Estimating carbon dioxide exchange rates at contrasting northern peatlands using MODIS satellite data

Northern hemisphere peatlands play an important role in the global carbon (C) cycle, accounting for about 30% of global soil C and ~ 10-25% of global natural methane (CH4) emissions. Satellite remote sensing has the potential for extracting continuous information related to C exchange rates at regional and global extents, yet, few studies have focused on peatlands. In this study we examined the potential of moderate resolution imaging spectroradiometer (MODIS) vegetation indices (normalized difference vegetation index, NDVI and simple ratio, SR), MODIS light use efficiency (LUE) based gross primary production (GPP) and a MODIS derived phenological index (annual peak photosynthetic rate) for the estimation of eddy covariance (EC) flux-derived GPP and net ecosystem production (NEP) at four contrasting northern peatlands. At the four sites of this study MODIS NDVI and SR explained between 39% and 71%, and between 42% and 69% of the variation in EC-derived GPP, respectively; and between 25% and 53%, and between 29% and 39% of the variation in EC-derived NEP, respectively. The relationships were mostly consistent across sites and within sites, suggesting that data may be pooled across years and sites, which could simplify the prediction of gross and net C dioxide (CO2) uptake over large areas dominated by northern peatlands based on MODIS data. MODIS GPP explained between 68% and 89% of the variation in EC-derived GPP at the four study sites. The root mean square errors ranged between 0.62 and 1.16 g C m− 2 d− 1 and were similar to errors from ecosystem process model estimates reported in the literature. Annual peak MODIS GPP, NDVI and SR rates explained up to 50% of the variations in annual cumulative EC-derived GPP and NEP at two of the study sites. Our results show the potentials and limitations of MODIS data to monitor the C dynamics of northern peatlands; among the three studied approaches the MODIS LUE-based GPP approach showed better performance as a predictor of GPP and NEP. The other approaches (VIs and phenology) can provide important input data for LUE models.