Evaluation of a Linear Mixing Model to Retrieve Soil and Vegetation Temperatures of Land Targets

Jinxin Yang,Massimo Menenti,Yaokui Cui,Li Jia,Jie Zhou

doi:10.1088/1755-1315/17/1/012272

Abstract

A simple linear mixing model of heterogeneous soil-vegetation system and retrieval of component temperatures from directional remote sensing measurements by inverting this model is evaluated in this paper using observations by a thermal camera. The thermal camera was used to obtain multi-angular TIR (Thermal Infra-Red) images over vegetable and orchard canopies. A whole thermal camera image was treated as a pixel of a satellite image to evaluate the model with the two-component system, i.e. soil and vegetation. The evaluation included two parts: evaluation of the linear mixing model and evaluation of the inversion of the model to retrieve component temperatures. For evaluation of the linear mixing model, the RMSE is 0.2 K between the observed and modelled brightness temperatures, which indicates that the linear mixing model works well under most conditions. For evaluation of the model inversion, the RMSE between the model retrieved and the observed vegetation temperatures is 1.6K, correspondingly, the RMSE between the observed and retrieved soil temperatures is 2.0K. According to the evaluation of the sensitivity of retrieved component temperatures on fractional cover, the linear mixing model gives more accurate retrieval accuracies for both soil and vegetation temperatures under intermediate fractional cover conditions.

Highlights

Multi-angular TIR remote sensing data are considered capable of reflecting sub-pixel structure as well as component temperatures and provides a new information source in terrestrial energy balance studies [1,2,3,4]
The most previous studies resorted to inversion of the physical thermal model to retrieve the component temperatures, which is usually time consuming and difficult to apply to satellite image data
Ignoring the reflected atmosphere long wave radiance, the thermal infrared radiance of a soil-vegetation canopy can be simplified as a linear mixing of exitance of the soil and vegetation components weighted by their fraction of areas as below, B(Tb(θ)) = Fc(θ)εf(θ)′B(Tf) + (1 − Fc(θ))εs(θ)′B(Ts) where B is the Planck function, Tb(θ) is the brightness temperature of a scene or a pixel, θ is the observing zenith angle, Fc(θ) is the fractional cover of vegetation in a scene or pixel, εf(θ)′ is the effective emissivity of vegetation calculated by Eq (2), Tf is the mean radiometric temperature of vegetation, (1 − Fc(θ)) is the fractional cover of soil, εs(θ)′ is the effective soil emissivity calculated by Eq (3), and Ts is the mean radiometric temperature of soil

Summary

Introduction

Multi-angular TIR remote sensing data are considered capable of reflecting sub-pixel structure as well as component temperatures and provides a new information source in terrestrial energy balance studies [1,2,3,4]. Using multi-angular observations and physicsbased models, the retrieval of component temperatures became possible [3, 17,18,19]. A lack of good-quality remote sensing data in suitable scale and angular setting is still the major obstacle in the applications of component temperatures retrieval [2, 20]. The most previous studies resorted to inversion of the physical thermal model to retrieve the component temperatures, which is usually time consuming and difficult to apply to satellite image data.

Methods

Results

Conclusion