Photosynthetic efficiency of northern forest ecosystems using a MODIS-derived Photochemical Reflectance Index (PRI)

Abstract

Abstract This study evaluates a direct remote sensing approach from space for the determination of ecosystem photosynthetic light use efficiency (LUE), through measurement of vegetation reflectance changes expressed with the Photochemical Reflectance Index (PRI). The PRI is a normalized difference index based on spectral changes at a physiologically active wavelength (~ 531 nm) as compared to a reference waveband, and is only available from a very few satellites. These include the two Moderate-Resolution Imaging Spectroradiometers (MODIS) on the Aqua and Terra satellites each of which have a narrow (10 nm) ocean band centered at 531 nm. We examined several PRI variations computed with candidate reference bands, since MODIS lacks the traditional 570 nm reference band. The PRI computed using MODIS land band 1 (620–670 nm) gave the best performance for daily LUE estimation. Through rigorous statistical analyses over a large image collection (n = 420), the success of relating in situ daily tower-derived LUE to MODIS observations for northern forests was strongly influenced by satellite viewing geometry. LUE was calculated from CO2 fluxes (mol C mol− 1 absorbed quanta) measured at instrumented Canadian Carbon Program flux towers in four Canadian forests: a mature fir site in British Columbia, mature aspen and black spruce sites in Saskatchewan, and a mixed deciduous/coniferous forest site in Ontario. All aspects of the viewing geometry had significant effects on the MODIS-PRI, including the view zenith angle (VZA), the view azimuth angle, and the displacement of the view azimuth relative to the solar principal plane, in addition to illumination related variables. Nevertheless, we show that forward scatter sector views (VZA, 16°–45°) provided the strongest relationships to daily LUE, especially those collected in the early afternoon by Aqua (r2 = 0.83, RMSE = 0.003 mol C mol− 1 absorbed quanta). Nadir (VZA, 0° ± 15°) and backscatter views (VZA, − 16° to − 45°) had lower performance in estimating LUE (nadir: r2 ~ 0.62–0.67; backscatter: r2 ~ 0.54–0.59) and similar estimation error (RMSE = 0.004–0.005). When directional effects were not considered, only a moderately successful MODIS-PRI vs. LUE relationship (r2 = 0.34, RMSE = 0.007) was obtained in the full dataset (all views & sites, both satellites), but site-specific relationships were able to discriminate between coniferous and deciduous forests. Overall, MODIS-PRI values from Terra (late morning) were higher than those from Aqua (early afternoon), before/after the onset of diurnal stress responses expressed spectrally. Therefore, we identified ninety-two Terra-Aqua “same day” pairs, for which the sum of Terra morning and Aqua afternoon MODIS-PRI values (PRIsum) using all available directional observations was linearly correlated with daily tower LUE (r2 = 0.622, RMSE = 0.013) and independent of site differences or meteorological information. Our study highlights the value of off-nadir directional reflectance observations, and the value of pairing morning and afternoon satellite observations to monitor stress responses that inhibit carbon uptake in Canadian forest ecosystems. In addition, we show that MODIS-PRI values, when derived from either: (i) forward views only, or (ii) Terra/Aqua same day (any view) combined observations, provided more accurate estimates of tower-measured daily LUE than those derived from either nadir or backscatter views or those calculated by the widely used semi-operational MODIS GPP model (MOD17) which is based on a theoretical maximum LUE and environmental data. Consequently, we demonstrate the importance of diurnal as well as off-nadir satellite observations for detecting vegetation physiological processes.