Broad scale strategic analysis of flood risk requires a rather different approach to hydraulic modelling compared to more conventional engineering design studies (Evans et al., 2002). Strategic analysis involves looking over wide spatial scales, which precludes very detailed representations of topography and other boundary conditions. Strategic analysis over long time scales implies the need for large numbers of runs, in order to explore uncertainties and future management options.
The Taihu Basin in China has been a particularly challenging application for broad scale strategic analysis of flood risk. It has a total area of about 37 000 km2 and is located in the delta region of the Yangtze River, and so is very low-lying. The channel network is extensive and subject to complex arrangements for flow control.
This paper describes how a novel hydraulic modelling system was selected and implemented for the Taihu Basin. The broad scale flood simulation system was able to generate credible results which provided inputs to a flood risk analysis system which estimated regional flood impacts for a range of climate change, socioeconomic and flood risk management scenarios.
Selection of modelling approach
The first stage in a flood simulation study is to select the overall flood modelling approach that is most appropriate for the particular circumstances of the project. There are many factors influencing selection of the modelling approach; the key criteria (Figure 1) are discussed in this section and include:
- overall project deliverables and objectives;
- modelling output requirements (locations, variables, durations, epochs, accuracy);
- relative importance of flow and flooding mechanisms for the study area/objectives;
- data availability;
- resource availability (staff, software, computer hardware) and programme constraints.
The aim of the formal approach to model selection for the Taihu Basin study was to design a modelling approach that was proportionate to the supporting role that flood simulation modelling took in the project. The aim was not to build the ‘best possible’ model, but rather to develop a modelling approach that was just sufficient to provide the evidence necessary to support the higher-level objectives of the project.
For the Taihu Basin project, the overall project objectives were to predict how flood risks may change in the Taihu Basin over a 50-year period and to assess the options for responding to the future challenges (see Cheng et al., this edition).
Modelling outputs required
The overall methodology developed to address the project objectives resulted in specific output requirements for the flood modelling project component which were, in summary, to generate peak water levels and flood volumes for the ‘plains’ area of the Taihu Basin to feed into the quantified flood risk assessment (see Yu et al., this edition). The objective of the flood simulation component of the project was therefore to develop a broad scale flood simulation model of the Taihu Basin and to use it to assist in the scenario analysis through prediction of water levels and flood volumes. The flood simulation model was not a fully detailed, relatively localised model such as would be used for design purposes but a wide-area, sparse-data model which, while being fast enough to permit the running of many cases needed for scenario analysis, reproduces at a sufficient level of accuracy the broad features of flooding and approximate flood levels and extents.
Important flow and flooding mechanisms
An important step in defining the requirements of the modelling approach was the qualitative analysis of important flood generation mechanisms and likely responses that was undertaken early in the project programme. Expert elicitation was used to create a conceptual model of the flooding system and identify important drivers for change and potential responses. The flooding mechanisms, drivers and responses were ranked and the project team and key stakeholders decided which key processes were needed to be included in the quantified risk analysis. This technique of deciding which processes to model was considered more efficient than alternatives such as modelling all potential flood influencing processes or relying on the experience and skills of a small team of modellers. An important secondary benefit of the technique was that it facilitated acceptance by the wider project team and key stakeholders of the modelling approach and acknowledgement of the contribution of the modelling to the overall project objectives.
The flooding processes in the Taihu Basin that were considered most important for the project and were therefore included in the model selection criteria were:
- inflows from the hills in the west of the Basin;
- inflow/outflow exchange with the Yangtze River to the north of the Basin;
- tidal boundary effects at the east and south boundaries of the Basin;
- direct rainfall in the ‘plains’ area;
- the main channel system within the ‘plains’ area;
- Taihu lake storage/water level relationship;
- gates and other major control structures;
- main flood embankments/dikes;
- internal polders and pumping
The data available to support the Taihu Basin flood simulation modelling included historic and predicted rainfall rates, channel locations and dimensions, broad scale terrain data, primary control structure operation data and historic water level and flow data. In general, all data were at a ‘broad scale’, for example, some channel cross sections were composite sections representing the combined conveyance of parallel channels. There were significant gaps in the data, both for historic observational data used for calibration/validation and in the dimensions/elevations of the flood management system (for example, many flood embankment crest levels had to be inferred).
