoring and knowledge of spatial and temporal variability of surface water resources are of high importance (Klein et al. 2021b).

Despite the plethora of studies in LCB, scientific knowledge on surface water duration on a daily time scale is still not documented. Many studies on climate variability and hydrological shifts have been conducted (Wald 1990; De Zborowski & Lemoalle 1996; Birkett 2000; Coe & Birkett 2004; Gao et al. 2011; Candela et al. 2014; Okonkwo et al. 2014; Zhu et al. 2017; Buma et al. 2018; Pattnayak et al. 2019; Pham-Duc et al. 2020; Mahamat Nour et al. 2021; Fu et al. 2023; Gbetkom et al. 2023; Lemenkova 2023). While most studies have focused on Lake Chad (Zhao et al. 2023), fewer studies have attempted to understand dynamics in other open water bodies within the basin, as demonstrated by (Djimadoumngar & Adegoke 2018; Poulin et al. 2019; Abba Umar et al. 2021; Christine et al. 2021). Most previous studies (Coe & Birkett 2004; Lemoalle 2004; Gao et al. 2011; Lemoalle & Magrin 2014; Okonkwo et al. 2014; Buma et al. 2018; Pham-Duc et al. 2020; Gbetkom et al. 2023) employ satellite altimetry in the study of lake water dynamics. How dense time series derived from optical remote sensing could help reconstruct surface water dynamics is not well documented. Also, not only the seasonal patterns of open water resources are not fully understood, but scientific details on their daily dynamics are completely missing in the research landscape. The extent to which daily and seasonal surface water varies with precipitation is also not fully understood. According to the Food and Agricultural Organization (FAO), one main problem of the LCB is the variability of the hydrological regime and the dramatic decrease in freshwater availability (FAO 2009).

The advancements in remote sensing in the past decades have enabled large-scale and long-term analyses to reconstruct land surface dynamics amongst others. As such, powerful methods and tools have been developed for analyzing Earth observation (EO) data (Klein et al. 2017). Several studies (Klein et al. 2014, 2015; Mueller et al. 2016; Pekel et al. 2016; Zhou et al. 2017; Huang et al. 2018; Li et al. 2020; Ling et al. 2020; Pickens et al. 2020; Bonnema et al. 2022; Liu et al. 2022; Yang et al. 2022) have demonstrated such space-based estimation of open surface water dynamics using dense EO data. We employed a temporally dense, daily cloud and gap-free global dataset on water bodies' dynamics, the Global WaterPack (GWP) (Klein et al. 2024, 2021a, 2017; Uereyen et al. 2022) for reconstructing the daily and seasonal patterns of open water bodies in the most active part of the LCB. The GWP is derived from daily remote sensing image time series at a spatial resolution of 250 m using a dynamic thresholding methodology that captures rapid changes in surface water extent at a high temporal frequency (Klein et al. 2024).

The objectives of this study are: (1) to analyze the daily and seasonal dynamics of open surface water in the LCB, (2) to estimate change in open surface water area extent, including trends and change points, and (3) to assess the relationship between surface water extent and seasonal rainfall variation in the basin. Our study represents a significant contribution to the hydrological balance of the lake system as it focuses on the daily and detailed seasonal surface water dynamics. Unlike previous studies that analyze long-term trends, this study provides a granular view of how the water bodies' surface extent fluctuates daily and seasonally derived based on medium spatial resolution remote sensing data. This work not only improves our knowledge of the basin's hydrology but also provides vital information for real-time resource management and landscape planning in the context of climate change.

Lake Chad is an endorheic lake in Sahelian Africa, with an extension to hyper-arid areas of the Tibesti Mountains. The most vegetative part of the LCB constitutes the southern portion (Figure 1(a)). Open surface water bodies and groundwater offer better opportunities for rain-fed agriculture for over 60% of the population (Vaquero et al. 2021) in LCB. Both surface and groundwater variations in the LCB are not fully understood. The lake is sourced by two main contributing areas, the Chari-Logone and the Komadugu-Yobe headwaters, direct surface rainfall, and groundwater flows. According to Bader et al. (2011) and GIZ (2018), the Komadugu-Yobe Basin contributes 1.8–2% of the flow to Lake Chad.

