Spatial Distribution Of Inversion To Soil Fertility Factors In The Low Mountain Districts Of Northeast Sichuan On TM Data

In order to save the cost to predict the soil fertility and space distribution features in the intensive and convenient landscape type complex areas. This study establishing the prediction model of the remote sensing spectral information from Landsat-5 TM and the pH, digital elevation model, normalized difference vegetation index, soil color index, Normalized difference wetness index, slope position, aspect, slope gradient, buffers of river, land use and soil type with the soil fertility in the low mountain districts of Northeast Sichuan. The soil fertility was corrected by Monte Carlo Method with the ground factors which didn’t participate in the modeling, then comparison between measured and predicted values of soil fertility for test sample through trend analysis comparison and error statistics. And predict the distribution features of the measured data by using ordinary kriging method.The following results are:soil fertility was poor in the low mountain districts of Northeast Sichuan, The soil organic matter (SOM) was 17.78±5.13 g kg-1, total nitrogen (TN) was 0.64±0.40 g kg-1, and available nitrogen (AN) was 82.66±31.66 mg kg-1, available phosphorus (AP) was 10.08±8.42 mg kg-1, available kalium (AK) was 69.37±19.94 mg kg-1. Study area SOM was gradually decreasing distribution strip from the northwest to the southeast, TN was gradually decreasing in plaque from the northeast to the southwest, AN was gradually decreasing and increasing from the middle to the surrounding, AP and AK were patchy gradually decreasing trend from the west and east to the middle.The accuracy of forecasting model and the inversion profile stated that remote sensing spectral information can be better used for predictive modeling (P=0.000). After bringing in ground factors, the R2 of SOM, TN, AN, AP and AK raised 81.9%,167.4%,109.7%, 97.1%,87.3% and precision and inversion effect of establishing prediction model were more efficiency by leading into the ground parameters than using spectral information model alone, which indicated that ground parameters were the optimization of the model.The soil fertility 4 kinds of estimation value which corrected by Monte Carlo Method were more close to the measured values than the predicted value by tendency distribution, the root mean square Error of the estimation value was lowered 6.22%,29.07%,22.83%, 31.56%,22.23% than the predicted value, that Monte Carlo model can improve the prediction accuracy, but correction of the different correction factors had different effect.The predicted value indicated the spatial variation of the soil fertility at the designated area, the measured of SOM corrected by the spectral reflectance of band 1 was the best prediction model, the measured of TN corrected by SCI was the best prediction model, the measured of AN corrected by the soil type was the best prediction model, the measured of AP corrected by NDWI was the best prediction model, the measured of AK corrected by the spectral reflectance of band 2 was the best prediction model.