The Oti river basin in Togo is located in the north of the country, in the Volta river basin, and drains part of the catchment areas of the countries bordering Togo, namely Burkina Faso, Benin and Ghana. At the Mango outlet, the Oti basin covers an area of 3.652 square kilometres and is located between 11˚05' and 10˚96' N and 0˚12' and 0˚95' E ( Figure 1 ).

The topography of the watershed is based on a morphology that includes the Gourma peneplain, the Dapaong-Bombouaka plateaus and the Oti plain. The total elevation ranges from 106 to 506 meters. Its relatively dense hydrographic network is controlled by the Oti River. The Oti river, the main water collector (167 kilometers long in Togo), rises in northern Benin on the eastern slopes of the Atacora chain, under the name of Pendjari. He follows a south-west-north-east orientation, crosses the Atacora region, and then reverses and returns south-west, arriving in Togo, where it takes the name of the Oti ( Mul et al., 2015 ). The Oti river forms the natural border between Togo and Benin from its confluence with the Koumongou River, south of the Galangachi classified forest, to the Ghanaian commune of Sabari, south of which the Oti river flows into Lake Volta. Its tributaries are the Wabga, Namiélé and Sansargou on its right bank. Its left bank tributaries are the Koumongou, Kéran and Kara. The riverbed of the Oti is cut into the clay formations and here and there bordered by a forest gallery. When the Oti overflows its banks during the rainy season, it causes major flooding with serious consequences for the people living along its banks.

The map shows the meteorological stations (Barkoissi, Borgou, Dapaong, Mandouri, Mango, Tankpamba and Toaga) used for this research in the Oti watershed.

According to the 5th General Census of Population and Housing (5thRGPH) in 2022, the Oti watershed had a population of 1.143.520. Several activities are practiced by these populations. These include agriculture, livestock, fishing, hunting, handicrafts, commerce, tourism, sand and gravel extraction, and transportation. However, agriculture seems to be the main activity of the people living in the watershed.

The data used for this study are cumulative daily rainfall and air temperatures (maximum and minimum). They are collected from seven meteorological stations (Barkoissi, Borgou, Dapaong, Mandouri, Mango, Tankpamba, Toaga) of the National Agency for Meteorology in Togo (ANAMET) covering the period 1961-2022.

Table 1 summarizes the weather station characteristics of the observed and simulated data used for this study.

To analyze extreme weather events in the Oti watershed, it is necessary to use the results of climate model data in comparison with observed data. Data from climate models were used to analyze extreme hydroclimatic events in comparison with historical and observed data. They are also used to assess the performance of data from these models in analyzing extreme rainfall events as indicators of flooding in the study area. Simulation data from two models (CCLM4.8 and REMO) from the Coordinated Regional Climate Dawnscaling Experiment (CORDEX) West Africa program, which provides high-resolution data in the form of 0.44˚ × 0.44˚ grids, i.e. approximately 50-km × 50-km, are also used. These data cover the period from 2006 to 2100. The data is simulated in two phases, taking into account the historical aspect from 1961 to 2022 without the RCP (Representative Concentration Pathways) scenarios and the future aspect from 2025 to 2069 with the combination of RCP 4.5 and RCP 8.5 for the simulations. The priority simulations carried out as part of the CORDEX project concern only two of the RCPs defined, namely the RCP 4.5 and RCP 8.5 scenarios. These particular scenarios correspond to the B1 and A1B scenarios of the old Greenhouse Gas evolution profiles (SRES = Special Report on Emissions Scenarios) ( Nka Nnomo, 2016 ). They represent optimistic and pessimistic climate change scenarios for the 21st century, respectively ( Herbst & Rautenbach, 2016 ). For this reason, both scenarios are used in this study.

In order to ensure the homogeneity and reliability of the climatic data chronicles, data criticism and control are the basic processes to which this research is dedicated. For example, data from the raw precipitation and temperature files were systematically checked. This check eliminates values with gaps and does not guarantee the accuracy of the data retained.

The missing data from the stations selected for this research were then reconstituted to obtain the most complete series possible, constituting the operational database used for subsequent calculations. Stations with significant data gaps (several months or years with >5% missing data) were simply deleted. On the other hand, stations with <5% missing data were used without filling in, as they have a daily time step. This method avoids the artificialization of data, while eliminating any risk of introducing vitiated and fanciful figures ( Klassou, 1996 ; Totin, 2010 ).

