This study investigated over Côte d’Ivoire, a West African country located between (4˚N-11˚N, 8˚W-2˚W), covering a surface area of 322,462 km² (about 1% of the total surface area of the African continent). The country is bordered to the south by the Gulf of Guinea, to the east by Ghana, to the west by Liberia and Guinea, and to the north by Mali and Burkina Faso. There are three main types of climate in Côte d’Ivoire equatorial, tropical and mountainous [18] . The equatorial climate, located in the southern part of the country, is characterized by a high humidity (82% in Abidjan), a mean temperatures (about 26˚C) and abundant rainfall (about 1922 mm/year in Abidjan) [19] . This region is governed by two dry and wet seasons during the year. The tropical climate predominates in the northern part of the country. With mean annual temperatures varying around 26˚C, with a relatively higher daily temperature ranges around 32˚C

and low humidity (63% in Korhogo) compared to the southern region. This region is dominated by a long dry season and a short rainy (wet) season, as well as the presence of a cool and dry wind originating from the Sahelian regions known as the “harmattan” blowing during December-February. Mean annual rainfall is generally less than 1300 mm/year. The mountainous climate in the western part of the country is characterized by abundant annual rainfall, compared to the northern region with two seasons: a rainy and a dry season. Mean annual temperatures vary around 24˚C, and relative humidity remains very high throughout the year, reaching about 98% in Man [20] .

In this study, we used simulated daily precipitation data derived from an ensemble of 14 simulations of the Coordinated Regional Climate Downscaling EXperiment (CORDEX) phase 2 [21] [22] available at the Earth System Grid Federation (ESGF) website (https://esg-dn1,nsc,liu,se/projects/esgf-liu/) with a horizontal resolution of 0.44˚ (about 50 km). These simulations are composed of 8 Regional Climate Models (RCMs) forced by 7 Global Climate Models (GCMs) from the Coupled Intercomparison Project phase 5 (CMIP5) (see Table 1). The data span the historical (1950-2005) and future (2006-2100) periods, under RCP4.5 and RCP8.5 radiative forcing scenarios. They are useful for providing a regional climate analysis for many meteorological parameters (temperature, precipitation, etc.,) for a given region (e.g. Africa) and at different periods [23] [24] . Moreover, CORDEX simulations have been evaluated in terms of their ability to simulate mean climate (precipitation and temperature) as well as climate extremes

in Africa with satisfactory results [23] [24] [25] [26] . In addition, high resolution (about 5 km) daily precipitation data from the “Climate Hazards Infrared Precipitation” (CHIRP) of the University of California [27] covering the period 1981-2005, are also used as observation data. These data have the advantage of providing reliable information on the climatic variability over the study area due to their high resolution [5] [6] .

The analysis of climate zoning and its projections in Côte d’Ivoire is carried out using the multi-model ensemble mean of the 14 CORDEX simulations based on the grid clustering method. The multi-model ensemble approach consists of using the mean of an ensemble of simulations to reduce the difficulties associated with characterizing the uncertainties existing between them. Clustering is a multivariate statistical technique designed to explore structures within a data set whose prior properties are unknown [28] . This technique is regularly used in various forms to determine rainfall structures and climatic regions in West Africa, particularly in Côte d’Ivoire [14] [15] [29] . This technique aims to assign data to significant classes by increasing the similarity within each cluster while maximizing the differences between clusters. Although various clustering algorithms exist, we used hierarchical clustering on principal components (HCPC). The advantage of this approach is to combine the three standard methods used in multivariate data analysis, in particular, the principal component analysis (PCA), the hierarchical clustering and the k-means [30] . Taken individually, k-means and hierarchical clustering are less effective when applied to a large number of dimensions. However, the results are better as the dimensions decrease [31] . In the HCPC, PCA is used to reduce the dimensions of the multivariate input data for two or three principal components, or even more, depending on the selected criterion, while losing as much information as possible. PCA axes use the Kaiser criterion [32] stipulating that only axes associated with eigenvectors and eigenvalues greater than or equal to 1 are considered significant. The hierarchical classification of the variables deduced from the PCA will define the different clusters over the historical (1981-2005), the near future (2031-2060) and the far future (2071-2100) periods under RCP4.5 and RCP8.5 scenarios. The similarity between the variables for the different climatic zones derived from CORDEX and CHIRPS during the historical period is evaluated using Spearman correlation coefficients. The similarity is significant if the p-value associated with the correlation coefficient is less than 0.05 [33] . Spatial variability of the climate zones over future periods is estimated by the variation rate (in percentage %), following Equations (1) and (2):

Rate ( % ) = Pop zone ( i ) projection − Pop zone ( i ) historical Pop Total * 1 00 (1)

Variation = ( Pop zone ( i ) projection − Pop zone ( i ) historical ) ∗ surface grid (2)

Rate: rate of change in the area of zone i, for a given period,

Pop zone ( i ) projection : number of grids in zone i, over the projection period,

Pop zone ( i ) historical : number of grids in zone i, over the historical period,

Pop Total : total number of grids over the study area,

Variation: variation in the area of zone i, for a given period,

surface grid : surface area of a grid (50 km * 50 km).

