A dramatic increase in the frequency and intensity of floods due to the Niger
River in the city of Niamey (Niger) has been observed in the last decade.
Previous studies highlighted the role of the land use changes on the flood
increase since 1970s. In the last decade, observations have raised the issue
of a possible increase in extreme rainfall in the Sahel, which may have
caused the recent and extreme floods in Niamey in 2010, 2012 and 2013. The
study focuses on the 125 000 km
The Tropics is a region of the world where global changes and the associated risks are increasing. The rising anthropic pressure, massive urbanization, land use changes and development of intensive agriculture are among the factors that increase the hydrological impacts of climate variability. In such a changing context, understanding and attributing the observed changes in the continental hydrological cycle is difficult. In many tropical regions, as in most African river basins, the sparseness of the observation network is an additional challenge.
The Niger is the main river running through West Africa. The upper part of the basin, between its source (situated in humid Guinea) and the city of Niamey, crosses a region which has seen major changes since the 1950s: the Sahel. This region has undergone one of the strongest climatic signal ever recorded, with a period of dramatic droughts in the 70s and the 80s. These sustained droughts had drastic consequences on the population, with many famines. The dramatic decrease in rainfall had also some irreversible (or at least long lasting) consequences on the eco-systems and on hydrological processes. Many studies have reported the so-called 'Sahelian paradox': decrease in rainfall but increase in runoff because of changing properties of the soil (soil crusting etc.). The latest rainfall records tend to show that the Sahel region is gradually recovering, since the late 90s and that the annual rainfall amounts are getting back to the level they were before the 70s – at least in the Eastern part of the Sahel (Lebel and Ali, 2009). The recent years have also seen the occurrence of many floods in the Sahel and particularly in Niamey where the Niger River caused dramatic inundations in 2010, 2012 and 2013 (Sighomnou et al., 2013).
Several studies based on the long series of river height data, available for the Niamey Niger station have analysed the changes in the hydrograph and the recent increase of floods (Descroix et al., 2012; Sighomnou et al., 2013). The river flow in Niamey exhibits two peaks. The first peak occurs between August and September and is due to the heavy monsoon rainfall occurring relatively close from Niamey (within a few hundred km). The second and longer lasting peak, is due to the delayed arrival of water drained by the upper river basin. In the past, the two peaks used to be merged into a single one (Amani and Ngueta, 2002; Descroix et al., 2012). But since 1970s, the first peak gradually became distinct from the second, arguably because of the increase in runoff in the Sahelian zone (Albergel, 1987; Amani and Nguetora, 2002; Mahé and Paturel, 2009; Amogu et al., 2010; Descroix et al., 2012). Previous studies have highlighted the role of the land use changes. Land clearing and soil crusting lead to increase the runoff and the endoreic ruptures, which in turn caused an increase of the first peak in Niamey. In the last decade, these hydrological changes have become more dramatic with the occurrence of extreme floods. In 2010, 2012 and 2013 the discharge levels were the highest ever recorded since 1920s. In addition to land-use changes, recent observations have mentioned a possible increase in extreme rainfall in the Sahel (Panthou et al., 2014) which may had an important role in these extreme floods.
To study the inter-annual variability of the Niger floods in Niamey, understand their genesis and their recent increase, the hydrological response to rainfall variability must be analysed. Combining observations and hydrological modelling is a useful strategy to understand the sensitivity and test the impact of various factors or processes on the observed hydrological changes. The sparseness of the operational gauge network in the studied region makes satellite rainfall estimate attractive especially since the quality of these products have improved, and their usability for hydrology has been recognized (Bitew et Gebremichael, 2011; Thiemig et al., 2013; Gosset et al., 2013).
