Mid-Century Climate Change Impacts on Ouémé River Discharge at Bonou Outlet (Benin)

Abstract

This work focuses on impacts of climate change on Ouémé River discharge at Bonou outlet based on four global climate models (GCM) over Ouémé catchment from 1971 to 2050. Empirical quantile mapping method is used for bias correction of GCM. Furthermore, twenty-five rain gauges were selected among which are three synoptic stations. The semi-distributed model HEC-HMS (Hydrologic Modeling System from Hydrologic Engineering Center) is used to simulate runoff. As results, HEC-HMS showed ability to simulate runoff while taking into account land use and cover change. In fact, Kling–Gupta Efficiency (KGE) coefficient was 0.94 and 0.91 respectively in calibration and validation. Moreover, Ouémé River discharge is projected to decrease about 6.58 m³/s under Representative Concentration Pathways (RCP 4.5) while an insignificant increasing trend is found under RCP 8.5. Therefore, water resource management infrastructure, especially dam construction, has to be developed for water shortage prevention. In addition, it is essential to account for uncertainties when designing such sensitive infrastructure for flood management.

1. Introduction

Among natural resources, water is the largest and only 3% of it is freshwater [1]. Moreover, only 1/3 of the available volume is accessible for use in agriculture and cities. The remaining third is frozen in glaciers or hidden too deep in underground layers. Furthermore, due to population growth, the increasing demand for water-intensive usage such as agriculture, fisheries and industries, has stressed global freshwater resources especially in West Africa [2]. In addition, climate change is projected to impact temperature and rainfall with consequences on runoff especially in developing countries [3,4]. Therefore, there is a need for quantifying climate change impacts on river discharge for better planning.

Model projections based on ensemble median show no substantial threat to future river discharge availability with changes in the range of ±5% for large parts of West Africa [5]. A 10% decrease is project in Nigeria and a 10% increase for most of Guinea, Sierra Leone, Liberia and Côte d’Ivoire [5]. Runoff decrease of more than 10% is projected for Senegal, Gambia and Guinea-Bissau, whereas a strong increase of over 10% is expected for the regions along the border of Liberia and Côte d’Ivoire [5]. However, there is unclear change in Ouémé River discharge in Benin.

Moreover, most previous works focusing on quantifying impacts of climate over river discharge hardly integrate land use and land cover change aspects as part of anthropogenic impacts while using climate projection. Land use and land cover are physical aspects which changes impact on surface runoff [6]. Taking them into account is essential in quantifying impacts of climate change on river discharge [7]. Numerous researchers have shown evidence of impact of forest conversion into cropland due to food security in a growing population especially in developing countries with considerable impacts on stream flow [8,9,10,11,12]. In addition, change in climate as a result of greenhouse gas emission has to be taken into account with climate models based on scenarios. Model projections exist essentially in four scenarios based on the representative concentration pathways RCP 2.6, RCP 4.5, RCP 6, and RCP 8.5. The names are labeled after possible range of radiative forcing values of respectively 2.6, 4.5, 6.0, and 8.5 W/m² in the year 2100 [13]. RCP 4.5, which is the business as usual scenario and the pessimist scenario RCP 8.5 are considered in this study since African countries are expected to be engaged in industrialization for their development. Thus, quantifying impacts of climate change on water resources with regard to anthropogenic activities is important for resources planning and management.

As climate models are key elements in weather projection, they have been promoted and used during the last decades [6,14,15]. However, climate models hold some biases that need to be corrected before any impact study. Numerous bias correction methods exist and are applied around the world [16,17,18,19,20,21,22], including delta change, linear scaling, distribution mapping, empirical and adjusted quantile mapping [18,23,24]. M’Po et al. [18] showed that empirical quantile mapping (EQM) was the best among other bias correction methods over Ouémé catchment. Similarly, efficiency of this method in climate model bias correction has been shown worldwide [16,22,25,26,27]. Consequently, EQM is the method used in this work for climate model bias correction.

