1. Introduction
Recent years have been characterized worldwide by an increasing number of flood events and flood-related effects [
1,
2]. An alarming case is the high frequency of flooding events in the Middle Niger River Basin (MNRB) during the last 20 years [
3,
4,
5]. Sahelian rivers have shown an increase in discharge since the early 1990s leading to the catastrophic floods registered in the past decade [
6,
7]. Since Tarhule at al. [
8] highlighted that extreme floods represent a crucial issue for Sahelian countries, many researchers claimed that the increasing occurrence of extreme rainfall events and the reduced water-holding capacity of the soil are at the origin of more severe floods [
9,
10,
11]. Flow time series of Sahelian rivers showed a remarkable increase of the annual maximum of discharge that necessarily needs to be considered in the evaluation of hydrological variables over time [
12,
13]. The change in frequency of extreme events was firstly addressed using non-stationary methodologies by Aich at al. [
14] and then by Wilcox at al. [
15] for the main rivers of the MNRB.
The high pace of population growth generated a negative feedback such as a systematic deforestation that converted the savannah into fields providing cereals, firewood and building timber [
16,
17] and the occupation of the flood-prone areas with new settlements.
In Niger, recent outlines report that during the last 15 years the number of people and their property affected by flooding events have drastically increased [
18]. These events have raised the awareness of administrations towards the necessity of developing flood management strategies such as hazard mapping and early warning systems [
19].
The state of the art of the flood hazard literature in European countries is very unbalanced compared to that one of developing countries. African studies often focus just on economic aspects of flood risk [
20,
21] or evaluate the flood hazard on a regional level [
22,
23]. Only in a few cases there is an accurate analysis based on hydraulic aspects [
24]. European rivers are specifically analyzed following the European Flood Directive 2007/60 [
25] and statistical, hydraulic and GIS tools are used for flood hazard and flood risk mapping [
26,
27,
28,
29].
The present work provides the flood hazard assessment on 108 km of the Nigerien reach of the Sirba River which is the most populated stretch of the main tributary of the MNRB. To achieve the objective, an in-depth local hydraulic analysis based on the application of advanced statistics and a hydraulic numerical model was carried out [
30].
This methodology followed two main steps: the definition of hazard thresholds and the flood hazard mapping of each hazard scenarios. Firstly, hazard thresholds were statistically calculated and referred to field effects adopting the color codification of the Nigerien guidelines [
31]. The statistical analysis allowed to connect the probability of occurrence to each threshold. The non-stationary method was fundamental to evaluate the evolution of return levels over time, implicitly considering hydrological changes highlighted in the updated flow time series of Garbey Kourou [
32]. Secondly, a hydraulic numerical model was implemented to simulate different hazard scenarios related to each threshold. The model was based on a detailed topography and was calibrated with data obtained from topographical surveys and field measurements of discharge. Each simulation allowed to delineate the flood-prone areas summarized in flood hazard maps. Furthermore, the hydraulic model was used to evaluate the flood propagation time along the analyzed reach and validate the rating curves of the two Nigerien gauging sites of the Sirba River.
Adopting advanced statistical approaches and the consolidate hydraulic practice, the methodology used in this work could represents a guideline for the flood hazard evaluation of that countries affected by intensive hydrological changes. This study has a strategic importance because provides the first flood hazard mapping on the Sirba River and offers basic data for the implementation of an Early Warning System for the flood alert. These outputs will be a precious resource both for the research community and for decision makers to develop risk reduction plans and urban planning.
The work is structured as follows:
Section 2 presents the materials focusing on the study area, hydrologic dataset, and topography used in this research.
Section 3 describes methods used for hazard threshold definition and hydraulic numerical modeling.
Section 4 sets out results and related discussions.
Section 5 contains the conclusions and future perspectives.
4. Results and Discussion
4.1. Hazard Thresholds
The hazard thresholds fixed for the different flow conditions of the Sirba River respected both the statistical analysis and the field effects on the main riverine settlements according to the classification of national flood hazard scenarios used by the Niger Republic [
31]. The hazard thresholds in
Table 5 are shown with the maximum discharge related to each scenario and the statistical index related to both FDC [
32] and (stationary and non-stationary) GEV. These thresholds are related to the magnitude of the flooding event and damages on human settlements and are referred the following values:
- -
The normal condition was considered for the flow below the threshold of 600 m3/s that correspond at the Q15 (the discharge meanly exceeded 15 days a year) and the S-RT5;
- -
The frequent flood comprised the flow in the range 600–800 m3/s where 800 m3/s is related both to FDC (Q5) and GEV analyses (S-RT10 and NS-RT2);
- -
The severe flood varied from 800 to 1500 m3/s and corresponded to a flood with S-RT30 years that became a NS-RT5, a little greater than the major measured flood event of 2012;
- -
The catastrophic flood covered up to 2400 m3/s that is the flood event with S-RT equal to 100 years. After the non-stationary analysis, it was ten times more likely (NS-RT10).