The Chari River rising from the southern ridge contributes more than 80–95% (GIZ 2018)) of the lake's water while surface rainfall contributes 15% (Olivry 1996), mostly as runoff from the southwestern section. The lake's water cycle is rhythmed by monsoon with minimum levels around April–May and maximum levels around December–January (Bader et al. 2011), thus exhibiting a significant lag between rainfall and lake surface water area. Vaquero et al. (2021) established that most rainfall occurs between April and October, while potential evapotranspiration (PET) values of 2,000–3,000 mm/year are common. Precipitation rates vary from an average of 1,400 mm/year in the south to 10 mm/year in the north (Global Water Partnership 2013). With a mostly flat topography, runoff is only produced along the basin's southern periphery, making groundwater a vital component in the preservation of Lake Chad (Geerken et al. 2012; GWSP IPO 2012).

Surface flow in the LCB is mostly ephemeral, with the exception of rivers Komadugu-Yobe and Ebeji (El-Beïd) (Figure 1(a)) in the southwestern section (Goni et al. 2021), and rivers Logone and Chari from the southeastern section that drains directly into Lake Chad (Goni et al. 2021). Goni et al. (2021) found this to be linked to the basin's geological setting; deeply weathered crystalline rocks. The underlying geological setting of the lake's basin, mostly a succession of marine and continental sediments deposited on the Precambrian basement of the Mesozoic (Cretaceous), Cenozoic and Quaternary (Vaquero et al. 2021), makes it an excellent reservoir for groundwater. BGR (n.d.) reported that 11,395 boreholes tap groundwater from the Quaternary aquifer of the basin.

In this study, we employed the GWP (Klein et al. 2017, 2024), a daily time series of open surface water dynamics for the period 2003–2022. The GWP dataset is produced by using a pixel-based water classification approach based on dynamical thresholding and temporal interpolation (Klein et al. 2021b), consisting of daily open water/no water layers which were generated by the combination of one-class classifications derived from the latest Moderate Resolution Imaging Spectroradiometer (MODIS) reflectance products of the Terra (MODGQ09) and Aqua (MYDGQ09) satellites of the same day (Klein et al. 2021a, 2021b). Details on data quality and accuracy can be found in Klein et al. (2017, 2021b, 2024).

For this research, we selected open water bodies above 10 km² area using HydroLAKES shapefile (Messager et al. 2016). These include Lake Chad, Lake Fitri, Lake Iro, Maga Dam, Alau Dam, Tiga Dam, Challawa Dam, Wattari Dam, Kano River Irrigation Project (KRIP), and Hadejia Valley Irrigation Project (HVIP) (Figure 1(a)). We prepared a 10 km buffer (here seen as the approximate maximum extent of surface water during flood events) for each water body's shapefile and used it to extract the zonal statistics and construct the daily time series. To reduce outliers and get a smoothened time series of GWP, we decomposed the time series into trends, seasonality and residuals/noise components using an additive model. At a later stage, we used the Savitzky-Golay filter (Savitzky & Golay 1964), window length = 50 and polyorder = 3 to smoothen the residuals, before recomposing (trend, seasonality and cleaned residuals) to get a cleaner time series from 2003 to 2022 (Figure 1(b)).

Further, we acquired and processed ERA5-Land reanalysis dataset (Muñoz-Sabater et al. 2021) via Google Earth Engine (Gorelick et al. 2017), for daily precipitation (rainfall) time series (2003–2022) the same way as GWP (Figure 1(b)). Spatiotemporal maps of open surface water duration and extent for all water bodies with an area above 100 km² were produced.

When β_i is positive, it implies a positive trend, and when β_i is negative, the trend is negative. Because the original MK test is sensitive to serial auto-correlation typical of hydrological and meteorological time series, MMK tests are highly recommended. To ensure a high certainty of our results, we employed the original Mann–Kendall (Mann 1945; Kendall 1948) and MMK (Hamed & Ramachandra Rao 1998; Yue & Wang 2004). The MMK and Theil–Sen Slope estimator were employed using the pyMannKendall python package (Hussain & Mahmud 2019).

For significant statistics, the change point in the series occurs at K_T (Parajuli et al. 2021). The Pettitt test (Pettitt 1979) was implemented in Python using the pyHomogeneity package (Hussain et al. 2023).

Pearson's correlation analysis was employed to estimate how surface water responds to seasonal precipitation variation over the basin (Figure 1(b)). Mean monthly catchment precipitation and surface water area for each water body were correlated for six temporal lags at 0.05 and 0.01 significance levels.

We also calculated monthly and yearly rainfall anomalies for the entire active basin. The formula of Goswami et al. (2023) is used for the calculation and classification of excess and deficit years with respect to precipitation and percent deviation of the mean (climatology). A year is classified as an excess year when the percent deviation of the mean is >10%, and as a deficit year when it is less than −10%; otherwise, it is considered a normal year (Goswami et al. 2023). We further implemented the Köppen–Geiger (Köppen 1900) climatic zonation (Beck et al. 2023) for the period between 1991 and 2020 over LCB to spatially relate precipitation seasonality with surface water variations.

