Studies On The Spatial Simulation And Estimation Of Soil Resistivity Based On Remote Sensing

Soil resistivity is not only a important parameter of electric mechanism but also basis of modern lightningproof grounding engineering which directly decide proper design for the grounding equipment and lightningproof engineering. Moreover, as characterization of soil electrical properties, soil resistivity is an important aspect of soil science, many other physical and chemical factors, involve its soil moisture, salt, cationic and organic content, temperature, texture and so on, influencing soil resistivity to different extents. Therefore, by virtue of influencing factors, the spatial simulation and estimation of soil resistivity is of great significance in applied research for lightning disaster mitigation, engineering geology survey, hydrology geology exploration and environmental monitoring fields.Taking a case study of Nanjing municipality and other15counties (municipalities) around in the lower-middle Yangtze Plain area, soil physi-chemical properties of the sampling points were analyzed by field survey, indoor chemical analysis and remote sensing retrieval in terms of three vegetation types of grassland, farmland, forestland. First of all, areas of the three vegetation types were extracted by remote sensing image classification within the study area in order to fulfill follow-up research of spatial simulation and estimation of topsoil resistivity. Then, the overall spatial statistics of the soil resistivity were acquired in different depth layers under different cover types using classical statistical methods. Meanwhile, the distribution characteristics of the main influence factors and their effects on soil resistivity were explored; The5spatial difference related models of reciprocal, logarithmic, linear, partial least squares (PLS), partial least squares quadratic polynomial (PLSQM) were used to capture the quantitative relationship between soil resistivity and its main influence factors under different vegetation types, and the models’fitting precision were compared as well. Retrieval of soil moisture and temperature spatial distribution were accomplished covering the study area while ordinary kriging interpolation was used for analyzing the spatial distribution of TSS and CEC, so as to attain the main parameters acquired by the simulation process; Finally, the partial least squares quadratic polynomial model and BP neural network (BPNN) were chose to simulate spatial variation of soil resistivity under different land cover types, compared with the results of ordinary kriging (OK) and inverse distance weighted (IDW) methods.As a consequence, The results show that there are obvious differences among soil resistivity in layer of0～20cm under the three vegetation types with the order of grassland> farmland> forestland. And the soil resistivity in0～10cm soil is generally a little higher than that in10-20cm. The PLSQM model has a relatively highest model fitting precision among the5multivariate regression models. Models according to Grassland, forestland and farmland have the determination coefficients of0.837,0.800,0.605. The geostatistical analysis results show that the best fitting model for TSS, CEC and soil resistivity are exponential model, Gaussian model and the exponential mode respectively. The retrieval results of the spatial distribution data of soil moisture present high accuracy, its correlation coefficients between the observed value and the predicted value is0.73, mean relative error (MRE) is19.66%, root mean square error (RMSE) is4.76respectively, while the correlation coefficients between the observed value of soil temperature and the predicted value of land surface temperature (LST) is0.78, Precision assessment result indicates that, compared with OK, IDW and PLSQM model, BPNN method performed relatively better, whose correlation coefficient between the observed value and the predicted value was increased by0.12,0.15and0.05, MRE was decreased by11.63%,7.74%and3.86%, RMSE was decreased by2.59,2.23and1.13. Furthermore, the spatial simulation map produced by BPNN method is closer to the actual situation so as to have a better application potential to a certain extent.