^{1}

^{*}

^{2}

^{1}

^{3}

This study assessed the rainfall trends and changes over Mono river basin under the highest greenhouse gas emission scenario RCP8.5. Simulations of eight regional climate models (RCMs) provided by Africa-CORDEX program were considered. To analyze the performance of a set of regional climate models, the MBE (mean bias error), the RMSE (root mean square error), the volume bias (VB), correlation coefficient (R
^{2}) and the t-Test statistics were calculated. The precipitation concentration index (PCI), Mann-Kendall trend test, Theil-Sen’s slope estimator (
*β*), and relative percentage change methods were also adopted for data analysis. Changes from the baseline period 1981-2010 were computed for far future (2061-2090 and 2071-2100). As results, the analysis herein highlighted the multi-models’ mean ability to simulate the Mono river basin rainfall adequately. Two distinct patterns emerged from the calculated PCI indicating that stations in southern basin will have moderate, irregular, and strongly irregular rainfall concentrations, whereas stations in northern basin will have irregular and strongly irregular rainfall concentrations. Significant declining in the rainfall was detected in most stations for the future period. The evolution of the monthly average rainfall amounts will be broadly characterized by a decrease and increase between 32.4 and 12% with late rainy seasons. It is understood that future changes in rainfall distribution and trends will affect the availability of water for crops, which should affect the productivity of rain fed agriculture.

Global warming and climate variability are emerging as the foremost environmental problems in the 21st century, particularly in developing countries. The research from scientific community on climate change has estimated and predicts that the frequency, intensity and extent of these phenomena (drought on continental regions and violent cyclonic storms) are and will be likely to increase across the world which will increase the exposure of populations in the years to come [

Africa is one of the most vulnerable continents to climate change across the world. This situation will aggravate by different interactions between population and ecosystems and low adaptive capacity [

It is thus common that the developing countries’ low-income populations are likely to be affected by factors related to global warming, as West Africa one of the regions in the world that are most vulnerable to climate change. This is particularly true for rural areas in West Africa where agriculture is the most prominent instrument for securing income and overcoming poverty. This region experiencing exponential population growth is already facing the consequences of climate change through gradual land degradation and loss of croplands and ecosystem services [

For the specific case of Togo, according to the report of Togo Republic [

To better understand climate change and its consequences, different types of models have been developed. The first generation of models is the General Circulation Models (GCMs). The second is the regional climate models (RCMs). Global climate models (GCMs) face enormous difficulties and cannot represent the climate in the region because of their very rough spatial resolution (200 to 300 km). Therefore, those models are only important when it comes to representing climate on a very large scale of space. But at regional level, they present many limitations [

The aim of this study is therefore threefold: 1) to evaluate the performance of some RCMs in simulating rainfall over Mono basin in Togo (West Africa); 2) to exanimate trends of future rainfall over Mono basin in Togo (West-Africa).; 3) to project future changes in rainfall in the period 2061-2090 and 2071-2100 using a regional climate model (Africa-cordex) under highest emission scenarios (i.e. RCP8.5).

The study area, used data and methods are described in the next Section 2. Section 3 was only reserved to the results. The Section 4 represents the discussions and conclusion is in the Section 5.

Located on the coast of the Gulf of Guinea in West Africa, Togo (TG) has a surface area of 56,600 km^{2}, bordering the Atlantic Ocean in the south, Burkina Faso in the north, Benin in the east and Ghana in the west. The Togolese population was estimated at 6.3 million inhabitants in 2007 according to the report of UNDP in 2007. The main important rivers are Oti, Mono, Kara, Keran, Koumongou, Anie, Zio and Haho. Mono basin, which is the study area, is located in the Gulf of Guinea region, more precisely between 6˚16'N and 9˚20'N and 42˚E and 20˚25'E (^{2}, the Mono river is the major one of Togo [

average discharge of the main watercourse of this basin is approximately 106 m^{3}/s at Nangbeto hydrometric station from 1981 to 2010. The annual rainfall average is 1915 mm from 1981 to 2010. The mean maximum temperature for the whole basin varies between 28.4˚C and 33.7˚C while the mean of minimum temperature fluctuates between 24.3˚C and 27.8˚C from 1981 to 2010. According to the report of WAEMU in 2006, the population of the basin is more than two million, with an annual increase of 2.9%. This population also distributed at high densities in the south of the basin has as main activities, agriculture (mainly rain fed); in the lower valleys, fishing and salt-farming are the major activities [

