5:15 PM - 6:45 PM
[HGG02-P05] Satellite Remote Sensing for Soil Erosion Assessment in Semi-Arid Area of Central Kenya
Keywords:Remote sensing, Soil erosion, Kenya, NDVI, Landsat, Spectral Mixture Analysis
The Il Polei sub-location, located in Laikipia County, Central Kenya, is situated in a semi-arid grazing area at an elevation of 1,750 to 1,850 meters, where concentrated rainfall during the rainy season has caused soil erosion, including gully erosion. This study was applied two methods to Landsat 7, 8, and 9 imageries; calculation of normalized indices (NDVI/SAVI/EVI/NDSI) and spectral mixture analysis; to assess the extent of soil erosion over the past decades (2000-2023). The objectives of our presentation are to discuss 1) the differences in properties of each normalized index, 2) the indices capable of accurately detecting soil erosion, and 3) the relationship between erosion areas and geomorphic variables.
According to the long-term trends of NDVI/SAVI/EVI/NDSI, pixels showing increased vegetation dominated over the past 24 years, which is associated with Opuntia Stricta, one of the invasive alien plant species (IAPS). However, areas where soil erosion has continued to progress were distributed in gullies and riverbeds. The distribution of the strength of erosion trends (Sen’s Slope) of NDVI/SAVI/EVI was very similar, reflecting similar surface changes. This is thought to be partly due to the relatively low proportion of vegetation each the pixels, suggesting that the adjustment factors in SAVI and EVI to mitigate the effects of soil may not be effectively functioning. However, due to the similarity between the spectra of low shrubs and bare soil, NDSI could not accurately assess the expansion of erosion. The strength of erosion trends was significantly higher in river channels, indicating the progression of lateral erosion and recent human-induced impacts such as sand harvesting.
Trends detected by spectral mixture analysis could more accurately detect the expansion of riverbed and bare soil areas than NDVI. While NDVI could not assess erosion in areas with weak lateral erosion, indices derived from spectral mixture analysis originating from riverbed sediments detected the expansion of all riverbeds. The expansion of bare soil areas was effectively detected by indices derived from low shrubs. It seems that the expansion of soil erosion could be evaluated by mitigating the effects of the temporary rapid growth of herbaceous vegetation during rainy years and detecting the gradual retreat of low shrubs. When detecting specific changes in land cover, spectral mixture analysis is more suitable than traditional normalized indices.
Results from logistic regression analysis with the presence of gullies as the dependent variable indicated that steep slopes and concave topography compared to the surroundings favor the development of gullies. As the erosion intensity of gullies increases with the convergence of runoff and runoff volume, gully distribution is concentrated on concave slopes. In considering of, however, the spatial scale of gullies is remarkably smaller compared to Landsat resolution (30m), the above results of the logistic regression analysis may also encompass the effects of erosion occurring in bare soil pixels surrounding the gullies. Moreover, areas, where sheet erosion and rill erosion causing expansion of bare soil are significant, are presumed to have active gully erosion associated with them. Although the correlation between Sen's Slope, derived from normalization indices or spectral mixture analysis, and geomorphic variables generated from DEM, was very weak, this result quantitatively suggests that erosion in this area is a complex combination of sheet erosion, rill erosion, and gully erosion, making it difficult to explain solely based on simple geomorphic variables.
According to the long-term trends of NDVI/SAVI/EVI/NDSI, pixels showing increased vegetation dominated over the past 24 years, which is associated with Opuntia Stricta, one of the invasive alien plant species (IAPS). However, areas where soil erosion has continued to progress were distributed in gullies and riverbeds. The distribution of the strength of erosion trends (Sen’s Slope) of NDVI/SAVI/EVI was very similar, reflecting similar surface changes. This is thought to be partly due to the relatively low proportion of vegetation each the pixels, suggesting that the adjustment factors in SAVI and EVI to mitigate the effects of soil may not be effectively functioning. However, due to the similarity between the spectra of low shrubs and bare soil, NDSI could not accurately assess the expansion of erosion. The strength of erosion trends was significantly higher in river channels, indicating the progression of lateral erosion and recent human-induced impacts such as sand harvesting.
Trends detected by spectral mixture analysis could more accurately detect the expansion of riverbed and bare soil areas than NDVI. While NDVI could not assess erosion in areas with weak lateral erosion, indices derived from spectral mixture analysis originating from riverbed sediments detected the expansion of all riverbeds. The expansion of bare soil areas was effectively detected by indices derived from low shrubs. It seems that the expansion of soil erosion could be evaluated by mitigating the effects of the temporary rapid growth of herbaceous vegetation during rainy years and detecting the gradual retreat of low shrubs. When detecting specific changes in land cover, spectral mixture analysis is more suitable than traditional normalized indices.
Results from logistic regression analysis with the presence of gullies as the dependent variable indicated that steep slopes and concave topography compared to the surroundings favor the development of gullies. As the erosion intensity of gullies increases with the convergence of runoff and runoff volume, gully distribution is concentrated on concave slopes. In considering of, however, the spatial scale of gullies is remarkably smaller compared to Landsat resolution (30m), the above results of the logistic regression analysis may also encompass the effects of erosion occurring in bare soil pixels surrounding the gullies. Moreover, areas, where sheet erosion and rill erosion causing expansion of bare soil are significant, are presumed to have active gully erosion associated with them. Although the correlation between Sen's Slope, derived from normalization indices or spectral mixture analysis, and geomorphic variables generated from DEM, was very weak, this result quantitatively suggests that erosion in this area is a complex combination of sheet erosion, rill erosion, and gully erosion, making it difficult to explain solely based on simple geomorphic variables.