Casse et al. (2015) have already shown that satellite rainfall estimates associated with hydrological modelling could be a proxy to follow and diagnose the first flood in Niamey between 2000 and 2013. Here, the study is extended to a longer period (1983–2013) with the benefit of new data sets – both rainfall and discharge. The first objective is to analyse whether or not a hydrological model forced with a satellite rainfall product can reproduce the observed changes in the Niger Hydrograph in Niamey over the last 30 years. A further objective is to understand the detailed role of rainfall in the recent flood increase in Niamey.
The work is based on a long series of observations (discharge and rainfall)
and the modelling set up introduced in Casse et al. (2015; C2015 here after).
C2015 focused on the recent period (2000–2013) and tested 6 rainfall products
as forcing. The aim was to assess the skill of model and satellite rainfall
combination to simulate a realistic discharge and diagnose the occurrence of
floods. Here, the aim is to analyse the hydrological changes over a longer
period. The only satellite rainfall product with sufficient resolution
(0.5
The study area consists in a 125 000 km
The Niger River basin, modified from Casse et al. (2015). The black arrows represent the stream flow direction in TRIP, for the 49 pixels of the hydrological modelling domain between Ansongo and Niamey.
The daily discharge in Ansongo and Niamey (Fig. 1) station from 1983 to 2013
are provided by the Niger Basin Authority (NBA or ABN in French), which
records the daily discharge in several stations of the Niger River. Ansongo
is characterized by a single flood occurring around December or January. The
delay for the flows between Ansongo and Niamey is around 1 day. Niamey is
characterized by two floods, the “first” due to local monsoon rainfall and
the “second” due to upper basin contribution. According to the NBA, the
threshold of 530 cm, equivalent to a river discharge of 1700 m
Annual rainfall accumulation
The main focus is on the analysis of the long term series based on PERSIANN-CDR (Ashouri et al., 2014). As PERSIANN-CDR is recent, no work has been reported yet on the validation of this product in Sahel. Figure 2 compares some characteristics of PERSIANN-CDR with 4 products that have been tested previously (Gosset al., 2013; Casse et al., 2015): (i) the gauge only product from the Climate Prediction Centre (CPC) (Xie and Arkin, 1997) (ii) the near real time satellite only product PERSIANN(Sorooshian et al., 2000) (iii) the global gauge adjusted satellite product from the Tropical Rain Measurement Mission (TRMM) Multi-platform Algorithm (TMPA) version 7, shortly named 3B42v7 (Huffman et al., 2007) and (iv) the regional gauge adjusted satellite product developed by the CPC, the African Rainfall Estimate (RFE2) (The NOAA Climate Prediction Centre, 2002). PERSIANN-CDR presents the same annual rainfall accumulation as 3B42v7, RFE2 or CPC (Fig. 2a). This product does not overestimate rainfall quantities as PERSIANN near real time product. However PERSIANN-CDR produces the highest number of rainy days. As CPC or RFE2, PERSIANN-CDR tends to produce a lot of rainy days and spatially extended rainfall events with relatively small intensities, leading to smooth rain field.
Mean observed
As in C2015, the hydrological simulation is based on the land surface model
(LSM) ISBA (Interactions between Soil, Biosphere, and Atmosphere) coupled
with the routing scheme TRIP (Total Runoff Integrating Pathways) within the
SURFEX platform ( 3 h time step atmospheric data (exclusive of rainfall) based
on 2002-2012 WATCH Forcing Data methodology applied to ERA-Interim data
(WFDEI, Weedon et al., 2011) reanalysed by MeteoFrance. Sensitivity tests
realised on the 2002–2012 period, show no significant difference in
modelling results between the raw atmospheric forcing and its inter-annual
mean. Based on these results the 2002–2012 inter-annual mean is used to force
the model between 1983 and 2013 for the preliminary runs presented here. 3 h time step rainfall data based on PERSIANN-CDR (Sect. 2.2.2).