A common limitation of the methods and dataset lies in the use of gridded precipitation dataset instead of in situ data for analyzing rainfall events. However, the choice of ERA5 reanalysis is quite suitable for the study as this has proven high performance in detecting rain events over the Sahel and West Africa (Gbode et al. 2023; Saley & Salack 2023). Despite its strengths, ERA5 has also shown poor performance in some parts of West Africa (Quagraine et al. 2020; Goudiaby et al. 2024). Despite the robustness of GWP, we noted some underestimations of daily water extent in Lake Fitri partly linked to spectral reflectivity of clay dominated muddy water and the shallow water levels in some seasons. Also, between 2019 and 2021, we found possible underestimations (artifacts due to clouds and missing data) of Lake Iro by GWP.

The wide remarkable difference between surface water extent in the Northern pool as opposed to the Southern pool could be explained by three aspects. First, the Northern pool is sourced by the Komadugu-Yobe River which has a low annual discharge linked to its long distance flow across the Sahelian and desertic part of the basin and increased dam construction upstream. This river system only contributes about ∼2% of the total annual discharge to Lake Chad. Secondly, the Northern pool is exposed to hyper-arid climatic conditions with relatively high evaporation rates. Recent satellite images also show that the Northern pool is more encroached by desert sand than the Southern pool, which has resulted to some sort of infilling of the pool (Supplementary material, Figure S2). Lastly, the development of the Great Barrier between the two pools has hindered the free circulation of water that formerly existed within the lake system. This barrier is only bridged when the Southern pool is flooded.

Figure 4 shows the highly variable nature of the Northern pool. For some years, e.g. 2012, the Northern pool only contains water for about the first 30 days of the year (i.e. in January), but in the last three years (2020, 2021 and 2022), the Northern pool is slowly recovering. It should be borne in mind that the water duration in the Northern pool is also highly controlled by surface water volume and areal extent of the Southern pool. Though the Northern pool is directly sourced by the Kamadugu-Yobe River system which contributes ∼2% surface flow into Lake Chad, as well as direct rainfall and groundwater, the Northern pool receives a significant amount of water via the Great Barrier following flooding of the Southern pool around December-January, especially in wet years. This partly explains why seasonal water (∼1–150 days) concentrates between the Great Barrier and the Northern pool (Figures 3 and 4). In the Southern pool, the shallow depths of the archipelagoes also harbour water that stays only for about half a year, though the recent years show almost a complete recovery of the Southern pool (Figures 3 and 4). In contrast to the Northern pool, the Southern pool is not subjected to continuous desert sand intrusion (Supplementary material, Figure S2).

In Lake Fitri, we observed that the maximum water area extent is reached between the 1st and 59th day and between the 305th and 365/6th day of the year (Figure 5).

Though these maxima are experienced, our analysis reveals that the areal coverage is quite negligible (Figure 5), as such Lake Fitri exists in a larger part of the year (∼60–304 days) as a shallow water body. Surface water area between the 1st and ∼150th day around the lake also corresponds to flood recession as a result of flash floods from river Bartha and early torrential rains in May–June in this Sahelian section of the basin. Thus, the maximum surface water area reached in the later days of the year (∼305th–365/6th) partly reflects the time lag between precipitation, stream flow and lake water levels. With the exception of 2012, 2016, 2018, 2019, and 2022, fewer years in the monitoring period showed a duration of water that reached 365 or 366 days in Lake Fitri (Figure 6). This seasonal pattern in the cycle of Lake Fitri is also well explained by the seasonal nature of its source river (Bartha), as well as the seasonal progression of rainfall from the south to the northern arid zone of the basin. The extension of its contributing basin to hyper-arid areas also accounts for huge water loss via evaporation. However, we noted some underestimations of GWP over the surface water extent of this lake due to its shallowness and muddy nature in some seasons of the year. For example, between April and May, field reports reveal that the lake depth reduced to about less than 2 m and became increasingly muddy. This poses a problem to ambiguous spectral reflectance between clay soil and shallow water with high sediment content or shallowness that the reflectance of lake's ground is dominantly registered by the optical sensors used in GWP computations.

In contrast, seasonal patterns of Lake Iro (Figure 6) in the south sub-humid zone of the basin are less pronounced (between 2003 and 2016), with surface water lasting for 365/366 days/year. From 2017 to 2021, the lake illustrates very pronounced seasonal patterns, shrinking and reduction in surface water duration and areal coverage.

Between 2018 and 2021, much of the lake's water showed to only stay for less than half a year, ∼170–290 days (Figures 2 and 6), but recovered in 2022.