Two sources of data have been used in this paper. The first one is CORDEX program archives. The CORDEX program archives outputs from a set of RCM simulations over different regions in the world. In this study datasets from CORDEX Africa are accessed from https://esgf-data.dkrz.de/projects/esgf-dkrz/. The datasets are quality controlled. The spatial grid resolutions of all CORDEX Africa RCMs were set to longitude 0.44˚ and latitude 0.44˚ using a rotated pole system coordinate. These models operate over an equatorial domain with a quasi-uniform resolution of approximately 50 km by 50 km. The components of the model are based on the scenario or Representative Concentration Pathway (RCP). Detailed description of CORDEX RCMs and their dynamics as well as their physical parameterization are given by [

In this study, to evaluate the accuracy of the estimated data, from the models described above, the following statistical estimators were used: MBE (Mean Bias Error), RMSE (Root Mean Square Error), MPE (mean percentage error) and the correlation coefficient (R^{2}), to test the linear relationship between simulated and observed values. For higher modeling accuracy, these estimators should be closer to zero, and the correlation coefficient, (R^{2}), should approach to 1. However, these estimated errors provide reasonable criteria to compare models but do not objectively indicate whether the estimates from a model are statistically significant. The t-Test statistic allows models to be compared and at the same time it

Global Model Name | Institute ID | Model Short Name | RCM Model Name |
---|---|---|---|

GFDL-ESM2M | NOAA-GFDL | NOAA | SMHI-RCA4 |

NorESM1-M | NCC | NCC | SMHI-RCA4 |

MPI-ESM-LR | MPI-M | MPI | SMHI-RCA4 |

MIROC5 | MIROC | MIROC | SMHI-RCA4 |

IPSL-CMA5-MR | IPSL | IPSL | SMHI-RCA4 |

EC-EARTH | ICHEC | ICHEC | KNMI-RCAMO22T |

CNRM-CM5 | CRNM-CERFACS | CNRM | SMHI-RCA4 |

CanESM2 | CCCma | CCCMA | SMHI-RCA4 |

indicates whether a model’s estimate is statistically significant at a particular confidence level. So, the t-Test was carried out on the models to determine the statistical significance of the predicted values. Lawin et al. (2019) used these methods in a similar study [

Mean Bias Error (MBE)

To evaluate the accuracy of the predicted data from the models described above, this test provides information on the long-term performance of a model. The validation period is from 1999 to 2008. A low MBE value is desired. A negative value gives the average amount of underestimation in the calculated value. The MBE value is computed by Equation (1). The subscript i refer to the i^{th} value of the daily rainfall; n is the number of the daily rainfall. The subscripts “s.” and “o.” refer to the simulated and measured daily climatic parameters (rainfall) values. A percentage error between −10% and +10% is considered acceptable [

MBE = 1 n ∑ 1 n ( X i s − X i o ) × 100 (1)

Root Mean Square Error (RMSE)

The value of RMSE given by Equation (2) is always positive and is zero in the ideal case. The RMSE gives information on the short-term performance of the correlations by allowing a term-by-term comparison of the deviation between the simulated and observed values. The smaller the value, the better the model’s performance is [

RMSE = [ 1 n ∑ 1 n ( X i s − X i o ) 2 ] 1 2 (2)

The correlation coefficient (R^{2})

In statistics literature, it is the proportion of variability in a data set that is accounted for by a statistical model, where the variability is measured quantitatively as the sum of square deviations. A high value of R^{2} is desirable as this shows a lower unexplained variation. R^{2} is a statistic that gives some information about the goodness-of-fit of a model. In regression, the R^{2} coefficient of determination is a statistical measure of how well the regression line approximates the real data points. An R^{2} of 1.0 indicates that the regression line perfectly fits the data, which is never valid in any climatic parameters estimation model [

The relative error (E)

The relative error (EP and ET), respectively for precipitations and temperature of the estimated values of the rainfall or temperature (minimum or maximum) may be calculated from the following:

EP = X o − X s X o (4)

ET = X o − X s (5)

Volume bias (VB)

Volume bias (VB) which indicates over and underestimations is representing by equation:

VB = ∑ t = 1 n ( RCMs − OBS ) ∑ t = 1 n ( OBS ) (6)

RCMs and OBS designed the values of observed and simulated data for validation period (1999-2008).

