TIGGE and the multi-model approach in hydrology (a brief review)

tigge1The multi-model approach is well established and popular in hydrological forecasting and modelling. For example, Shamseldin et al. (1997) showed different methods of combining the output of different hydrological models and this HEPEX post raises a number of challenges of multi-model approaches.

Therefore, it comes as no surprise that the THORPEX Interactive Grand Global Ensemble (TIGGE, Bougeault et al., 2010) archive, which contains ensemble forecasts by many different centres around the globe (access here), has been attracting increasing attention among hydrological forecasters and even featured in a white paper for HEPEX (Cloke and Pappenberger, 2005, Cloke et al., 2009). Three main areas in which HEPEX contributes to THORPEX are identified: evolution and predictability of weather systems, data assimilation, and benefits of improved forecasts of high impact weather (see Table here).

The opportunities offered by the multi-model TIGGE database in hydrologic ensemble prediction call for a review of past studies and a summary of achievements in view of contributing to HEPEX SIP.

The first published example using TIGGE in hydro-meteorological forecasting environment was by Pappenberger et al. (2008)Do you have an unpublished study or an earlier reference? Please comment below! – In this paper, the forecasts of nine centres where used within the setting of the European Flood Awareness System (EFAS, Thielen et al., 2009) for a case study of a flood event in Romania in October 2007. The study found that, through the use of multiple forecasts, flood warnings could have been issued 8 days before the event, whereas warnings based on a single ensemble system would have allowed for a lead time for the warning of only 4 days.

These results were confirmed for other climatic regions. For example, Bao and Zhao (2012a, b) used ensemble weather predictions based on the TIGGE data from the Canadian Meteorological Centre (CMC), the European Centre for Medium-Range Weather Forecasts (ECMWF), the UK Met Office (UKMO), and the US National Centers for Environmental Prediction (NCEP) to drive the Xinanjiang model in the upper reaches of the Huaihe River (China). A significant probability of flooding could be detected up to a few days in advance. The authors state that TIGGE is a promising tool for forecasting flood inundation, comparable with that driven by raingauge observations.

He et al. (2010) and Zhao et al. (2012, 2010) have demonstrated this already for the same catchment with a different hydrological model proving that results are fairly independent from the hydrological model applied. However, there is a clear sensitivity to catchment size. He et al. (2009) used TIGGE in a meso-scale catchment (4062 km2) located in the Midlands region of England. In addition, in this study, not only a hydrological model which predicts discharge in a river was used, but also a flood inundation model which predicts flood extent. The study proved that the numerical weather prediction (NWP) ensembles did not represent the spatial variability adequately, indicating a clear need for post-processing techniques.

This was also pointed out by He et al. (2010), who showed that individual systems of the multi-model forecast have systematic errors in time and space requiring a temporal and spatial post-processing. Such a post-processing must carefully maintain spatial, temporal and inter-variable correlations; otherwise they lead to deteriorating hydrological forecast skill (Pappenberger, 2010).

Liu et al. (2013) developed such a technique and demonstrated it successfully in the Huai River basin. The forecast skill is highly dependent on seasonality, with relatively lower skill for the wet summer season, when convective storm patterns dominate, in comparison with other seasons. Hydrological models act as non-linear filter and integrator of meteorological predictors and present the ideal opportunity to understand whether a certain NWP ensemble deficiency in the forecast does actually matter.

Xu et al. (2012) demonstrate the usefulness of hydrology as diagnostics for another catchment in China (the Linyi watershed) and showed how under-prediction in precipitation of the ensemble forecasts by ECMWF, NCEP and the China Meteorological Administration (CMA) has only moderate impact on discharge forecasts.

All the previous studies used hydrological models (the main aim is to close the water balance), which are traditionally different from land surface models (the main aim is to close the energy balance). In the not so far future, the difference between hydrological and land surface models will have lessened or disappeared (Beven and Cloke, 2012). Therefore it is encouraging that the results reported for hydrological models can be reproduced with land-surface schemes (Zheng et al. 2011).

The TIGGE archive has been of incredible value for research in hydro-meteorological forecasting. Within the GEOWOW project, the archive gets currently extended to include local area models. Bogner et al. (in press) already shows the value of combining global and local area models as they combine forecasts using forcing of two high resolution (ECMWF-HighRes, DWD-COSMO), one regional ensemble (COSMO-LEPS) and one global ensemble (ECMWF-ENS) demonstrating significant improved predictability in a hydro-meteorological forecast chain. The latter has been so successful that it forms an integral part of EFAS (Bogner and Pappenberger, 2009 and Bogner et al., 2012).

Did I miss YOUR reference? Did I miss something else? Add it in the comment box highlighting its main findings.

For References click here



Leave a Reply

Your email address will not be published. Required fields are marked *

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