Lake Chad only starts regaining its actual size after the lag time between precipitation and effective stream discharge from September through December (either from 245 to 365/366 days). Under semi-humid conditions, Lake Iro does not reflect marked seasonal fluctuations, though between July and August, the areal coverage of surface water is low (Figure 9).

Despite the location of the reserviors/dams in the basin, they all show an increasing water area around 180 days (around July) each year. This period corresponds to the onset of the wet season. They also each contain high water levels until the last quarter of the year, corresponding to the dry season. Here, water is used for off-season irrigation (Figure 1(a)). Though the Tiga Dam is located in a sub-humid zone with relatively high rainfall as opposed to the Alau and Maga dams in the Sahelian zones, they all illustrate similar seasonal patterns (Figure 10).

Inter-annual variations are also registered within the investigated period. Mean annual water coverage in Lake Chad varied from 2,952.85 to 3,114.43 km² between 2004 and 2021, respectively (Supplementary material, Table S1). Meanwhile, between 2005 and 2012 and between 2016 and 2019, the surface water area is below 2,500 km². This corresponds to dry years as opposed to 2004, 2013, 2014, 2015, 2020, and 2021 which are correspondingly wetter, with mean annual surface water coverage above 2,800 km² (Supplementary material, Table S1). However, this inter-annual trend is not identified in all Lake Chad pools. In the Southern pool, the mean annual surface water area varied from 2,030.44 km² in 2004 to 2,051.49 km² in 2021. Between 2005 and 2017, with the exception of 2010 and 2012, the Southern pool is limited at ≤1970 km², while in 2010 and 2012, the pool is between 1,600 and 1,723 km² (Table S1). The Northern pool demonstrates recovery and dwindling phases. Between 2005 and 2012 and between 2016 and 2019, the Northern pool does not contain water above 600 km², with just ≤350 km² of surface water area for most of the year (Figure 3 and Supplementary material, Table S1). However, it recovered in 2004, 2013, 2014, 2015, 2020, and 2021, though mean annual water coverage is not above 1,060 km². This period of recovery also corresponds to the same period in which Lake Chad recovers from dry periods as rainfall increases in the Sahel.

Our results revealed that seasonal surface water is increasing in Lake Chad in all seasons (Supplementary material, Table S2) and annually (Table 1), either 0.0782 km²/daily or 28.54 km²/year. However, the trend analysis shows an increasing trend for Lake Chad, with exceptions for some seasons, i.e. January and February for the Northern pool and April for the Southern pool, where surface water showed no significant trends (Supplementary material, Table S2). However, both pools are increasing almost at similar trends (Table 1). Similarly, with the exception of July and September, all seasons showed significant increasing trends in surface water area in Lake Fitri over the study time scale. Meanwhile, the annual increase rate is 1.095 km²/year. In contrast, Lake Iro shows a declining trend (0.7665 km²/year) across seasons and years (Table 1). The slight recovery of these Sahelian lakes is also partly explained by the recent increase in rainfall reported across the Sahel.

Table 1

Estimated trends and change magnitude of open water bodies

^aIn bold, significant at 0.05 level, light blue (increasing trend), and light orange (decreasing trend).

View Large

For reservoirs/dams, the trend is less clear than in natural lakes (Supplementary material, Table S2 and Table 1). Surface water duration in Maga Dam shows a continuous decline across all seasons and years, with an annual decrease rate of −1.168 km²/year. This is going to have far-reaching implications on irrigated rice and off-season crop cultivation in the dry northern regions of Cameroon. The causes for such decline are unclear. In this same vein, the Challawa and Wattari dams in Northern Nigeria also showed a declining trend, though most of the seasons do not show any significant change trends in their surface water extent, but the annual rate of decrease is quite minimal (Table 1). The Tiga Dam showed significant increasing trends with seasons and years. Alau Dam also demonstrated seasonal and annual increase trends, but the rate of increase is quite minimal (Table 1). Though the Challawa Dam, KRIP and HVIP upstream of Lake Chad do not show significant seasonal trends, some seasons in Challawa Dam and HVIP showed quite significantly increasing trends in surface water duration and extent (Supplementary material, Table S2). In HVIP, surface water duration in August, October, and November showed significant increasing trends (Supplementary material, Table S2). The same scenario is observed in Challawa Dam for November, and KRIP in December (Supplementary material, Table S2).

Within the cycle of Lake Chad, a decreasing trend is observed between 2003 and 2012, with an average surface water area extent of 2,274.82 km². From 2012 to 2022, an increasing trend is noticed with an average surface water extent of 2,744.91 km². A significant change point in the time series is observed on 2012-10-14.