The statistical descriptions of the variables are summarized in Table 1(a) and Table 1(b). The results reveal that monthly rainfall over the period 1981-2005 varies between 0.95 and 462.81 mm for CHIRPS variables (respectively between 0.17 and 432.9 mm for CORDEX variables) while monthly mean rainfall varies between 9.62 and 184.07 mm for CHIRPS (respectively between 6.59 and 244.1 mm for CORDEX).

1) Principal component analysis (PCA)

Table 3(a) and Table 3(b) present the eigenvalues and variances of the various factors (axes) obtained from the principal component analysis (PCA) of the mean daily cumulative precipitation over the period 1981-2005. Principal components with eigenvalues equal or greater than 1 were considered significant according to the Kaiser criterion [32] . Only the first ten components fulfil this criterion for CHIRPS variables and the first seven for CORDEX. The results indicate that these components explain 81.81% and 92% of the total variance of CHIRPS and CORDEX rainfall, respectively. The principal components of the axes dim.1 to dim.10 for CHIRPS and dim.1 to dim.7 for CORDEX are retained for analysis (see Table 2).

2) Hierarchical Classification of Principal Components (HCPC)

The hierarchical classification of principal components (HCPC) of CHIRPS and CORDEX variables during the period 1981-2005 defines three (3) distinct major classes in Côte d’Ivoire (Figure 2(a) and Figure 2(b)). The first class (C1) includes 23 grids for CHIRPS and 18 grids for CORDEX, located along the coast between latitudes 4.75˚N and 6.25˚N for CHIRPS and latitudes 4.75˚N and 5.25˚N for CORDEX. The second class (C2) comprises 50 grids for CHIRPS and 39 grids for CORDEX. This class includes the central grid for CORDEX with a maximum latitude of 7.75˚N, and the central and northeastern grids with a maximum latitude of 8.75˚N. Finally, the third class (C3) consists of 39 grids for CHIRPS and 55 grids for CORDEX. It contains grids in the northern zone with minimum latitudes of 8.75˚N and 7.75˚N respectively.

Figure 3 shows the spatial distribution of the homogeneous climatic zones obtained from the CHIRPS (a) and CORDEX (b) principal components hierarchical classification analysis. These representations show that Côte d’Ivoire can be subdivided into three (3) homogeneous climatic zones:

- The first region corresponds to the sub-equatorial climate (zone1) including grids at low latitudes along the coast. The abundant rainfall in this region can be explained by a combination of factors linked to its proximity to the Atlantic Ocean, such as the strong influence of the West African monsoon [12] .

- The second climate zone is the humid tropical climate (zone 2), located in the centre of the country. The rainfall in this zone is relatively lower compared to the rainfall in the zone 1.

- The third zone represents the transitional tropical climate (zone 3), including grids located at higher latitudes. This climate zone dominates the northern part of Côte d’Ivoire influenced by Harmattan with the lowest cumulative rainfall during a year. Indeed, the air masses coming from the Atlantic Ocean towards the Sudanian zone of Côte d’Ivoire (zone 3) warm up, as they cross the

Sahelian zone, leading to a reduction in condensation and relative humidity resulting in an increase in temperature in this climate zone.

3) Analysis of the mean daily cumulative rainfall derived from CHIRPS and CORDEX over the period 1981-2005

The box-and-whisker plots of mean daily rainfall for each climate zone obtained from the hierarchical classification of the principal components of CHIRPS and CORDEX over the period 1981-2005 are illustrated in Figure 4(a) and Figure 4(b). The important climatic feature is the same number of zones three (3) for each of the climate profiles derived from CHIRPS and CORDEX. In addition, a comparison of the box plots of mean daily precipitation for the climate zones shows small interquartile differences between CHIRPS and CORDEX. For all the climate zones derived from the HCP, CORDEX simulations overestimate the median of daily precipitation with relatively smaller biases about 1.9% for the south (zone 1); 4.8% for the centre (zone 2) and 3.8% for the north (zone 3).

Furthermore, analysis of Spearman correlation coefficients between accumulated daily rainfall from CHIRPS and CORDEX for the different climate zones over the historical period (1981-2005) reveals the significant similarity between them (see Table 3) with p-values lesser than 0.05.

