Study on the Effect of Fractional Derivative on the Hyperspectral Data of Soil Organic Matter Content in Arid Region

Cheng-Biao Fu,Hei-Gang Xiong,An-Hong Tian

doi:10.1155/2019/7159317

Cheng-Biao Fu, Hei-Gang Xiong + Show 1 more

Open Access

https://doi.org/10.1155/2019/7159317

Copy DOI

Abstract

Discussion on the application of fractional derivative algorithm in monitoring organic matter content in field soil is scarce. This study is aimed at improving the accuracy of soil organic matter (SOM) content estimation in arid region, and the undesirable model precision caused by the missing information associated with the larger discrepancy between conventional integer-order, i.e., first order and second order, derivative, and raw spectral data. We utilized fractional derivative (of zeroth order to second order in 0.2-order interval) processing on the field spectral reflectance (R) of the salinized soil sample from Fukang, Xinjiang, and its square root-transformed (R), log-transformed (lgR), inverse-transformed (1/R), and inverse log-transformed (1/lgR) values. The correlation coefficient of each fractional derivative of transformed value with SOM content was calculated. The simulation showed the derivative reflectance value approximates zero. When increasing from zeroth order to first order, the derivative curve gradually aligns to the first-order curve, and the destination alignment was also seen while increasing from first order to second order. The significance test of 0.05 showed initial increase and later decay of bands in the five spectral transformations as the order increases. For specific bands, the derivative algorithm clearly justifies the correlation between soil spectra and organic matter content, and all of the absolute highest correlation coefficient values were obtained at fractional orders. When compared with integer-order derivative, fractional derivative is significantly better in improving correlation, showing overall superiority. The result supports the application of fractional derivative in the hyperspectral remote monitor of SOM in arid zone, which may in turn realize the timely and accurate SOM monitor in arid zone, and provides the basis for ecological restoration.

Highlights

Soil organic matter (SOM) refers to all carbon-containing organic matter found in soil, including animal and plant debris, microorganisms, and the organic products resulted from biological activities; it helps improve soil structure and maintain porosity and is an important indicator of fertility [1,2,3]
Hou et al [16] combined the lab SOM content and hyperspectral data to compare the modeling precision of simple linear regression, multiple linear stepwise regression, and PLSR; they confirmed that second-order derivative PLSR is the optimal model, and the best sensitivity bands were found at 640∼790 nm
As no application of fractional-order derivative in field SOM content testing has been reported to date, the purpose of this study is to explore the feasibility of utilizing fractional-order derivative in monitoring SOM content by coupling field hyperspectral data with the lab SOM content data and provide reference for the development of mass-scale, efficient quantitative SOM monitoring with hyperspectral satellite-remote sensing technology

Summary

Introduction

Soil organic matter (SOM) refers to all carbon-containing organic matter found in soil, including animal and plant debris, microorganisms, and the organic products resulted from biological activities; it helps improve soil structure and maintain porosity and is an important indicator of fertility [1,2,3]. When SOM mass fraction is less than 2%, because of the absorption of spectral reflectance of organic matter is greatly reduced, there is a big challenge to accurately determine the response band of SOM. The spectral response band determination is the key to estimate SOM content accurately and effectively. There are few studies on the hyperspectral estimation of organic matter in desert soil, and the traditional integer-order derivative algorithm has been widely applied in studies building models for hyperspectral monitoring SOM. Hou et al [16] combined the lab SOM content and hyperspectral data to compare the modeling precision of simple linear regression, multiple linear stepwise regression, and PLSR; they confirmed that second-order derivative PLSR is the optimal model, and the best sensitivity bands were found at 640∼790 nm. Its wide applicability is due to the fact that integer-order derivative models lack the precision when presenting the fractional order-based systems in real life

Objectives

Results

Discussion

Conclusion