In the Southern pool, the change point is observed in 2013-10-16 with a shift in the surface water area from 1,858.22 to 2,028.40 km² after the change point. In contrast, in the Northern pool, the change point occurred on 2012-08-14. Here, the surface water area shifted from 408.04 to 721.61 km² after the change point. In Lake Fitri, a decreasing trend is observed during 2003–2012, with the average surface water area being 123.84 km². From 2012-03-20, a significant change point is observed until 2022 with an increasing trend and average surface water extent of 144.35 km². In comparison, Lake Iro depicts a decreasing trend throughout the investigated period. It significantly shifted from 105.56 km² between 2003 and 2013 to 91.48 km² between 2013 and 2022. In Lake Iro, a significant change point with an increasingly declining shift was registered on 2013-02-27 (Figure 11). In reservoirs/dams, significant change points were also observed.

Other lakes of natural origin showed significant but less than 50% correlation at the 4th lag. This is partly explained by the influence of desertic and Sahelian climatic conditions of high temperatures and evaporation which results in a loss of surface water from precipitation in rivers before it arrives in the lakes (Figure 13). Both rivers Bartha and Bahr Azoum (Figure 13) which feed the lakes Fitri and Iro, respectively, are subject to great seasonal contrast. Lake Fitri is highly exposed to hyper-arid conditions, thus explaining its marked seasonality (Figures 5 and 13).

On the contrary, reservoirs/dams showed strong and significant correlations at lags two and three. This implies that precipitation could contribute to the maximum surface water area in most reservoirs/dams in LCB with a temporal lag of 2–3 months (Figure 13).

Open surface water variation and anomalies also align with periods of climatic extremes in LCB. Figure 14 reveals that years of excess and deficit correspond closely to similar signals of monthly surface water extent and duration following the corresponding lag time. Across the monitored period, four periods of wet and three periods of dry conditions were identified (Figure 14). Following the identified time lag between rainfall and surface water extent in open water bodies, the impact of wet and dry years is quite visible across the lakes in the LCB (Figure 13).

However, in every case, reservoirs/dams irrespective of their climatic location (Figure 13) do not portray the same signals as natural lakes (Figure 14). For example, the excess or wet year of 2005 is not reflected in the Alau and Tiga dams. Likewise, the deficit years of 2009 and 2014 are not reflected in the Alau and Maga dams, and the Tiga, Alau and Maga dams, respectively (Figure 14). It is also important to note that the droughts or rainfall deficits observed in 2021 did not affect Lake Chad, Lake Fitri, and the Tiga Dam. This may be partly explained by the role of groundwater concentration on the water balance of Sahelian lakes as well as less human influence within that year. Though Tiga Dam recovered, this anomaly is not as strong as in the wet years. Thus, this might be explained by the combined influence of human controls on dam water dynamics and the relatively wetter climatic location of the dam (Figure 13).

The Lake Chad, Lake Fitri, and Lake Iro are characterized by different seasonal cycles reaching their maximum extents in December and January. This has been reported in previous studies. To this inter-annual variability, we must add a strong seasonality, characterized by lake flooding which occurs between December and January (Southern pool) and between February and March (Northern pool) and by low water in August (Magrin et al. 2015). The daily and seasonal dynamics observed in Lake Chad's cycle are also explained by the variability of rainfall and streamflow into the lake system. The Chari River contributes 95% of the surface flow to Lake Chad and explains much of its seasonal cycle. Magrin et al. (2015) reported that the Chari River regime is a tropical type with an annual flood marked by a peak toward the beginning of November and a pronounced low flow in March–April. This is also reflected in the annual cycle of Lake Chad, with high waters observed from November to February (Figures 3, 4, and 9).

River Kamadugu-Yobe, a second contributor to Lake Chad's waters was estimated to have brought approximately 0.5 km² of water per year to the lake for the period 1990–2000, i.e. 1.8% of Chari's contributions (Magrin et al. 2015). The construction of two dams (Tiga in 1974 and Challawa in 1992) upstream of the Yobe River resulted in a modification of the flow to Lake Chad. Magrin et al. (2015) recently reported that the reduced flooding of the Hadejia-Nguru floodplain, which leads to less water loss through evaporation, partially compensates for the water withdrawals from the dams. The El-Beïd River contributes to the rest of Lake Chad's surface flows. Its streamflow reaches its maximum between November and December (Bdliya & Bloxom 2007.; Magrin et al. 2015), but the flow rates vary from one year to the next depending on the Logone floods (Magrin et al. 2015). The highest water levels observed in Lake Chad between December and January (Figures 3, 4, and 9) is a combined response of the lag time resulting from the discharge of the aforementioned river systems.

