Ouedraogo, Issoufou
[UCL]
Defourny, Pierre
[UCL]
Vanclooster, Marnik
[UCL]
Groundwater is a crucial natural resource supporting the development of the African continent, but it is subjected to many pressures. To support Africa policy in groundwater management in such context, nitrate indicators at the pan-African scale are required. Therefore, it is important to identify which aquifer systems/groundwater resources and settings are most vulnerable to degradation. To this regard, we addressed a significant knowledge gap for groundwater pollution at the continental scale in Africa by developing methods for assessing groundwater pollution risk at the African scale. We produced the first groundwater vulnerability map by using the generic DRASTIC vulnerability indicator (Ouedraogo et al., 2016). However, in order to consider effective tools that could use in environmental planning and management, the end products of DRASTIC method (i.e., groundwater vulnerability map indicator) must be scientifically sound, meaningful and reliable. For this reason, statistical models of nitrate pollution in the African scale were developed. We evaluated a machine-learning method, Random Forest (RF), for its ability to predict nitrate in groundwater with environmental variables, and compare it with multiple linear regression (MLR). As predictors, we collected a comprehensive GIS database of thirteen spatial attributes related to land use, soil type, hydrogeology, topography, climatology, regions types and nitrogen fertilizer application rate. We used nitrate dataset from various sources (Ouedraogo and Vanclooster, 2016). RF was found highly capable of predicting nitrate at the African scale and outperformed MLR in all performance statistics that were compared. For example, RF and MLR yielded respectively an R2 of 97 % and 64 % of the observed logtransfrormed mean nitrate concentration observations. The results show that RF is an effective and versatile machine-learning method for nitrate prediction at continental scale for its high performance, ease of use, and utility in data analysis. This study will help to awareness the managers International Basin Authorities or Transboundary Basin Organizations (TBO) in Africa and constitutes essential tools for transboundary groundwater management. Because the nitrate issue in Africa needs special attention in order to avoid the problems experienced in some areas mainly in Europe.


Bibliographic reference |
Ouedraogo, Issoufou ; Defourny, Pierre ; Vanclooster, Marnik. Modelling nitrate concentrations at the pan-African scale: A random forest approach.The 1st Atlas Georesources International Congress (AGIC) (Hammamet, Tunisia, du 20/03/2017 au 22/03/2017). In: Book of abstract, CERTE, Tunisia2017, p. 137 (O5-T3) |
Permanent URL |
http://hdl.handle.net/2078.1/183775 |