The hazard thresholds identified for the Sirba River are consistent with European Flood Directive, but some differences can be highlighted. The three thresholds identified in the European Flood Directive [
25,
61,
62,
63]: (1) totally depend on hydrological analyses; (2) are related to return levels of 20/50 years (high frequency), 100/200 years (low frequency) and 500 years (rare frequency). The European Flood Directive overlooked the lower thresholds (normal condition and frequent flood) while the computed thresholds considered field effects and neglected rare events. The lower thresholds are very important in developing countries with high population growth and lack of urban planning since they may consistently affect rural activities [
64]. The events with high return level were not considered because of the connected discharges were not representative of the physical behavior of the river basin.
4.1.1. Generalized Extreme Value Analysis
The selected NSGEV model, by Wilcox et al. [
15], for the original Garbey Kourou time series was double linear (i.e., with different slopes before and after a break point) for both the μ (central tendency) and σ (dispersion) parameters. The model choice was confirmed for the updated series with high significance (alpha < 0.001) over fewer complex models.
The analysis of GEV time variation shown the break point for μ was 1996 for both series. The break point for σ was 2003 for the original series and 2006 for the new series. This may be due to either the new data or boundary effects; the methodology separated the break point from the end of the data series by 10 years, so the break point 2006 would not have been identified in the original series which arrive only to 2015.
Both μ and σ have a rising trend since 1996, which indicates an increase in magnitude and variability of extreme events as discussed by Amogu at al. and Descroix at al. [
3,
4]. The magnitude of the trend is much greater in the updated series. The shape parameter (ξ) is higher in the updated series but has a larger confidence interval. This indicates the most extreme (rare) values risk to be even greater than before, but with more uncertainty (
Figure 9).
The influence of considering non-stationary parameters can be seen in
Figure 10. This figure shows the trends for the return levels of 2 and 20 years. The return levels calculated with the NSGEV show an increase of several hundred m
3/s even for short return time (2-year) in the last 20 years. In all cases, non-stationary return levels for recent years were well above stationary return levels (
Figure 10 and
Figure 11). Confidence intervals were large and increasingly large for longer return levels. With the 2-year and 5-year return levels, there was high confidence that the current non-stationary values exceeded the return levels calculated as if the river discharge were stationary. Results were less clear starting from the 10-year return level. The range of uncertainty for the 20-year return level, given a non-stationary model, indeed encompassed the range of values included from the stationary model. This demonstrated the challenge of estimating the value of a given return level in a non-stationary context due to the exceptionally high uncertainty.
The comparison between stationary and non-stationary return levels underlined that with equal discharge, non-stationary values were ten times higher than the S-RT (
Figure 11) [
15,
65]. This behavior is justified by the substantial increase of the AMAX values in the last two decades and by the very steep trend of this increase [
32].
The application of the non-stationary methodology allowed to implicitly consider ongoing changes of the drivers that are strongly affecting the hydrological behavior and the change in frequency of extreme events [
15,
32]. Until now, the scientific community did not really understand which driver, between climate or land use/cover changes, more influence the hydrological evolution of Sahelian rivers [
3,
4,
5,
7,
9,
14].
4.1.2. Historical Flood Analysis
Floods impacts on the human settlement are a complex system based not only on physical factors (hydrology and river behavior) but also on social factors (urban development, people growing rate and awareness of the flood risk) [
66,
67,
68].
An analysis, based both on the ANFD and on the time series of maximum flow, allows identifying the worst events on the riparian villages of the Sirba River. The effects on the villages are not directly proportional to the magnitude of the flood, considering the first discharge threshold of 600 m
3/s (
Table 6).
Even though a lack in the record for the events of 1998 and 2012 some significant considerations could be made:
most of the events (six out of seven) in recent years (nine out of 18) are emphasizing the increased significance of recent flood events also for the Sirba River;
adaptation and flood awareness produce a decrease of the damage related to a certain discharge: the flooding event of 2010 (with the maximum recorded discharge until that year) affected more than 20,000 people while the higher event of 2013 affected half of the people (10,000). The same behavior is noticeable compared to the flood of 2007 and the higher event of 2015. The decrease of damages was due to the displacement of settlements after flooding events. The reallocation was noticed both during local surveys and participative meetings with inhabitants.
