The effect of uncertainty and systematic errors in hydrological modelling

Monday, 15 December 2014
Ingelin Steinsland1, Kolbjørn Engeland2, Stian Solvang Johansen3, Asgeir Øverleir-Petersen3 and Sjur A. Kolberg4, (1)Norwegian University of Science and Technology, Trondheim, Norway, (2)Norwegian Water Resources and Energy Directorate, Oslo, Norway, (3)Statkraft, Oslo, Norway, (4)SINTEF Energy Research, Trondheim, Norway
The aims of hydrological model identification and calibration are to find the best possible set of process parametrization and parameter values that transform inputs (e.g. precipitation and temperature) to outputs (e.g. streamflow). These models enable us to make predictions of streamflow. Several sources of uncertainties have the potential to hamper the possibility of a robust model calibration and identification. In order to grasp the interaction between model parameters, inputs and streamflow, it is important to account for both systematic and random errors in inputs (e.g. precipitation and temperatures) and streamflows. By random errors we mean errors that are independent from time step to time step whereas by systematic errors we mean errors that persists for a longer period. Both random and systematic errors are important in the observation and interpolation of precipitation and temperature inputs. Important random errors comes from the measurements themselves and from the network of gauges. Important systematic errors originate from the under-catch in precipitation gauges and from unknown spatial trends that are approximated in the interpolation. For streamflow observations, the water level recordings might give random errors whereas the rating curve contributes mainly with a systematic error.

In this study we want to answer the question “What is the effect of random and systematic errors in inputs and observed streamflow on estimated model parameters and streamflow predictions?".

To answer we test systematically the effect of including uncertainties in inputs and streamflow during model calibration and simulation in distributed HBV model operating on daily time steps for the Osali catchment in Norway. The case study is based on observations from, uncertainty carefullt quantified, and increased uncertainties and systmatical errors are done realistically by for example removing a precipitation gauge from the network.We find that the systematical errors in precipitation input and streamflow observations has large impact, both for prediction and for water balance parameters while the dynamic parameters are less effected.