The t-Test statistic (t)

The tests for mean values, the random variable t with n − 1 degrees of freedom may be written here as follows. The smaller values of t-statistic the better the performance of modeling.

t = [ ( n − 1 ) ( MBE ) 2 ( RMSE ) 2 − ( MBE ) 2 ] 1 2 (7)

Mann-Kendall Trend Test:

Many techniques can be used for analyzing the series trends, yet the most commonly used technique by meteorologists is the Mann-Kendall (MK) test [

S = ∑ i = 1 n − 1 ∑ j = i + 1 n sgn ( X j − X i ) (8)

Positive S value indicates an increasing trend and negative value indicates a decreasing trend in the data series. The sign function is given as:

sgn ( X j − X i ) = { 1 if X j − X i > 0 0 if X j − X i = 0 − 1 if X j − X i < 0 (9)

In cases where the sample size n > 10, the statistics S is approximately standard normal distribution with the mean zero and variance is denoted by the following:

V A R ( S ) = n ( n − 1 ) ( 2 n + 5 ) − ∑ i = 1 m t i ( t i − 1 ) ( 2 t i + 5 ) 18 (10)

where n is the number of data points; t_{i} are the ties of the sample time series; and m is the number of tied values (a tied group is a set of sample data having the same value). Then Equations (8) and (10) were used to compute the test statistic Z from the following Equation (11):

Z = { S − 1 V A R ( S ) if S > 0 0 if S = 0 S + 1 V A R ( S ) if S < 0 (11)

A positive value of Z indicates an upward trend; a negative value indicates a downward trend, and a zero value indicates no trend.

Sen’s slope estimator and relative percentage change (RPC):

β which is Sen’s slope estimator, sign reflects data trend reflection, while its value indicates the steepness of the trends (Equation (12)):

β = Median ( x j − x k j − k ) (12)

Equation bellow (13) was used to compute the relative percentage change (RPC) of the annual as well as monthly rainfall:

RCP = ( n β X m e a n ) ∗ 100 (13)

where n is the length of the trend period; β is the magnitude of the trend slope of the time series determined by the Theil-Sen median estimator; and X is the absolute average value of the time series. This method was used in a similar study by Tabari et al. (2013) [

Precipitation Concentration Index (PCI)

In order to study monthly and annual variability of rainfall in the study area of each weather station, a Precipitation Concentration Index (PCI) was used. The guidelines for interpretation of PCI are presented in

PCI = [ ( ∑ i = 1 12 P i 2 ) ( ∑ i = 1 12 P i ) 2 ] ∗ 100 (14)

PCI Value | Distribution of Precipitation |
---|---|

<10 | Uniform precipitation distribution (i.e., low precipitation concentration) |

11 - 15 | Moderate precipitation concentration |

16 - 20 | Irregular distribution |

>20 | Strongly irregular distribution (i.e., high precipitation concentration) |

where P_{i} is the monthly precipitation in month i. An interpretation of the PCI classification as suggested by Oliver (1980) is represented in

Changes in precipitation from the reference period:

Finally, the changes from the reference period are assessed as shown by Equation (15) for precipitations:

I p = ( FF − HIST HIST ) × 100 (15)

where FF and HIST represent respectively the mean for the far future (2061-2090 and 2071-2100) and the historical or reference period (1981-2010) [

The seasonal averages as well as the annual averages were calculated using the Climate Data Operator Commands and the free software R is used to compute all the statistical parameters and the plots presented in the following [

This section is divided into two main parts. The first part focuses on statistical tests of the performance of the eight models used. The second part analyses the results of the seasonal and annual distribution of rainfall, changes and trends in rainfall over Mono basin in Togo.

