An enhanced RPV model to better capture hotspot signatures in vegetation canopy reflectance observed by the geostationary meteorological satellite Himawari-8

Abstract

The hotspot effect denotes a special case of the Bidirectional Reflectance Distribution Function (BRDF) when the solar direction coincides with the sensor viewing direction, which is essential for remote estimation of canopy structure information. In contrast to polar-orbiting satellites, third-generation geostationary (GEO) meteorological satellites provide a new opportunity to investigate the hotpot effect at a modest spatial resolution (∼1 km) due to their extremely high observation frequencies. Nevertheless, modeling of the hotspot effect observed by GEO satellites is a significant challenge because their observations of Bidirectional Reflectance Factor (BRF) are usually off the principal plane. Among the existing semi-empirical BRDF models, the Rahman-Pinty-Verstraete (RPV) model has been widely applied to simulate intricate fields of canopy BRF. However, the RPV model has also been criticized for underestimating the hotspot signatures. Consequently, an enhanced version of the RPV model (i.e., ERPV) was proposed in this study to improve its capability for modeling the hotspot signatures of canopy reflectance. To verify the proposed ERPV model, a reflectance dataset of hotspot effect for different vegetation types was created using the atmospherically corrected Hiwamari-8 reflectance, and the ERPV model was applied to estimate foliage Clumping Index (CI) through constructing hotspot and dark spot within the principal plane. Validation results demonstrated that the land surface reflectance of Himwari-8 could measure the hotspot effect properly for each vegetation type. The EPRV model yielded satisfactory accuracies in capturing the hotspot signatures with Root-Mean-Square-Error (RMSE) of 0.0034 and 0.0056, and Bias of −0.0019 and −0.0028, for the red and near-infrared (NIR) bands, respectively. In contrast, the RMSE and Bias for the original RPV model and three existing kernel-driven BRDF models ranged from 0.0187 to 0.125, and from −0.0149 to −0.114, respectively. Moreover, the estimated CI based on the ERPV model (0.66) was closer to the field measurement (0.65) for a mixed forest site than the RPV-based CI estimate (0.72) and the MODIS CI product (0.705). The findings demonstrate that our ERPV model can not only improve the modeling accuracies of hotspot signatures, but also has the potential to construct reliable BRF within the principal plane for CI retrieval.