In the process of quantifying climate change impacts on stream flow, rainfall runoff simulation has to be well handled in order to reduce uncertainties. According to Box [28], “all models are wrong, but some are useful”. Moreover, Field [29] emphasized that not only all models are wrong, but their relative usefulness varies depending on location and interest. As the objective here is to assess climate impacts on river flow, the model that will be used has to integrate land use, to suit large catchment and to be able to simulate both low and high flow. HEC-HMS (Hydrological Engineering Center for Hydrologic Modeling System) is the one used. It has been widely applied worldwide with satisfactory results. HEC-HMS is a semi-distributed model that integrates land use and cover change [11,30,31,32,33,34], like SWAT [8,12,35,36,37,38,39]. However, comparative study based on model’s hydrological processes, governing equations used, minimum data required and spatial as well as temporal scale of both HEC-HMS and SWAT concluded that HEC-HMS has the best efficiency and lightness [40,41].

The main objective of this work is to quantify impacts of climate change on river discharge in Ouémé catchment at Bonou outlet till 2050 taking into account land use and land cover change. More specifically, this work aims at bias correcting four global climate models (GCMs), simulating discharge over Ouémé River catchment at Bonou outlet from 1971 to 2010 and integrating land use and land cover using HEC-HMS. In addition, future climate change impacts on Ouémé River discharge based on the Representative Concentration Pathways (RCP 4.5 and RCP 8.5) is quantified over the period 2020–2050.

2. Materials and Methods

2.1. Study Area

2.1.1. Ouémé Catchment Description

Ouémé catchment at Bonou outlet is in Benin Republic and positioned between 6.9° and 10.3° latitudes, and 1.5° and 3.5° longitudes in West Africa. Benin is bordered in the west by Togo, by Nigeria in the east, by Niger in the north, by Burkina Faso in north-west and Atlantic Ocean in the south (Figure 1).

Ouémé catchment area is about 50,000 km² with 0.1% of this in Togo and 8% in Nigeria [42]. It is under tropical climate with three climate zones from the subequatorial climate to Sudanian climate northward. Ouémé catchment receives annually between 724 and 1396 mm rainfall [43].

2.1.2. Ouémé Catchment Hydrological Soils Groups, Land Use and Land Cover Types

Soil types and hydrological soil groups over Ouémé catchment, based on the Harmonized World Soil Database (HWSD) [44], are shown on Figure 2a,b.

Luvisols are the dominant soil type in Ouémé catchment, reflecting stable geological conditions with high activity of clay throughout followed by lixisols (Figure 2a). Moreover, productive nitisols are developed on the alluvium of the coastal region. On the sandbars and lagoons in the coastal region, there are gleysols. The vertisols lay on Zagnanado plateau within the Lama depression. In addition, 5% of the catchment is recovered with plinthosols.

The texture of the various soil types encountered is mainly composed of loam, clay, clay-loam, sandy-loam and sandy-clay-loam as illustrated in Figure 3.

In addition, Ouémé catchment lies on mio-pliocene aquifer, whose top is essentially made of clay.

Based on the hydrological soil groups classification, Ouémé catchment holds two hydrological soil groups, which are C and D, respectively qualified as moderately high and high runoff potential (Figure 2b). These conditions imply least infiltration into deep aquifer but considerable surface retention due to the clay cover. In fact, based on the infiltration rate established by the soil conservation service, the hydrological soil groups C and D infiltration rate is respectively from 1.27 to 3.81 mm/h and 0 to 1.27 mm/h [45].

The land use and cover map of the years 1975, 2000 and 2013 of the Ouémé catchment extracted from the West African land use and cover map provided by Tappan et al. [46] is illustrated in Figure 4. Land use and cover degradation is observed in Ouémé catchment.

In fact, from 1975 to 2013 savanna represents respectively 77.01% in 1975, 64% in 2000 and 60% in 2013 of Ouémé catchment as shown in

Table 1. Percentage of change in land use and land cover within 1975–2000 and 2000–2013.

Land Use Types	Area Percentage			Percentage of Change
	1975	2000	2013	1975–2000	2000–2013
Forest	1.2	0.7	0.6	−0.5	−0.2
Savanna	77.0	64.6	60.4	−12.4	−4.2
Wetland—flood plain	0.4	0.6	0.8	0.2	0.2
Plantation	0.3	0.4	1.8	0.1	1.3
Agriculture	7.1	23.1	31.4	16.0	8.3
Water bodies	0.1	0.1	0.1	0.0	0.0
Settlements	0.2	0.5	0.9	0.3	0.4
Irrigated agriculture	0.0	0.0	0.2	0.0	0.1
Gallery forest	5.1	4.7	4.1	−0.4	−0.6
Degraded forest	0.7	0.1	0.2	-0.6	0.1
Woodland	6.4	3.0	1.9	−3.4	−1.0
Cropland and fallow with oil palms	1.5	2.4	2.8	0.9	0.5

. In addition, agriculture occupied 7.1, 23.1 and 31.4% of the catchment area respectively in 1975, 2000 and 2013. Thus, savanna has decreased over the years to the benefit of agriculture. Percentage of change in each land use and cover classes are computed and shown in Table 1.