1) Principal component analysis (PCA)

Table 5(a) and Table 5(b) show the PCA for the projected variables derived from CORDEX under RCP4.5 and RCP8.5 scenarios over the near future period (2031-2060). The PCA for the far future period (2071-2100) is also reported in Table 6(a) and Table 6(b). Following Kaiser criterion (see section 2.3), the first fifteen and fourteen components for CORDEX variables under RCP4.5 and RCP8.5

scenarios are retained for the near future period (2031-2060). For the far future period (2071-2100), only the first fourteen components are significant for both scenarios, with cumulative variances of 96.69% and 96.45%, respectively.

b) Analysis of the Hierarchical Classification of Principal Components (HCP)

Figure 5 and Figure 6 show the HCP of the projected variables derived from CORDEX for the near (2031-2060) and far (2071-2100) future periods. The structure and composition of the classes vary according to the scenarios and the period. For the near future, four (4) different classes are identified for RCP4.5 and three (3) classes for RCP8.5, with numbers ranging between eleven (11) and fifty-seven (57), while during the far future (2071-2100), both scenarios (RCP4.5 and RCP8.5) indicate four classes with the number varying between twelve (12) and fifty-four (54) grids.

The spatial distribution of climate zones of Côte d’Ivoire under RCP4.5 and RCP8.5 scenarios for 2031-2060 and 2071-2100 periods is presented in Figure 7 and Figure 8.

There are four (4) climate zones resulting from the hierarchical classification of principal components (HCPC), which are maintained over the two future periods under the RCP4.5 scenario, while under RCP8.5 there are three (3) and four (4) climate zones over the near and far future periods, respectively. Thus, except

RCP8.5 scenario for the near future period, the coastal climate zone (zone 1), the humid tropical mountainous climate zone (zone 2), the transitional tropical climate zone (zone 3) and a new climate zone (zone 4) combining the humid tropical and transitional tropical climates are found for each scenario from the south

to the north. The main observed changes in the surface areas of the different climate zones over the future periods are due to the latitudinal displacement of the inter-zone boundaries and the appearance of the fourth climate zone. These changes are the results of a reduction or an extension of the surface area of the climate zones, depending on the scenario and the period. Extended surface areas are generally observed in the northern climate regions. The climate zone 4 reaches its maximum extension over the period 2071-2100 under RCP8.5 scenario (Figure 8(b)). Surface area reductions are more marked in the southern climatic regions of Côte d’Ivoire. The maximum reduction is observed over zone 1 during far future period (2071-2100) and under RCP8.5 scenario (Figure 7(b)).

3) Trends in climate zones for short (2031-2060) and long (2071-2100) terms

Figure 9 illustrates the projected rates of spatial changes of the climate zones derived from CORDEX simulations under RCP4.5 scenario over the near and

the far future periods. Analysis reveals significant changes in the surface areas of the different climate regions of Côte d’Ivoire. During the near future period, climate zones 1, 2 and 3 will experience a reduction of their surfaces areas with negative rates of changes estimated about −6.3%, −6.2% and −15.2%, respectively; corresponding to 17,640 km², 17,360 km² and 42,560 km², surface areas, respectively. This situation favors the appearance of a fourth intermediate climatic zone (zone 4) with an appearance rate of 27.7% (i.e., 77,560 km²). This zone extends from west to east, between the northern and central zones. In the far future, zones 2 and 3 will experience a significant reduction of their surface areas, with respective rates of −16% and −20.4%, corresponding to surface reduction of about 44,800 km² and 56,000 km², respectively. Zone 1, with a zero rate of change, preserves the same surface area as during the historical period. Zone 4, in addition to the centre, will extend to the northeast with a rate of appearance of 36.5%, corresponding to 102,200 km².

These results clearly indicate that climate change will result in changes in the extent (surface area) and structure of Côte d’Ivoire’s climate zones. Northern and central zones are the most impacted, with the appearance of an intermediate fourth zone with a surface area varying between 77,560 km² and 102,200 km², depending on the period.

Figure 10 shows the projected rates of spatial changes of the climate zones derived from CORDEX simulations under RCP8.5 scenario over the near and far future periods. Under RCP8.5, Zones 1 and 3 experience an increase in their surface area of 6.2% and 1.8%, corresponding to 17,360 km² and 5040 km², respectively, while Zone 2 will experience a reduction of 8%, corresponding to 22,400 km². During the far future period 2071-2100, zones 1, 2 and 3 will experience a decrease of their surface areas estimated about −5.4%, −9.8% and −33%, corresponding to 15,120 km², 27,440 km² and 92,400 km², respectively. Moreover, Zone 4 will appear at a rate of 48.2% (i.e., 134,960 km²). In addition, RCP8.5 scenario projects a reduction of the surface area of the southern zone, particularly in the south-western part, over the far future period.