Effects of bias in solar radiation inputs on ecosystem model performance

Solar radiation inputs drive many processes in terrestrial ecosystem models. The processes (e.g. photosynthesis) account for most of the fluxes of carbon and water cycling in the models. It is thus clear that errors in solar radiation inputs cause key model outputs to deviate from observations, parameters to become suboptimal, and model predictions to loose confidence. However, errors in solar radiation inputs are unavoidable for most model predictions since models are often run with observations with spatial or/and temporal gaps. As modeled processes are non-linear and interacting with each other, it is unclear how much confidence most model predictions merits without examining the effects of those errors on the model performance. In this study, we examined the effects using a terrestrial ecosystem model, DayCent. DayCent was parameterized for annual grassland in California with six years of daily eddy covariance data totaling 15,337 data points. Using observed solar radiation values, we introduced bias at four different levels. We then simultaneously calibrated 48 DayCent parameters through inverse modeling using the PEST parameter estimation software. The bias in solar radiation inputs affected the calibration only slightly and preserved model performance. Bias slightly worsened simulations of water flux, but did not affect simulations of CO₂ fluxes. This arose from distinct parameter set for each bias level, and the parameter sets were surprisingly unconstrained by the extensive observations. We conclude that ecosystem models perform relatively well even with substantial bias in solar radiation inputs. However, model parameters and predictions warrant skepticism because model parameters can accommodate biases in input data despite extensive observations.