Joint research project

Hydrological modeling optimization in poorly gauged and small sized basins - Lebanon as a case study

Project leaders
Luca Brocca, Chadi Abdallah
Agreement
LIBANO - CNRS-L - National Council for Scientific Research of Lebanon
Call
CNR-CNRS-L 2016-2017
Department
Earth system science and environmental technologies
Thematic area
Earth system science and environmental technologies
Status of the project
New

Research proposal

Of all natural disasters, floods affect the greatest number of people worldwide and have the greatest potential to cause damage. In fact, floods are responsible for over one third of people affected by natural disasters; almost 190 million people in more than 90 countries are exposed to catastrophic floods every year. Nowadays, with the emerging global warming phenomenon, this number is expected to increase, therefore, flood prediction and prevention has become a necessity in many places around the globe to decrease damages caused by flooding (Abdallah et al., 2012)

Floods, in Lebanon, normally take place during wet season, generally after a strong storm or at the beginning of the spring with snow melting. During floods, rivers burst their bank causing damages to buildings and agricultural land. The consequences of such events are tragic including annual financial losses, casualties, loss of livestock, destruction to houses and agricultural lands, damage to structures, utilities and public services, in addition to triggering landslides

Flood prediction in most countries, including Lebanon, is a challenging task due to the scarcity of hydrometeorological data (historical and real time), as well as, the complex physical interactions, including spatially and temporally variable meteorological processes. Aggregating such processes into a forecasting model brings uncertainties to all components of the flood forecasting chain (e.g., inputs, model structure). The nonlinear nature of models and the correlation of errors make uncertainty quantification and disaggregation a difficult task in hydrology (Gupta et al., 2005; Vrugt et al., 2009; Tobin, 2012).

Nowadays, the distributed hydrological models are widely applied in hydrology to simplify the understanding of the hydrological processes. Due to the complexity of these models, several problems arise and need to be solved. Some of these problems include the problem of nonlinearity, the problem of scale, the problem of uniqueness, the problem of gaps in data, and the problem of uncertainty (Beven, K., 2001). Most of the utilized models have some defects related to the integration of the meteorological and terrain parameters (e.g. rainfall intensity, and runoff coefficients) which may have a major influence for flood prediction. Similarly, the contribution of space images is still limited and conditioned by indirect reasoning related to the different platform types. Despite much research, developing a hydrological model that best reflects the hydrological processes of poorly gauged watersheds is still questioned.

Recently two of the flood prediction hydrological models have been tested on some of the Lebanese watersheds The Coupled Routing and Excess STorage model (CREST, jointly developed by the University of Oklahoma and NASA SERVIR) and The Continuum model developed by the CIMA Research Foundation. CREST is a distributed hydrological model developed to simulate the spatial and temporal variation of surface and subsurface water fluxes and storages by cell-to-cell simulation. The major problem of this model is its suitability for the application in small scale watersheds and it needs to be downscaled and calibrated before being used in the forecasting chain. The Continuum model is a complete and distributed model that allows to simulate the main hydrological processes. It is designed with simple and robust processes schematization but the hydrologic characteristics of the Lebanese terrain in addition to the high karstification and poorly gauging showed limitation in its application.

Thereof the proposed mmethodology shall articulate on two major folds:
- Implement available hydrological models in literature (also using the continuous and distributed MISDc hydrological model developed by IRPI-CNR) and tune them to best reflect the hydrological processes of poorly gauged watersheds and scarce data environments that best fit the Lebanese context.
- Develop a Bayesian hierarchical modeling (BHM) approach for complex environmental problems in order to obtain a joint distribution model process and parameters.
These parameters will be integrated in a well-designed hydrological model. The model will be run and calibrated with the available historical events until the parameters are best correlated and the uncertainty in the model is minimized.

Accordingly, the first step shall be conducted through structured procedure, starting with data acquisition and preparation; vector data (land cover/ Land use, Soil, Drainage, Temperature, Rainfall, Discharge...), and raster data like DEMs and satellite images( near real time data). This step will be followed by data analysis to establish the required model parameters.
Rainfall and flow data will be analyzed for gaps and their corresponding probability of occurrence will be established. Soil data will also be analyzed to establish a weight-rate model that will be combined with landuse data to establish the corresponding runoff coefficients. Supplementary and complementary information from space born platform related to soil moisture (e.g., SMOS, SMAP, ASCAT products), precipitation (e.g., GPM) and snow will be utilized. These parameters will be integrated in a well-designed hydrological model and shall be calibrated (with the available historical events) to obtain the optimal correlation and the minimal uncertainty.
The second step consists of deriving an integrated hierarchical framework analogous to the BHM approach. Accordingly this approach shall adequately quantify the hydrologic predictive uncertainty of the model and reduce it to its maximum degree.

Research goals

Thereof, this research aims to develop an integrated uncertainty framework for hydrological modeling in order to improve operational flood forecasting by accurately characterizing hydrological processes and by associating an uncertainty with the model outputs.

It requires adequate understanding of all the different uncertainty sources and interrelated relationships, to conduct uncertainty quantification and reduction in a meaningful way. This is because different uncertainty sources may introduce significantly different error characteristics that require different techniques to deal with; and missing important uncertainty sources may lead to misleading uncertainty predictions in the hydrologic outputs.

The research also aims to distinguish model uncertainty from predictive uncertainty. While modeling uncertainty comes mainly from the imperfect fit to the truth of the past, predictive uncertainty can also arise from extrapolation errors or temporal prediction errors due to the fact that the future typically does not look exactly like the past (e.g., Morgan et al., 1990; Krupnick et al., 2006). In the context of hydrological data assimilation, addressing modeling uncertainty is of primary interest, which, in turn, will have an impact on predictive uncertainty.

Last update: 27/11/2021