Conversion of woodland, gallery forest, forest and savanna into agriculture land is noticeable. Such changes may result in increase in direct runoff as well as erosion. Build-up has also increased slightly over both periods. This change is supposed to increase runoff in Ouémé catchment.

2.2. Data

Twenty-five rain gauges among which are 3 synoptic stations are selected over Ouémé catchment. Details about geographical position of each station are found in

Table 2. Rain gauges and synoptic stations geographical position details.

Station Name	Longitude (Degree)	Latitude (Degree)	Station Name	Longitude (Degree)	Latitude (Degree)
ABOMEY	1.98	7.18	INA	2.73	9.97
AGOUNA	1.7	7.55	KETOU	2.61	7.36
AKLAMPA	2.02	8.55	KOUANDE	1.68	10.33
BANTE	1.88	8.42	OKPARA	2.73	9.47
BASSILA	1.67	9.02	PARAKOU_AIRPORT **	2.6	9.35
BEMBEREKE	2.67	10.2	PENESSOULOU	1.55	9.23
BETEROU	2.27	9.2	PIRA	1.72	8.65
BIRNI	1.52	9.98	SAVALOU	1.98	7.93
BOHICON **	2.07	7.17	SAVE **	2.47	8.03
BONOU	2.5	6.93	TCHAOUROU	2.6	8.87
DASSA_ZOUME	2.17	7.75	TCHETTI	1.67	7.82
DJOUGOU	1.67	9.7	ZAGNANADO	2.33	7.25
GOUKA	1.95	8.13

Station names with (**) represent names of synoptic stations.

.

Rainfall data over the 25 rain gauges and temperature at the 3 synoptic stations are provided by the National Meteorological Agency of Benin from 1971 to 2010. Moreover, Ouémé River discharge at Bonou outlet is provided by the general Benin Directorate of water. Observed data went through quality control where missing data are filled using double curve mass [47] apart from discharge data in the entire years 2006 and 2009, which are left unfilled. The global climate models (GCM) are extracted at the twenty-five station points. The four GCM used for rainfall and temperature projection are shown in

Table 3. Details of the regional climate models used.

Model Name	Institute	Driven Model
CanRCM4	Canadian Centre for Climate Modeling and Analysis	CCCma-CanESM2_CCCma
RACMO22T	Royal Netherlands Meteorological Institute, De Bilt, The Netherlands	ICHEC-EC-EARTH
HIRHAM5	Danish Meteorological Institute	NCC-NorESM1-M
REMO2009	Helmholtz-Zentrum Geesthacht, Climate Service Centre, Max Planck Institute for Meteorology	MPI-ESM-LR

. The historical period considered is from 1971 to 2005 while future projection is taken from 2020 to 2050. Climate scenarios used were based on the Representative Concentration Pathways 4.5 and 8.5.

2.3. Methods

Impacts of climate change over stream flow in Ouémé River at Bonou outlet are quantified using the following steps. First of all, regional climate model bias correction is done. Then, rainfall over the catchment is averaged using the robust and non-biased kriging method [48], while temperature was averaged using the mean value of the three synoptic stations. In addition, rainfall runoff simulation is done using the semi-distributed model HEC-HMS based on the curve number loss method in order to take into account land use and land cover change. The calibration period is 1971–1990 and the validation period is 1991–2010. The future stream flow is projected over the period 2020–2050. Change in Ouémé River peak discharge is quantified using flow duration curve.

2.3.1. Climate Models Bias Correction

Climate projection data are corrected using quantile mapping bias correction method. Comparing different bias correction methods such as Delta change method, linear scaling and empirical and adjusted quantile mapping over Ouémé catchment, M’Po [18] proved that the empirical quantile mapping performed better than others in correcting biases of daily precipitation. Details of mathematical equations are found in [18,49]. Models’ bias correction efficiency is measured using two efficiency coefficients: the Kling–Gupta Efficiency (KGE) coefficient and the percentage of bias (PBIAS) as detailed in

Table 4. Model performance criteria.

