These data were entered, coded, and analyzed in SPSS v23 and Jamovi v2.3 software and Excel 2013. These data were analyzed in two ways. The first consisted of descriptive statistical analysis, allowing us to identify trends expressed in numbers and percentages, and the second consisted of a cross-tabulation of the responses according to age and level of study. For aspects of endogenous climate knowledge, the term before (years < 1990) and actual (years ≥ 1990) are used to refer to the standard climate normal (1991-2020) of the World Meteorological Organization ( OMM, 2017 ).

Climatic data were entered and analyzed using Excel 2013, Instat+.3.36, Xlsat2019 software. The types of analysis included the Standardized Precipitation Index (SPI), Lamb’s Index, and statistical tests, rainy season analysis (Onset, length, end, dry sequence) for rainfall data. Statistical tests and anomalies on temperatures and wind speed.

· Lamb’s standard index: It is used to highlight trends in time series and to estimate deficit and surplus periods on an annual scale ( Dao et al., 2010 ; Alamou et al., 2016 ). It is calculated according to the following equation:

V c r = x i − x ¯ σ (2)

x i is data i for the annual data series; x ¯ is the average of the annual data series for the reference period; σ is the standard deviation of the annual data of the series of the reference period.

· Standardized Precipitation Index 9SPI): is calculated using the arithmetic mean and standard deviation of the precipitation series. It is used to illustrate and categorize drought periods. In the model, negative values obtained from this equation indicate precipitation deficits (drought events) while positive values illustrate precipitation excesses (wet events) (Table 1). Four different categories of drought are defined by McKee et al. (1993) . It is calculated from the following equation:

S P i = X i − X ¯ S x (3)

X i is the cumulative rainfall for year i; X ¯ is the cumulative average rainfall; S x standard deviation of precipitation in the series.

Statistical tests used

· Mann-Kendall trend test (1945)

This non-parametric test allows to characterize a trend or a sudden jump. It is based on the correlation between the ranks of a time series and their order. Its null hypothesis is that there is no trend. For each pair of values y_i and y_j, where i > j, the test asks whether y_i > y_j (P) or whether y_i < y_j (M). Its statistic is equal to the sum of the differences between two observations, i.e., S = P − M. The significance of a trend can be tested by comparing the standardized value U to the test variable at the desired level of significance ( Cédric, 2008 ; Hamed, 2008 ).

· Pettitt test ( Pettitt, 1979 ): The Pettitt test is defined as the point at which a sudden change or jump occurs in a data series. It is used to determine the year of the break in precipitation or temperature for a station under consideration. The Pettitt approach is non-parametric and is derived from the Mann-Whitney test. The absence of a break in the series (X_j) of size N constitutes the null hypothesis. The implementation of the test assumes that for any time t between

Source: ( McKee et al., 1993 ).

1 and N, the time series (x^ i = 1 to t and t + 1 to N belong to the same population. The variable to be tested is the maximum of the absolute value of the variable U_t,N defined by:

U t , N = ∑ i = 1 t ∑ j = t + 1 N D i j (4)

where D_ij = sgn(X_j − X_j) wth sgn(x) = 1 if x > 0, 0 si x = 0 et − 1 if x < 0.

If the null hypothesis is rejected, an estimate of the break-up date is given by the time t defining the maximum in terms of the number of days in the week ( Servat et al., 1998 ). Its null hypothesis being non-breakage ( Paturel et al., 1998 ).

· Student’s t test for comparison of means: it is used to compare two means. It tests the equality of two independent samples and thus makes it possible to prove that the difference between the two means is significant or not. T follows a Student’s law with v degrees of freedom where v= n₁ + n₂ − 2, where n₁ and n₂ are the sizes of the first and second samples respectively ( Cédric, 2008 ).

Analysis of rainy season parameters: The analysis of the characteristic parameters of the season (start date, end date and duration) was carried out using INSTAT + v3.36 software according to the criteria selected for the Sudano-Sahelian zone. These criteria are as follows:

· The start date, upon receipt of 20 mm of rain collected in one (1) or two (2) consecutive days after May 1, without a dry period of more than 20 days within the following 30 days ( Stern, 1981 ).

