Land degradation along a climatic gradient in Mali: Farmers' perceptions of causes and impacts

Land degradation (LD) in Mali is prevalent and leads to an enduring environmental and humanitarian crisis. Farmers' ecological knowledge has proven to be a valuable tool in addressing its challenges. How farmers perceive LD affects how they deal with induced risks, and their responses to these perceptions will shape restoration options and outcomes. Therefore, this study assessed farmers' perceptions of LD along a climatic gradient in three regions of Mali. We interviewed 270 farmers, and we analyzed their responses using descriptive statistics and Spearman rank‐order correlation. We found that the respondents were aware of LD and have identified its key indicators and its impacts on their livelihoods. Moreover, we found that farmers' perceptions are not influenced by gender, age, or education level, but rather by agricultural training, participation in agricultural labor, the practice of fallowing, shortage of firewood, livestock, household size, appearance of some plant species and famine. Additionally, farmers' perceptions of LD vary along the climatic gradient as they correlate to different variables in each agro‐ecological zone. LD's impacts, however, decrease in severity along the north–south gradient, although they are linked to the same variables. As LD is seen through a reduction of ecosystem services provisioning capacity because of the local communities' heavy dependence on natural resources, actions should be geared towards agronomic and vegetative land management options. Such actions should prioritize context‐specific soil and water conservation techniques and proven indigenous practices.

Sudanian and Guinean zones. Situated in the western part of the Sahel Zone, Mali is seriously affected by LD which is causing an agricultural GDP loss estimated at 2-4% per year, with reduced crop yields (Monitoring African Food and Agricultural Policies [MAFAP], 2013) and subsequent degradation of ecosystem services (Millennium Ecosystem Assessment [MEA], 2007;Sidibé, 2013;UNCCD, 2017). LD also drives an average annual deforestation of 500,000 ha per year (MEA, 2007;UNCCD, 2019). Furthermore, erosion affects soil fertility and crop production; it directly impacts the livelihood of local communities. LD increases food insecurity, and it forces farmers to shorten the fallow time, to expand to marginal lands, and to adopt non-sustainable practices (Spiekermann, Brandt, & Samimi, 2014). Rainfall variability, erosion and human activities, such as agricultural expansion, overgrazing, overharvesting of fuelwood and charcoal production are considered to be the key characteristics of LD in Mali (Mbow, Brandt, Ouedraogo, de Leeuw, & Marshall, 2015;MEA, 2007). To cope with LD at the local scale, farmers have been using their ancestral ecological knowledge and practices, such as fallowing, use of stone bunds and planting pits (traditionally known as zai or towalen) in an attempt to reverse LD and increase crop production (Saïdou, Kuyper, Kossou, Tossou, & Richards, 2004). At the national scale, since 1960, the government of Mali has made sustainable land management and soil fertility improvement programs a priority (MEA, 2007;Sidibé, 2013). Additionally, more than 1,000 local non-governmental organizations (NGOs) are implementing various land restoration projects throughout the Country (MEA, 2007).

