Flood Forecasting using Artificial Neural Network

D.P Gamage 1, Kalani Ilmini2

1,2 Department of Computer Science.

General Sir John Kotelawala Defence University, SriLanka.

1D.P. Gamage;

2Kalani Ilmini;

Abstract – Flood is a natural disaster that cause human hardship and economic loss. To mitigate these effects, we need a better flood forecasting and warning system for communities which suffer from floods. However, establishing a precise flood forecasting system is still challenging because of the lack of knowledge about the effective variables in forecasting. The present study has showed that the use of artificial intelligence, especially neural networks is suitable for flood forecasting systems.

Keywords- Natural disasters, Artificial intelligence,

Artificial Neural Network, Flood

I.INTRODUCTION

Flooding impact on individuals and communities, with consequences including risk to human life, disturbance of transport and communication networks, damage to property, the loss of agricultural crops and deterioration of health conditions owing to waterborne diseases. Therefore, prevention and protection are required that aim to reduce the damage to people and public and private property.

Flood forecasting and prediction abilities evolved slowly during the 1970s and 1980s. Recent technological advances have had a major influence on forecasting methodologies. For instance, hydrological models use physical detection systems to forecast flood conditions based on predicted or measured parameters like rainfall. River flow models are used as components in actual flood forecasting schemes, where forecasts are required to issue warnings and to permit the evacuation of populations threatened by rising water levels. The basis of such forecasts is always observation and predictions of rainfall in the upper catchment area or river flows at upstream points along main rivers or tributaries. Forecasts about the discharge are obtained in real-time, by using the model to transform the input functions into a corresponding discharge function time.

Prediction of flood flow with suitable lead time is important to provide time for taking safety measures and removal of population being threatened by flood. Artificial Neural Network (ANN) is a powerful computational tool having the capability of capturing underlying characteristics of any physical process from the data set and can extract the relation between the input(s) and output(s). Some of the earliest application of ANN in hydrology has been reported by Daniel (1991).

ANN has a strong capacity to identify the underlying nonlinear relationship of a physical process. So, it may be suitable for the problems of estimation and prediction in hydrology.

The importance of Artificial Neural Networks (ANNs) to flood forecasting can be described in three points. Firstly, ANNs can represent any random nonlinear function given sufficient complexity of the trained network. Secondly, ANNs can find relationships between different input samples and, can group samples in similar fashion to cluster analysis. Finally, ANNs are able to generalize a relationship from small subsets of data at the same time, remaining relatively robust in the presence of noisy or missing inputs and can adapt or learn in response to changing environments. However, in spite of these important advantages, ANNs have found rather limited application in hydrology and related disciplines. For example, French et al. (1992) used a neural network to forecast rainfall intensity fields in space and time, while Raman & Sunilkumar (1995) used an ANN to synthesize reservoir inflow series for two sites in the Bharathapuzha basin, South India. Similarly, Hewitson & Crane (1994) described a range of climatological ANN applications such as snowfall prediction, classifying arctic cloud and sea ice, precipitation and, more recently, climate change impacts modelling (Hewitson & Crane, 1996).

However, the use of artificial neural networks for flood forecasting is an area which has yet to be fully discovered (Cheng & Noguchi, 1996). Until now the most of work in this area has been mainly theoretical, concentrating on neural network performance with artificially generated rainfall-runoff data (Minns & Hall, 1996). These theoretical approaches tend to supervise the difficulty in converting and applying actual data to artificial neural network topologies. Hall & Minns (1993) go some way to address this criticism by applying neural networks to a small urban catchment area. However, their argument is limited to the performance of a neural network on a small number of events.

Adequate warning time may save lives and belongings by allowing time to effect various structural and other adjustment. Earlier advanced warning can be achieved through advances in mathematical modelling. There are many rainfall-runoff models being developed and employed for flood forecasting which lead to the issue of flood warnings. An ANN is a quick and flexible approach which gives very promising results. Unfortunately, the inability to predict beyond the limits of the training range (extrapolation problem) was found to be a serious limitation of this approach.

At the current moment in SriLanka there is no proper system for flood forecasting using ANN.