Data quality control has been performed to help identify any recording errors that may exist in the daily data ( Balliet, 2017 ; Atcheremi et al., 2018 ). Thus, the principle consists of:

There are several methods for interpolating rainfall data (simple arithmetic mean, linear interpolation, Thiessen polygon interpolation and ordinary block Kriging). The most robust and most widely used in recent years in tropical regions, particularly in West Africa, are linear interpolation and interpolation by Thiessen polygons ( Kodja, 2018 ). For the purposes of this study, the Thiessen polygon interpolation method was used for both observed and model-simulated rainfall data. The weighted average precipitation P for the basin is then calculated by summing the precipitation Pi from each station, multiplied by their weighting factor (area Si), divided by the total surface area S of the basin. The average rainfall over the basin is therefore written as:

$P=\frac{{\displaystyle \sum P_i{S}_i}}{S}$(1)

With P: average rainfall (mm); P: rainfall at the station within the polygon (mm); S_i: surface area of the polygon (km²) and S: total surface area of the watershed (km²).

The various zones of influence are determined by geometrically dividing the basin on a topographical map. Once the available stations have been plotted on a map, a series of straight lines is drawn between adjacent stations. Perpendiculars are drawn at the center of each of these lines (perpendicular bisectors); the intersections of these perpendiculars form polygons. Within each polygon, the precipitation level selected is that recorded at the station located within it. The sides of the polygons and/or the watershed represent the limits of the area (and weight) assigned to each station. The area of each S_i polygon is determined planimetrically or numerically.

Following analysis of the two precipitation models simulated in CORDEX, data from the 50-km × 50-km grids of the two regional models were interpolated. These grids cover a variable surface area, but have only one geolocalized value, generally at the center of the grid. Spatial interpolation consists in calculating the value of a simulated field on a destination grid point (also known as the target grid), based on the values of the simulated field on the meshes of the source grid, which may be closer or further away. Several interpolation methods are available, based on weighting methods that take into account the distance between source and target grid points, or the altitude of the source grid point.

The Pettitt (1979) test was applied to determine the break year and the comparison of means in order to analyze the temporal stability of rainfall series over the 1961-2022 period, using Khronostat 1.01 software. This test was confirmed by Buishand’s test and Hubert’s segmentation. Pettitt’s method for detecting breaks in the stationarity of time series was used in this study to identify current trends in interannual rainfall compared with the average for the period under consideration. It consists in splitting the main series of N elements into two sub-series at each time t between 1and N − 1.

The main series is broken at time t if the two subseries have different distributions. The null hypothesis H0 of no break is tested using a non-parametric test.

The Pettitt test has been chosen for its use in numerous studies of stationarity change detection in West Africa ( Hubert & Carbonnel, 1987 ), its power, especially with regard to the break test on the mean ( Lubes-Niel et al., 1998 ), and its robustness ( Lubès et al., 1994 ). It was then applied in the Oti basin using Khronostat 1.01 software. Then (Xi) I = 1 at t and t + 1 at N belong to the same population ( Lubès et al., 1994 ). This test is based on the variable U_t and N is defined by:

$U_{t,N}={\displaystyle \underset{i=1}{\overset{t}{\sum }}{\displaystyle \underset{j=t+1}{\overset{N}{\sum }}{D}_{ij}}}$(2)

where D_ij = sgn(xi-xj) with sgn(Z) = 1 if (Z) > 0; 0 if Z = 0 and -1 if Z < 0. Let KN be the variable defined by the maximum absolute value of U_t,N for t varying from 1 to N − 1. If K denotes the value of KN taken from the series studied, under the null hypothesis, the probability of exceeding the K value is given approximately by: Prob(KN > K) is approximately equal to 2exp(-−6K²/N³ + N²).

For a given first-species risk a, if Prob(KN > K) is less than a, the hypothesis is rejected.

