2.3.1. Trend Analysis

Trends and breakpoints were assessed using the non-parametric test of Mann-Kendall first, followed by the Pettitt and SNH tests The Mann-Kendall test is used in order to establish whether there is a trend (increasing or decreasing) in the time series. It is done with a confidence level of 95%, and the hypotheses are:

H0: there is no trend in rainfall time series;

H1: there is a trend in rainfall time series.

For both Pettitt and SNH tests, significance level α = 0.05, and the hypotheses are:

H0: there is no change in annual rainfall data;

H1: there is a date at which there is a change in the data.

Moreover, in order to assess the trend of annual rainfall at watershed scale, and because rainfall data are not measured in every single grid of the watershed, spatial interpolation was required. In the scope of this study, ordinary kriging (OK) was chosen over other methods—such as arithmetic mean, Thiessen polygon, inverse distance weighting—because (i) it takes into account not only the distance between observation stations and estimation point but also the distance between stations taken two by two; (ii) it is a stochastic method which provides the best linear unbiased predictions; (iii) the interpolation error can be estimated [23]. Nonetheless, one of the limitations of kriging is that it is not suitable when there are few observation points. Kriging was basically developed for geostatistics purposes [24] but is widely used in climatology. It is worth noting that the 'backbone' of kriging is the variogram which explains the variance of the studied variable with respect to distance between observation points. Equation (1) presents the formula of variogram

$$\chi(h) = \frac{1}{2N(h)} \sum\_{i=1}^{N(h)} \left( z(p\_i) - z(p\_i + h) \right)^2,\tag{1}$$

where G(*h*) is the variogram, *N*(*h*) the number of coupled points separated by the distance *h*, *z*(*pi*) the observed rainfall at location *pi*, and *z*(*pi* + *h*) the observed rainfall at location *pi* + *h*.

Furthermore, considering the fact that rainfall regime in Mono watershed is not homogenous, rainfall trend analysis is carried out with respect to three latitude-based regions, as done in previous studies [15,18]. The regions are defined as follows: latitude < 7, 7 ≤ latitude ≤ 8 and latitude > 8. Hereinafter, these regions are respectively referred to as southern part, central part and northern part of the Mono watershed.

Analysis of temperature trends over the watershed was conducted on the arithmetic mean from the stations of Tabligbo, Sokodé and Atakpamé.
