Design and analysis of numerical experiments to compare four canopy reflectance models

A method designed to study the relative effects of the input parameters of any model has been investigated with canopy reflectance (CR) models. Traditionally, sensitivity analyses are performed by changing one input parameter at a time. Such an approach is limited because it lacks strategy. A promising alternative is in the use of design of experiments, a statistical method that allows defining a structured and restricted number of simulations for which all input parameters vary simultaneously. This approach is especially helpful in multidimensional parameter spaces. It is demonstrated using four 1D radiative transfer models that are compared in direct mode. These models are combinations of the PROSPECT leaf optical properties model with the four CR models, SAIL (Scattering and Arbitrarily Inclined Leaves), KUUSK, IAPI, and NADI (New Advanced DIscrete model). The sensitivity studies were conducted in the visible/near-infrared on the following parameters: the leaf structure (N), the chlorophyll-a and -b content (Cab), the leaf area index (LAI), the mean leaf inclination angle (θl), the hot spot (sl), and the soil brightness (αsoil). We compared simulated reflectances for a given set of measurement geometries and two wavebands of the POLDER (Polarization and Directionality of the Earthʹs Reflectances) spaceborne instrument. The relative effects of the biophysical parameters are assessed as well as their contribution to reflectance, allowing us to rank the most influential ones. Their interactions were also studied from the perspective of improving inversion procedures. Globally, the four models agree well in terms of computed reflectances and parameter effects, nevertheless with some discrepancies due to the implementation of different leaf angle distribution (LAD) functions.