Characterization, controlling, and reduction of uncertainties in the modeling and observation of land-surface systems

Uncertainty is one of the greatest challenges in the quantitative understanding of land-surface systems. This paper discusses the sources of uncertainty in land-surface systems and the possible means to reduce and control this uncertainty. From the perspective of model simulation, the primary source of uncertainty is the high heterogeneity of parameters, state variables, and near-surface atmospheric states. From the perspective of observation, we first utilize the concept of representativeness error to unify the errors caused by scale representativeness. The representativeness error also originates mainly from spatial heterogeneity. With the aim of controlling and reducing uncertainties, here we demonstrate the significance of integrating modeling and observations as they are complementary and propose to treat complex land-surface systems with a stochastic perspective. In addition, through the description of two modern methods of data assimilation, we delineate how data assimilation characterizes and controls uncertainties by maximally integrating modeling and observational information, thereby enhancing the predictability and observability of the system. We suggest that the next-generation modeling should depict the statistical distribution of dynamic systems and that the observations should capture spatial heterogeneity and quantify the representativeness error of observations.