Quantitive Estimation Of Soil Characteristics Based On Hyper Spectral Data And Modis Images

As a non-renewable natural resource, soil is one of the crucial conditions that supports the survival and development of human society. Recent years, over deforesting and over developing with incomplete understanding of soil conditions caused a series of problems such as the declining of farmland, the soil degradation, and environment deterioration. Hyper spectral remote sensing technique with special advantages of high spectral resolution and strong band continuity. It can monitor and analyze crops vigor and soil environmental factors that affect crops production. These possess practical significance for agricultural production and soil quality monitoring, eco-environment management and maintenance.This research is taken place in Daqing of Heilongjiang province. The soil condition in this area is studied by integrating soil spectral characteristic (indoor reflectance measurements among350nm-2500nm and MODIS image. In this research,126soil samples was collected, SOM, total phosphorus (P), potassium (K), nitrogen (N),Heavy Metal(Co、Cd、F、Hg、 V、Se、Cr、C、As、Pb、Ni、Mn、Zn),Total salt,Total Alkalinity,Exchange Sodium Percentage(ESP) concentration were estimated by using the Partial Least Squares Regression (PLSR) and Back-Propagation Neural Network (BPNN) model. After that, MODIS images were used to create spatial distribution maps for soil contents. The results showed:(1) According to the different soil types and land use types, collected soil samples, analysis of SOM, N, P, K Heavy Metals (Co、Cd、F、Hg、V、Se、Cr、Cu、As、Pb、 Ni、Mn、Zn), Total Salt Total Alkalinity and ESP content.(2) Soil mechanical composition and soil moisture, soil organic matter and so on are the important factors that affect soil spectral curve features. When soil moisture content of70%is near saturation, The two absorption peaks of the moisture band width than before in band1400nm and1900nm.(3) Soil spectral preprocess play a decisive role for the establishing model. In the research we choice19kinds of preprocess.including Standard Normal variables, Multiplicative Scatter Correction, Mathematics, Continuum Removed and it’s the derivative value.The correlation index significantly increased between the index content and soil spectral after treatment.(4)Estimated SOM.Total N P K, Heavy Metal(Co、Cd、F、Hg、V、Se、Cr、Cu、 As、Pb、Ni、Mn、Zn),Total Salt, Total Alkalinity, ESP concentration by using the Partial Least Squares Regression (PLSR) and back-propagation neural network (BPNN) model.Model is high precision, low RMSE, it s feasible of accurate estimates for soil content.(5) Using hyper spectral bands to simulate MODIS band,then established SOM, Total N, P K, Heavy Metals (Co、Cd、F、Hg、V、Se、Cr、Cu、As、Pb、Ni、Mn、Zn). Total salt, Total Alkalinity, ESP by PLSR model.Most soil index estimation accuracy above0.7, Total alkalinity, Hg and P estimation accuracy is low, just only be used as a rough estimating.(6)Based on MODIS images and result of SMLR model, creating spatial distribution maps for s SOM, Total NPK, Heavy Metals (Co、Cd、F、Hg、V、Se、Cr、Cu、As、 Pb、Ni、Mn、Zn),Total Salt, Total Alkalinity and ESP in Daqing region.