Efficiency Coefficient	Definition and Utility	Optimal Value	Expression
Nash–Sutcliffe Efficiency (NSE) [51]	NSE is a normalized statistic that determines the relative magnitude of the residual variance or noise compared to measured data variance. It runs from –inf to 1.	Value of 1	$NSE=1-\frac{{\displaystyle \sum _{i=1}^N{\left(S_i-O_i\right)}^2}}{{\displaystyle \sum _{i=1}^N{\left(O_i-\overline{O}\right)}^2}}$
rPearson (r) [52]	rPearson estimates the degree to which two series are correlated and runs from 0 to 1	Value of 1	$r=\frac{{\displaystyle \sum _{i=1}^N{O}_i{S}_i-N}\overline{O_i}\overline{S_i}}{\sqrt{\left({\displaystyle \sum O_i{}^2-N}{{\displaystyle \overline{O_i}}}^2\right)}\sqrt{\left({\displaystyle \sum S_i{}^2-N}{{\displaystyle \overline{S_i}}}^2\right)}}$
Percent bias (PBIAS) [53]	Percent bias (PBIAS) measures the average tendency of the simulated values to be larger or smaller than their observed ones. Positive values indicate overestimation bias, whereas negative values indicate underestimation bias.	Value of 0	$PBIAS=100\frac{{\displaystyle \sum _{i=1}^N\left(S_i-O_i\right)}}{{\displaystyle \sum _{i=1}^N{O}_i}}$
Kling–Gupta Efficiency (KGE) [54]	KGE provides a diagnostically interesting decomposition of the Nash–Sutcliffe Efficiency (NSE), which facilitates the analysis of the relative importance of its different components such as correlation, bias and variability in the context of hydrological modeling	Value of 1	$KGE=1-\sqrt{{\left(r-1\right)}^2+{\left(\alpha -1\right)}^2+{\left(\beta -1\right)}^2}$ where r = rPearson, α is the ratio between simulated variance and observed variance, and β is the bias (the ratio between simulated mean and observed mean)
Where Si simulated discharge, Oi observed discharge, N sample size

. The Kling–Gupta Efficiency (0 ≤KGE ≤ 1) has the advantages of taking into account the Nash–Sutcliffe Efficiency (NSE) as well as the correlation coefficient [50]. In addition, the PBIAS is used to quantify the overall difference between observation and simulation.

2.3.2. Rainfall Runoff Modeling

The main parameters used here are: meteorological, rainfall loss, baseflow, basin and surface runoff components. Part of the rainfall is lost through evapo-transpiration, interception of canopy and storage areas. Pervious areas are those into which rainfall infiltrates with no loss [32,33]. These are the areas that feed the baseflow. In contrast, the impervious areas are considered as those areas that do not allow infiltration, such as paved areas and urban areas. These contribute more to direct runoff [45]. Both pervious and impervious depend on the land use and land cover types. Thus, in quantifying impact of land use and cover change on stream flow, the loss method based on the runoff curve number (CN) is the best. Actually, the runoff percentage over a grid cell rely on CN which varies as a function of hydrological soil group, land use and land cover as surface condition, and antecedent moisture condition features. Only C and D were the hydrological soil groups (HSG) identified in Ouémé catchment (Figure 2) that are qualified respectively as moderately high and high runoff potential. They are characterized by similar curve numbers considering each land use and cover type [45]. In addition, savanna is the largest land cover type followed by agricultural lands that represent respectively more than 60% and 25% of the catchment, which also have similar curve numbers. Therefore, it is assumed that the average curve number over the catchment is representatives. The CN can either be computed or calibrated considering homogeneous watershed [55]. In this work, it has been calibrated since Ouémé catchment here is taken with no subdivisions.

In the model calibration process, model parameter optimization helps to determine the set of parameters whose values make the model best simulate the observations. After selection of the optimized set of parameters, a sensitivity analysis is proceeded in order to identify the most sensitive parameters. In fact, the method of one parameter at a time that is detailed by Ouédraogo et al. [56] is applied. The value of change in each parameter is varied from −30% to +30% with increments of 10%, while keeping all other parameters constant. The objective function used is the Nash–Sutcliffe Efficiency (NSE). In addition, an elasticity ratio is computed in order to rank model parameters according to sensitivity level. Once the most sensitive parameters are detected, an uncertainty analysis is conducted to estimate their precise value [45].

