The Fifth Phase of the Radiation Transfer Model Intercomparison Exercise (RAMI-V): Experiment Description and Results on Actual Canopy Scenarios

This paper presents the latest results of the radiation transfer model intercomparison (RAMI) of the realistic vegetation scenarios. RAMI-V included the same one-dimensional (1D) and 3D scenes of RAMI-IV phase and 2 new realistic ones, defined through a semiparametric (Savanna) and an empirical (Wytham Woods) approaches. The measurements to simulate were the bidirectional reflectance factor, directional–hemispherical reflectance, and bidirectional–hemispherical reflectance. In addition, the radiant flux transmission and absorption through and below the canopy and digital hemispherical photography were also proposed. The spectral bands were defined to mimic not only the ones of Copernicus optical missions, e.g., for the Sentinel-3 Ocean and Land Colour Imager (OLCI) and Sentinel-2 Multispectral Instrument (MSI), but also the Moderate Resolution Imaging Spectroradiometer (MODIS). New solar and viewing geometry configurations were adopted from realistic satellite overpasses for different seasons and geographical locations. The role of internal consistency checks was reinforced to provide more reliable feedback to the participants in the early stage of the experiment and reduce the role of outliers in the model-to-model comparison and the identification of a surrogate reference. Over 4 of the 8 scenarios proposed, a set of models agreed within 2% uncertainty thresholds for most of the virtual measurements defined in the experiment. Specifically, they were the birchstand both leaf-on (HET09) and leaf-off (HET15) versions, and the structured canopy models consisting of a citrus orchard (HET14) and a poplar forest (HET16). It is noteworthy that less was among the models designated to set a reference benchmark across all chosen instances. Besides, dart , raytran , and wps were contributing to the benchmark in most of the experiment proposed, especially referring to total BRF and DHR, and total absorption, while for the transmittance the results were more dispersed. Dart , less , raytran , and wps contributed by submitting 100%, 83.9%, 99.4%, and 86.2% of the experiment proposed, respectively. The proficiency testing of the models was performed by means of the z ′ metric defined in ISO-13528. A custom reference, based on a selection of models that showed the best agreement, as well as a reference based on robust statistic were adopted. Above the aforementioned selected scenes, and assuming a compliance threshold of 3% (5%) for bidirectional reflectance (albedo) measurements, dart , less , and raytran were in agreement in all (more than 95%) cases. The approach based on the robust statistic described in ISO-13528 confirmed its relevance in interlaboratory comparison exercises where the benchmark is not defined a priori, allowing us to obtain proficiency results equivalent to those defined against the customized references.