Hubert’s segmentation procedure is also used in this work. Its principle is to divide the rainfall series into segments (m) significantly different from the mean of the neighbouring segment(s). This approach involves searching for multiple changes in the mean. Thus, if the procedure does not produce an acceptable segmentation of order greater than or equal to 2, the main hypothesis of stationarity is accepted. By means of a specific algorithm, this segmentation provides one or more break dates separating contiguous segments whose means are significantly different ( Hubert et al., 1998 ). In fact, the hydrometeorological series segmentation procedure proposed by ( Hubert et al., 1989 ) is interpreted as a stationarity test, the main hypothesis being that the series studied is stationary.

The parametric t-test is used to compare two means and two sub-series of respective numbers ${\bar{X}}_1$ and ${\bar{X}}_2$ obtained on either side of the breakpoint. This parametric test is used to assess the significance of the regression coefficient.

The statistic used in this test is: $t=\frac{{\bar{X}}_1-{\bar{X}}_2}{\sqrt{\left(S_2\left(\frac{1}{n_1}+\frac{1}{n_2}\right)\right)}}$ (3)

${\bar{X}}_1$ = Average before year of breakup;

${\bar{X}}_2$ = Average after breakup;

n₁ = Number of elements in the first series;

n₂ = Number of elements in the second series.

where denotes the weighted variance of the whole group of the two samples, i.e.:

$S_2=\frac{{\displaystyle \underset{i=1}{\overset{n_1}{\sum }}{\left(X_{1,i}-{\bar{X}}_1\right)}^2}+{\displaystyle \underset{i=1}{\overset{n_2}{\sum }}{\left(X_{2,i}-{\bar{X}}_2\right)}^2}}{n_1+n_2-2}$(4)

In order to assess the performance of the two models (CCLM and REMO) of the CORDEX programme in reproducing as well as possible the interannual variability of annual rainfall on the one hand, and for the analysis of extreme rainfall events in the future on the other hand, the calculation of biases is necessary. The bias calculation method involves calculating the difference in means between simulated and observed rainfall ( Yèkambèssoun et al., 2016 ; Kodja, 2018 ). It is used to identify the best simulation model that best matches the analysis of future projections of mean rainfall and extremes ( Taïbi-Freddal, 2016 ; Yira et al., 2016 ). It evaluates the deviation of simulated rainfall from observations and is obtained by the following formula:

$Bias=\frac{PSim-PObs}{PObs}$(5)

where PSim = simulated precipitation and PObs: observed precipitation.

Several indices were used to analyze extreme rainfall events. These indices are relatively simple ways of assessing changes in extreme rainfall events that affect the natural and human environment ( Frich et al., 2002 ). It was in this context that the CLIVAR (Climate Variability and predictability) group ( World Meteorological Organization, 2012 ) of the WCRP (World Climate Research program) of the World Meteorological Organization proposed a list of twenty-seven (27) indices calculated from daily series of surface variables (rainfall and temperature). The purpose of these indices is to characterize extremes, frequency, amplitude and persistence. In the context of this study, the calculation of these indices is based on rainfall, and nine indices have been calculated to analyze extreme rainfall events as proposed by the ETCCDI (Expert Team on Climate Change Detection and Indices) expert group ( Table 2 ). This analysis method is commonly used to study extreme events ( Hountondji et al., 2011 ; Filahi et al., 2015 ; Djossou et al., 2020 ). These indices have already been used in several studies to analyze extreme climatic events, such as those by ( Gachon et al., 2005 ; Dubuisson & Moisselin, 2006 ; Issaou, 2014 ; Gbohoui et al., 2018 ).

Trends were analyzed using linear regression between the different rainfall indices and time over the entire series. The slopes estimated in this way were grouped into three classes ([p > 0.05] insignificant, [0.01 < p < 0.05] significant and [p < 0.01]) highly significant, indicating upward or downward trends. The threshold is defined on the basis of the Student’s t statistic, which is used to test the hypothesis of a slope equal to 0. The trend is therefore qualified as significant if the probability p of the t-test applied to the regression slope is less than 0.05, while it is not significant if it exceeds the 0.05 threshold.

To analyze extreme values, methods derived from the calculation of SPI “Standardized Precipitation Index” were used. SPI indices are used to measure the deviation from the long-term average at a point scale, based on station data. The Standardized Precipitation Index (SPI) ( McKee et al., 1993 ) is used in this study. It has advantages in terms of statistical consistency and has the ability to describe both short-term and long-term extreme rainfall events across different time scales ( McKee et al., 1993 ). It should be noted that the probabilistic nature of the SPI index allows it to be comparable between different sites ( McKee et al., 1993 ). Based on daily rainfall for multiple time scales, these standardized indices ( WMO, 2012 ) were calculated. In this case, the standardized index is based on the equiprobability of the transformation of rainfall values, aggregated at k-months, into normal standard values, with k generally set according to the objectives of the analysis, for example: $k=1,3,6,9,12,24,36$ months ( Cancelliere & Bonaccorso, 2009 ).

