Time stability of catchment model parameters: Implications for climate impact analyses

TLDR

Climate impact analyses typically rely on hydrological models driven by future climate scenarios, assuming that parameters calibrated to past runoff remain valid for the future. In this paper we calibrate a conceptual rainfall‑runoff model to six consecutive 5‑year periods (1976–2006) for 273 Austrian catchments and analyze how the calibrated parameters change over time. The study finds that snow and soil‑moisture parameters exhibit significant, climate‑driven trends—e.g., the runoff‑generation parameter doubled over three decades—leading to substantial biases (≈15 % median, 35 % peak) when using time‑invariant parameters, with errors increasing as the time lag grows, underscoring the need to account for parameter change in hydrologic prediction and climate impact analyses.

Abstract

Climate impact analyses are usually based on driving hydrological models by future climate scenarios, assuming that the model parameters calibrated to past runoff are representative of the future. In this paper we calibrate the parameters of a conceptual rainfall‐runoff model to six consecutive 5 year periods between 1976 and 2006 for 273 catchments in Austria and analyze the temporal change of the calibrated parameters. The calibrated parameters representing snow and soil moisture processes show significant trends. For example, the parameter controlling runoff generation doubled, on average, in the 3 decades. Comparisons of different subregions, comparisons with independent data sets, and analyses of the spatial variability of the model parameters indicate that these trends represent hydrological changes rather than calibration artifacts. The trends can be related to changes in the climatic conditions of the catchments such as higher evapotranspiration and drier catchment conditions in the more recent years. The simulations suggest that the impact on simulated runoff of assuming time invariant parameters can be very significant. For example, if using the parameters calibrated to 1976–1981 for simulating runoff for the period 2001–2006, the biases of median flows are, on average, 15% and the biases of high flows are about 35%. The errors increase as the time lag between the simulation and calibration periods increases. The implications for hydrologic prediction in general and climate impact analyses in particular are discussed.