Mapping Soils of the Uasin Gishu Plateau, Kenya with Limited Data Using a Knowledge-based Inference Mapping Approach
Kenya has a current population of ~47 million people living on an arable land area of ~56,000 km2 (~1.7 times the area of Indiana). With an annual growth rate of ~800,000 people, there is tremendous pressure on the soil resource to produce more food. Understanding the available soil resources is important to increase and sustain agricultural productivity. Soil surveys made with traditional approaches are too expensive, and new surveys must rely on limited data, legacy information and new digital mapping techniques. We used three knowledge-based inference soil mapping approaches to predict soil types and properties for the Uasin Gishu Plateau in western Kenya. Available data included legacy soil surveys, expert knowledge, aerial imagery, terrain attributes, and landform pattern recognition and classification based on geomorphons. Terrain attributes, including slope gradient, multi-resolution valley bottom flatness index, multi-resolution ridgetop flatness index, topographic position index, elevation, and profile curvature, were calculated from the Shuttle Radar Topographic Mission (SRTM) 30 m digital elevation model (DEM) and then used to quantify soil-landscape relationships. The k-means clustering and fuzzy logic soil mapping approaches were utilized to model soil-landscape relationships to produce raster-based maps of predicted soil types, effective soil depth, soil moisture storage capacity, and soil drainage classes. The fuzzy logic soil type map performed slightly better (kappa coefficient, k = 0.68; overall accuracy = 0.76) than the map based on k-means clustering analysis (k =0.59; overall accuracy=0.68). The accuracy for the effective soil depth map based on fuzzy logic was better (R2 = 0.56; RMSE = 11; ME = 1.1) compared to the best existing soil map (R 2 = 0.34; RMSE = 27; ME = 8). The third mapping approach was based on landform pattern recognition and classification using geomorphons calculated from the 30 m SRTM DEM using the module r.geomorphons “add-on” in GRASS GIS with look up distance (L) values of 10, 20, 30, 40, 50, 60, 70, 80, 90 and 100 cells and a flatness threshold of 0.5 degrees. An L value of 20 cells performed best (k = 0.52; overall accuracy = 0.62) followed by an L value of 30 cells ( k = 0.50; overall accuracy = 0.58). The L value of 30 cells, however, better captured the geomorphology and soil-landscape relationships based on expert knowledge, legacy data, and the fact that the bottomlands pattern was more continuous than for an L value of 20 cells. Of the three approaches, the fuzzy logic approach performed better and produced a map that best represents the soil-landscape relationships on the Uasin Gishu Plateau. The results of these studies produced more spatially detailed, higher resolution soil maps compared to the existing soil maps, and these new maps are likely to be more useful for soil, crop and land use management decisions.
Schulze, Purdue University.
Off-Campus Purdue Users:
To access this dissertation, please log in to our