Assessing the Effects of Land Use/Land Cover Change on Groundwater Recharge in a Sudano-Sahelian Zone: Case of Koda Catchment, Mali, West Africa

Groundwater is the main source of water in the studied area; therefore, it is significantly requested in all the activities of the inhabitants. These natural resources are affected by some drivers especially Land Use/Land Cover (LULC) and Climate Change. A Land Use/Land Cover (LULC) dynamics study is crucial for any global environmental change evaluation. For instance, for a given place, its change could affect considerably water cycle components. There-fore, the knowledge of the effects of LULC on groundwater recharge is then the key in water resources management system, in particular for the decision makers of the Koda Catchment where the scarcity of the water availability for vannah areas in Koda catchment is converted to agricultural land and urban area due to human activities. The decline of 8.4% in groundwater recharge might become so far obvious in the future if the current rate of deforestation continues in the Koda catchment. There is a need to closely monitor the changes in LULC for sustainable development. The results of this study could help to well understand the recharge pattern across Koda catchment under a changing LULC.


Introduction
In many developing countries one of the principal driving forces of global environmental change is Land use and land cover (LULC) change (Botlhe et al., 2019). According to Wondie et al. (2011), the LULC change is impacting many sectors of the economy. Changes in the LULC component could be observed spatially and temporally. This is mainly due to the intensity of land use and extent of area. On the temporal scale, LULC changes from a few months to several years, characterized by short term and long term changes, respectively (Lambin & Ehrlich, 1997). The long-term change is of major concern and the most significant for global environment change. Furthermore, groundwater resources and regional hydrology are controlled by many factors including the LULC change (Stonestrom et al., 2009;Ashaolu et al., 2019). Many authors have investigated the long term LULC change to evaluate the sustainability of natural resources (Scanlon et al., 2005;Lin et al., 2018;Ashaolu et al., 2019). Most of the results of the previous studies show clearly that the causes of LULC change are many being of natural and anthropogenic effects (Tamba & Li, 2011;Pervez & Henebry, 2015;Yin et al., 2017;Diallo et al., 2019). Some of human effects are the causes of increases of population growth rate, rural-urban migration, agricultural expansion, deforestation and climate change (Toure et al., 2017). Urbanization is the most irreversible form of land use. In developing countries, especially in Africa, urban land expansion has been observed since the 1980s which is more related to urban population growth than the growth in the Gross Domestic Product (GDP) (Ashaolu et al., 2019). In West African countries, large of extents of natural land cover classes have been replaced by human influenced landscape mainly dominated by agriculture (CILSS, 2016). Most of the rural population are migrating to look for better survival opportunities (Pandey et al., 2013). Consequently to feed the growing population, the agricultural land area has been increased in order to meet the demand for food (Jamtsho & Gyamtsho, 2003). All these phenomena could lead to LULC changes. In Mali, due to the increase of population growth rate leading to an important pressure O. Diancoumba et al. Journal of Geoscience and Environment Protection on agricultural sectors in order to satisfy the food demand, the portions of savannah and forests land have decreased by 23% from 197523% from to 201323% from (CILSS, 2016. The southern part of Mali including the study catchment is the most populated region over the country. Population is continuing to expand and it is strongly connected to the high demand of fresh water. The population of the Koda catchment registered an increase of 26% from 2009 to 2019 with 3.6% as population growth rate (RGPH, 2009). The principal land use in the catchment is agriculture. The quantification of LULC changes is crucial to better understand the variability and its ecological effects on the natural resources (Turner, 2005).
The main objective of this study is to assess the effects of LULC change on groundwater resources over 1990-2016 period in Koda catchment. To achieve this objective, the remote sensing coupled with GIS is used to study the dynamics of the LULC over Koda catchment between 1990 and 2016. The supervised method has been used to classify the LULC classes existing in Koda catchment while the Gardenia model developed by Roche and Thiery (1984) has been used to estimate the recharge pattern. The study reveals that the potential effects of LULC can be properly quantified for a given catchment using a hydrological model.
In the current study, the effects of LULC change on groundwater were assessed for the period 1987-2016 and the results can be used as water resources management tool.

Study Area
The study area is located in the semi arid zone of West Africa. The study cathment, Koda, occupied the southern part of Mali at 120 km in the North of Bamako, the capital of Mali. Koda catchment is lies between in Latitude: 13˚56'00"N and 12˚57'80"N and the Longitude: 7˚30'8"W and 8˚28'5"W (

LULC Dynamics
The Supervised Classification method, using Envi 4.5 Software coupled with Arc-GIS 10.3, was applied to subset Landsat images from 1990 to 2016.
In order to produce the land cover map, the coordinates of some points were taken throughout the catchment scale on May. Based on the fact that there is no cloud form the images in this period of the year. These points have been selected as unit pixel of the study area to make the land cover classification. The supervised classification method was selected using the Envi 4.8 software, and the Landsat images downloaded from USGS website: https://www.usgs.gov/tools/earthexplorer was used for that. Three years Landsat images (March 1990, March 1998    The main objective of this work is to evaluate the dynamics of the land cover change in the past 27 years. The choice of those years has been based on the availability of the Landsat data set with good quality.