The remainder of the paper is organized in the following sections:

II.LITERATURE REVIEW

A. Related Works

Applications of ANNs in hydrology are forecasting daily water demands (Zhang, Watanabe, & Yamada, 1993) and flow forecasting (Zhu & Fujita, 1993). Zhu and Fujita used NNs to forecast stream flow 1 to 3 hours in the future. They used three situations in applying ANNs: (a) off-line, (b) online, and (c) interval runoff prediction. The off-line model represents a linear relationship between runoff and incremental total rainfall. The on-line model assumes that the predicted hydrograph is a function of previous flows and precipitation. The interval runoff prediction model represents a modification of the learning algorithm that gives the upper and lower bounds of forecast. They found that the on-line model worked well but that the off-line model failed to accurately predict runoff (Zhu & Fujita, 1993).

The United Nations General Assembly declared the 1990s the International Decade for Natural Disaster Reduction. A prominent element within this programme has been the development of operational flood forecasting systems. These systems have evolved through advances in mathematical modelling (Wood and O’Connell, 1985; O’Connell, 1991; Lamberti and Pilati, 1996), the installation of telemetry and field monitoring equipment at critical sites in drainage networks (Alexander, 1991), through satellite and radar sensing of extreme rainfalls (Collier, 1991), and through the coupling of rainfall and runoff models (Georgakakos and Foufoula-Georgiou, 1991; Franchini et al., 1996). But, a successful real-time flood forecasting system often depends on the efficient integration of all these separate activities (Douglas and Dobson, 1987). Under the auspices of the World Meteorological Organization (1992) a series of projects were implemented to compare the characteristics and performance of various operational models and their updating procedures. A major conclusion of the most recent inter-comparison exercise was the need for robust simulation models to achieve consistently better results for longer lead times even when accompanied by an efficient updating procedure.

An ANN, using input from the Eta model and upper air surroundings, has been developed for predicting the probability of precipitation and quantitative precipitation forecast for the Dallas-Fort Worth, Texas, area. This system provided forecasts that were remarkably accurate, especially for the quantity of precipitation, which is paramount importance in forecasting flooding events (Hall & Brooks, 1999).

An application related to flood forecasting is a study done to model rainfall-runoff processes (Hsu et al., 1995). They developed an ANN model to study the rainfall runoff process in the Leaf River basin, Mississippi. In the study, the network was compared with conceptual rainfall-runoff models, such as Hydrologic Engineering Center (HEC)-I ((HEC), 2000), the Stanford Watershed Model, and linear time series models. In the study, the ANN was found to be the best model for one step ahead predictions. From the research and applications that are currently available, the addition of ANN learning abilities would be priceless to disaster planners, disaster logistics, mitigation, and recovery as well as many business, community, and transportation decisions.

Lately, functional networks were added to the ANN tools. Bruen and Yang (2005) investigated their use in real-time flood forecasting. They applied two types of functional networks, separate and associatively functional networks to forecast flows for different lead times and compared them with the regular ANN in three catchments. They demonstrated that functional networks are comparable in performance to ANNs, as well as easier and faster to train (Bruen & Yang, 2005).

The feasibility of using a hybrid rainfall-runoff model that used ANNs and conceptual models was studied by Chen and Adams (2006). Using this approach, they studied the spatial variation of rainfall and heterogeneity of watershed characteristics and their impact on runoff. They demonstrated that ANNs were effective tools in nonlinear mapping. It was also determined that ANNs were useful in exploring nonlinear transformations of the runoff generated by the individual sub catchments into the total runoff of the entire watershed outlet. They concluded that integrating ANNs with conceptual models shows potential in rainfall-runoff modeling (Chen & Adams, 2006).

