The observational data used are global monthly precipitation and temperature data from the Climatic Research Unit (CRU) at a resolution of 0.5˚ × 0.5˚ over the period from 1901 to 2013 (http://catalogue.ceda.ac.uk/uuid/3f8944800cc48e1cbc29a5ee12d8542d). These data were extracted and analyzed in central and southern areas of Côte d’Ivoire over the period from 1951 to 2005. They were used for validation and the choice of simulations for the study.

Six simulations from the CORDEX-Africa database were used in this study (Table 1). CORDEX-Africa is a set of multi-model simulations at 50 km spatial resolution, from regional climate models (RCM) forced by global models from CMIP5 [17]. They are carried out according to four greenhouse gas emission trajectories represented by the RCP2.6, RCP6, RCP4.5 and RCP8.5 scenarios. However, in the context of this study, we refer to the two scenarios RCP4.5 (so-called optimistic scenario) and RCP8.5 (so-called pessimistic scenario) which are well good for impact studies. It has been shown that CORDEX-Afrique reproduces the seasonal averages and the annual cycles of temperature and precipitation

of several climatic regions of Africa quite well [18] [19], however with some bias [20]. Thus, the simulated rainfall and temperature from CORDEX-Africa used here has been subdivided into historical (1951 to 2005) and projection (2006 to 2100) periods.

Statistical analysis based on calculations of the mean, standard deviation (STD), minimum and maximum of the time series of both CORDEX simulations and CRU observations was used for comparative studies between these two datasets. The details are shown in Table 2 and Table 3.

In order to study the sensitivity of the cocoa tree to extremes of rain and temperature, the climatic indices impacting this crop were calculated. Thus, within the framework of this study, four climatic indices from the team of experts on the detection and indices of climate change (ETCCDI) and identified as being relevant for cocoa-culture were considered. Those are:

The CDD index which corresponds to the number of consecutive days without rain. When within the rainy season, dry spells occur; they have a negative effect on agricultural production. Indeed, the dry spells create water stress at the level of the plant when they occur at the beginning or in the middle of the growing season. Given a time series of daily precipitation amount (RR), then the largest number of consecutive days where RR is less than R is counted. R is an optional parameter with default R = 1 mm. Another output variable is the number of dry spells longer than N days. Here, taking into account the specificities and needs of the cocoa tree, the following thresholds have been set for the CDD index, for the cumulative amount of rain we have 70 mm and the period of for the number of days we have 90 days have been fixed for the CDD calculation. A period is considered dry if during a period of 30 days the precipitation received is less than 70 mm.

The number of rainy days (CWD) is defined as the number of 5 consecutive days where at least a given amount 5 mm of rain has occurred. The CWD

determines the length of the rainy-agricultural seasons and helps guide farmers in the choice of varieties to grow. The increase in drought linked to rainfall variability not only reduces production, but increases harvesting difficulties. Faced with this vulnerability, agricultural households are adopting resilient strategies [21]. For the calculation of the CWD index, the thresholds of 5 mm for the amount of rain and 5 days for the number of days have been set.

For the cultivated cocoa tree to have regular growth, well-distributed leaf flushes throughout the year, abundant flowering and fruiting, annual rainfall between 1200 mm/year and 1500 mm/year is required [15];

The duration of the great dry season less than 3 months with a total of more than 70 mm [22].

A maximum temperature (T_max) is the maximum daily temperature reached. Given that the ideal conditions for the cultivation of cocoa trees are maximum daily temperatures between 30˚C and 32˚C, a threshold of T_max at 33˚C was fixed in the present study. This index was calculated over the reference period (1971 to 2000) and the future period (2021 to 2050) and makes it possible to estimate the number of days per year when the maximum temperature is above 33˚C in the central areas and southern Côte d’Ivoire.

The total amount of rain during the main rainy season (April-May-June-July) is defined as the accumulation of precipitation received during this main rainy season. The threshold of 700 mm being specific to the cocoa tree, since quantities of rain lower than 700 mm are not favorable to the good development and production of the cocoa tree (Mian, 2007).