· The end date, based on the water balance as a criterion, for example, the first day after September 1, when the water balance is less than or equal to 0.05 mm. Thus, when plant water consumption and climatic demand exhaust the soil water supply ( Stern et al., 1982 ; Stern et al., 2006 ).

· Season length is obtained by calculating the difference between the end date and the start date of wintering ( Sivakumar, 1988 ; Stern et al., 2006 ).

· Analysis of the frequency of dry spell: A dry spell for a given period is defined as the maximum number of consecutive days without rain after the start of the rainy season. The frequency and duration of dry spells are calculated for the months of May, June, July and August. The threshold chosen is 0.85 mm ( Stern et al., 1982; 2006 ).

The analysis of the Mann Kendall test on rainfall from 1950 to 2022 shows a decreasing trend in rainfall. This decrease is significant with a p-value of 0.05. However, analysis of the Pettitt test showed a rupture in the series in 1966 with 0.023 as p-value. Comparison of the two means using Student’s t Test reveals a reduction in rainfall of 213 mm from 1950 to 2022 (Figure 2(a)). The analysis of the SPI shows severe droughts in 1973 with an index of -1.5 and 2002 with an index of −1.9; moderate drought for the years 1968 and 2006 with an index of −1.2, 1972 and 1987 with an index of −1.1, 1984 with an index of −1.4. The Long-Term-Average (LTA) 5-year and 10-year illustrates a high alternation of wet and dry years after the 1970s (Figure 2(b)).

The Analysis of the season’s parameters shows that the two recent 30-year periods (1950-1979; 1980-2009) were marked by a delay in the start of the rainy season (on average 6 days), an early cessation of rainy season (on average 4 days) and a reduction in the number of rainy days (on average 10 days) (Table 2). The

Source: Personal work. ¹Average Onset of Rainy Season; ²Average End of Rainy Season; ³Average Length of Rainy Season.

dry spell of May, June, July and August were analyzed. Figure 4 revealed that frequent of dry spells greater than or equal to five days was present in all months but very low in May. However, the frequency of those greater than or equal to 10, 15 and 20 days are very high for the month of May. This summarizes that July and August constitute the rainiest months (Figure 3).

The standardized maximum temperature anomalies indicates that the years 2013, 2016 and 2021 were years of warmer days because the annual average maximum temperatures corresponding to these years are higher than the annual average maximum temperature (34.1˚C) over the reference period (1950-2022). However, the years 1950 to 1995 were years of less hot days due to the fact that the annual average of maximum temperatures for these years is lower than those over the reference period. At the same time, the moving average indicates an increase in maximum temperatures (Figure 4(a)). The Man-Kendall test shows a significant upward trend in maximum temperatures with a P-value of 0.0001. The Pettitt test indicates a rupture in the time series from 1950 to 2022 with a P-value of 0.0001. This rupture occurred in 1995. Comparison of the two means using Student’s t-Test indicates an increase in maximum temperatures of 1.33˚C for the reference period (Figure 4(b)).

The standardized minimum temperature anomalies indicate that the years 1951, 2008 to 2010; 2021 were years of warmer nights as the annual average minimum temperatures corresponding to these years are higher than the annual average minimum temperatures (21.2˚C) over the reference period (1950-2022). In contrast, the years 1952 to 2007; 2011 to 2022 were years of less warm nights due to the fact that the annual average minimum temperatures corresponding to these years are lower than the annual average minimum temperatures over the reference period (Figure 4(c)). The Man Kendall test shows a significant upward trend in minimum temperatures with a p-value of 0.0001. The Pettitt test shows a rupture in the time series from 1950 to 2022 with a p-value of 0.0001. This rupture was observed in 1995. Thus, the application of the Student’s test indicates an increase in minimum temperatures of 1.24˚C for the reference period (Figure 4(d)).