“Artificial neural network approach to flood forecasting in the River Arno” (2003) is a neural network model predicting water-level variation up to 6 h in advance is presented and discussed. The model is built using data collected on the basin of the River Arno. A methodology is described for the selection of model inputs based on analytical procedures (cross-correlation analysis) and hydrological expertise, rather than on extensive model and sensitivity analysis to input data. This procedure allows the identification and calibration of the model to be speeded up because the relevant rainfall, power production, and water level data to be used for prediction and the possible limit to the prediction lead time can be identified in advance. Furthermore, it was found that the model predicts satisfactorily the evolution of water levels during floods. The percentage error in flow rate is less than 7% for the 6-h ahead prediction when the water level is higher than 5.0 m. (Campolo, Soldati & Andreussi ,2003)

“Flood Forecasting Using Artificial Neural Networks in BlackBox and Conceptual Rainfall-Runoff Modelling “(2002) is consisted with black-box application and a conceptual model. The black-box application confirms the importance of the addition of the exogenous input. This appears to be more remarkable for increasing lead-times, as could be expected given the stronger influence of rainfall. The best performing networks have moderate hidden layers dimensions and are relatively parsimonious. In fact, the analysis of the forecasts with varying number of input nodes indicates as most influencing the past four hourly values of observed flow and the past four values of rainfall. As far as the conceptual model is concerned, the gain allowed by the introduction of ANN-based rainfall forecasts is sensible, but not as remarkable as the one given by the addition of the discharge updating, as it may be seen considering the differences in the goodness-of fit criteria between conceptual models. (Toth & Brath ,2002).

“River flow model using artificial neural networks” (2015) An artificial neural network model named, “River flow model using artificial neural networks” (2015), show good ability to model hydrological process. They are useful and powerful tools to handle complex problems compared with the other traditional models. In this research, the results obtained show that the artificial neural networks are capable of model rainfall-runoff relationship in the semiarid and Mediterranean regions in which the rainfall and runoff are very irregular, therefore, confirming the general enhancement achieved by using neural networks in many other hydrological fields. The results and comparative study indicate that the artificial neural network method is more suitable to predict river runoff than classical regression model. The ANN approach could provide a very useful and accurate tool to solve problems in water resources studies and management. (AichourI et al, 2015)

“Real-time multi-step-ahead water level forecasting” (2014) has three ANN models (one static, two dynamic) are developed to make forecasts on the evolution of water level at floodwater storage pond (FSP) as a function of current FSP water level and rainfall information based on the inputs extracted by an advanced factor selection method for allowing sufficient time advance to warm up the pumping system and enhancing secure pumping operations to prevent the city from flooding. The time-based resolution of water level and rainfall data is 10 min, and the forecasting horizon is 60 min (i.e., 6-time steps ahead). (Chang et al ,2014)

“Flood Forecasting Using Artificial Neural Networks: An Application of Multi-Model Data Fusion Technique” (2016) is a research that establishing an accurate flood warning system is important for flood management. Artificial neural network is typically considered as a suitable method to model physical phenomena such as riverine flood forecasting. But, identification of the effective inputs for ANN modelling is important. In this research a neural network model and a data fusion technique were used to predict flood flows in the Karun River. Five ANN models were developed with different inputs combinations. Data for the fourth and fifth model were combined based on input variables for discharge, precipitation and the precipitation index for the previous seven days. Despite variety of input data, the resulting model showed greater accuracy in predicting floods compared to models with less variety of input variables. Sensitivity tests were utilized to omit inputs with less effects on prediction of floods resulting in even greater accuracy.

In 2017 Sri Lanka floods resulted from a heavy southwest monsoon, beginning around 18 to 19 May 2017. According to the Disaster Management Center, SriLanka; flooding was worsened by the arrival of the precursor system to Cyclone Mora, causing flooding and landslides throughout Sri Lanka during the final week of May 2017. The floods affected 15 districts, killed at least 208 people and left a further 78 people missing. As of 3 June, 698,289 people were affected, while 11,056 houses were partially damaged, and another 2,093 houses destroyed.

The flooding severely affected Sri Lanka’s Western Province, Sabaragamuwa Province, Southern Province and part of Central Province. The worst-affected districts were Kalutara, Matara and Ratnaputra. In Kalutara, flooding of the Kalu River also triggered several mudflows. Agalawatte, a town within Kalutara District, reported 47 deaths and 62 people missing as of 29 May, with many areas made inaccessible by landslides. The Ratnaputra District had recorded 79 deaths by 30 May.