Summarized values of statistical parameters over the period 1999-2008 are given by ^{2}, t-test and VB. The MBE between RCMs simulations and observations are from −18.75% to 6.72% for precipitation. The largest MBE is observed with models MIROC5 (mean value −18.75) and the smallest MBE is given by the multi-models’ mean (Model-Mean) where the value is −2.33. With regards to RMSE, the models MIROC5, GFDL-ESM2M and MPI-ESM-LR exhibited the largest biases for precipitation. On the other hand, the highest values of R^{2} for rainfall are presented by the models EC-EARTH, IPSL-CMA5-MR, NorESM1-M and Model-Mean with 0.863, 0.899, 0.921 and 0.952, respectively. Among the four latter models, the t-test lowest value is given by the Model-Mean (1.87 mean value) followed by IPSL-CMA5-MR, NorESM1-M and EC-EARTH, respectively. However, the model EC-EARTH was excluded because of the t-test value was high (4.35 mean value). Therefore, the three models can be considered as the best to estimate rainfall.

Model name | MBE | Rainfall | |||
---|---|---|---|---|---|

RMSE | R^{2} | t-test | VB^{ } | ||

GFDL-ESM2M | −8.17 | 5.63 | 0.723 | 3.42 | −8.13 |

NorESM1-M | 5.84 | 2.52 | 0.921 | 2.61 | 15.01 |

MPI-ESM-LR | −4.56 | 8.23 | 0.812 | 2.18 | −13.33 |

MIROC5 | −18.75 | 6.52 | 0.736 | 2.13 | −3.66 |

IPSL-CMA5-M | −3.82 | 2.13 | 0.899 | 7.13 | 7.55 |

EC-EARTH | −6.33 | 4.11 | 0.863 | 4.35 | 13.33 |

CNRM-CM5 | −6.13 | 1.63 | 0.833 | 6.71 | −7.55 |

CanESM2 | 5.35 | 4.32 | 0.679 | 5.52 | 6.52 |

Model-Mean | −2.33 | 1.31 | 0.952 | 1.87 | 11.53 |

Models | Relative error | J | F | M | A | M | J | J | A | S | O | N | D |
---|---|---|---|---|---|---|---|---|---|---|---|---|---|

NOAA | E (mm) | 0 | 0 | 1.1 | −1.9 | 3.7 | 0.8 | 1.3 | 1.6 | 3.7 | 1.9 | 0.3 | 0 |

E (%) | 0 | 0 | 6 | −15 | 29 | 5 | 7 | 8 | 18 | 15 | 5 | 0 | |

NCC | E (mm) | 1.6 | 1.4 | 1.1 | 3.7 | 3.4 | 2.3 | 2.6 | 2.2 | 3.9 | 0.8 | 0.9 | −1.3 |

E (%) | 6 | 9 | 6 | 51 | 34 | 14 | 14 | 10 | 15 | 12 | 37 | −6.3 | |

MPI | E (mm) | 0.4 | 1.0 | 1.1 | 0.3 | 1.2 | 2.5 | 3.8 | 2.7 | 2.6 | 1.1 | 2.1 | 1.3 |

E (%) | 4 | 12 | 7 | 4 | 11 | 16 | 20 | 12 | 10 | 5 | 11 | 3 | |

MIROC | E (mm) | 1.2 | 2 | 1.3 | 0.5 | 0.5 | 3.1 | 2.1 | 3.2 | 4.5 | 1.4 | 0 | 1.1 |

E (%) | 6 | 6 | 15 | 15 | 17 | 20 | 12 | 12 | 16 | 6 | 0 | 2 | |

IPSL | E (mm) | 1.7 | 1.7 | 1.2 | 3.0 | 4.8 | 3.9 | 3.3 | 1.0 | 1.3 | −0.8 | 0.7 | 1.8 |

E (%) | 8 | 8 | 4 | 32 | 37 | 24 | 22 | 13 | 17 | −4 | 3 | 2 | |

ICHEC | E (mm) | 1.8 | 0.8 | 1.3 | −2.7 | 3.8 | −4 | 0.5 | 1.3 | 1.8 | 2.2 | −0.9 | 3.1 |