The normalized average wind speed anomalies indicates that the years 1950 to 1954; 2016 to 2020 were higher average wind speed years because the corresponding annual averages for these years are higher than the annual average wind speed (1.28 m/s) over the baseline period (1950-2022). In contrast, the years 1957 to 1994; 2004 to 2015; 2021 and 2022 were years of lower average wind speed because the corresponding annual averages for these years are lower than the average annual wind speed over the baseline period (Figure 5(a)). The Man Kendall test revealed a non-significant upward trend in mean wind speed with a p-value of 0.228. The Pettitt test indicated a rupture in the time series from 1950 to 2022 with a p-value of 0.002. This rupture was observed in 1994. Thus, application of Student’s t-test indicates an increase in mean wind speed of 0.59 m/s for the baseline period (Figure 5(b)).

The observation of the demographic characteristics of the agropastoralists shows that 52% of the respondents are between 30 and 49 years of age; 49.5% are not educated; 68.6% have fewer than 5 working persons; 76.7% have not received any training on pasture management (Table 3).

Source: Individual Survey (December, 2022)

Table 4 shows the endogenous perceptions of agropastoralists regarding rainfall. It indicates that the season is increasingly dry (63.4%). The onset of the rainy season before (years < 1990) is at onset of May (62.4%) and actual (years ≥ 1990) is at onset of June (51%) with an increasingly later start of the rainy season. In contrast, the rainy season ended in October before (30.9%) and actual (20.2%) with an increasingly erratic rainy season duration. A large majority of the surveys noted a frequency of dry spell (93.6%) particularly those greater than or equal to 20 days (55.7%). They have noted that it is getting hotter (93.1%) and the winds are increasingly violent (97%) (Table 4).

Source: Individual Survey (December, 2022).

Table 5 is based on the effects of climate hazards on pastoral resources before (years < 1990) and actual (years ≥ 1990). It shows that the condition of watering points for animals was good before (56.2%) compared to poor at present (44.8%). The condition of pastures was similar to watering points (44.6%), but very poor

Source: Individual Survey (December, 2022).

at the moment (54.5%). The 57.4% of agropastoralists perceived that the saline lands were in very good condition before, but are in poor condition currently (63.1%). In the Bougouni district, agropastoralists’ observations show that there is an increasing number of highly palatable species (herbaceous and woody) for all the different categories of animal species. However, most of them are disappearing due to climatic risk-related hazards (Table 6). In particular for the herbaceous species, it is about Andropogon guyanus (Kunth), Cymbopogon giganteus (Chiovenda, Kerhar), Diheteropogon amplectens (Nees), Rottboellia exaltata (Robuta hook), Digitaria horizontalis (Willd), Sporobolus pyramida-lis (P. Beauv). For woody plants, these are Afzelia africana (Smith ex Pers), Pterocarpus erinaceus (Poir), Khaya senegalensis (Desr A. juss), Ficus gnaphalocarpa (miq, steud.ex A. Rick), Securinega virosa (Roxb. Ex Willd. Baill), Baissea multiflora (Stapf) (Figure 6).

Source: Individual Survey (December, 2022).

The main causes of degradation of pastoral resources cited by agropastoralists are the increase in long dry spells (55.4%), agricultural pressure (25.2%), erratic rainfall (10.6%), and livestock pressure (overgrazing…) (4.2%) (Table 6).

Agropastoralists in the research sites have adopted response strategies to reduce their vulnerability to climate risks related to hazards. Thus, to adapt to the frequency of droughts and erratic rainfall, 49.5% practice collection and storage of crop residues; 25.7% practice adapted forage cultivation. In face to temperatures increasing, regular watering of animals (39.6%) and changing of animal rhythm driving (35.9%) are the most common strategies used. Concerted reforestation (37.9%) and the practice of protection and pruning species (31.7%) are the most important strategies to adapt to frequency of violent winds (Table 7).

Source: Individual Survey (December, 2022).