Model performance is assessed using the Kling–Gupta Efficiency (KGE) coefficient and the percentage of bias (PBIAS) between observed and simulated flows. Moreover, the flow duration curve of both observed and simulated flows is used to assess model efficiency based on quantile curves as indicators of various hydrologic conditions [57]. Quantile curve intervals could be subdivided into several groups. A standard approach is to divide the quantile curve into five classes: high flows (0–10%), moist flows (10–40%), mid-range flows (40–60%), dry conditions (60–90%), and low flows (90–100%) [57].

2.3.3. Climate Change Impacts on Water Resources in Ouémé Catchment

Impact of climate change over stream flow in Ouémé catchment is quantified as change between observed and projected trend slope. This change is assessed in discharge, rainfall, rainfall loss, potential evaporation and temperature, based on the Mann–Kendall trend analysis [58].

3. Results

3.1. Climate Models Bias Correction

3.1.1. Rainfall

Result of catchment averaged rainfall after bias correction is shown on Figure 5. On graphical basis, seasonal regime of the corrected models fits well that of observations.

The agreement between observed and corrected data is confirmed by the efficiency coefficients computed below. Bias correction performance is assessed using the KGE and the PBIAS and summarized in

Table 5. Bias correction efficiency at seasonal scale.

	KGE		PBIAS
	Before	After	Before	After
REMO	0.01	0.95	60.7	−4.6
RACMO22T	0.73	0.99	23.6	−0.8
HIRHAM	−0.06	0.98	71.7	−1.6
CanRCM4	0.50	0.98	46.5	−1.1

.

After bias correction, the KGE of all the models is equal to or more than 0.95 with an absolute percentage of bias between 0.8 and 4.6. On average, the quantile mapping method used here is good at the models’ correction. However, RACMO22T is the best corrected with a KGE of 0.99 and 0.8% of underestimation of observation followed by CanRCM4 and HIRHAM, whose KGE is 0.98 with respectively 1.1 and 1.6 of underestimation.

The results of bias correction at daily scale are shown on Figure 6. Bias corrected and observation superposed well.

Efficiency coefficients computed for the bias correction at daily scale is summarized in

Table 6. Bias correction efficiency at daily scale.

	KGE		PBIAS
	Before	After	Before	After
REMO	−0.08	0.89	60.7	−4.6
RACMO22T	0.70	0.91	23.6	−0.8
HIRHAM	−0.22	0.89	71.7	−1.6
CanRCM4	0.50	0.88	46.5	−1.1

.

It is noticed that before bias correction, prediction of observations by CanRCM4 and RACMO22T was somewhat good with KGE respectively of 0.5 and 0.7. Moreover, the percentage of bias was respectively 46.5 and 23.6. After correction, KGE is of 0.88 and 0.91 respectively for CanRCM4 and RACMO22T, whereas it is 0.89 for both REMO and HIRHAM. Furthermore, percentage of bias is 4.6, 1.6, 1.1 and 0.8% of underestimation of observation respectively for REMO, HIRHAM, CanRCM4 and RACMO22T. Therefore, at daily scale, RACMO22T still remains the best corrected model.

3.1.2. Temperature Projection

The results of bias correction of averaged temperature are exhibited in Figure 7. As previously noticed, corrected data fits observation well. CanRCM4 raw data is closer to observation compared to others.

The efficiency coefficients of models before and after bias correction are summarized in

Table 7. Efficiency of temperature bias correction.

	KGE		PBIAS
	Before	After	Before	After
REMO	0.86	0.98	−5.9	−0.1
RACMO22T	0.50	0.98	−13.9	−0.03
HIRHAM	0.74	0.99	−5.2	−0.1
CanRCM4	0.94	0.99	2.9	0.02

.

The KGE before correction is respectively 0.5, 0.74, 0.86 and 0.94 for RACMO22T, HIRHAM, REMO and CanRCM4. In addition, the PBIAS is 13.9, 5.9 and 5.2% of observation underestimation respectively for RACMO22T, REMO and HIRHAM, whereas CanRCM4 overestimated observation by 2.6%. After correction, the KGE of all four models is between 0.98 and 0.99 with the absolute value of bias percentage between 0.02 and 0.1.

3.2. Rainfall Runoff Modeling

3.2.1. Model Calibration

The model parameters optimization results are detailed in

Table 8. Optimized model parameters.

