MODIS PRI performance to track Light Use Efficiency of a Mediterranean coniferous forest: Determinants, restrictions and the role of LUE range

The relationship between the Photochemical Reflectance Index (PRI) and Light Use Efficiency (LUE) is well established at leaf and small canopy scales, but upscaling to ecosystem level is still a challenge. Only few studies have applied satellite-derived PRI to estimate LUE, mostly using MODIS, and although the results are promising, many external factors have been identified affecting PRI performance. The present study investigates determinants and restrictions of MODIS-derived PRI potential to follow the LUE variability of a Mediterranean coniferous forest.Daily and half-hour LUE values were calculated from eddy covariance measurements, dividing GPP by either Photosynthetically Active Radiation (PAR) or the absorbed fraction of PAR (APAR). Also, various PRI datasets were created based on different sensor (Terra, Aqua, Both), reference band (1, 12, 13) and observation/illumination angles. Overall, PRI correlated better with LUE calculated using PAR instead of APAR and Aqua PRI yielded better results than Terra. Restricting acquisitions according to observation/illumination angles improves the PRI:LUE relationship (maximum R-2 = 0.512), with backscatter observations yielding the best correlations. Our findings suggest that MODIS-derived PRI is more sensitive to relatively large seasonal LUE changes, but is unable to closely follow severe drought events. Among the tested reference bands, the best results were derived using band 12 (546 - 556 nm), although the optimum reference band seems to depend on viewing conditions. The PRI:LUE relationship was further improved when half-hour LUE of the satellite overpass was used instead of daily LUE. However, it was found that the PRI:LUE relationships for the different datasets were strongly affected by the range of LUE values corresponding to each PRI group, with lower LUE variability resulting to weaker PRI:LUE correlations. LUE range effect should be accounted for in future studies, when different PRI datasets are compared and might explain the contradicting findings in the existing literature.