| dc.description.abstract | The Universal Soil Loss Equation (USLE) is a widely used mathematical model that describes soil erosion processes. Among six different soil erosion risk factors of the equation, the cover-management factor (C-factor) is related to the land use type. Commonly used values of C-factor are from 0.001 to 1, so C-factor might cause a thousandfold difference in USLE calculation. The traditional methods for estimating C-factors are in situ experiments, soil physical parameters models, look up tables with land use map and regression equations between vegetation indices and C-factors. However, the methods described above have limitations in data acquisitions or updating; they may also result in unreasonable estimations. Thus, this research tries to construct a relationship between C-factor and spatial data with a data mining algorithm.
The study area is Shimen reservoir watershed in Taiwan. The major land cover here is forest. The SPOT-4, SPOT-5 images, 10m DEM, a land use map and other spatial data were used in this research. The SPOT images were processed with basic radiometric correction and topographic correction with Minnaert constant, which can reduce the difference of apparent radiance between phototropic and apheliotropic regions. After that, Gray Level Co-occurrence Matrix (GLCM) was used to build statistical feature indicators for data mining. A decision tree classifier was used to rank influential conditional attributes in the preliminary data mining. Then, factor simplification and separation were considered to optimize the model and the random forest classifier was used to analyze 9 simplified factor groups.
The overall accuracy of data mining model check points is about 73% - 79% with a kappa value of 0.7-0.758. According to the analysis results, the overall agreement between the look-up table approach and the data mining generated C-factor is about 60.5% - 62.4%. The calculated total soil erosion amount in 2004-2008 according to the data mining results is about 74.032 - 129.756 ton/ha-year with the sediment delivery ratio and correction coefficient. Comparing with the sediment and dredging data published by the Shimen watershed administration authority, the experimental results indicated that using spatial analysis to determine C-factor can produce more reasonable soil erosion estimation. | en_US |