PERSIANN-CDR is provided as a daily product. In the Sahel (Mathon et al.,
2002) most of the rainfall is due to organized convection. The associated
rain events at ground level have a typical duration of a few hours, with most
of the rainfall brought by intense rain rates during the short lasting
convective part (Mathon et al., 2002). In order to test the sensitivity of
the simulations, two simple assumptions are tested: (i) the daily rainfall is
uniformly distributed and (ii) the daily rainfall is assumed to have fallen
within a single time step (in the 6th, 15:00 to 18:00 GMT). Note that the
second assumption is more consistent with the convective nature of Sahelian
rainfall. Observed daily discharge at the entrance point of the simulated part
of the Niger basin. As in C2015 only the Ansongo-Niamey reach is simulated
and the model need to be constrained with the observed discharge at Ansongo
(details in C2015).
The hydrological modelling domain (Fig. 1) covers 49 pixels
(0.5
Figure 2a displays the PERSIANN-CDR annual rainfall and number of rainy days over the studied region from 1983 to 2013. The series is consistent with the recent recovery of Eastern Sahel precipitation since the 90s compared to the 70s and 80s (Lebel and Ali, 2009). The PERSIANN-CDR decadal mean increases from 327 mm (1983–1992), to 419 mm (1993–2002) and 455 mm (2003–2013); with a marked jump between the first and the second decade (Fig. 1a). Figure 2b shows the mean annual number of rainy days. It rises between the first (59 days) and the second (71 days) decade, and stays quite steady for the last decade (71 days).
Figure 3a presents the observed changes in the mean decadal hydrograph over
the 31 years studied period, divided in three decades: 1983–1992, 1993–2002
and 2003–2013 (11 years). The mean value of the first peak, changes from
1079 to 1412 m
Observed and simulated flooding periods in day in Niamey between 1983 and 2013. Years in bold font correspond to observed floods. For simulations years in bold font and equal to 0 correspond to non-detection and years in normal font correspond to false-alarm.
In addition to this mean increase in the first and second peak of the hydrograph, another phenomenon has been observed and is reported in Table 1: the first peak has been reaching extreme values and caused many floods in Niamey in the last decade. Table 1 (1st row) shows that only 1 flood occurred during the 1983–1992 decade, 2 during the 1993–2002 decade and 5 during the 2003–2013 period (4 for the 2003–2012 decade). C2015 and Fig. 2a show that the extremes floods are associated with relatively high rainfall amounts in the area.
Figure 3b shows the mean decadal simulated discharge in Niamey for the three
decades: 1983–1992, 1993–2002 and 2003–2013 (11 years). Due to the
restricted domain used for the simulation and because the discharge is
constrained at the entrance of the domain, we concentrate the analysis on the
first peak – the most directly influenced by the rain forcing over the
domain. Both the simulation based on uniform rain within the day and the
simulation with rainfall concentrated in one 3 h step, reproduce the change
in the mean decadal peak. The shape of the peak is well reproduced by the
“3 h rain” simulation and too smooth for the daily uniform forcing. For
the “3 h rain” simulation the mean first peak increases from 1159 to
1652 m
Figure 3b also shows that the simulations reproduce well the observed changes in the timing of the high water season, with an earlier rise of the water, which shift from August to July.
