| The prediction accuracy of soil organic matter can be improved to an extent by observed data combined with remote sensing images, especially in the prediction process of soil organic matter in large scale, it exists deviation in local special topography units if predicting soil organic matter sole depend on observed data. in this study, AWIFS(Advanced wide field sensor), MODIS(Moderate resolution imaging Spectroradiometer) and SRTM(Shuttle Radar Topography Mission) data, whose spatial resolution are 56 meters, 250 meters and 90 meters, were used to predict spatial distribution of soil organic matter in Shaanxi province, combined with 501 observed data by RF(Random Forest) model,. The influence of resolutions and acquired data of remote sensing data to soil organic matter prediction were analyst, then the spatial distribution of soil organic matter in six topographical area were summarized, and the prediction accuracy of soil organic matter based RF model and OK(Ordinary krig) model were compared, the prediction result of this study were compared with pre-existing study.The result shows that:1. the discrepancy of soil organic matter is obvious in north-south direction in Shaanxi province, it is highest in Qinling region and Daba mountain, most of soil organic matter content in this region are bigger than 25 g kg-1, the soil organic matter content belongs to medium in the south of Loess Plateau region and the value of which are in 22-30 g kg-1, the content of soil organic matter is lower in Guanzhong Plain and Hanzhong low mountains and hills region, most value of which are in 13-33 g kg-1, and lower than 10 g kg-1 in local area in the east of Guanzhong Plain, and soil organic matter content is lowest in the north of Loess Plateau and the most region of sandy-windy Plateau, it can hardly reached 10 g kg-1 in this region.1. Depended on the relative importance and correlation coefficients of factors to soil organic matter, factors were selected to join the process of soil organic matter prediction in Shaanxi, selected factors were elevation, slope, longitude, lantitude and vegetation factors(NDVI、TVI、RSR and RVI) obtained by different remote sensing data with muti-resolution and muti-dates.2. The discrepabcy of soil organic matter is obvious from south to north in Shaanxi, it was highest on Qinling and Daba Mountain, the content are bigger than 25 g kg-1, and it was in 22-30 g kg-1 on Loess Plateau, the soil organic content belongd to medium in this area, it was lower in most area of Guanzhong Plain and Hanzhong low mountains and hilly area with the value of 4.43-16 g kg-1 and 9.48-30 g kg-1, in local of the east of the Guanzhong Plain, north of Loess Plateau and most part of Windy Sandy Plateau, the content of soil organic matter is below than 10 g kg-1.3. There exists influence on soil organic prediction accuracy for the spatial resolution of remote sensing images, in Shaanxi province, the prediction results were better based AWIFS data than MODIS data. In Dingbian County, the prediction accuracy is better based TM data than AWIFS data.4. It has little influence for soil organic prediction by remote sensing images acquired in different dates, in total, the prediction value of soil organic matter is lower by remote sensing images acquired in spring than autumn.5. Under the condition of different sampling set, the prediction effect of soil organic matter is better based RF model than Ordinary kriging model. Meanwhile, compared with existing results, the prediction accuracy in this study is under expect, for most of mean error in independent validation set are no more than 3g kg-1,the correlation coefficient of prediction value and observed value of soil organic matter if around 0.7.6. Elevation is the most importance factors which influence the prediction of soil organic matter in Shaanxi, when the spatial resolution of remote sensing images lower, the influce degree of graphic location of sampling sits and slope increased, and the influce degree of vegetation factors decreased. |
PDF Full Text Request