Global sensitivity analysis of the spectral radiance of a soil-vegetation system

作者：Mousivand Alijafar^*; Menenti Massimo; Gorte Ben; Verhoef Wout

来源：Remote Sensing of Environment, 2014, 145: 131-144.

DOI：10.1016/j.rse.2014.01.023

• Global sensitivity analysis
• Coupled radiative transfer model
• SLC
• MODTRAN
• Winding stairs sampling
• Soil-vegetation system
• TOA radiance
• Imagine spectroscopy

We analyzed the sensitivity of Top-Of-Atmosphere (TOA) radiance and surface reflectance of a soil-vegetation system to input biophysical and biochemical parameters using the coupled Soil-Leaf-Canopy radiative transfer model SLC and MODTRAN. We applied variance-based global sensitivity analysis for different atmospheric conditions and observation configurations. Among 23 input parameters, crown coverage, leaf area index, leaf inclination distribution function and soil moisture were found to be the most influential parameters driving the output variance of the radiance between 400 and 2500 nm with a few exceptions. Hapke%26apos;s soil parameters and the canopy layer dissociation factor were recognized to have marginal influence on the output radiance. It is also found that a large portion of uncertainty in the output radiance is driven by the interaction effects among input parameters in the visible (similar to 550 nm), whereas the red-near infrared (similar to 670 nm), seems to have fewer interaction effects. The effect of solar/view direction is found to be significant on TOA radiance sensitivity to the input parameters. The results also confirmed that the sensitivities of surface reflectance are comparable to the TOA radiance sensitivities when the atmosphere is clear and visibility is high. Since coupled surface-atmosphere RT models can be computationally intensive, this work also introduces an improvement to the design and sampling of screening methods for efficient sensitivity analysis of computationally expensive models. The improvement is based on three elements: a) generating sample points by Sobol%26apos;s sequence generator; b) variational analysis in the parameter space using the winding stairs method; c) use of mean and variance sensitivity measures. The results with 1200 model runs demonstrated high correlation (92%) with variance-based global sensitivity analysis using 49,152 model runs, in determining the most influential and non-influential parameters.

• 出版日期2014-4-5