The staff resources available to the project included skilled modellers and those with access to knowledge of the major operational activities within the Basin. However, involvement in the project was a secondary activity for most of those involved and this impacted on the actual staff resource available for the modelling. For example, the knowledge held by the Taihu Basin Authority (TBA) members of the project team was particularly important for the project; however, these team members also held important operational roles and had restricted availability to the project.
Following consideration of the main factors introduced earlier, it was decided to apply an approach which consisted of broad scale hydrological rainfall–runoff modelling of the hilly areas feeding a broad scale hydrodynamic model representing the main conveyance channels in the ‘plains’ area (using a full solution of the 1D Saint-Venant equations), linked to a very broad scale representation of the polder regions (using a storage cell concept representing mass conservation of inflows/outflows over embankments or from direct rainfall). Alternative simpler and more complex approaches were considered but rejected. Simpler approaches (such as 10 to 20 linked lumped conceptual ‘reservoir’ components) would not have been able to represent the main flooding processes in the Basin. More complex approaches (such as a 2D hydrodynamic representation of the floodplain) would have been severely constrained by the major data gaps and staff resource issues. The implementation of the selected approach is described in the following section.
The expert elicitation workshops early in the project had identified that the main cause of flooding that needed to be analysed in the project should be the long-duration rainfall events typified by the June to August flooding in 1999. The modelling system was therefore designed to be able to represent this type of flooding. Key boundary inputs were rainfall inputs to the rainfall–runoff models of the hilly areas and direct rainfall to the polder areas together with water level boundaries for the Yangtze River and coastal boundaries. Water entered the polder regions either through direct rainfall or through overtopping of the levees. In addition, a decoupled representation of potential breaching inflows to the polders was provided as part of the Taihu Basin Risk Assessment System (TBRAS) risk analysis model that was run following the hydrological and hydraulic model simulations.
Figure 2 illustrates the modular, loosely coupled approach to modelling in which the following three main groupings of models can be identified:
- hydrological models – modelling fluvial inflows [variable infiltration capacity (VIC) model] and direct rainfall inputs [Soil Conservation Service Curve Number (SCS CN) method];
- hydrodynamic model (ISIS) – modelling the movement of water through the network of channels and flood storage cells in the basin; and
- A risk analysis model (TBRAS) – modelling depths of inundation in the flood cells and calculating associated economic damages.
It was decided to drive the models using data based on the 1998–1999 rainfall season transformed to a series of critical return period events ranging from 2 years to 1000 years. A useful body of existing knowledge had been built surrounding this event (e.g. Taihu Basin Authority and Hohai University, 2000) and, as such, model performance could be readily assessed and acceptance criteria established. For example, it is widely acknowledged by the wider project team and key stakeholders that the 1999 rainfall event is approximately equivalent to a 200-year return period 30-day duration critical storm.
Rainfall depth-duration-frequency estimates were derived using a Pearson-III distribution with observed rainfall (1961–1990) at nine locations in the Taihu basin. The Chinese Academy of Agricultural Sciences (CAAS) Providing REgional Climates for Impact Studies (PRECIS) model rainfall data, provided on a 1° by 0.5° grid basis, were used to estimate both baseline (1961–1990) and climate change scenarios [Special Report on Emissions Scenarios (SRES) A2 and B2 scenarios for the 2030s]. The resultant depth-duration-frequency statistics across the basin were compared with existing estimates of the 100-year rainfall depth (TBA and Hohai University, 2000) to ensure that both baseline and climate change statistics were calibrated. Spatial variability in the depth-duration-frequency estimates was greatest in estimates derived from use of observed data. This variability was however not considered to be a true representation of rainfall heterogeneity across the basin and can be attributed to a variable Pearson-III model performance. Accordingly, a ‘regionalised’ model averaging local depth-duration-frequency statistics was adopted in subsequent modelling (Table 1).
