HEPEX-SIP Topic: Hydrologic data assimilation (2/2)

Contributed by Liu Yuqiong, Albrecht Weerts and Hamid Moradkhani

The following challenges and research needs are based on the paper by Liu et al. (2012).

The need for transitioning of hydrologic DA research into effective operations has become increasingly recognized in the wake of frequent occurrences of extreme events in recent years and increasing availability of new observations.

  • Several theoretic or mathematical challenges need to be addressed before hydrologic DA can fully benefit operational forecasting. Issues include the high nonlinearity in hydrologic processes, the high dimensionality of the state or parameter vectors of hydrologic models, the skewness and heteroscedasticity in the probabilistic distribution of hydrologic variables, the need for impractically large samples in ensemble approaches, and the limited observations of extreme events. Emerging opportunities include localized and transformation-based ensemble approaches and decomposing of the hydrologic forecast system into smaller subcomponents for separate DA solutions. It is recommended that bias correction precede or accompany DA applications.
  • The success and outcome of a DA scheme depends critically on the characterization of uncertainties. Uncertainty in precipitation can be quantified by stochastically perturbing precipitation inputs or through conditional simulation methods. Determining model uncertainty can be complicated by interactions among uncertainty sources, poorly constrained inference problems, and difficulty in constructing a reliable multi-model ensemble. These issues can be addressed by disentangling uncertainty sources, intelligent use of available data for inverse modeling, using integrated multi-model and multi-parameterization frameworks, and combining the strength of DA and multi-model ensembles.
  • Hydrologic forecasting can potentially benefit from integrating, via DA, newly emerging observations such as remote sensing data. Some of the difficulties (or emerging opportunities) in effectively assimilating these data include, for  example, developing proper modifications of an operational hydrologic model to assimilate “raw” satellite data (i.e., radiance observations), constructing proper “mapping” relationships between remotely observed and modeled variables, providing appropriate specification of uncertainty in remote sensing data, and building an efficient computing infrastructure for retrieving remote sensing data to support satellite DA in operational forecasting.
  • DA has played an important role in real-time control of water resources systems and hydraulic structures. DA for real-time control is often conducted within a parameter optimization framework and uses optimization-based techniques (rather than sophisticated DA approaches such as EnKF) for state updating. It is recommended that the hydrologic and control communities work together to learn from each other’s experiences to more efficiently address issues encountered in DA applications.
  • Besides the computational and technical aspects, other operational aspects like transparency and clearness of the outcomes of DA applications are at least as important for the uptake of automated DA methods in operational practice. For operational forecasters who are used to manual interactions with the forecast system, automated DA of the black-box type may be too adventurous a step to take in the short run. Hence, efforts are needed to speed up the operational implementation of automated DA, by enabling the operational control (with manual interaction) over noise models and observations to be used in DA, and developing DA-guided systems, where the DA results will be presented next to the current forecast to guide operational forecaster.

It is important to note that comprehensive and robust verification of DA results is necessary to demonstrate the value of DA for operational forecasting and to build trust in DA among operational forecasters. One important goal of DA in operational forecasting is to provide an improved analysis of the model initial conditions to produce improved hydrologic forecasts. However, the link between the accurate characterization of the initial conditions and the sensitivity of forecast skill at different lead times to this characterization is still uncertain, largely due to lack of proper verification of the potential gain from DA in a forecast context. Some have also argued that statistically based post-processing of hydrologic forecasts may outperform DA, since the latter aims at improving initial conditions that may not have a sufficiently long memory to improve forecast skill at longer lead times. All of these point to the need for robust forecast verification (e.g., Demargne et al., 2010) that will identify and quantify the sensitivity of forecast skill to accuracy of initial conditions and hence help quantify the value of DA for operational hydrologic forecasting.

The various issues described above call for the need of a community-based approach to hydrologic DA, which aims at  providing  a  set  of  generic  modeling,  DA,  and  verification  tools  to  serve  the  diverse  needs  of  the  community and  to  facilitate  effective  and  efficient  advances  through community contribution and feedback. This also opens  a  promising  pathway  for  the  cost-effective  transition of hydrologic DA research into operational forecasting while  at  the  same  time  facilitating  the  communication  of new hurdles encountered in operational DA back to the research community.

In summary, it is recommended that cost- effective transition of hydrologic DA from research to operations should be helped by developing community-based, generic modeling and DA tools, and through fostering collaborative  efforts  among  hydrologic modellers,  DA  researchers,  and operational forecasters.


This post is a contribution to the new HEPEX Science and Implementation Plan.

See also in “HEPEX SIP Topic: Hydrologic data assimilation”:

0 comments

Leave a Reply

This site uses Akismet to reduce spam. Learn how your comment data is processed.