A Study Of High Spatial And Temporal Remote Sensing Fusion Model For Soil Erosion Risk Assessment

Vegetation is one of the major soil erosion controlling factors, varying significantly throughout the year. Remote sensing technology has become the primary means of estimating vegetation cover factor in soil erosion assessment studies. However, key challenges are still imposed by technological limitations for soil risk mapping at the regional scale due to the trade-off has to be made between the spatial and temporal resolutions of remote sensors, which often preventing the acquisition of the multi-temporal high spatial resolution remote sensing data. Data fusion of different spatial and temporal resolution data would be a good alternative for generating high spatial and temporal resolution remote sensing data, which can contribute more valuable information for regional soil erosion risk assessment.In the study presented herein, we developed a new MODIS and Landsat data fusion model (MLDFM) based on the in-depth analysis of existing relevant algorithms. MLDFM is aimed at eliminating the BRDF effects, and resolving the mixed pixels problem simultaneously. Also the MLDFM model was compared with other most common models (GAIFM, STARFM, and Semi-physical model) in order to verify the effectiveness of the proposed new model. Furthermore, we produced high spatial and temporal resolution vegetation index dataset for the purpose of depicting intra-annual vegetation dynamics. The main results were as follows:(1)Through incorporating the MODIS BRDF/Albedo parameters and land cover data, the proposed MLDFM simultaneously resolve the BRDF effect and mixed pixels problem, which ensured its superiority from a theoretical design standpoint.(2)In the study area, MLDFM produced a better result in homogeneous region as well as heterogeneous landscapes. Firstly, the predicted image from MLDFM has natural texture as a whole, which is unlike other results with more or less patchiness, smoothing and sharpening phenomenon. Secondly, taking the green, red, and NIR bands as examples, the ratios of prediction difference to temporal difference of MLDFM result are0.51,0.55, and0.34respectively, by contrast, which were greater than0.6, and even more than1at the red band when other models were used. The scatter plots of real reflection and the predicted ones shows that the correlation coefficient of MLDFM results are0.9,0.91,0.92for green, red, and NIR-infrared band respectively, while others are less than0.8.(3)The intra-annual changing curve of vegetation index (NDVI), which calculated with time-set reflectance predicted by MLDFM, show that the trend is consistent with MODIS NDVI on the whole, which could reflect the seasonal changes of typical vegetation in study areas. In addition, the result from MLDFM is more sensitively to the disturbance of land cover types, which has important significance for monitoring soil erosion.