WMO/CAS/WWW


FIFTH INTERNATIONAL WORKSHOP ON TROPICAL CYCLONES


Topic 2.3 : Hydrological Models of Precipitation

Rapporteur : Chong Sun Fatt,
Chairman,
Working Group of Hydrological Component,
Typhoon Committee,
c/o Department of Irrigation and Drainage,
KM 7, Jalan Ampang,
68000 KUALA LUMPUR, MALAYSIA.

E-mail : chongsf@did.moa.my

Fax : 00603-4256-3735

Working Group : Peter Baddiley, Reggina Garza

Abstract:

Every year tropical cyclones result in severe flooding and causes enormous economic damage, social disruption, and even loss of lives. Accurate forecasting of floods induced by tropical cyclones is therefore instrumental to the reduction of flood impacts. This report summarizes various operational flood forecasting models used in a number of countries that are affected by tropical cyclone flooding. Issues related to model performance and evaluation criteria are discussed. Limitations and problems in model selection, model calibration and real-time operation of flood forecasting models are deliberated. The report also suggests the priority areas of research and development for the improvement of meteorological inputs and hydrological considerations.

2.3.1 Introduction

Tropical cyclones are one of the most destructive weather phenomena to mankind. Damages caused by tropical cyclones are generally associated with wind damage, storm surge, and flooding. Accurate forecasting of floods induced by tropical cyclones requires adequate meteorological inputs such as real-time rainfall, quantitative precipitation forecasts (QPF), and the cyclone landfall location. Hence, close interaction and cooperation between flood forecasters and meteorologists is of significant importance to improving flood forecasting.

As tropical cyclones affect directly or indirectly many countries globally, a wide spectrum of forecasting models have been deployed based on national resources available. In the last few decades, the Tropical Cyclone Programme (TCP) under the World Meteorological Organization (WMO) has helped many countries in improving their national capabilities in flood forecasting and warning. The Typhoon Operation Experiment (TOPEX) from 1982 to 1983 was a classical example in which six countries in the Typhoon Committee area had improved their flood forecasting systems. Similar efforts were seen through the other Regional Bodies under the TCP.

2.3.2 The Existing Hydrologic Forecasting Models

Hydrology was defined by Penman (1961) as the science that attempts to answer the question, “What happens to the rain?” Hydrologic models are designed to answer Penman’s question and are often employed in a wide spectrum of applications that range from watershed management to engineering design (Singh 1995). A Flood Forecasting Model is one specific type of hydrologic model developed to simulate catchment responses to precipitation and generate forecasts of the water levels and streamflows.

Flood Forecasting Models can be categorized into three broad types (Wood et al. 1985), namely:

  1. Distributed physics-based models (e.g., SHE, TOP-model)
  2. Lumped Conceptual Models (e.g., Sacramento, Tank, CLS)
  3. Black box models (e.g., stage-regression, Unit-hydrograph)

Over the last few decades, there has been a proliferation of flood forecasting models since the advent of computers. Many different forecasting models which range from simple-stage regression techniques to sophisticated physically-based distributed hydrologic models, have been developed and used to forecast floods.

In the Typhoon Committee area covering the North Pacific, Malaysia, and the Philippines, countries are using the regression techniques, Unit-hydrograph methods, and conceptual models to forecast floods. China has developed and deployed many types of conceptual and Black Box models as the operational forecasting tools. In 2001, China has developed the National Flood Forecasting System, which could perform standardized data processing, model calibration, real-time forecasting and image displaying. The Republic of Korea has also developed the Standard Flood Forecasting Model using the intranet on the Web User Interface to ensure consistency at five flood control offices.

Japan uses a Storage Function Model at 366 flood forecasting points (75%) and Black-box method at 49 points. Rainfall forecast is required as a model inputs in 329 forecasting points out of a total 483.

Bangladesh has developed and utilized the MIKE-II flood forecasting model to provide river forecasts for an area covering 82,000 sq. km.

In the United States of America, the National Weather Service (NWS) generates forecasts for America’s rivers and streams at 4,000 locations. The NWS River Forecast System (NWSRFS) is the operational river forecast model, which consists of a comprehensive suite of programs and algorithms covering the entire forecast process, from real-time data ingest to the generation of river forecast hydrographs. This model needs to be fed continuously by meteorological and hydrologic data, mainly precipitation and river-level data. Other Simplistic Models are being developed to forecast for basins and catchments with short lead times (flash floods). This model is based on the development of Flash Flood Headwater Tables and the use of Antecedent Precipitation Index (API) methods.

In Australia, the primary hydrologic model used in flood forecasting operations is a distributed network storage routing model (named URBS). A typical configuration for the tropical cyclone-affected river systems has these elements:



2.3.3 Performance of the Flood Forecasting Models

Performance of flood forecasting models is normally evaluated based on forecasting errors, which are the difference between the observed and forecast values. Many evaluation criteria could be used to assess model performance. WMO (1975) used the following six evaluation criteria in the project on “Intercomparison of Conceptual Models used in Operational Hydrological Forecasting:”

  1. Coefficient of variation of residual of errors;
  2. Ratio of relative error to the mean;
  3. Ratio of absolute error to the mean;
  4. Arithmetic Mean;
(v) Phase coefficient’ and
(vi) Coefficient of Persistence

The degree of acceptance of the forecasting error is much dependent on many factors such as forecast lead time, catchment concentration time, rate of rise of flood level and bund-overlapping level. Forecast error increases with lead time, but the lead time must not be too small to enable meaningful warning dissemination and evacuation. Generally a forecasting error of ±1.0 metre could be the limit for rivers with larger fluctuation ranges. Otherwise, an error of ±0.3 metres should be the limit, especially for high stage flooding where bund overflow may occur.