|Return period||Average: observed (1961–1990)||Average: PRECIS (1961–1990)||Average: estimated A2 2030s||Average: estimated B2 2030s|
The VIC macroscale hydrological model (Liang et al., 1994) was used to estimate inflows from 19 hilly catchments into the hydrodynamic model. Setup and calibration of this model are described in Liu et al. (this issue).
Effective rainfall inputs into the ISIS model flood cells were modelled using observed 1999 rainfall processed for 16 individual rainfall districts using SCS curve numbers that differentiated between water, paddy, dry land and urban land use categories to estimate losses. The resultant estimates of effective rainfall were validated against existing estimates of effective rainfall produced by TBA. The regionalised depth-duration-frequency statistics were then applied to produce the required effective rainfall for all modelled return period events.
The hydrodynamic model of the Taihu Basin was built using the ISIS software (Evans et al., 2007 and http://www.halcrow.com/isis) with data principally provided by the TBA. The main reasons for the decision to use ISIS were that the ISIS software enabled an appropriately scaled channel and flood-cell model to be constructed (as opposed to a solely in-bank model that was possible using TBA's existing HOHY2 model) and that ISIS allowed flexible control rules to be developed for sluice and pump operation. The existing data and schematisation held within the HOHY2 model were extracted and used to build the initial ISIS model of the network.
The model used inputs of direct net rainfall, upland inflows, Yangtze River and coastal tide levels, Taihu lake initial water levels, sluice gates control rules and polder pumping rules. Simulations took about 0.5 h to run a 90-day period (based on June to August rainfall profiles). Outputs included channel water levels and flood volumes in the floodplain cells; these data, together with calculated expected flood volumes from breaching (see Xie et al., this edition), were passed to a Geographic Information System (GIS) model of the flood cells and receptors (the TBRAS system) for use in the calculation of risk estimates (see Yu et al., this edition).
The schematisation of the system was based on that previously used in the HOHY2 in-bank model but extended onto the floodplain and updated for recent flood control projects (Figure 3).
Not all channels were explicitly included in the ISIS model as the HOHY2 model used a process in which smaller channels are concatenated into equivalent channels. The concatenated channels have the same capacity as their component channels, and, thus, the overall conveyance capacity was preserved. After modification, the ISIS model for the Taihu basin consists of 2394 river cross sections, 22 inflow boundaries (19 nodes are western upland inflow nodes and the other three are fixed discharge inflow nodes) and 42 water level boundaries.
The floodplain was represented by ISIS flood cell units, each covering an area of about 100 km2, connected to the channel system by overbank spill units (weir equations representing flow over the flood banks). The approximate level–volume relationship in the flood cell units was derived from STRM-90 DEM data (projected to GCS_WGS_1984) using the ISIS Mapper tool to extract level-area data sets.
In the ISIS model, the flows into and out of the flood cells consist of (Figure 4):
- net direct rainfall;
- flows spilling over the flood banks surrounding the flood cells;
- pumped and/or gravity flows from the flood cells into the channel system.
The (in-bank) HOHY2 model did not contain bank crest data and surveyed or design bank crest data were not available for parts of the network. Where bank crest data were not available, approximate levels were inferred based on calculated extreme water levels, with freeboards supplied by TBA decided by judgement.