This climatic index was determined for the reference (1951 to 2005) and future (2006 to 2100) periods.

1) Temperature

Comparison of CORDEX-Africa simulations with CRU observational data reveals that the models reproduce the seasonal temperature profile quite well with maximums in February, March and April and minimums in July and August. However, compared to that of CRU, the models underestimate the temperature in both the southern and central zones. The ensemble mean of the models reproduces the temperature profile with some bias. From January to but these differences are smaller than those observed between May and December. This is the same in the south as well as in the center. In general, the overall mean is closer than CRU values in the south (not shown in this study).

The comparison with the observations was deepened with a statistical study based on parameters such as standard deviations; the averages, minimums and maximums of the annual temperature in the central and southern zones, over the 1971-2000 period. These results are recorded in Table 2.

In the central zone, the standard deviation varies from 0.39˚C to 0.41˚C, reflecting a great variability in the average annual temperatures simulated by the models. Simulations S1 and S3 have a higher standard deviation than that of the observation while that of simulations S2, S4, S5 and S6 is lower. The overall mean of the models is lower than that of the observation, explained by the fact that the average, maximum and minimum values simulated in the models are all lower than those of the observation. These underestimates of simulated temperatures (average, maximum and minimum) could be explained by the conditions of the simulations, the various hypotheses put forward.

Concerning the southern zone of Côte d’Ivoire, the comparison of the simulated average and maximum temperatures with those of CRU, reveals an underestimation of the CORDEX Africa models considered in this study, although the minimum simulated temperature is similar to that observed. Moreover, in this southern zone, the dispersion of temperatures around the average varies between 0.23˚C and 0.30˚C, implying a low dispersion of temperatures around the average.

2) Precipitation

The comparisons of the simulated annual rainfall cycles and those of the CRU observation for each of the southern and central zones are presented. These comparisons reveal the bimodal nature of the rainfall distributed over the long and short rainy seasons in these two zones. From the country, this bimodal characteristic of precipitation is more or less fairly well reproduced by the different individual models, although the peaks do not occur at the same time. This non-coincidence of the peaks in the different models impacts the overall mean of the models, where the break point in the precipitation tends to disappear for a long unimodal rainy season, moreover, in the southern zone, there is a lag between the simulated peak (July) and that observed (June) of the long rainy season. Thus, the break in the precipitation is less marked at the level of the models although the overall mean retains its bimodal character.

Table 3 presents the comparison of annual precipitation between models and observation over the historical period from 1971 to 2000. The average annual precipitation simulated in the two areas (south and center) are overestimated compared to observation. However, the overall mean is underestimated compared to the observation in the two zones considered. There is a great variability in precipitation illustrated by high values of standard deviations between 107 mm and 160 mm for the models considered and 129.94 mm for the observation. This high variability between the models influences the overall mean and would explain the overestimation of precipitation by the models.

The CDD index determines the maximum number of consecutive dry days. Figure 2(a) and Figure 2(b) represent the number of consecutive days for the

historical period, the RCP4.5 and RCP8.5 scenarios and the difference historical period and scenarios respectively. In general, we note that the CDD index is more important in the scenarios. This index varies from area to area. Thus, for Ayamé and Tabou, compared to the historical period, we have a surplus of 18 and 28 times for RCP4.5 and 23 and 29 times for RCP8.5 where the threshold set for CDD is exceeded.

Ayamé and Tabou are experiencing an increase in the number of consecutive dry days. Increasing the number of consecutive dry days within a rainy season can have a strong impact on cocoa production. This increase in the CDD index creates an increase in dry seasons and a shortening of rainy seasons in cities. The dry sequences create a lack of water at the level of the plant, but for good production, cocoa needs an annual rainfall of 1200 mm/year [15]. Figure 2(b) shows the difference in the number of consecutive dry days between the scenarios and the models. Note that this difference is positive.