Inter-annual surface water extent also varied considerably across the basin. We observed that the mean annual water area for Lake Chad between 2004 and 2021 stood at 2,475.64 km². However, significant differences exist between area estimates of water extent from various optical sensors. In comparison with Landsat-derived water extent datasets, we found a difference in annual area estimates ranging from ≤249 to ≥901.6 km² (Supplementary material, Figure S5) between the JRC water product and MODIS-based GWP of this study. Nonetheless, similar spatiotemporal seasonal patterns of surface water variations are observed by both optical sensors Supplementary material, Figures S3 and S4. More so, during the drought years of 2005–2012 and 2016–2019, both products effectively estimated lower water coverage over the lake, both below 2,500 km² (Supplementary material, Figure S5) as observed in this study. Despite this, some Landsat-based studies (e.g. Ekpetere et al. 2023) rather underestimate surface water extent in Lake Chad as compared to Pekel et al. (2016), Buma et al. (2018) and Pickens et al. (2020). It should also be noted that surface water extent derived from Landsat sensors over the LCB was flawed during the period between 31 May 2003 and 2012 following the failure of Landsat 7 Scan Line Corrector (SLC). This resulted in a significant loss of valid pixels leading to huge data gaps and, thus, affecting the estimated water area extent during these periods. The aforementioned gaps are well observed over some Landsat surface water products (e.g. Pekel et al. 2016; Pickens et al. 2020) over LCB. Also, Herndon et al. (2020) observed that the JRC surface water dataset (Pekel et al. 2016) greatly underestimated the surface water extent in the Nigerian Sahel landscape. Mindful of sensor differences in surface water extent area estimates, our study revealed an increasing trend in Lake Chad water extent, which is in line with a previous study (Buma et al. 2018) that used Landsat for seasonal and annual surface water area extent estimates. Though surface water area extent may vary when estimated from both optical sensors, we found that seasonal and annual surface water extent in Lake Chad aligns with Global Landsat-based products (Pekel et al. 2016; Pickens et al. 2020) (Supplementary material, Figures S3 and S4). When compared with global Landsat-based surface water extent datasets of Pekel et al. (2016) and Pickens et al. (2020), MODIS-based surface water extent estimation is shown to be more robust to capture intermittent surface water, though smaller waterbodies may be underestimated due to the coarse resolution of the sensor (Liu et al. 2022). We also found more or less similar estimates of open water areas in Lake Fitri with findings by Talikun et al. (2019) using Landsat.

Similarly, our findings revealed interesting daily and seasonal dynamics of surface water extent in Lakes Fitri and Iro. As commonplace with optical sensors, we, however, noted some underestimations of GWP over Lake Fitri as compared with ground realities. This is partly linked to its shallow depth (at times < 2 m) and concentration of muddy water in some parts of the year. This poses a serious challenge to spectral reflectance registered by channels of optical sensors. Such underestimations from optical sensors were common between May and July (days 121–212), as observed also by Talikun et al. (2019). Also, possible artifacts were detected over Lake Iro between 2019 and 2021, though we observed similar patterns within the aforementioned period by other optical sensors products, e.g. the Landsat surface water product by Pickens et al. (2020). Consequently, our estimated trends and temporal correlations for these lakes may be flawed, but not incorrect as Talikun et al. (2019) also found that Lake Fitri doubled in size between 1972 and 2015.

Surface water in reservoirs depicted increasing trends extending beyond the season as opposed to lakes in the basin. Their seasonal pattern was explained more by human manipulation of infilling and water withdrawals (Figures 10, 12 and 14). It is largely reported that upstream dam construction has profound effects on downstream river systems and flow modifications. The Tiga and Challawa dams each control 80% of the flows into the Hadejia-Nguru wetlands (Thomas 2008) of the Kamadugu-Yobe Basin. The average annual flow in the upstream part of the Hadejia River has been reduced by 33% after the completion of the Tiga Dam (1979–1989) (Chiroma et al. 2005; Sanni 2018). In the Chari-Logone Basin, Zilefac (2010) reported a 70% reduction of water supply by the Maga Dam to the floodplain from the Mandara Mountains. Between 1960 and 2000, Olowoyeye & Kanwar (2023) reported that the reduction in surface water of Lake Chad was entirely linked to climate change, population pressure, and water abstraction for irrigation.

