Creating initial digital soil properties map of Afghanistan

Afghanistan is a country with a population of more than 31 million people and is located in south central Asia. The total arable land in the country is 12%, 5% is irrigated and the remaining 7% is rainfed. Due to the lack of available soil information, poor farming practices and land management planning severely affect the yield of agriculture products. In order to ensure sustainable agriculture and prevent land degradation problems, understanding spatial variability of soil is crucial. The overall objective of this research study was to use digital soil mapping techniques to identify the soil resources and generate a spatially explicit soil map of a 8,358,160 ha pilot study area. The specific objective is to develop a version 1 map of the six Northern provinces of Afghanistan. Several techniques such as artificial neural networks, multiple regression analysis, hybrid geostatisitcal approaches are developed to create digital soil maps. However, most of these procedures required large amounts of data to create digital soil maps at a useful resolution. Countries like Afghanistan have limited available data and it is difficult to develop the map based on the aforementioned procedures. For this research, we utilized a knowledge based approach utilizing fuzzy logic to create a version 1 map with limited point data. The fuzzy logic maps are developed based on five soil forming factors; therefore soil knowledge and soil landscape relationship is required. From the ecoregion map of the study area, we assumed that climate, organisms and time were constant and geology and topography were the deriving factors of soil formation. Therefore, the fuzzy property map of the study area was developed from geology and geomorphon composition. In ordered to capture the variability of the soil, we used those terrain attributes which have close relationship with water redistribution. geomorphon was used to classify the landforms of the study area. As a part of the fuzzy process, membership curves are required to define the soil similarity vectors. Traditionally, the membership curves are manually defined by the soil scientists based on their tacit knowledge of the soil and landscape. Even though, the manual method adequately predicts soil properties, it is time consuming and limits the application of fuzzy logic. In order to make fuzzy logic an easy and time effective approach for developing functional property maps, it is essential to use the Automatic Landform Inference Mapping (ALIM) model to automatically generate the accurate membership functions. Purdue University developed ALIM model was used for this research to define the membership functions. To generate the membership functions, ALIM model combines the digital elevation model derived terrain attributes to the soil classes. The determined membership values and soil property values were then assigned to the Zhu (1997), equation to predict the soil property maps of the pilot area. The overall results showed that predicted properties generally followed the landscape patterns but in some cases, they did not. The accuracy test of Normalized Root Mean Square Prediction Error (RMSPEr) also showed that the model prediction was insignificant. Several factors such as few data points, inaccurate coordinate location of the data points and low 90 m resolution DEM were assumed to be the reason for inaccurate assessment. Overall, the methods did produce a spatially explicit map that will be useful for the next map version. More data and a higher resolution DEM is necessary for improving the soil property predictions of the pilot area