WMO (1996) introduced a point-rating system called “Management Overview of Flood Forecasting System (MOFFS)” that provides a consistent method to evaluate the performance of flood forecasting systems. However, the MOFFS rating system is considered as rather subjective in evaluating model performance.

It is therefore suggested that the MOFFS be modified to serve as a more objective and quantitative way in evaluating model performance. Besides, a flood forecasting model should be evaluated based on its real-time operational performance and also its simulation capability, i.e., using known rainfall as the model’s QPF input during post-flood evaluation. River basins could be designated by interested members to participate in the model evaluation exercise.

2.3.4 Limitations of Flood Forecasting Models

Forecasting of flood caused by landfalling tropical cyclone still contains much uncertainty due to a number of constraints. Basically, these constraints could be categorized into three types, namely the model adequacy, calibration process, and real-time operation problems.

(a) Adequacy of a Selected Flood Forecasting Model

The adequacy of a selected forecast model to simulate the catchment characteristics and response is of significant importance. For the same lead time, a larger river is relatively easier to forecast than a small catchment, by virtue of its inertia. However, a tropical cyclone typically brings short-duration, intense rainfall (<12 hours) with high spatial variation. Hence, the model chosen is required to consider catchment characteristics such as catchment size; shape; topography; slope; land use; river network configuration; channel characteristics; man-made structures; flood concentration time; and flood travel time.

Singh (2001) pointed out thsy most models perform little or no error analysis and from the standpoint of a user, it is not clear how reliable a particular model is. Consequently, a user runs into difficulty when selecting a model.

(b) Adequacy of Model Calibration

An adequate forecasting model may not perform well if it is inadequately calibrated. Model calibration is often constrained by the lack of adequate calibration data, especially for small river basins with high spatial rainfall variability. Rainfall station network may not be dense enough to accurately estimate the catchment rainfall. High discharges during flood events are subject to significant error due to flow gauging difficulty. Dynamic changes in catchment conditions (e.g., urbanization) invariably introduce non-homogeneity in the water level and flow data.

(c) Real-time Operation Constraints

An adequate forecasting model, which has been well calibrated, may not yield satisfactory forecasts due to the following constraints during real-time operation:


The list of factors above present limitations to operational forecast modelling accuracy and the overall flood warning service quality. Despite the recent efforts to develop more complex physics-based distributed models (e.g., Radar spatial rainfall input module coupled with grid hydrologic cum hydraulic modules) for flood forecasting, the improvements in forecasting accuracy over the simpler models are only marginal. The two major constraints of physics-based distributed models are QPF and the difficulty to simulate representatively the heterogeneity of river catchments.

2.3.5 Priority Areas in Research and Development

(a) Meteorological Inputs

With respect to the forecasting of floods caused by tropical cyclones, the following areas should be given due priority concerning the meteorological inputs to the hydrologic modelling process. There are associated mostly with the limitations of estimation of real-time rainfall, QPF, and forecast of tropical cyclone landfall locations.





(b) Hydrological Considerations

In the domain of hydrology, focus should be given to the following areas to improve flood forecasting:









2.3.6 Summary

Forecasting of floods resulting from tropical cyclones is an important component in reducing flood damages and loss of lives. In the last few decades, a wide variety of forecasting models have been developed and used by national forecasting agencies. Cooperation between countries and assistance from Regional Bodies of WMO Tropical Cyclone Programme have helped many countries in improving their forecasting capability.

Despite the rapid advancement in real-time data collection and hydrological modelling technologies, the flood forecasters are presently confronted with three main problems: choice of suitable forecast models; the model calibration process; and the real-time operation constraints. Performance of forecasting models must be thoroughly evaluated with suitable criteria so that improvements to forecasting could be made.

Recognizing that meteorological inputs related to the tropical cyclone such as landfalling location, spatial rainfall estimation, and quantitative precipitation forecasts are fundamental to the accurate forecasting of floods (especially flash floods, landslides and debris flows), joint research between flood forecasters and tropical cyclone researchers should be carried out to address those issues.

Bibliography

Wood, Eric F. and P. E. O’Connell (1985). “Chapter 15: Real-time Forecasting”. Hydrological Forecasting, M. G. Anderson and T. P. Burt, ed., John Wiley and Sons Ltd., 505-558.

Penman, H. L. (1961). “Weather, Plant and Soil Factors in Hydrology”. Weather, 16. 207-219.

Singh, V. P. (1995). “Chapter 1: Watershed Modelling”. Computer Models of Watershed Hydrology, V. P. Singh, ed., Water Resources Publications, Littleton, Colo., 1-22.

Singh, V. P., and Woolhiser D. A. (2002). “Mathematical Modelling of Watershed Hydrology”. Journal of Hydrologic Engineering (Jul.–August 2002), 270-292.

WMO Technical Reports in Hydrology and Water Resources No. 55 (1996). “Development and Use of Management Overview of Flood Forecasting Systems (MOFFs)”, World Meteorological Organization, Geneva.