McKee et al. (1993) originally proposed a Gamma transformation for calculating the SPI index. This index is used at monthly or fortnightly time steps. Thus, the classic standardized index (noted Z) was used, but with a change of time step to the daily scale. It is a reduced-centered index, as indicated by the following formula:

$Z=\frac{Xi-M}{S}$(6)

With Xi: rainfall in year i; M: mean interannual rainfall over the reference period and S: standard deviation of interannual rainfall.

These indices are used to define thresholds for extreme rainfall events, based on the classification of threshold values for the intensity of extreme rainfall events in Table 3 of McKee et al. (1993) .

Source: Adapted from Aghrab (2003) and Mckee et al. (1993) .

In the context of climate change and the recurrence of extreme weather events, it is important to analyze trends in maximum daily rainfall so as to be able to take measures to prevent extreme rainfall events from damaging human and environmental systems.

Figure 2 illustrates the interannual variability of maximum daily rainfall in the Oti watershed in Togo.

Analysis of Figure 2 reveals interannual variability in maximum daily rainfall in the Oti watershed in Togo. This indicates a non-statistically significant upward trend in maximum daily rainfall, with a positive slope. The calculated p-value (0.20) is above the significance level at the alpha value threshold of 0.05. Maximum daily rainfall varies from 53.22 to 124.90 mm in the watershed. The maximum daily rainfall for the entire series was 124.9 mm in 1999. The lowest value for maximum rainfall is 53.22 mm recorded in 1976. It should be noted that it is the accumulation of rainfall maxima that amplifies extreme rainfall events in the Oti watershed in Togo. This in turn increases run-off followed by flooding.

Detection of breaks in the rainfall series by the Pettitt, Buishand and Hubert segmentation tests shows a change in the evolution of the rainfall chronicles. These three tests show a break in stationarity in the series. The null hypothesis of no break was rejected at the 95% and 90% confidence levels, with a probability of exceeding the critical value of 1.30E−02. Figure 3 shows the Pettitt test applied to the annual maximum rainfall series (1961-2022) in the Oti watershed.

Analysis of Figure 3 shows that, at the 99% threshold according to the Pettitt tests, annual rainfall levels show interannual variability, with a break in rainfall in 1998. The break was observed in 1998, showing a change in the rainfall series, dividing the series into two sub-periods, confirmed by Hubert segmentation with Scheffé test significance level of 1%. The first subperiod (1961-1998) is marked by a dry phase, with a mean rainfall of 72.72 mm and a standard deviation of 13.44. In fact, the 1961-1998 sub-period is divided into two phases, the first of which is characterized by a wet period (1961-1979) with an average rainfall of 79.19 mm. The second phase (1980-1998) was a dry period, with maximum daily rainfall dropping by 12.95 mm to an average of 66.24 mm. This decline is followed by a slight upturn in rainfall that begins in 1999 and continues until 2022. By contrast, the second sub-period (1999-2022) in the series is characterized by a wet phase, with mean precipitation of 86.26 mm and a standard deviation of 17.08, confirming a slight upturn in precipitation in the watershed. This upturn in precipitation is accompanied by extreme rainfall events.

The need for information on climate change at regional and local scales is one of the crucial concerns for assessing the impact of climate change on human and natural systems, and for developing sound national adaptation and mitigation strategies. The “Coordinated Regional climate Downscaling Experiment” (CORDEX) was set up to fill this gap and reduce uncertainties in the analysis of climate information. This program aims to produce reliable climate scenarios for impact studies over most of the world’s landmass, using Regional Climate Models (RCMs).

Figure 4 shows the interannual variability of observed and simulated rainfall in the Oti watershed in Togo, based on data from the African Cordex program.