4.2. Flood Hazard Scenarios
Flood hazard mapping of the Sirba River represents the direct assessment of the degree of hazard exposition to which riverine populations are subjected.
Results from hydraulic simulation show an extension of flood-prone areas between 30 km
2 of the green scenario up to the 77 km
2 of the red scenario (
Table 7). These outcomes demonstrate that in any case floods cannot watered vast areas beside the river due to the carved morphology of the Sirba valley.
The bulk of the receptors are on the left bank of the Sirba River (houses, community services, infrastructures, fields and vegetable gardens) while on the right bank the receptors are essentially made up of fields and orchards, but no settlement is affected.
The normal condition (green scenario) does not interfere at all with human activities. The frequent flood (yellow scenario) could create some problems usually related to the livestock safety or to the damage on the human tools used on or near the river (fish nets, water pumps, pirogues). The severe flood (orange scenario) affects gardens, orchards, gardens, the lower sector of the villages and it is very significant also because in some cases it could compromise the use of the wells for drinkable water. The catastrophic flood (red scenario) inundates the lowest portion of the riverine villages and affects hundreds of houses even if the percentage of affected villages is never above 10% of their surface. Flood exposure of the main settlements along the Sirba River is quite the same perhaps because the memory of the severe flood of 2012 is still alive. This flood destroyed a large part of the low-altitude houses and caused damages until the limit of the orange threshold.
Tallé presents different features than other settlements and is the only case where the red scenario interest more than 10% of the populated surface (
Figure 12). This village is located a few kilometers from the national (paved) road Niamey-Téra and near the confluence Sirba-Niger, where access to irrigation water is possible throughout the year, this settlement is naturally suited to commercial horticultural activities and rice farming. Here demographic growth is restless, and the new households are often built in the flood exposure areas.
Analogous case is the catastrophic floods occurred in Niamey in 2012 and 2013, both events had a similar magnitude. Even if the dynamics are slightly different in the capital city, the flood in 2013 watered the same areas as the year before and destroyed the new settlements immediately re-built as a consequence of the strong population growth [
7,
69].
A more detailed analysis of the vulnerability of receptors will be the subject of future work and will allow the definition of the mitigation measures. The materialization of graduated mast (with the level of the four scenarios) and the draft of risk reduction plan will be performed for each village to improve the awareness of flood hazard. Mitigation or adaptation measures will be determined on the base of a cost-benefit analysis and will include: (1) the reallocation of settlements or gardens; (2) the improvement of building techniques; (3) the creation of limited sections of levees.
A supplementary scenario was performed to evaluate the amount of the backwater caused by the high-water level of Niger River during the Guinean flood of January 2019. This level indicates that Niger River level could regurgitate the 8-last km of the Sirba River reach (until the downstream boundary of the village of Garbey Kourou). This extent, confirmed by field measurements, allowed to identify the flood hazard caused by the Guinean backwater coming from the Niger River. This flood did not affect the household but destroyed many gardens between the Tallé village and the Sirba-Niger confluence.
4.3. Evaluation of Hydrometric Station Rating Curves
The hydraulic model allowed to validate the rating curves and evaluate the application range of the computed equations. This technique was applied for two rating curves: Garbey Kourou, calculated in Tamagnone et al. [
32], and the new one of Bossey Bangou, computed following a power-law with two parameters [
32,
71]:
Table 8 shows the parameter figures of the Rating Curve (RC) equation for Garbey Kourou e Bossey Bangou. The comparison between the computed rating curves and the curves extracted from the hydraulic model shows a good match between the two curves (
Figure 13). It allows appreciation of the hydraulic model as a reliable simulator of the natural behavior of the river also for flood conditions. The alignment between the two curves enables extension of the RC ranges of application above the major field measurement (
Figure 13). The application range of the GK rating curve can be considered valid up to 5 m, while the range of BB rating curve is extendable up to 8 m. The behavior of BB RC has an optimal alignment with the model outcome, while the GK RC underlines a mismatching possibly causing an overestimation of discharges above 5 m. The unaligned trend displayed in
Figure 13a is probably caused by the complex geometry downstream the GK station. It is characterized by the presence of a stone island that influences the river flow by creating a backwater. It caused complications also in model calibration. To overpass these problems, high-flow field measurements must be done to improve the reliability of the GK RC with for a wider application range.
