On the ability to infer spatial catchment variability using streamflow hydrographs

Spatially distributed models can potentially provide improved hydrologic predictions because of their ability to exploit spatially distributed data while providing estimates of hydrologic variables at interior catchment locations. However, attempts to estimate spatially distributed parameter fields via model calibration have been fraught with difficulty. This paper examines the factors that can influence the ability to infer spatial properties of a distributed model when the only information available for model evaluation is catchment streamflow response. In particular, we investigate the conditions under which spatial variability in parameters and rainfall cause sufficiently strong variations in the streamflow hydrographs to justify their representation in catchment models and whether such information can be detected via commonly used model performance measures. Our results show that spatial variability in parameter and precipitation fields can, indeed, have a detectable impact on the properties of the streamflow hydrograph but that this impact can be so greatly diminished by the damping and dispersive effects of routing that it is virtually nondetectable by conventional performance measures by the time the water reaches the catchment outlet. And although measures based on information theory may be able to detect subtle variations of this kind, the information may not ultimately be useful in the face of model structure and data errors. The only reasonable way forward therefore is to explore other kinds of catchment information (including multiple interior flow gauging locations) for use in estimation of spatially distributed parameter fields.