Analysis of Figure 4 shows that the cumulative rainfall simulated by the models and that observed have almost the same trend. Figure 4 shows that both models overestimate observed rainfall in the study area. It should be noted that the CCLM4.8 model overestimates observations less than REMO. This is justified by the difference values of 1.95 for CCLM4.8 and 8.53 for REMO. Cumulative annual rainfall varies from 715.3 mm to 1762.2 mm for REMO, while the CCLM4.8 shows cumulative precipitation ranging from 646.7 to 1402.1 mm. At the same time, observations vary from 878.4 mm to 1233.3 mm.

The Oti watershed in Togo has a tropical Sudanian climate. It is the displacement of the Intertropical Front (ITF) that explains the seasonal nature of the climate, marked by alternating rainy and dry seasons. The rainfall regime provides information on the amount of monthly rainfall recorded in a given year and the periods of rainfall extremes in the watershed. The aim here is to compare the observed and simulated rainfall patterns of the models in order to see which model best represents the reality of observations in the Oti watershed in Togo.

Figure 5 shows the rainfall pattern of observed and simulated rainfall amounts from 1961 to 2005. This will enable the REMO and CCLM4.8 models and the mean climatological regime of the study area as a whole to be evaluated to see whether these models are suitable for the seasonal study.

Figure 5 shows that the two models are close to the observed rainfall and tend to represent the rainfall regime towards a unimodal rainfall regime for the observed rainfall in the Oti watershed, which peaks in August. The REMO model overestimates the observations, while the CCLM4.8 model underestimates them by 6.27 and 0.01 respectively. It should be pointed out that they reproduce the seasonal variability of rainfall within the same region and in the same sector in different ways, which is likely to influence the variability of cumulative rainfall and the choice of the appropriate model. From this observation, it should be noted that CCLM4.8 reflects the reality of observations and is suitable for seasonal analysis in the Oti watershed in Togo. The CCLM4.8 model was therefore chosen to project rainfall in the Oti watershed in Togo.

Climate projections are needed to analyze future rainfall extremes under the RCP 4.5 and RCP 8.5 scenarios for the period 2025-2069 in the Oti watershed. The CORDEX program in West Africa offers data from several models and concerns historical data and data projected to 2100. For this study, the analysis of extreme rainfall data is carried out taking into account the period 2025-2069 with the RCP 4.5 and RCP 8.5 scenarios for the Oti watershed in Togo. This period covers 44 years and is linked to the reference period (1961-2005) for evaluating the performance of the RCM.

Figure 6 shows the precipitation trends for 2025-2069 in the CCLM4.8 model according to the RCP 8.5 and RCP 4.5 scenarios.

Analysis of Figure 6 reveals interannual variability in future rainfall for the RCP 4.5 and RCP 8.5 scenarios and the reference rainfall for the period 2025-2069 in the Oti watershed in Togo. The RCP 4.5 and RCP 8.5 scenarios underestimate the reference rainfall, with differences of −0.15 for each RCP. This shows that with RCP 4.5 and RCP 8.5, rainfall in the CCLM4.8 model will decrease by 2025-2069.

Figure 7 shows daily rainfall variability in the Oti watershed in Togo from 2025 to 2069.

Figure 7 shows the daily rainfall variability projected from 2025 to 2069 in the Oti watershed in Togo. The simulated daily rainfall varies from 1 mm to 137.95 mm with an average of 2.36 mm for the RCP 8.5 scenario and from 1 to 196.52 mm with an average of 2.38 mm for the RCP 4.5 scenario. The change in daily rainfall by 2069 will therefore be accompanied by an increase in run-off and, consequently, recurrent flooding over this period.

A study of the projected interannual variability of precipitation from 2025-2069, using rainfall indices and their smoothing by five-year rolling averages in the Oti basin ( Figure 8 ) with RCP 8.5 and RCP 4.5, shows that positive anomalies characterize wet years and negative anomalies present dry years.

Analysis of Figure 8 reveals that the period from 2025 to 2069 is marked by two rainfall phases in the two scenarios (RCP 8.5 and RCP 4.5). For RCP 8.5, the first phase runs from 2025 to 2038, reflecting a wet period. Of the 14 years in this period, 10 have recorded positive standardized precipitation indices and only 4 negative anomaly indices. Average rainfall during this period for RCP 8.5 is 962.05 mm. At the same time, for RCP 4.5, the first phase runs from 2025 to 2037, i.e. 13 years with 8 years of positive anomalies and 5 years of negative anomalies. The average rainfall over this period for RCP 4.5 is 954.56 mm. Generally speaking, positive anomalies are frequently observed during this wet period.

