The role of local‐scale heterogeneities in terrestrial ecosystem modeling

The coarse‐grained spatial representation of many terrestrial ecosystem models hampers the importance of local‐scale heterogeneities. To address this issue, we combine a range of observations (forest inventories, eddy flux tower data, and remote sensing products) and modeling approaches with contrasting degrees of abstraction. The following models are selected: (i) Lund‐Potsdam‐Jena (LPJ), a well‐established, area‐based, dynamic global vegetation model (DGVM); (ii) LPJ‐General Ecosystem Simulator, a hybrid, individual‐based approach that additionally considers plant population dynamics in greater detail; and (iii) distributed in space‐LPJ, a spatially explicit version of LPJ, operating at a fine spatial resolution (100 m × 100 m), which uses an enhanced hydrological representation accounting for lateral connectivity of surface and subsurface water fluxes. By comparing model simulations with a multivariate data set available at the catchment scale, we argue that (i) local environmental and topographic attributes that are often ignored or crudely represented in DGVM applications exert a strong control on terrestrial ecosystem response; (ii) the assumption of steady state vegetation and soil carbon pools at the beginning of simulation studies (e.g., under “current conditions”), as embedded in many DGVM applications, is in contradiction with the current state of many forests that are often out of equilibrium; and (iii) model evaluation against vegetation carbon fluxes does not imply an accurate simulation of vegetation carbon stocks. Having gained insights about the magnitude of aggregation‐induced biases due to smoothing of spatial variability at the catchment scale, we discuss the implications of our findings with respect to the global‐scale modeling studies of carbon cycle and we illustrate alternative ways forward.