Accuracy Assessment
The coefficient of KAPPA (K) has been evaluated in the view to better assess the classification of LULC units done over Koda catchment. The Kappa coefficient (K) is computed based on the error matrix.
According to Amler et al. (2015) and Ren et al. (2018) the K value is used to assess the accuracy of remote sensing data using Equation (1).
where P ij is error matrix,  i P + row total pixel,  j P + colum total pixel,  ii P corrected mapped pixel of a particular class i and N total number of pixel.
The quantification disagreement and allocation disagreement method have been proposed by Pontius & Millones (2011) and showed the limitations of the use of the K values in comparing the maps viewed. Nevertheless, in several studies, K is still considered as a vital tool for accuracy assessment measurement (Biondini & Kandus, 2006;Ren et al., 2018). The value of K that shows the consistency of data classification has been statistically classified by Fitzgerald and Lees (1994)

Estimation of Groundwater Recharge
Gardenia model is used in this study to estimate the recharge for the period 1987-2016 through a global hydrological modelling rainfall-groundwater level. All files in Gardenia model are free format, they are easy to edit by the user. It required Potential Evapotranspiration (PET) data as input data, which were estimated using Blaney Criddle (1962) method. Refer to Blaney & Criddle (1962) for the detail of the PET calculation process. The Groundwater Level (GWL) data, used from three (3)

Calibration and Validation Process
This section describes methods for evaluating the calibration and validation results. This includes a discussion of calibration acceptance criteria and descriptions on various qualitative and quantitative methods for comparing field measurements to the same parameter as calculated with the model. Two (2) quantitative statistics R 2 and NSE were used in evaluating the performance of the GARDENIA model with respect to GWL in the calibration and validation periods.
The correlation coefficient is calculated using Equation (2): The NSE was proposed by Nash and Sutcliffe (1970). The Nash coefficient quantifies the relative magnitude of the relative magnitude of the residual variance compared to the observed data variance (Equation (3)

Dynamics of LULC
The Koda catchment is mainly characterized by five (5)   The graphic representation of the results is given by Figure 4. This figure   clearly shows the increase in the areas of cultivated land and bare land and the decrease of degraded savannah and shrubby savannah areas.
Based on these results, there was an increase of 1% and 9.78% of Bare land and Cultivated land respectively. The increase of these two LULC units over Africa has been explained by many authors such as (Koglo et al., 2018). The same trend has been examined in Mali by (CILSS, 2016) where the increase of the cultivated land increased by a factor of 2.3 for the period 1975-2013 (38 years) equivalent to an average annual increase of 3.5 percent.
For many authors, this increase could be related to the increase in population  and their food demand (Koglo et al., 2018;CILSS, 2016;Daou et al., 2019). The decrease of different types of Savannah area is represented by a decrease of herbaceous savanna area of 24.4% degraded Savannah 10.32% and 3.6% for shrubby savannah. According to Koubodana et al. (2019) and Aziz (2017), the decrease of savannah is further amplified by using wood as energy sources and the lack of forest management. In the Koda catchment, deforestation is still occurring.
No water body is observed in the study area as a land cover unit, that lack of water bodies can be explained by the following reasons: the Landsat image used is dated of February (dry period) corresponding to the period where there is no surface water because only few surface water characterize the Koda catchment are present during the rainy season. The existence time of all these surface water bodies is driven by the amount of the rainfall received within the study area.

Accuracy Assessment
In this study, all the values of Kappa coefficient (K) are greater than 75%, therefore, the results are considered excellent. The results of K values are outlined in Table 2.

Model Performance Evaluation
The model was calibrated on the basis of monthly time step of GWL. Three piezometer data were used for this calibration purpose. Based on the availability of data, the calibration was carried out using ten years (2008-2017) for piezometer F1, three years (2016-2018) for Nossombougou N1 and five years (1987)(1988)(1989)(1990)(1991) for Kossaba K1 observed monthly GWL. The calibration process was ended when satisfactory values of the coefficient of determination-R 2 (2) and the Nash-Sut-Journal of Geoscience and Environment Protection cliffe efficiency-NSE (3), were achieved. Thereafter, the model was validated for GWL by using the calibrated model to simulate GWL for periods other than those used for the calibration and without any further changes to the model GWL parameters. The GWL was validated O. Diancoumba et al. with two years (2018Diancoumba et al. with two years ( -2019 for piezometer F1 and one year (2019) for Nossombougou N1 monthly observed GWL data. There was no recent available data for the piezometer Kossaba K1 which could be used in its validation. Therefore, only the first two piezometers were used for the validation. The model performance was satisfying and the values are outlined in Table 3.  From 1990 to 2016, the groundwater recharge has been considerably varying over Koda catchment. The mean annual recharge decreases from 1990 to 1998 while its increase is observed from 1998-2016. The mean annual groundwater recharge of the Koda catchment for the LULC in the years 1990, 1998 and 2016 are 117, 93 mm/y and 109 mm/y, respectively for the piezometer F1; 278, 21 mm/y and 257 mm/y respectively for piezometer K1 and finally 118, 87 mm/y and 108 mm/y for the piezometer N1. The variation of the recharge pattern for the years 1990, 1998 and 2016 in the three piezometers (F1, K1 and N1) over Koda catchment is shown in Figure 7.

Effects of Land Use/Land Cover Change on Groundwater Recharge over Koda Catchment
The overall mean recharge variation from 1990-2016 over the entire catchment of Koda is outlined in Table 4. There was decrease of 24% in groundwater recharge between 1990 and 1998, while it was 21.4% increase between 1998 and 2016. Generally, for the 27-year period of investigation, the change in land use/land cover accounts for only 8.32% reduction in mean groundwater recharge occurrence from 1987 to 2016.
There exists a variation in recharge pattern in the study area during the period 1990-2016. The change of the groundwater recharge is influenced by the change in Land Use/Land cover which is manifested by the transition from one LULC class to another. According to Ashaolu et al. (2019) the degree of transition that took place is controlling the influence of groundwater recharge pattern over the Koda catchment. The characteristics of the dominant LULC units at a particular area of the catchment is also one of the principal parameters that controls the groundwater recharge.

Conclusion
The effect of LULC change on groundwater recharge in Koda catchment is that there are only little changes in recharge pattern that can be associated with to change in LULC. The overall mean annual recharge revealed a decrease of 8.4%