Parameter	Optimized Value	Unit
Recession—Initial Discharge	31.373	m3/S
Recession Constant	0.9314
Recession—Threshold Discharge	4.9428	m3/S
Curve Number	35.721
Initial Abstraction	0	mm
Lag Time	23,292	min
Simple Canopy—Initial Storage	1	%
Simple Canopy—Max Storage	116.93	mm
Simple Surface—Initial Storage	35	%
Simple Surface—Max Storage	598.6	mm

. The optimal value of the lag time, the maximum canopy storage and the maximum surface storage are respectively 23,292 min, 116.93 mm and 598.6 mm.

The model is found to be sensitive to these three parameters as showed by the dashed lines on Figure 8. In addition, the model is insensitive to the remaining parameters.

Moreover, the maximum surface storage (Surface—Max Storage) is revealed to be the most sensitive parameter followed by the lag time and the maximum canopy storage (Canopy—Max Storage) as illustrated on Figure 9.

Results of statistics of the uncertainty analysis based on Monte Carlo are shown in

Table 9. Results of Monte Carlo Analysis with 500 Trials.

Statistics	Canopy—Max Storage	Lag Time	Surface—Max Storage
Mean	115.2	23,286.6	599.1
Number of trials of observations	500.0	500.0	500.0
Minimum	85.1	23,091.0	571.0
Maximum	129.8	23,487.0	630.9
Amplitude	44.7	396.0	59.9
1st Quartile	109.7	23,251.8	592.7
Median	116.1	23,286.0	598.7
3rd Quartile	122.0	23,324.0	605.4
Mean	8.2	60.2	10.0
Standard deviation	0.1	0.0	0.0
Variation coefficient	0.1	0.1	0.1

. The average value of the lag time, the maximum canopy storage and the maximum surface storage are respectively 23,286.6 min, 115.2 mm and 599.1 mm. These particular values of the sensitive parameters are then used to calibrate the HEC-HMS model

3.2.2. Hydrological Model Performance

Results of HEC-HMS model calibration and validation of Ouémé River discharge at Bonou outlet are plotted respectively in Figure 10a,b. In calibration, simulated discharge better reproduces that observed with an efficiency of 0.94 based on the KGE and 7% overestimation based on the PBIAS. In validation, the efficiency was 0.91 based on the KGE with 1.3% of underestimation of observation. Therefore, we can confirm the high performance of HEC-HMS over Ouémé catchment.

3.2.3. Comparison of Observation and Simulation Flow Duration Curve

Calibrated and observed flow duration curve is compared according to the flow quantile values. Results are shown respectively in Figure 11a,b. Model performance judgment is based on the five standard classes: high flows (0–10%), moist flows (10–40%), mid-range flows (40–60%), dry conditions (60–90%), and low flows (90–100%).

Considering calibration and validation, high flow is mostly underestimated. In calibration (Figure 11a) the underestimation is about 200 m³/s whereas in validation (Figure 11b) underestimation is about 100 m³/s. However, simulated moist flow fits that of observed. Mid-range and dry flow are slightly underestimated, whereas low flow is better simulated. Therefore, this underestimation has to be taken into account in designing hydraulic infrastructures for high flow attenuation.

3.3. Climate Change Impacts

At annual scale, change in annual rainfall, peak discharge, temperature and potential evapotranspiration from observed (1971–2010) to future projection (2020 to 2050) based on the RCP 4.5 and RCP 8.5 are illustrated in Figure 12. On a visual basis, the four variables show an increasing trend during observation period. Considering RCP 4.5 projection, rainfall is projected to decrease, whereas the trend stays constant for RCP 8.5 (Figure 12a). As a result, peak discharge follows rainfall trend as shown in Figure 12c. Moreover, temperature is projected to increase comparing both RCP 4.5 and RCP 8.5 to observation (Figure 12b). Consequently, potential evaporation also shows an increasing trend for both future scenarios (Figure 12d).

Mann–Kendall trend analysis resulting at 0.05 confidence level, over annual rainfall, peak discharge, temperature and potential evapotranspiration is summarized in

Table 10. Trend in annual rainfall, discharge, temperature and potential evapotranspiration.

