Evaluating Ecohydrological Model Sensitivity to Forcing Variability with an Information Theory-Based Approach

Abstract:

Understanding how a complex ecohydrological model behaves under different forcing conditions and under temporal variability is an important factor in model validation. While many studies focus on model sensitivity to parameters, model structure, or forcing errors, we focus on the impact of forcing precision, or quantization, on model outputs and performance. We demonstrate an analytical method based on rate-distortion theory, a branch of information theory, to optimally approximate possible representations of input variables at different precision. In other words, we apply quantization as a tool for complex simulation models to yield useful sensitivity analysis. We apply this method to a sophisticated ecohydrology model, MLCan, forced by an extensive eddy covariance flux tower dataset at Goose Creek in Central Illinois. We identify the effect of individual and multiple quantized input variables on the model results by comparing the performance of the quantized model with both the original model and flux tower observation data. We find that this particular model is relatively insensitive to the precision of wind speed, air temperature, and humidity, indicating that the model acts as a filter of many inputs. This method could be applied broadly to other models beyond the ecohydrological scope demonstrated here to better understand model behavior and the importance of input data at a given precision.