Considering that discharge observations are strongly related to the quality of the rating curve equation, the uncertainty of the rating curve might lead to significant errors [
32,
72,
73]. For these reasons, the quality of rating curves should be periodically controlled and updated with on-site measurements, as well as, confirmed by hydraulic models. Moreover, the hydraulic modeling was able to extend the stage range of validity of the calculated equation above the maximum field measurement. In particular, the GK and BB rating curves were validated up to respectively 20% and 40% above the maximum field measurement.
4.4. Flood Propagation Time
The flood propagation time is the flow peak transit time from the gauging station of Bossey Bangou until the downstream villages. These times are crucial for the implementation of the early warning system of the Sirba River. The peak propagation velocity depends not only on geometric factors but also on the river discharge: the higher the discharge, the faster the flux, and hence the less the peak transit time. In the Sirba River, the mean flow velocity varies from 0.97 m/s, in the normal condition, to 1.13 m/s of the catastrophic scenarios. The yellow and orange scenarios were characterized by a velocity of respectively 0.98 and 1.01 m/s, little higher than the normal condition.
Table 9 shows the peak transit time for the severe inundation (Q = 1500 m
3/s) which can reflect what happened during the flood in 2012. The flow peak transit time is 29 hours for the entire reach. This means that the warning can be sent with an anticipation of about one day for the biggest villages in the downstream river reach (Touré, Guidare, Larba Birno, Garbey Kourou and Tallé) and of about 10 hours for the upstream villages. This notice was considered enough for the flood procedure alert but not adequate for the traditional communication system of the national structures devoted to flood alert. In other words, to make faster the communication of the alert to the Sirba riverine villages using the Bossey Bangou notice, the early warning system needs to also include a direct village-to-village communication mechanism.
The limited waring time from the Bossey Bangou gauging station to the main villages is expected to be the biggest challenge of an Early Warning System on the Sirba River. This weak point could be emphasized by the feeble internet network in the country. To solve this problem two different ways could be feasible: (1) the installation of new gauging stations in the headwater; (2) the use of hydrological models able to predict the river flow some days in advance [
74,
75]. For the first option, additional field work and supplementary investments would be needed but the system might guarantee higher accuracy than the use of hydrological models. In fact, these models are strongly affected by the uncertainty of meteorological input [
76,
77].
5. Conclusions
Considering increasing-attention that flooding events attract on the Sahelian countries, this research was designed to evaluate the flood exposure of the riverine population of the Sirba River. This represents an important step for the operability of an early warning system against floods. After the identification of the hazard thresholds, this study computed the flood hazard scenarios using a hydraulic numerical model on the 108 km river reach.
The hazard thresholds were defined, in accord with the national code of Nigerien flood alert, for four scenarios (green, yellow, orange, and red). They were identified on the field effects and then related to an index based on statistical analysis of the flow series. The application of the non-stationary method to compute the occurrence probability of flood events became fundamental in the Sahelian framework where the climate changes are drastically modifying the river hydrology. The results of the analysis showed that the probability of occurrence of an extreme event deriving from the non-stationary approach was about ten times higher than that considering the stationary approach (e.g., S-RT = 100 years ≅ NS-RT = 10 years). This resulted even though the confidence level of the non-stationary analysis was very low for high RT.
The flood scenarios showed that only the low sector of the villages was exposed to the flood hazard and a flooding event smaller or equal to the historical maximum flood of 2012 would not cause considerable damages on the villages. This effect is due to adaptation to the flood events caused by the recent phenomenon. Flood scenarios can be used by local administrations for urban planning avoiding new settlements in flood-prone areas. Indeed, flood hazard maps, with the scenario extents, will be shared with local administrations as a base for local planning. This information is very useful in a developing context where the urban system is not completely fixed and the population growth is very high, thus enhancing the awareness of the new inhabitant. Moreover, the association of flood scenarios with the national alert code constitutes a powerful decision-making tool for early warning.
Additional results of this study were the validation of the rating curves of the Sirba hydrometric stations and the computation of the flow peak transit time. The rating curve validation and the extension of the application range enhanced the reliability of the flow computed in the gauging stations of Bossey Bangou and Garbey Kourou. The flood propagation time was an essential result for the early warning system that will be implemented on the Sirba River. The short notice between the Bossey Bangou gauging station and the downstream villages is the major challenge that could be solved only with the installation of new gauging stations in the headwater. Alternatively, a possible solution could be the use of a hydrological model that predicts the river flow discharge some days before.
Future developments of this study will explore the implementation of the early warning system and the integration with the hydrological model results to increase the notice time in the villages. Moreover, a detailed study on the exposure and vulnerability will be conducted on riverine villages to exhaustively evaluate the flood risk in terms of economic and social losses.