B. Technology Adopted

The concept of ANNs dates back to the third and fourth Century B.C. with Plato and Aristotle, who formulated theoretical explanations of the brain and thinking processes. Descartes added to the understanding of mental processes. W. S. McCulloch and W. A. Pitts (1943) were the first modern theorists to publish the fundamentals of neural computing. This research initiated considerable interest and work on ANNs (McCulloch ; Pitts, 1943). During the mid to late twentieth century, research into the development and applications of s accelerated dramatically with several thousand papers on neural modeling being published (Kohonen, 1988).

(Muller ; Reinhardt, 1990) wrote one of the earliest books on ANNs. The document provided basic explanations and focus on ANN modeling (Muller ; Reinhardt, 1990). Hertz, Krogh, and Palmer (1991) presented an analysis of the theoretical aspects of ANNs (Hertz, Krogh, ; Palmer, 1991).

In recent years, a great deal of work has been done in applying ANNs to water resources research. Capodaglio et al (1991) used ANNs to forecast sludge bulking. The authors determined that ANNs performed equally well to transfer function models and better than linear regression and ARMA models. The disadvantage of the ANNs is that one cannot discover the inner workings of the process. An examination of the coefficients of stochastic model equations can reveal useful information about the series under study; there is no way to obtain comparable information about the weighing matrix of the (Capodaglio, Jones, Novotny, ; Feng, 1991).

III.CONCLUSION

The literature shows that ANN methodology has been reported to provide reasonably good solutions for circumstances where there are complex systems that are (a) poorly defined and understood using mathematical equations, (b) problems that deal with noisy data or involve pattern recognition, and (c) situations where input data is incomplete and ambiguous by nature. It is because of these characteristics, that it is believed an ANN could be applied to model the daily rainfall-runoff relationship. It was demonstrated that the ANN rainfall-runoff models exhibit the ability to extract patterns in the training data. The testing accuracy that is compared favorably to the training accuracy, based on the ratio of standard error to standard deviation and percent prediction of peak flows, supports this belief.

As in the literature review In SriLanka mainly Kaluthara and Ratnapura districts are flooded during heavy rainfall.

REFERENCES

Aichouri, I., Hani, A., Bougherira, N., Djabri, L., Chaffai, H., Lallahem,S.(2015). River flow model using artificial neural networks. International Conference on Technologies and

Materials for Renewable Energy, Environment and Sustainability, TMREES15

Alexander, D. (1991) Information technology in real time For Monitoring and managing natural disasters.Progress in Phys.Geogr, 15,238-260.

Biragani,Y.T., Yazdandoost,F., Ghalkhani,H.(2016) Flood Forecasting Using Artificial Neural Networks: an Application of Multi-Model Data Fusion Technique. Journal of Hydraulic Structures J. Hydraul. Struct., 2016; 2(2):62-73 DOI: 10.22055/jhs.2016.12853

Bruen, M., ; Yang, J. (2005). Functional networks in realtime flood forecasting–a novel application. Advances in Water Resources, 28(92), 899-909.

Campolo.M. , Soldati , A. ; Andreussi ,P . (2003) Artificial neural network approach to flood forecasting in the River Arno, Hydrological Sciences Journal, 48:3, 381-398, DOI:

10.1623/hysj.48.3.381.45286

Capodaglio, A. G., Jones, H. V., Novotny, V., ; Feng, X.

(1991). Sludge Bulking analysis and forecasting: Application of system identification and artificial neural computing technology Water Research, 25(10), 1217-1224.

Chang, F., Chen,P. , Lu, Y., Huang, E. , Chang, K. (2014).Realtime multi-step-ahead water level forecasting by recurrent neural networks for urban flood control.

Journal of Hydrology 517 ,836–846

Chen, J., ; Adams, B. J. (2006). Integration of artificial neural networks with conceptual models in rainfall-runoff modeling. Journal of Hydrology, 318(1-4), 232-249.

Cheng, X. ; Noguchi, M. (1996) Rainfall—runoff modelling

By neural network approach. Proc, Int. Conf. on Water Resource

Collier, C. G. (1991) Problems of estimating extreme rainfall from radar and raingauge data illustrated by the Halifax storm, 19 May 1989. Weather 46, 206-210

Daniell, T. M, 1991. ”Neural networks—Applications in hydrology and water resources engineering.” Int. Hydrology and Water Resources Symposium, Perth, 2–4 October 1991, 791– 802.