E (%) | 9 | 3 | 4 | −28 | 31 | −26 | 31 | 27 | 19 | 11 | −15 | 4 | |

CNRM | E (mm) | 2.1 | 0.2 | 1.4 | 2.0 | 2.9 | 2.2 | 1.3 | −2 | 3.3 | 3.2 | 0.4 | 0.5 |

E (%) | 2 | 26 | 5 | 33 | 28 | 13 | 18 | −22 | 15 | 13 | 11 | 6 | |

CCCMA | E (mm) | 1.2 | 0.8 | 1.3 | 3.5 | 3.6 | −3 | 2.5 | 1.4 | 1.2 | 0.3 | −1.8 | 0.1 |

E (%) | 4 | 2 | 3 | 34 | 25 | −16 | 28 | 12 | 9 | 14 | −8 | 12 | |

Model-Mean | E (mm) | 0 | 0.1 | 0 | 1.9 | 1.3 | 2.8 | 3.1 | 2.5 | 2.3 | 2.7 | 0.2 | 0.3 |

E (%) | 0 | 9 | 0 | 11 | 10 | 8 | 9 | 9.3 | 10 | 7 | 2 | 3 |

performance. In fact, the (E) between RCMs simulations and observations varies between −15% and 37% for precipitation. We notice from the above discussion that the models are performed differently for different months. Some models performed well, while others performed poorly, and vice versa. This is explained by compensation or cancellation of the biases between the different RCMs.

In the next section the multi-models mean only will be considered. To bias-correct the model future projection, the cumulative distribution function matching (CD FM) of the model in the reference period is used to identify the corresponding percentile values of the future period [

Seasonal Changes of Rainfall

Seasonal variations were investigated to provide more detail about the changes in climatological monthly precipitation, averaged across Mono basin using the multi-models mean. The analysis of the results obtained by the rainfall changes is presented in

results from the multi-models mean suggest a decrease in precipitation in general, with no significant differences between the stations and the horizons considered. Thus, the magnitude of the decrease peaks in rainy season (April to October), with the smallest increase in dry season (November to March).

Future Annual and seasonal rainfall patterns

This section analyzes future annual and seasonal rainfall patterns. According to the rainfall data analysis, the long-term mean annual rainfall from 2061 to 2100 in the region is 1784.3 mm. However, as shown in

The PCI is computed and presented in the same

Station | PCI^{1} | CV^{1} | Mean^{1} | PCI^{2} | CV^{2} | Mean^{2} |
---|---|---|---|---|---|---|

Atakpamé | 21.09^{d} | 0.11 | 233.5 | 15.18^{c} | 0.19 | 1923.54 |

Tabligbo | 11.53^{b} | 0.22 | 158.04 | 33.52^{d} | 0.22 | 1643.01 |

Anié | 31.86^{d} | 0.18 | 222.17 | 11.51^{b} | 0.38 | 1858.01 |

Sokodé | 16.52^{c} | 0.12 | 159.41 | 29.06^{d} | 0.55 | 1543.52 |

Tchamba | 33.08^{d} | 0.34 | 133.50 | 59.34^{d} | 0.52 | 1113.52 |

Soutouboua | 26.54^{d} | 0.38 | 177.38 | 38.56^{d} | 0.48 | 1018.02 |

^{a}Uniform distribution, ^{b}Moderate distribution, ^{c}Irregular distribution, ^{d}Strong irregular distribution. ^{1}and ^{2}represent Annual and seasonal mean rainfall, respectively.

precipitation distribution and a strongly irregular distribution. The highest values of PCI were recorded at the Tabligbo rain gauge station. The PCI values for humid tropical climate stations (Sokodé, Sotouboua and Tchamba) were in the same range (between 16.36 and 59.34) in the climate subequatorial but with a predominant occurrence of irregular and strongly irregular distributions. The highest values of PCI were recorded at the Soutouboua and Tchamba rain gauge stations. As would be expected, these two stations are on the northwestern and northeastern flank of the study area where the mean annual total rainfall is lowest in Togo.

