Assessment of MODIS NPP algorithm-based estimates using soil fertility and forest inventory data in mixed hemiboreal forests

Optical remote sensing data-based estimates of terrestrial net primary production (NPP) are released by different projects using light use efficiency-type models. Although spatial resolution of the NPP data sets is still too coarse (500–1000 m) for single forest stands, regional monitoring of forest management and growth with 25–100 ha sampling units is feasible if the NPP_SAT estimates are sensitive to forest growth differences depending on soil fertility in the area of interest. In this study, NPP estimates for 2,914 mixed forest class pixels (according to the MODIS land cover map) located in Estonia were (1) obtained from three different NPP_SAT products, (2) calculated using an empirical soil potential phytoproductivity (SPP) model applied to a 1:10,000 soil map (NPP_SPP), and (3) calculated using stem volume increment estimates given in a forest management inventory data base (NPP_FIDB). A linear multiple regression model was then used to explore the relationships of NPP_SAT with the proportion of coniferous forests, the NPP_SPP and distance of the pixels from the Baltic Sea coast – the variables that have been found informative in previous studies. We found a positive moderate correlation (0.57, p < 0.001) between NPP_SPP and NPP_FIDB. The local or downscaled meteorological data-based NPP_SAT estimates were more consistent with the NPP_SPP and NPP_FIDB, but the correlation with NPP_SAT was weak and sometimes even negative. The range of NPP estimates in NPP_SAT data sets was much narrower than the range of NPP_SPP or NPP_FIDB. Errors in land cover maps and in estimates of absorbed photosynthetically active radiation were identified as the main reasons for NPP_SAT inconsistencies.