Assimilation Of Heterogeneous Uncertain Data, Having Different Observational Errors, In Hydrological Models

Accurate real-time forecasting of river water level is an important issue that has to be addressed in order to prevent and mitigate water-related risk. To this end, data assimilation methods have been used to improve the forecasts ability of water model merging observations coming from stations and model simulations. As a consequence of the increasing availability of dynamic and cheap sensors, having variable life-span, space and temporal coverage, the citizens are becoming an active part in information capturing, evaluation and communication. On the other hand, it is difficult to assess the uncertain related to the observation coming from such sensors. The main objective of this work is to evaluate the influence of the observational error in the proposed assimilation methodologies used to update the hydrological model as response of distributed observations of water discharge. We tested the developed approaches on a test study area - the Brue catchment, located in the South West of England, UK. The Ensemble Kalman filter is applied to the semi-distributed hydrological model. Distributed observations of discharge are synthetically generated. Different types of observational error are introduced assuming diverse sets of probability distributions, first and second order moments. The results of this work show how the assimilation of distributed observations, can improve the hydrologic model performance with a better forecast of flood events. It is found that different observational error types can affects the model accuracy.