The involvement of NGOs and other global initiatives in combatting LD
certainly underlines the urgency of the matter. Despite all these initiatives, Mali's ability to feed itself is still challenged nowadays, almost 50 years after the droughts and famines of the 1970s (Sacande & Berrahmouni, 2016;UNCCD, 2017). This situation is exacerbated by a socio-economic context driven by population pressure (Institut National de la Statistique [INSTAT],-2019;World Bank, 2016) and the absence of adequate policies addressing land tenure (Lal & Stewart, 2010;MEA, 2007; United States Agency for International Development [USAID], 2010).
However, in addressing LD, many of the adopted approaches have failed to take into account the context and the perceptions of local farmers (Olsson et al., 2019;Pulido & Bocco, 2014). According to Gray and Morant (2003), a contextual approach that takes into account farmers ecological knowledge could complement the predominant remote sensing methodologies. Assefa and Hans-Rudolf (2016) further noted that farmers that perceived soil degradation were more inclined to adopt management practices that reduced its impacts on their livelihood. Therefore, their opinions should not be ignored (IPBES, 2018;Mairura et al., 2008;Odendo, Obare, & Salasya, 2010;Olsson et al., 2019). Also, the literature suggests that farmers' perceptions are influenced by variables related to their socio-economic and demographic context, such as age, education, and household size (Barbier & Hochard, 2018;Davies, Pollard, & Mwenda, 2010;Hasan & Kumar, 2020;Tesfahunegn, 2018). Additionally, variables related to their agricultural practices, such as fallowing, use of fertilizers, tree planting (Biratu & Asmamaw, 2016;Gashu & Muchie, 2018;Sanogo et al., 2017) and other environmental and climatic factors (Lal, 2001;Orchard, Stringer, & Manyatsi, 2017), also influence them as well. Pulido and Bocco (2014) found that farmers tend to adopt mitigation options based on their agro-ecological conditions and prioritize the strategies that fulfill their basic needs. Similarly, the perceptions of Malian farmers living along a north-south climatic gradient, exposed to different agro-ecological conditions and experiencing different rates of LD, are essential in designing any successful mitigation strategy for their particular context (Davies et al., 2010;Dembele, 2006;Odendo et al., 2010). Such approaches are also necessary for undertaking actions in the framework of any land restoration initiatives, whether they are local, regional or global initiatives, such as the United Nations Sustainable Development Goal 15 (SDG 15), the Bonn Challenge, Land Degradation Neutrality (LDN) of UNCCD, and so forth. As LD is exacerbated by CC, previous studies on farmers' perceptions of CC in Mali, have suggested that farmers are aware of its adverse effects, and they consequently adopt adaptation measures (Sanogo et al., 2017;Traore et al., 2015). Concerning LD, until Ouedraogo et al., 2010;Perry, 2013), understanding the perceptions of local communities at the heartland of the region is therefore relevant for the rest of the region and is a necessary step towards addressing LD in the Sahel.
The present study, therefore, sought to address these knowledge gaps. We focus on understanding the variables that contribute to raising farmers' awareness of LD at the local context in Mali, in three separate regions belonging to different agro-ecological zones, characterized by a north-south increasing rainfall gradient which affects vegetation growth. LD impacts are different in drier versus wetter zones, and so are the mitigation options adopted by the communities living along a latitudinal climatic gradient (MEA, 2007;UNCCD, 2019;UNEP, 2012). As the mitigation options are based on their perceptions (Pulido & Bocco, 2014), we assumed that farmers' perceptions vary along the rainfall gradient according to the agroecological conditions and that it is important to capture the variations therein; this led us to conduct our study along a climatic gradient (Dawoe, Quashie-Sam, Isaac, & Oppong, 2012;Duguma & Hager, 2011;Pawluk, Sandor, & Tabor, 1992;Rist & Dahdouh-Guebas, 2006;Sternberg et al., 2011). We aimed to: (a) assess farmers' perceptions of LD in selected agro-ecological areass and determine the key factors that influence their perceptions; and (b) elucidate the impacts of LD on the livelihoods of the communities of these regions.

| Study sites
Mali is a landlocked country in the heart of West Africa; it covers 1,241,238 km 2 and is administratively divided into eight regions (Plan d'Action National d'Adaptation au changement climatique [PANA], 2007;World Bank, 2016). The study sites are located in the regions of Mopti, Segou, and Sikasso, along a north-south climatic gradient from the Sahelian Zone southwards towards the Sudano-Guinean Zone (Figure 1).
The population in the three regions has considerably increased in the past decades, Sikasso has a population of 3,148,819 inhabitants, followed by Segou (2,785,676) and Mopti (2,425,735) (INSTAT, 2015).
Agriculture and agropastoral activities are dominant. The mean annual temperature is 28.0 C for Mopti, 27.6 C for Segou, and 27.0 C for Sikasso region. Mali has four agro-ecological zones, characterized by a unimodal rainfall, with increasing averages from North to South.
The rainy season lasts for about 3-6 months, May to October in the South and July to September in the North, August being the peak (MEA, 2007), and the remaining months are dry (PANA, 2007). Rainfall is more abundant in Sikasso with an annual average of 1,121 mm; this yearly average decreases as we move north towards the drier