On the other hand, the second phase is marked by a dry period for both scenarios (RCP 8.5 and RCP 4.5), reflecting the rainfall deficit. The dry period runs from 2039 to 2069, i.e. 30 years of rainfall deficit for the RCP 8.5 scenario. This period is characterized by a high frequency of negative anomalies. Over the 30 years, 19 years have negative anomalies and 11 are positive. Average rainfall during this dry phase is 818.52 mm. RCP 4.5 is also marked by a long dry period from 2038 to 2069, with average rainfall of 837.46 mm. Over the 31-year period, 21 years recorded negative anomalies and 10 years positive anomalies.

In the case of the flood risk study, it is important to project the rainfall field in the Oti watershed in order to see the sectors that will record the most rainfall by 2069. This makes it possible to determine the amount of rain that will fall on the watershed under both RCP 8.5 and RCP 4.5.

Figure 9 shows the distribution of observed and simulated mean annual rainfall fields in the Oti watershed for 1961-2005 and 2025-2069.

Analysis of Figure 9 shows that rainfall increases from north to south, indicating a variation in mean annual rainfall in the watershed. The reference rainfall varies from 975 mm in the north to 1131 mm in the south and is distributed along an increasing north-south gradient. RCP 8.5 varies spatially from 816 mm in the north to 924 mm in the south and RCP 4.5 varies from 825 mm to 935 mm. In fact, the south is the wettest part of the basin, with average annual rainfall in excess of 1000 mm. The north-east (Mandouri) receives little rainfall, with average annual rainfall always below 1000 mm.

Generally speaking, the south of the watershed (downstream) will record more rainfall than the north (upstream) according to the RCP 4.5 and RCP 8.5 scenarios of this model. The models predict an increase in rainfall from the north to the south of the watershed, indicating a more accentuated decrease near Mandouri in the northeast and to the northwest at Dapaong.

Table 4 shows the nine climatic indices calculated to characterize extreme rainfall events observed in the Oti watershed in Togo. These are PRCPTOT, SDII, R10 mm, R20 mm, R95P, R99P, Rx1day, Rx5days and CWD.

Analysis of Table 4 shows that total annual rainfall in rainy days (PRCPTOT) in the Oti catchment in Togo is increasing and varies from 665.5 mm to 1373.5 mm. The positive value of the slope (0.93) confirms the non-statistically significant increase, with an estimated p-value of 0.40. In contrast, the daily rainfall intensity (SDII) showed a non-significant decrease with a p-value of 0.27. The indices for the number of days with rainfall of more than 10 mm (R10 mm) and 20 mm (R20 mm) show a non-significant downward trend, with p-values of 0.37 and 0.70 respectively. Rainfall amounts associated with the 95th percentile vary from 28.4 to 511.1 mm and are considered to be very heavy rainfall events in the study area. These 95th percentile rainfall amounts show a non-significant downward trend, with a p-value of 0.83. On the other hand, at the 99th percentile, rainfall amounts vary from 1 to 260.4 mm and are described as extremely heavy in the catchment area. This index shows a statistically insignificant upward trend, with a p-value of 0.49.

There is an upward trend in the maximum total rainfall indices for one rainy day (Rx1day) and five consecutive rainy days (Rx5days). The respective p-values are 0.20 and 0.90. Analysis of the maximum number of consecutive days of rainfall in excess of 1 mm in the watershed provides a better understanding of the extreme rainfall phenomenon. The CWD shows a non-significant upward trend, with a p-value of 0.34.

Table 5 shows the characteristics of the trends in extreme rainfall events simulated using nine climate indices under the RCP 8.5 and RCP 4.5 scenarios from 2025 to 2069.

Analysis of Table 5 shows that, according to the CCLM4.8 model, climate indices such as Rx1day, SDII and R99P will show a non-significant upward trend from 2025 to 2069 for RCP 8.5, as shown by their slope and p-value. The rest of the indices will show an insignificant downward trend. For the RCP 4.5 scenario, on the other hand, the Rx1day, Rx5days and R99P will rise over the same period and the rest of the climate indices will fall.