Besides flood control, the Tiga Dam stores water for the KRIP, and the HVIP, while the Challawa Dam provides water for domestic use in Kano and supplement water deliveries for the HVIP. In Borno State of Nigeria, the Alau Dam was constructed between 1986 and 1989 to supply domestic and industrial water to Maiduguri town. Since the completion of the Alau Dam, the natural inflow of the Ngadda River (see Figure 1(a)) into Jere Bowl no longer takes place (Babagana et al. 2015), thus reducing the usual inflow to Lake Chad.

However, it is also important to note that the reduction of surface flow for both perennial and seasonal rivers in the LCB is also compounded by climate and environmental changes. This is confirmed by a recent study by LCBC (2016), which affirms that an increase in the lake area could be a result of no major ongoing irrigation scheme around the Northern pool. This is also reported in Nilsson et al. (2016) and Buma et al. (2018). This partly confirms the role of surface water abstraction on the lake's water balance.

This study found a 4- to 5-month lag of Lake Chad surface water extent to regional rainfall of the contributing basin (Figures 13 and 14). This is in line with the findings of Gbetkom et al. (2023) who found a time lag of approximately 112 days between rainfall and Lake Chad water variations. Similarly, Pham-Duc et al. (2020) reported a 1-month lag between the peak of the Chari-Logone rivers and the peak of Lake Chad surface water. Zhu et al. (2017) reported a time lag of 1–2 months response of streamflow to precipitation in the LCB. As a result, most of the discharge is observed from July to December with a peak value in October (Zhu et al. 2017), thus explaining the cycle of the natural lakes in the basin (Figures 2–5, 9 and 12). Lemoalle & Magrin (2014) emphasized that despite surface water abstraction in the LCB, the balance of surface water in Lake Chad is largely explained by precipitation changes. Lemoalle & Magrin (2014) observed that a variation of ±10% in rainfall over the basin corresponds to a variation of the same magnitude of ±30% of the annual flow of river Chari, the main tributary of Lake Chad. Since the surface water of the lake is approximately proportional to the contributions of the Chari River, the surface water extent of the lake is particularly sensitive to variations in precipitation (Lemoalle & Magrin 2014; Mahamat Nour et al. 2021; Gbetkom et al. 2023).

The LCB shows a succession of wet (excess) and dry (deficit) years which is also reflected in surface water cycles (Figure 14). Though our study did not find a trend in the ERA5-Land precipitation time series, some studies Okonkwo et al. 2014; Zhu et al. 2017; Buma et al. 2018; Dan'azumi & Ibrahim 2023; Gbetkom et al. 2023) have reported a slight increase in rainfall in recent years over the LCB. The slight increase in lake area is in line with the findings of Buma et al. (2018) which showed improving wet conditions (Figure 14) in the last two decades. The slow recovery rate of Lake Chad has been confirmed by a previous study (Buma et al. 2018) applying remote sensing methods. Our findings showed that Lake Chad had an annual average increase rate of 28.54 km² following two phases, a low phase between 2003 and 2012 (2,275 km²) and a high phase from 2013 to 2022 (2,745 km²) (Figure 11, Supplementary material, Table S1), also confirmed by Buma et al. (2018). However, these changing phases of Lake Chad's surface water extent have been reported since the 1920s (Tilho 1928). For 60 years, Lake Chad has oscillated between three stages (small, average, and large Lake Chad), covering a surface water area of around 2,000 to 2,500 km² (Magrin et al. 2015). The current assessment period shows that Lake Chad exists as ‘small Lake Chad’, with varying levels of the Northern pool.

Besides rainfall controls on the hydrological balance of the LCB, underground water flows also play a vital role. Groundwater recharge also contributes to the lake's waters. Diffuse and focus recharge processes have been reported for the Chari-Logone Basin and its wetlands (e.g. Vassolo et al. 2016; Bouchez et al. 2019). Meanwhile, this was found impossible in the Komadugu-Yobe Basin as a result of clay layers beneath the river Yobe floodplain inhibiting focused recharge (Carter & Alkali 1996). Within the year, the surface water extent in Lake Chad is also partly maintained by groundwater. Vassolo et al. (2023) reported a high groundwater concentration along the Lake Chad shore and the Bahr el Ghazal corridor.

In this study, we reconstructed the daily and seasonal cycles of 12 open waterbodies in the LCB between 2003 and 2022 using the GWP. We also verified the relationship between basin-scale precipitation and surface water extent. Several methods including time series decomposition, trend analysis, and change point detection were used. We summarized the main findings as follows;

• A strong seasonality of the natural lakes was observed compared to man-made reservoirs/dams.

• Open water bodies either exhibit a monotonic or cyclic pattern.

• Several phases of dwindling and recovery were identified for Lake Chad, though for the entire study period the lake is recovering at a rate of 28.54 km²/year.

