Understanding the time-varying importance of different uncertainty sources in hydrological modeling using sensitivity analysis

Abstract:

Simulations from hydrological models are affected by potentially large uncertainties stemming from boundary and initial conditions, model parameters, and observational uncertainty due to errors in collection and post-processing, e.g. temporal and spatial interpolation, of the input (typically rainfall and temperature) and output data (streamflow). Understanding the relative importance of such sources of uncertainty and their possible interactions is essential to support model calibration, validation and diagnostic evaluation, and to prioritize efforts for uncertainty reduction. These relative influences and their interactions are likely to vary across time and between catchments. For instance, accurate characterisation of the temperature gradient might be essential in snow-dominated catchments and especially during the snow accumulation/melt period; errors in rainfall measurement and interpolation are expected to be mostly influential in the rising phase of a flood event; observational errors might become far less influential with respect to parameter uncertainty during the recession phase.

Traditional applications of sensitivity analysis (SA) rely on the aggregation of simulation results, typically by defining a measure of model performance like the root mean squared error. This aggregation of propagated uncertainties prior to a SA may lead to a significant loss of information and may cover up local behaviour that could be of great interest. Time-varying sensitivity analysis, where the aggregation and SA are repeated at different time-steps, is a viable option to reduce the loss of information and for obtaining a more accurate picture of the model behaviour. In this work, we address questions like: what are the specific time periods where different uncertainty sources (e.g. observations or parameters) dominate the model response and therefore uncertainty reduction would be most beneficial? How does this temporal uncertainty impact change in catchments with different characteristics (e.g. snow-dominated or intermittent)? Lessons learnt in this study will provide guidance for efforts towards uncertainty reduction and improve data gathering versus improving model diagnostics.