The number of wet days (CWD) determines the length of the rainy-agricultural seasons in a given place. Figure 3(a) shows us the evolution of the number of

wet days, the evolution of the index (CWD) increases faster at the level of the scenarios than at the level of the historical data in the cities. The longest wet days observed in Ayamé, Gagnoa and Abengourou with 309, 274 and 275 days respectively in the historical data; 518, 518 and 482 days in the RCP4.5 scenario; 508, 499 and 461 days in the RCP8.5 scenario. The increase in the number of wet days can have an impact on cocoa production. This increase leads to an increase in rainy seasons and a restriction of dry seasons in cities. The rainy sequences create an abundance of water at the level of the plant which will be favorable for good production. Indeed, the length of crop cycles is a very determining factor in cocoa production. It guides farmers in the choice of varieties to cultivate because a variety whose cycle length is clearly longer than the duration of the rainy-agricultural season will eventually experience serious adaptation problems [23]. Figure 3(b) shows the difference in the number of rainy days between the scenarios and the models. Note that this difference is positive.

Figure 4(a) and Figure 4(b) shows the interannual variations in cumulative rainfall during the rainy season in the central and southern zones. In general, the accumulation of precipitation is greater in the south than in the center during the rainy season. Also, apart from the historical period, where the total precipitation fluctuates around 700 mm (minimum threshold required for cocoa cultivation) for the central zone, the total precipitation is well above this threshold of

700 mm in the central zone. south. Indeed, this southern zone under the influence of the West African monsoon is characterized by high annual rainfall that can reach 1200 mm. In addition, the rains are almost continuous during the monsoon period where previous studies have shown that more than 80% of rainfall in West Africa occurs during this monsoon period. In addition, the analysis of the time series of annual rainfall data during the reference periods (1971 to 2000) and future (2021 to 2050) in the central zone, show great interannual variability (Figure 4(a)). This significant inter-annual variability in the central zone reveals years of both favorable (above 700 mm) and unfavorable (below 700 mm) rainfall for cocoa cultivation. In addition, the linear regression of the annual rainfall totals for the historical period and the projections (2021 to 2050 under the RCP4.5 and RCP8.5 scenarios), on the central zone, shows a regular increase in rainfall. Thus, the cumulative rainfall during the long rainy season, although variable over the period from 2021 to 2050 under the RCP4.5 and RCP8.5 scenarios, remains above the 700 mm threshold in the central zone, making it favorable for cocoa production.

The chronological series of rainfall accumulations during the rainy season in the southern zone reveals a large interannual variability associated with an increase in rainfall accumulations (Figure 4(b)). The linear regression of rainfall accumulations during the main rainy season, over the historical period (1971 to 2000), shows a drop in rainfall accumulations which were all greater than 700 mm in the southern zone, therefore favorable to the cultivation of cocoa. While the linear regression reveals an increase in rainfall over the entire period 2021 to 2050 under the RCP4.5 and RCP8.5 scenarios, maintaining the southern zone still favorable for cocoa production.

The spatial distribution of the number of days per year during which the maximum daily temperature is above 33˚C during the historical (1971-2000) and projected (2021-2050) periods is presented in Figure 5. It emerges that for all the periods considered (historical and projections under RCP4.5 and RCP8.5), the number of days per year when the maximum temperature is above 33˚C increases throughout the two periods. This increase in temperature due to radiative

forcings in the two scenarios RCP4.5 and RCP8.5, is greater in the north of Côte d’Ivoire with maximums of the order of 700 days, against maximums of 200 and 100 days on the central and southern zones respectively. Thus, according to the RCP4.5 and RCP8.5 scenarios, the central and southern areas of Côte d’Ivoire could experience fewer days with temperatures above 33˚C compared to the north of the country. This implies that the central and southern areas could experience small variations in maximum temperatures and remained with regard to maximum temperatures, favorable to cocoa production. Maintaining these southern and central areas favorable to the cultivation and production of cocoa is justified by previous studies which have shown for good cocoa production, average annual temperatures of 25˚C are needed [10], temperatures average maximums between 30˚C and 32˚C and minimum temperatures of 18˚C to 21˚C, with an absolute minimum of 10˚C.