Rainfall Trend Analysis

The result of the MK test was applied to analyze the future mean annual and seasonal rainfall trend for the period of 2061-2100 for six stations in the Mono basin in Togo. Similarly, Sen’s slope and percentage change were used to examine the magnitude and change of the future variables. The results of the MK trend, Sen’s slope and percentage change of rainfall are given in

Station | Z^{1} | β^{1} | % Δ^{1} | Z^{2} | β^{2} | % Δ^{2} |
---|---|---|---|---|---|---|

Atakpamé | 1.65 | 2.2 | 8.45 | 0.01 | 0.4 | 37 |

Tabligbo | −0.18a | −7.18 | −18.76 | −0.04a | −0.43a | −18.67 |

Anié | −0.85 | −0.45 | −2.85 | −0.25 | −0.78 | −32.6 |

Sokodé | −0.44a | −9.32 | −30.23 | −2.13a | −0.89a | −67.8 |

Tchamba | −0.15a | −4.8 | −12.3 | −1.3 | −0.35 | −68.3 |

Sotouboua | −0.27a | −6.09 | −17.7 | −2.56 | −2.48 | −22.35 |

^{a}and % Δ denote significant level at 5%, and percentage change, respectively. ^{1}and ^{2}denote Annual and Seasonal mean rainfall, respectively.

(0.89 mm/year) with percentage changes of 18.67 and 67.8, respectively (

We notice from the above discussion that the models are performed differently for different months. Some models performed well, while others performed poorly, and vice versa. This is explained by compensation or cancellation of the biases between the different RCMs. However, the analysis herein highlighted the multi-models’ mean ability to simulate the Mono river basin rainfall adequately and therefore can be used for assessment of future climate projections for this region. To have better comprehension of results of in this paper it is important to look at the results of other regions and countries. The disparity observed in the evolution of precipitation could be due to the type of forcing models or the convection pattern used in West Africa [

In this study, the precipitation decreases projected by the average multi-models are consistent with other studies in West Africa: Kouakou et al. [

However, we note that the divergence of climate change patterns is still uncertain over the world. Some climate projections suggest an increase in the rainfall and others indicate a decrease. Signiﬁcant positive trends have been observed in the USA [

The results indicate that the rainfall coefficient of variability is high in the area with low annual rainfall. This is supported by Birara et al. 2018 [

In RCP8.5 context, the agricultural sector, main economic activity in Mono basin watershed would be affected. Furthermore, a significant decrease in water availability (surface water and groundwater) due to a decrease in rainfall showed by [

In conclusion, the study analyzed the performance of a set of regional climate models. Some models performed well, while others performed poorly, and vice versa. This is explained by compensation or cancellation of the biases between the different RCMs. The PCI, Mann-Kendall trend test, Theil-Sen slope estimator and RPCs were calculated to investigate trends and changes in rainfall over Mono basin in Togo at annual, seasonal, and monthly time scales for the 1981-2010 and 2061-2100 periods.

Stations in the south have moderate, irregular, and strongly irregular rainfall concentrations, whereas stations in the North of basin have irregular and strongly irregular rainfall concentrations. In addition, annual and the monthly Mann Kendall trends indicate that both significant and insignificant negative and positive changes in rainfall have occurred across the Mono basin in Togo. With regard to the changes of the seasons, a decrease is found for the precipitations in general. In addition, it indicates that Mono Basin will be characterized by late rains in the future. It is understood that future changes in rainfall distribution and trends will affect the availability of water for crops, which should affect the productivity of rain fed agriculture. In short, further work is needed in order to improve the future impact assessments of the climate change on water resources and on some human activities such as agriculture, energy, fisheries and breeding.

We acknowledge the financial support from the Deutscher Akademischer Austausch Dienst (DAAD).

The authors declare no conflicts of interest.

Lamboni, B., Emmanuel, L.A., Manirakiza, C. and Djibib, Z.M. (2019) Variability of Future Rainfall over the Mono River Basin of West-Africa. American Journal of Climate Change, 8, 137-155. https://doi.org/10.4236/ajcc.2019.81008