• In Lake Chad, the maximum water area in all the pools is observed between December and January and the inter-seasonal surface water area varies between ∼1,500 and ∼3,800–4,000 km².

• The Southern pool shows a high water area above 2,400 km² at the start and end of each year with the exception of drought years (2006–2017). For wet years (2004, 2018–2022), water area between days 1 and ∼ 66, and between 301 and 365/366 days range between 2,200 and ∼2,400 km². Except for extremely dry years, the water area between the rest of 67–300 days of the year is between 1,600 and 2,000 km².

• In contrast, the Northern pool's maximum water area ranges between 1,600 and ∼1,700 km². With the exception of 2004, 2012, 2013, 2015, 2020, and 2021, the Northern pool only fluctuates between ≤200 and ∼800 km², staying only for a few days of the year.

• Mean annual water coverage in Lake Chad varied from 2,952.85 to 3,114.43 km² between 2004 and 2021, respectively. Meanwhile, between 2005 and 2012 and 2016 and 2019, the surface water area is below 2,500 km².

• While the Southern pool remains quasi-stable, the Northern pool shows recovery and dwindling phases, synonym to the historic ‘small Lake Chad’.

• Lake Fitri also depicted marked seasonality but showed a much slower recovery of 1.095 km²/year.

• In Lake Fitri, surface water showed to stay for a short part of the year, with a peak between 1st and 59th day and between 305th and 365/366th day of the year, while between the 60th and 304th/year (about eight months), the lake exists in a shallow state.

• The recent recovery of Lake Chad has been widely reported, and this is partly linked to recent increases in rainfall over the basin and a reduction in surface water abstraction for irrigation around the Northern portion of the lake.

• The current recovery trend of Lake Chad shows no certainty for stability as this is simply exhibiting the normal but usual cycles of present-day Lake Chad.

• In the sub-humid part of LCB, surface water in Lake Iro stays for 365/366 days/year, but showed a declining trend.

• On the contrary, reservoirs/dams respond to human controls of infilling and withdrawals of water for diverse uses.

• Despite this, a contrast exists between reservoirs/dams. In Northern Nigeria, while the surface water area in Tiga Dam remained quasi-stable, the Challawa Dam reflected extreme hydrological deficit, with a considerable shrinking of dam water area between 2004 and 2006, and between 2014 and 2020.

• In Sahelian Northern Cameroon, the Maga Dam between 2003 and 2010 was quite stable, then strong seasonality set in from 2010 to 2022, and declining area coverage between 2013 and 2019 was observed.

Our study provides additional insights for sustainable water management in LCB in line with FAO's recommendations and targets for the basin. As a response action to the water crisis in the region, FAO (2021) recommended mapping of existing resourceful features around the LCB (Buma et al. 2018), including land and water use at cross-border levels. This will assist in addressing the needs of the population with rising food insecurity in the region. Besides, some further research directions include but are not limited to:

• Estimation of surface water extent and dynamics from hyperspectral and microwave sensors may provide an increased understanding of the basin's hydrological cycle.

• Estimation of surface water quality and seasonal concentration of water quality parameters using hyperspectral remote sensing

• Incorporating remote sensing data and future climate models to predict future water variations in the LCB under different climate scenarios.

• Investigating the link between seasonal surface water variations and recurrent land use and armed conflicts in the basin.

We thank the German Aerospace Center (DLR) – German Remote Sensing Data Center (DFD) for providing the research infrastructure for realizing this study. We appreciate that MODIS data (from which the Global WaterPack is derived) from the National Aeronautics and Space Administration (NASA), ESA WorldCover of European Space Agency (ESA), and the ERA5-Land reanalysis dataset of the European Centre for Medium-Range Weather Forecasts (ECMWF) hosted at climate data store were provided free of charge. We would like to thank Professor Djangrang Man-Na, University of Moundou, Chad for providing in situ field observations of the seasonal situation of Lake Fitri. We gratefully thank the anonymous reviewers for their helpful and constructive critiques.

The research was supported by the Alexander von Humboldt Foundation through the International Climate Protection Fellowship.

All relevant data are available from an online repository or repositories. The datasets used in this paper are open source. GWP can be accessed at https://download.geoservice.dlr.de/GWP/files/, JRC surface water product is accessible at https://developers.google.com/earth-engine/datasets/catalog/JRC_GSW1_4_YearlyHistory and ERA5-Land dataset is accessible at https://developers.google.com/earth-engine/datasets/catalog/ECMWF_ERA5_LAND_DAILY_AGGR.

The authors declare there is no conflict.