Assessing sensitivity and persistence of updated initial conditions through Particle filter and EnKF for streamflow forecasting


 <p>Skillful streamflow forecasts provide a key support to several water-related applications. Ensemble forecasting systems are gaining a widespread interest, since they allow accounting for different sources of uncertainty. Because of the critical impact of the initial conditions (ICs) on the forecast accuracy, it is essential to improve their estimates via data assimilation (DA). This study aims at assessing the sensitivity of the DA-based estimation of forecast ICs to several sources of uncertainty and to the update of different model states and parameters of a conceptual rainfall-runoff model. The performance of two sequential ensemble-based techniques are compared, namely Ensemble Kalman filter and Particle filter, in terms of both efficiency and temporal persistence of the updating effect through the assimilation of observed discharges at the forecast time. Several experiments specifically address the impact of the meteorological, model state and parameter uncertainties over 232 catchments in France. Results show that the benefit of the DA-based estimation of ICs for forecasting is the largest when focusing on the level of the model routing store, which is the internal state the most correlated to streamflow. While the EnKF-based forecasts outperform the PF-based ones when accounting for the meteorological uncertainty, the representation of the model state uncertainty allows greatly improving the accuracy of the PF-based predictions, with a longer-lasting updating effect (up to 10 days). Conversely, the forecasting skill is undermined when accounting for the parameter uncertainty, due to the change in the hydrological responsiveness through the update of both the production and routing store levels. A further effort is focused on assessing the impact of the spatial resolution of the hydrological model on the predictive accuracy of DA-based streamflow forecasts.</p>


Skillful streamflow forecasts provide key support to several water-related applications. Because of the critical impact of initial conditions (ICs) on forecast accuracy, data assimilation (DA) can be performed to improve their estimation.
• sensitivity to several sources of uncertainty Daily discharge measurements at watershed outlets ( ) are assimilated. The uncertainty in observations is assessed as a function of the streamflow rate (Weerts and El Serafy, 2006;Thirel et al., 2010).
EnKF SIR-PF • Production store level (S) • Routing store level (R) • Unit hydrograph (UH) • Capacity of production store (X 1 ) • Capacity of routing store (X 3 ) #shareEGU20 Piazzi, Thirel, Perrin, Delaigue Assessing sensitivity and persistence of updated initial conditions through Particle filter and EnKF for streamflow forecasting > Uncertainty in meteorological forcings

Methodology
Probabilistic meteorological forecasts are generated by stochastically perturbing the SAFRAN meteorological reanalysis with multiplicative stochastic noise (Clark et al., 2008).

Model state variables
• Potential evapotranspiration (E) • Precipitation (P) • Production store level (S) • Routing store level (R) • Unit hydrograph (UH) • Capacity of production store (X 1 ) • Capacity of routing store (X 3 ) #shareEGU20 Piazzi, Thirel, Perrin, Delaigue Assessing sensitivity and persistence of updated initial conditions through Particle filter and EnKF for streamflow forecasting >

Methodology
After the analysis procedure, model states are perturbed through normally distributed null-mean noise (Salamon and Feyen, 2009).

Model state variables
• Potential evapotranspiration (E) • Precipitation (P) • Production store level (S) • Routing store level (R) • Unit hydrograph (UH) • Capacity of production store (X 1 ) • Capacity of routing store (X 3 ) #shareEGU20 Piazzi, Thirel, Perrin, Delaigue Assessing sensitivity and persistence of updated initial conditions through Particle filter and EnKF for streamflow forecasting > Uncertainty in model parameters

Methodology
Model parameters are jointly updated with state variables, according to the augmented state vector approach, and perturbed (Moradkhani et al., 2005).

Model state variables
• Potential evapotranspiration (E) • Precipitation (P) • Production store level (S) • Routing store level (R) • Unit hydrograph (UH) • Capacity of production store (X 1 ) • Capacity of routing store (X 3 ) #shareEGU20 Piazzi, Thirel, Perrin, Delaigue Assessing sensitivity and persistence of updated initial conditions through Particle filter and EnKF for streamflow forecasting >  Compared to PF, EnKF-based ICs guarantee a greater improvement in predictive accuracy (PF affected by ensemble shrinkage during no-rain periods).
 Both the EnKF and the PF schemes reveal an effective usefulness to improve predictive accuracy by the assimilation of observed discharges.  When dealing with a conceptual hydrological model, the main interest is on the routing dynamics to derive the most benefit from the DA-based ICs.
A comprehensive representation of both meteorological and state uncertainties allows for a more efficient improvement of predictive skill.
 PF-based ICs are greatly enhanced thanks to a larger spread of the ensemble simulations.  While the PF-based updating effect is longer lasting, the benefit of larger corrective terms for the EnKF rapidly decreases within a short lead time.
High sensitivity to the parameter estimation, as store capacities define the simulated hydrological responsiveness of the basin.
 Parameter values estimated at the forecast time may not be the optimal ones to represent the model response over the forecast horizon.  The equifinality issue can affect the parameter estimates, especially in PF. #shareEGU20 Piazzi, Thirel, Perrin, Delaigue Assessing sensitivity and persistence of updated initial conditions through Particle filter and EnKF for streamflow forecasting > Ongoing and future perspectives This study has been recently submitted to the Water Resources Research journal: Piazzi, G., Thirel, G., Perrin, C., Delaigue, O. Sequential data assimilation for streamflow forecasting: assessing the sensitivity to uncertainties and updated variables of a conceptual hydrological model.
An R package providing the DA schemes will be soon available.
The authors thank Météo-France and SCHAPI for providing climate and streamflow data. The first author received financial support from SCHAPI and the RenovRisk-Transfer project. This work contributes to the SPAWET project funded by the CNES-TOSCA program.

Introduction
Forecasting system

Results
Conclusions & perspectives 6 References