A Novel Microfacet Cosine Linear Kernel-Driven Bidirectional Reflectance Distribution Function Model

The algorithm of a four-parameter (isotropic, mixed cosine, normal zenith cosine square, and incident cosine square) microfacet cosine linear kernel-driven (MICOKE) bidirectional reflectance distribution function (BRDF) model is introduced. The MICOKE model was built from bidirectional reflectance factor data from a portable surface reflectance measurement system at $3 \times 3$ sample points (5-km spacing) at the Dunhuang site(longitude: 94.26°–94.38°, latitude: 40.09°–40.18°) in 2013. Traditional observation geometries were converted to microfacet observation geometries. Possible candidate models in multivariate power series form were tested and compared by the square of the correlation coefficients (SCCs) and the standard deviations (STDs). A model with a large SCC and small STD was selected as the MICOKE model. Using the Dunhuang site field campaign observation data, the mean of the SCCs of MICOKE was 4.73% higher than that of the Ross–Li BRDF model over 350–2500 nm for small observation geometries (SMGs). Using the Dunhuang site FY-2G/VISSR data, the SCCs of MICOKE were above 0.957 for large observation geometries (LAGs). In comparison, the SCCs of Ross–Li were only 0.020 (small field) and 0.357 (full field). The MICOKE model was compared with the Ross–Li model by the use of MCD43C1 and MCD12C1 products for 16 cover types. The SCCs varied from 0.930 to 1.000. The MICOKE BRDF model greatly improves the accuracy of the Gobi cover type for LAG and can be widely used in remote sensing.