Table 1 reports the number of flooded days per year (for the years where
either the observations or the simulation detected a flood). For the
simulations the flood threshold is defined either as the absolute threshold
defined by NBA 1700 m
The simulations have more difficulties in reproducing the length of the
flooded periods. All simulations overestimate the length of the flood period
for the first 3 flood years of the period: 1994, 1998 and 2003. The “3 h
rain” simulation with the 1700 m
Figure 4 provides 4 illustrative examples of the simulated discharges, for 3 years where floods were observed and 1 non flood year. Despite the relatively coarse and simple model used, and the uncertainties in the observed rainfall, the simulated discharge are globally in good agreement with the observations (the Nash Sutscliffe Efficiency is 0.89 for the 31 year simulation). The performance varies from year to year. In 1988 (Fig. 4a) and in 2012 (Fig. 4c) the simulation underestimates the highest relative maximum, and in 1988 the flood is not detected (Table 1). As expected the underestimation is worse for the simulation based on a uniform daily rain. In 1998 (Fig. 4b) on the contrary, the shape of the sharpest peak is well reproduced (and the value slightly overestimated). Further investigation is needed to understand if these differences are due to the rain forcing or to a misrepresentation of some processes in the model. The uncertainty in the Ansongo discharge used as a constraint or punctual problems such as a dyke rupture may also explain some differences. It is noteworthy that for both 1988 (Fig. 4a) and 1998 (Fig. 4b), the weak discharge variations at the very beginning of the season (June and July) are not reproduced by the model, even though rainfall is present. This could indicate that the Hortonian runoff is not well accounted for in the model. Figure 4d is an example of a non-flooded year where the rainfall and modelling system behaves well.
Mean daily rainfall on the study area (upper plot; solid grey
bars), observed discharge in Ansongo (solid black line) and Niamey (dotted
black line) and simulated discharge with uniform daily rainfall (dashed blue
line) or 3 h rainfall (solid blue line); in 1988
Scatter plot of the anomaly in the simulated peak discharge (Y)
versus the anomaly in the observed peak discharge (X). The last 2 digits of
the year are indicated for each dot (i.e. 12
Figure 5 illustrates the ability of the simulation to reproduce the
inter-annual variability of the first flood. The analysis is based on the
inter-annual series of the observed and simulated maximum discharge peak of
the first flood. Anomalies are computed in order to analyse the relative
rather than absolute variations. For any variable
Over the study area (Sahel) PERSIANN-CDR produces annual
rainfall amount comparable with those from gauge-adjusted satellite rainfall
estimates (3B42v7, RFE2) and gauge products (CPC). As CPC and RFE2,
PERSIANN-CDR smoothes the rain fields: a lot of rainy days with spatially
extended events of weak intensity. Both uniform and non-uniform daily rainfall simulations reproduce the trend
of the observed changes in the mean decadal hyetograph both in intensity and
timing (Sect. 3.2). The simulated discharges are in good global agreement
with the observations (NSE 0.89) but with some uncertainties, and a strong
sensitivity to rainfall distribution within the day. The PERSIANN-CDR based hydrological simulation presents a realistic
inter-annual variability, and detects flooded years, but not the exact
flooded period day by day.
A relatively simple and coarse hydrological simulation, forced by a satellite
rainfall product is able to reproduce the salient features of the changes in
the average hydrograph of the Niger River, in Niamey, as observed over the
last 30 years. This is quite remarkable, given the uncertainties associated
with both the model and the satellite product. It is worth noting that the
land use, soil properties, atmospheric forcing (other than rainfall) and
other model parameters are held unchanged during this 31 year long
simulation; only the rainfall forcing varies from year to year. Therefore the
results show that rainfall variability can be considered as the main driver
for the observed changes in the Niger hydrograph in Niamey.
More work is ongoing to study the changes over a longer period (from 1950) based on rain-gauges record. The sensitivity to the model parameters – and in particular land use and soil property changes- will be also tested. The conditions (on the discharge upstream of Niamey and on the rainfall in the area) that lead to flood, and their evolution over the decades will be analysed. It will be particularly interesting to assess if these conditions have changed between the 50–70s period (i.e before the long lasting droughts that impacted the Sahel in the late 70s and 80s) and the current period. If the rainfall-discharge relationship has changed, the reasons why should be investigated.
We warmly thank the Niger Basin Authority (ABN), for providing the Niger River discharge data and for their collaboration. We are particularly very grateful to Bachir A. Tanimoun (ABN). We would like also to thank all the team that provided the PERSIANN-CDR data sets online. This work has been co-funded by the French Centre National d'Etudes Spatiales (CNES) and the Midi-Pyrénées region.