The boundary conditions for the ISIS model consisted of:
- water level boundaries along the Yangtze River and coastline with data either obtained directly from gauged data or generated by interpolation from the nearest gauge stations;
- inflow boundaries from the hilly region in the west of the basin with the inflow hydrographs provided by the VIC model outputs at a daily time interval;
- rainfall on the ‘plains’ area supplied from the net rainfall from the 16 rainfall districts;
- gate operations represented using control rules of gate operation as a function of local in-channel or lake levels;
The ISIS broad scale model was calibrated using data from the June to August 1999 flood. The first step was to set up the model to represent conditions as close as possible to the system status at the start of the 1999 flood event. Due to lack of data and the broad scale nature of the model, the initial calibration model included many assumptions and approximations. An iterative process was undertaken of model running, comparing simulated and recorded water levels in the lake and channel system, making adjustments to the assumed values and then rerunning the adjusted (hopefully improved) model. The main parameters adjusted during the calibration were the sluice gate control rules, pumping abstraction rates and pump rules. The calibration targets were to achieve a good representation of lake and channel water levels and to derive floodplain water levels which, when combined with data on the potential property flood losses, resulted in credible floodplain event damage values. The calibrated modelled and observed water levels at six key locations are shown in Figure 5 which, given the project objectives and data constraints, demonstrates an acceptable level of fit (average discrepancy in water level of about 15 cm). Simulated flood extent maps for 1999 were produced and were considered to provide a reasonable broad scale representation of reported flooding. Given the project objectives of understanding potential economic impacts of future flooding, a more important measure of the suitability of the model was a comparison between official TBA estimates of 1999 flood damages and that calculated from the TBRAS system using the outputs of the ISIS model. The model-simulated damages of 10.3 billion Yuan compare favourably with the TBA estimate for the 1999 flood damages of 14.1 billion Yuan (noting that there is significant uncertainty in both estimates).
The evidence from the comparison of water levels, and from the assessment results derived from the TBRAS risk assessment model, suggests that the ISIS broad scale model is able to generate results of sufficient accuracy for the scenario analysis, bearing in mind that:
- the model is intentionally broad scale in nature and local processes occurring near the gauge sites may not be well represented in the model;
- the flood water drainage processes include pumping and gravity drainage, both of which may be tidally influenced. In general, the control rules for the pumping and sluice gate operation were not available and as such were key aspects of model calibration;
- pumping outflows are directed into representative channel confluences in the model, whereas in reality, the pump outflows will be distributed into the channels surrounding the flood cells;
- many data items have had to be assumed and, therefore, there is uncertainty surrounding these values;
- the accuracy of the observed values is unknown.
Simulations for flood risk analysis
The broad scale hydrology/hydraulic model of the Taihu Basin provides a key component of the flood risk analysis framework that was developed during the Taihu Basin project. During the final phase of the project, a set of initial scenarios were simulated to demonstrate the utility of the framework. These initial scenarios covered climate change and socioeconomic change to 2050. The simulations suggested that climate change or socioeconomic change acting in isolation could lead to fivefold increases in flood risk; acting together flood risk could increase by a factor of 25. Further analysis, including the assessment of adaptation and flood risk management responses, is required to provide a robust evidence base for long-term management planning. The broad scale model can support this future analysis as it is able to represent a range of key drivers and responses, including:
- changes in rainfall (duration, profile and quantity) – by adjusting the rainfall time series in rainfall inputs to the ‘plains’ region, and through the rainfall–runoff modelling of adjusted rainfall which will lead to revised upland inflow hydrographs to the ISIS model;
- sea level rise – by adjusting the tidal boundary along the coast;
- Yangtze River levels – by adjusting the Yangtze water level boundary;
- impact of storm surges – by adjusting the tidal time series along the coast;
- ground subsidence – by adjusting the water level-area function of flood cells in the model and the flood bank crest levels and sluice levels;
- changes in operational flood management through revised operation of pumps and gates – by adjusting the rules in flood control projects;
- newly built gates and pumps – by adding in corresponding flood control project components;
- dyke building – by adjusting the dyke crest levels;
- new channels – by adding the channels to the model;
- land use changes in the hilly region – by adjusting VIC model parameters.
Other drivers and responses can be simulated in other components of the flood risk analysis framework (for example, property and community resilience and resistance measures would be represented in the TBRAS system).