| Data collection
We have used a survey to assess farmers' perceptions about LD; this methodology is frequently used in LD research to inquire about land users' opinions (Kapalanga, 2008;Tesfahunegn, 2018). In the current study, we interviewed 270 farmers, using a combination of questionnaires, focus group discussions (FGD), and field observations to collect the primary data. We conducted surveys in May 2017; we purposively selected the nine villages that participated in the Drylands Development Program (drydev.org), which envisions to empower farmers to transition from subsistence farming to sustainable rural development, through improved soil and watershed management practices. In each village, we randomly selected 30 farmers from a list of the program beneficiaries, due to budget and time constraints. To administer the questionnaire, we recruited three enumerators for data collection based on their social data collection experience and familiarity with the study area. We conducted the interviews in the local dialects and recorded the translated responses using tablet computers loaded with the Open data toolkit application (Hartung et al., 2010). The questionnaire aimed at understanding the factors that would affect farmers' perceptions of LD in different agro-ecological zones. We assumed, based on the reviewed literature, that the variables affecting farmers' perceptions would belong to three main categories, socio-economic and demographic variables (age, education level, household size, plot size, livestock, off-farm employment), agricultural variables (level or participation in labor, land percentage under cultivation, fallowing, tree planting, crop yield, changes in plot size, soil and water conservation) and variables linked to the environmental context farmers live-in (appearance of new species, deforestation, rainfall regime, drought frequencies, erosion) (Doso, 2014;Geist & Lambin, 2002;Pulido & Bocco, 2014;Sanogo et al., 2017). Additionally, questions specifically related to erosion, deforestation, and soil fertility were also added to measure farmers' awareness of LD on a three-point Likert scale (no LD, moderate LD, Severe LD).
In each of the villages, we also set up FGD for adult women, adult males, and one for the youths. Their ages ranged from 16 to 19 years old for the youths and from 25 to 70 years old for the adults. In some cultures it is unusual for women and youth to express their opinions in the presence of elder members of the community; to avoid that any group would dominate the discussions, we proceeded to separate them in order to encourage members to freely express their opinions (Stewart, Shamdasani, & Rook, 2007). Each FGD had 6-10 persons purposively selected from youth associations, women organizations, and male elders.
FGDs were followed by field visits with local farmers to assess signs of LD in situ and to have a better understanding of the status of the natural resources (vegetation cover). In the fields, we also identified and documented land-use types, management practices, and farmers' on-going initiatives to mitigate LD. Through this exercise, we understood the scale of erosion and better contextualized farmers' opinions from one village to another along the climatic gradient.
In order to characterize the climatic gradient of our study zones, we have also acquired secondary data on monthly precipitations and

| Interview data
We used Spearman rank-order correlation to determine the strength and the direction of the relationship between selected independent variables (Table 1) and the dependent variable representing LD's perception. The Spearman correlation is less restrictive in that it does not require data to be linearly related or bivariate normally distributed (Salkind, 2012).
We formulated hypotheses on selected variables based on existing literature, the FGD, and our own field experience. Table 1

Mann-Kendall (MK) test and Sen's slope method
To test for the existence of a monotonic upward or downward trend for the precipitation variable over time, we applied the widely used Mann-Kendall (MK) non-parametric statistical test (Ahmad et al., 2015;Fahmi et al., 2015;Mallik, Chowdhury, Ahasan, Akhter, & Hasan, 2016). In this test, the null hypothesis H 0 is that there are no trends; the alternative hypothesis is that there is a trend that could either be negative, positive, or non-null (Mann, 1945). The MK test uses the following formula based on the S statistics to calculate the sum of differences: T A B L E 1 Hypothesis on selected independent variables, '+' and '−' represent the direction of the impact on farmers' perceptions

Variables description H 1 Justifications based on authors opinions and from sources in the literature
Age + We hypothesize that age pairs with experience; older farmers could have seen different land uses covers. Therefore, their capacity to identify LD is higher (Shiferaw & Holden, 1998;Tesfahunegn, 2018).
Education level + Educated farmers are expected to have more contacts with extension workers, thus they are more aware of risks (Asrat et al., 2004). Furthermore, lack of formal education has been linked to higher deforestation (Davies et al., 2010;Geist & Lambin, 2002;Hasan & Kumar, 2020;Shiferaw & Holden, 1998).
Participation to labour + Degraded lands usually require more labour and labour availability is a condition for both agricultural expansion and intensification. Therefore, this variable would have a positive impact (Gashu & Muchie, 2018;Reenberg et al., 1998).