Variable	Observed			RCP 4.5			RCP 8.5
	Tau	Sen’s Slope	p-Value	Tau	Sen’s Slope	p-Value	Tau	Sen’s Slope	p-Value
Rainfall	0.22	4.42	0.05	−0.18	−1.33	0.02	0.21	1.89	0.01
Discharge	0.22	9.56	0.05	−0.37	−6.58	0.00	0.09	1.59	0.23
Temperature	0.46	0.03	0.00	0.67	0.03	0.00	0.73	0.03	0.00
PET	0.03	0.07	0.79	0.63	4.51	0.00	0.66	4.92	0.00

. In this table, all significant trends at 0.05 confidence level are in bold. During the observation period, there is significant increase of 4.42 mm in rainfall, 9.56 m³/s in discharge and 0.03 °C in temperature. No trend is observed in observed potential evapotranspiration. Moreover, there is significant increase of 0.04 °C and 0.05 °C respectively in the temperature projection based on RCP 4.5 and RCP 8.5. Similarly, potential evapo-transpiration based on the RCP 4.5 and RCP 8.5 significantly increase respectively by 4.51 mm and 4.92 mm. Annual rainfall will decrease significantly by 1.33 mm according to RCP 4.5, whereas it will significantly increase by 1.89 mm based on RCP 8.5. As a result, peak discharge significantly decreases by 6.58 m³/s under RCP 4.5 and insignificantly increases by 1.59 based on RCP 8.5.

Apart from the significant decrease in peak discharge from 2020 to 2050 based on the RCP 4.5, there is also decrease in high flow whose probability of exceedance is less than 10% compared to observation (Figure 13). Similarly a decreasing trend is obtained considering discharge projection from 2020 to 2050 based on RCP 8.5. Moreover, there is less high flow in projected discharge under RCP 8.5 than that of RCP 4.5. However, more mid-range and low flow are observed according to RCP 8.5 than RCP 4.5 and observation.

Consequently, water resource management has to be developed to encounter future water shortage in an increasing demand context.

4. Discussion

The efficiency of bias correction using quantile mapping method corroborates that of M’Po et al. [18] over Ouéme catchment, Obada et al. [19] in Mékrou catchment and Badou et al. [49] in Benin portion of the Niger River catchment. In addition, Ouémé River discharge modeling using HEC-HMS showed very good results in calibration and validation. However, it exhibits an overall underestimation of high discharge. This has to be taken into account in setting up a hydrodynamic model especially for flood forecasting as well as dam construction scenarios. In addition, a soil moisture accounting method may be a good option in taking groundwater flow into account using HEC-HMS over Ouémé catchment, rather than the curve number method in order to consider the eventual delayed flow due to surface storage.

Moreover, the increase in annual rainfall during the observed period 1971–2010 showed in this work is also noticed in Oyerinde et al. [16] in the Niger sahelian catchment for the near future term. Current results are also in accordance with that of Oyerinde et al. [16], Biao et al. [20] and M’Po et al. [18] and showed increase for projected rainfall based on the RCP 8.5 and decrease for the RCP 4.5. Moreover, the increase in temperature noted here is also reported by Oyerinde et al. [16], Biao et al. [20], M’Po et al. [18], Hounguè et al. [58] and Lawin et al. [48] as proof of global warming. Similarly, increase in potential evapotranspiration is also highlighted by Oyerinde et al. [16] and Biao et al. [20]. All of the changes impede on discharge. As results, decrease of about 6.58 m³/s is observed in projected discharge based on the RCP 4.5 scenario and an insignificant increasing trend for the projected discharge based on the RCP 8.5 at mid-century term as obtained by Essou et al. [59] and Stanzel et al. [5]. These results are also in line with that of Benin country profile, which addressed climate change impacts on hydro-climatic variables based on projections made till year 2085 [60]. Therefore, it is essential to take adaptation measures for preventing possible drought or flood in Ouémé catchment as consequences of climate change impacts on Ouémé River discharge.

5. Conclusions

This work quantified change in Ouémé discharge for the period 1971–2050 based on four global climate models (GCM) as well as land use and land cover change. From 2020 to 2050, projection of Ouémé River discharge showed 0.94 and 0.91 of Kling–Gupta efficiency respectively in calibration and validation. Compared to previous works, HEC-HMS performed well and should be adopted in areas with limited soil data especially in developing countries. However, future work should explore the option of using soil moisture accounting methods for losses processing using HEC-HMS over Ouémé catchment. Furthermore, underestimation of high flow should be taken into account in hydraulic scenario development. Moreover, an increase in temperature is projected as proof of global warming. A significant decreasing trend is noticed in projected discharge based on the RCP 4.5 scenario with an insignificant increase in the projected discharge based on the RCP 8.5 at mid-century term. Thus resource planning has to be addressed in order to avoid potential future shortages.