The ability of the RossThick-LiSparseR BRDF model （Ross-Li model） to simulate the bidirectional reflectance is assessed by comparing the reflectance simulated using the Ross-Li model based on the MCD43A1 data with the observed POLDER-3 BRDF data for 9961 pixels selected from the global POLDER-3/PARASOL BRDF archive.It is found that the RossLi model is able to simulate the bidirectional reflectance of ground objects efficiently.The R 2 and root mean square error （RMSE） of simulated reflectance are 0.934 and 0.016

respectively.On average

simulated reflectance is 5.2% lower compared with the POLDER-3 data.However

the Ross-Li model underestimates the reflectance by 14% for hotspots

with R 2 equal to 0.824 and RMSE equal to 0.07.The errors in simulated reflectance for hotspots depend on land cover types.Coniferous forests have the largest underestimates of simulated reflectance for hotspots

followed by broadleaved forests

grasslands

croplands

shrubs and barren lands.The modification for the volume kernel in the Ross-Li model can noticeably improve reflectance simulation for hotspots （R 2 =0.839

RMSE=0.043）.The reflectance simulated by the Ross-Li model is relatively accurate at nadir and darkspot.The departure of simulated reflectance from the POLDER-3 observations increases with the increases in solar zenith and view zenith angles.For grasslands and croplands

the accuracy of simulated reflectance decreases with the increase in NDVI.However

for forests and shrubs

the discrepancy between reflectance simulated by the Ross-Li model and the POLDER-3 observations maximizes when NDVI approximately equals 0.5.