Household size −
We hypothesize a negative effect on perception as larger households could be less aware of LD and more concerned with increasing yields by all means and maximizing the use of available natural resources because of poverty (Barbier & Hochard, 2018;Gashu & Muchie, 2018;Qasim et al., 2011).
Land under cultivation (%) + Farmers with larger exploitation are more involved in soil and water conservation activities (SWC) and more aware of climate factors, (Biratu & Asmamaw, 2016;Sanogo et al., 2017).
Agricultural training + We hypothesize that trained farmers have been exposed to different SWC techniques, hence increasing their capacity to recognize LD (Assefa & Hans-Rudolf, 2016;Qasim et al., 2011;Tesfaye, 2017) Plots combined in one unit + Farmers with multiple units tend to abandon least productive units in favour of productive ones. Besides, applying management practices across several units requires additional labor with financial implications (Teshome et al., 2014) Practice of fallowing +/− Farmers are familiar with fallowing and its benefit for soil fertility (Dawoe et al., 2012;Gray, 1999). We do not propose a specific hypothesis for fallowing, those who practice it and those who have abandoned it use similar reasons.
Famine as a result of drought + We hypothesize this variable would have a positive impact on farmers' perception. Famine is sometimes perceived as a consequence of drought (Gautier et al., 2016) Use of chemical fertilizer + It is reported that farmers utilizing fertilizer would do so to compensate for soil fertility as a result of the shortened fallow period and LD (Gray, 1999;Saïdou et al., 2004;Thapa & Yila, 2012).
Growing trees on your farm + Trees on farms not only tend to improve fertility, but they also protect crops against wind erosion (Gray, 1999;Sterk, 2003). We assume farmers plant trees to mitigate LD amongst other benefits. .
Appearance of new species + Appearance of species that farmers relate to LD, would positively impact their perceptions (Tesfahunegn, 2018).
Firewood shortage + Excessive firewood extraction has in many cases lead to vegetation degradation and deforestation (Doso, 2014). We hypothesize that shortage of firewood, would positively affect farmers' perceptions on LD. Less firewood means fewer trees, consequently higher exposure to LD Use of marginal lands + We hypothesize this variable would have positive effect on farmers' perceptions (Lal, 2001;Reenberg et al., 1998) Livestock + Livestock grazing and trampling have been identified as cause of vegetation and soil degradation in the Sahel (Hiernaux et al., 1999;Savadogo, Sawadogo, & Tiveau, 2007). We hypothesize a positive a positive impact on farmers' perceptions of LD where n represents the number of data points, x j and x i are the annual observations. The variance of S is calculated as follows for n < 10: For n > 10, the variance of S is calculated as: where q represents the number of repeated sets or tied groups, t p rep- In a positive (upward) trend, Z is positive, and a negative Z value indicates a downward trend. Assuming there is a linear trend, we use Sen's non-parametric method to assess its magnitude (Mallik et al., 2016). The slope estimates of the data points are: where x j and x k are data points, with j > k. For n values of x j in the time-series, the median of N = n (n − 1)/2 is Sen's estimator of slope.
The Sen's estimator is: We used the MAKESENS software to apply the MK test and to perform the Sen's method to estimate the slope (Salmi, Maatta, Anttila, Ruoho-Airola, & Amnell, 2002).

Standardized precipitation evapotranspiration index (SPEI)
The Standardized Precipitation Index (SPI) is a precipitation-based universal meteorological drought index developed by Mckee, Doesken, and Kleist (1993)  proven to affect the intensity of droughts (Carré et al., 2019;Ndehedehe, Agutu, Ferreira, & Getirana, 2020); it is therefore, crucial to account for temperature and potential evapotranspiration (PET) in assessing drought conditions. To remedy this situation, we used the SPEI, which is a drought index based on temperature and PET (Vicente-Serrano et al., 2010), for a better assessment of drought events. SPEI is calculated as: improved seeds (65.5%), and only 37% of them practiced fallowing.
Amongst the perceived causes of LD and soil fertility decline, farmers have identified the overexploitation of land resources for agriculture as the leading cause across all three regions (Figure 3).
Of course, this cause appeared less critical in the Mopti region than the two others. Except for water erosion, which is more critical in Sikasso; other causes appear to decrease in percentages going from the north to the south. In the Mopti region, wind erosion and deforestation are perceived by many as the leading causes of soil fertility decline.

| Trends in climatic data
The SPEI revealed the existence of droughts and their severity ( Figure 7a). Although drought events occurred in the three regions, they ranged from moderate dry (−1 to −1.49) to severe dry (−1.5 to −1.99). However, for the specified period, only Mopti experienced extreme dryness (−2 and less). Using the rainfall data, the MK test and Sen's slope method produced significant Z score values for Mopti (2.87), Segou (2.94), and Sikasso (3.15), hence confirming the existence of a positive trend for rainfall in the three regions (Figure 7b). analyzing climate data and assessing drought severity in each region.

|
We used descriptive statistics to highlight the socio-economic and demographic contexts of the farmers in each of the regions, the causes of fertility decline, and the impacts of deforestation. Our study confirms that farmers across the three regions are aware of LD and they perceive it through a set of variables belonging to their socioeconomic and demographic context, their agricultural practices, and the environmental and climatic factors they live under.
Although the three regions belonged to separate agro-ecological zones, the farmers nevertheless shared the same socio-economic and demographic contexts. As most smallholder farmers, their primary concern is to produce enough food to sustain the household needs.
Thus, cropping is usually their main activity and sole source of income; this is the case for the majority of our respondents. With insufficient land, they find it challenging to produce enough food for the growing number of household members; at the same time, soil fertility of existing farmlands is decreasing. As an alternate means of livelihood, some engage in trading activities or animal rearing or migrate to urban areas in search of employment. Due to low literacy rates, options to diversify income are very limited. This socio-economic context the farmers live-in, is made-up of many concurring factors that together strongly determine their perceptions of LD and constitute an underlying driver for the eventual practices and measures they would adopt to improve their livelihood.
Farmers' agricultural practices certainly have an impact on the current conditions of their lands. In the three regions, farmers' perceptions of LD correlated with their participation in agricultural trainings. This was similarly found by Nigussie et al. (2017), who suggested that farmers who received training through extension agents were more susceptible to perceive LD; this corroborates with our findings for all regions. Independently of age and education level, training did increase farmers' perceptions of LD. In addition, in Mopti, livestock was perceived as a cause of LD. The role of livestock in relation to vegetation and soil degradation has been debated in the literature (Hosonuma et al., 2012;Orchard et al., 2017), but our results are comforted by findings of Savadogo et al. (2007), that livestock grazing has detrimental effects on vegetation and the soil by reducing its water infiltration rate, thus contributing to water-runoff and erosion. Participation in agricultural labor negatively influenced farmers' perceptions of LD in Segou and Sikasso; households that were more active in labor activities were less inclined to perceive LD. This could be due to the fact that additional labor effort could allow farmers to compensate for the decreasing productivity attributed to LD. Gashu and Muchie (2018) suggest that to meet their income needs, farmers increase labor partic- The environmental and climatic factors contain variables that are clearly perceived by farmers. According to Lal (Lal, 2001), the majority of farmers perceive LD through soil erosion, which is one of its main manifestations, and they consider it severe. This view is persistent across the three regions, regardless of socio-economic or agricultural practices. Davies et al. (2010), Assefa and Hans-Rudolf (2016) further found that farmers are aware of erosion through more apparent signs such as rills and gullies that appear in their farms. This is consistent with our findings, were farmers cite the appearance of rills, water T A B L E 3 Spearman rank-order correlations between selected variables ( stagnation, and silting of water channels as indicators of erosion. Those signs are generally linked to the biophysical and climatic conditions of the area. Our analysis of the climate data indicated that dryness is common along the gradient and more so in the Mopti region than the other regions. Agricultural drought reduces the amount of water available to crops and is usually caused by soil erosion and poor agricultural practices (Agnew & Warren, 1996). The rainfall pattern is erratic, but the overall rainfall is on an increasing trend. Increased precipitations (Mahé & Paturel, 2009) onto degraded soils without much vegetation cover, leads to water run-off, soil erosion, evaporation, and reduced water availability for optimal crop production (Stroosnijder, 2012 We attribute this to increased soil organic matter in the south, which improves soil fertility, structure, water infiltration, and it reduces soil erosion from water-run-off (Bayala & Ouedraogo, 2008;Kusimi, 2008;Lal, 2001;Mortimore & Turner, 2005;Neupane & Thapa, 2001). The results also highlight farmers' perceptions of indirect impact resulting from LD via deforestation. They drive the increased use of fertilizers, the shortening of fallow times, economic loss, and a reduction in the means of livelihood of the communities.
This study provides evidence that farmers perceive LD in the three studied regions across the climatic gradient. The results also suggest that farmers perceive LD differently across the climatic gradient. Their perceptions are correlated to separate variables even though they could be classified into three main categories. LD also impacts communities across the gradient through reduced productivity, deforestation, and erosion. However, as limitations of this study, we acknowledge the subjectivity of studying perceptions, and the seasonality could also have an effect on farmers' perceptions as many of the indicators are rainfall related, conducting the study at the onset of the rainy season may have affected their responses.

| CONCLUSIONS
The findings of this study reveal that farmers across a longitudinal cli- Meeting the growing demand for food products and other natural resources has certainly led many communities, on-the-one-hand, to engage in intensive agricultural practices, leading to shortened fallow times, and on-the-other-hand, to adopt extensive practices that could lead to deforestation and LD. All those paths lead to reduced fertility and lower crop production. Farmers should invest more in sustainable agricultural practices and soil and water conservation techniques they already know, such as agroforestry, stone bunds, zaï pits, and tie ridges that have proven their efficiency. All these existing practices should be scaled up through incentives on agricultural investment and policy enforcement. This would result in more resilient landscapes, more productive fields, and an improved livelihood for all the communities living along the climatic gradient.
The outcomes of this study could further be contextualized to serve many of the countries of the Sahelian Zone. As many of the ambitious global initiatives struggle to meet their targets, understanding farmers' perceptions of LD at the local level across the Sahel could facilitate the adoption of better-contextualized restoration strategies.