Douglas, J. R. ; Dobson, C. (1987) Real time flood forecasting in diverse drainage basins. In: Weather Radar and Flood Forecasting (ed. by V. K. Collinge ; C. Kirkby), 153-169. Wiley, London.

Franchini, M., Helmlinger, K. R., Foufoula-Georgiou, E. ;

Todini, E. (1996) Stochastic storm transposition couple With rainfall runoff modeling for estimation of exceedance probabilities of design floods. L Hydrol. 175(1-4), 511-532.

French, M. N., Krajewski, W. F. ; Cuykendall, R. R. (1992) Rainfall forecasting in space and time using a neural network. J. Hydrol. 137, 1-31.

Georgakakos, K. P. ; Foufoula-Georgiou, E. (1991) Real time coupling of hydrologic and meteorological models for flood forecasting. In: Proc. NATO Advanced Study Institute on Recent Advances in the Modelling of Hydrologic Systems (Sintra, Portugal, 10—23 July 1988) (ed. by D. S.

Bowles ; P. E. O’Connell). Kluwer, The Netherlands.

Hall, M. J. & Minns, A. W. (1993) Rainfall—runoff modelling as a problem in artificial intelligence: experience with a neural network. BHS 4th National Hydrology

Symposium, Cardiff, 5.51-5.57

Hall, T., & Brooks, H. E. (1999). Precipitation Forecasting Using a Neural Network. Weather Forecasting, 14(3), 338345.

Hertz, J., Krogh, A., & Palmer, R. (1991). Introduction to the Theory of Neural Computation: Addison-Wesley Publishing Company.

Hewitson, B. C. & Crane, R. G. (1994) Neural Nets: Applications in Geography Kluwer, Dordrecht, The Netherlands.

Hsu, K. L., Grupta, H. V., & Sorooshian, S. (1996). A superior training strategy for three-layer feed forward artificial neural networks (Technique report No. HWR no. 96-030).

Tucson: University of Arizona, Department of Hydrology and Water Resources.

Kohonen, T. (1988). An Introduction to Neural Computing. Neural Networks (1), 3-16.

Lamberti, P. & Pilati, S. (1996) Flood propagation models for real-time forecasting. L Hydrol. 175(1—4), 239—266.

McCulloch, W. S., & Pitts, W. (1943). A logical calculus of the ideas immanent in nervous activity. Bulletin of Mathematical Biology, 5, 115-133.

Minns, A. W. & Hall, M. J. (1996) Artificial neural networks As rainfall-runoff models. Hydrol. Sci. J, 41(3), 399417.

Muller, B., & Reinhardt, J. (1990). Neural Networks, an Introduction: Springler-Verlag

O’Connell, P. E. (1991) A historical perspective. In: Proc. NATO Advanced Study Institute on Recent Advances in the Modelling of Hydrologic Systems (Sintra, Portugal, 10—23 July 1988) (ed. by D. S. Bowles ; P. E. O’Connell). Kluwer, Dordrecht, The Netherlands.

Raman, H. & Sunilkumar, N. (1995) Multivariate modelling Of water resources time series using artificial neural networks. Hydrol, Sci. J. 40(2), 145—163.

Toth, E. and Brath, A. “Flood Forecasting Using Artificial

Neural Networks in Black-Box and Conceptual RainfallRunoff Modelling” (2002). International Congress on Environmental Modelling and Software. 106.

WMO (World Meteorological Organization) (1992) Simulated real-time intercomparison of hydrological models. Operational Hydrology Report no. 38, WMO 779, Geneva, Switzerland

Wood, E. F. & O’Connell, P. E. (1985) Real time forecasting.

In: Hydrological Forecasting (ed. by M. G. Anderson ; T. P. Burt), Chapter 15. Wiley, Chichester

Zhang, S. P., Watanabe, H., ; Yamada, R. (1993, June 1993). Prediction of Daily Water Demands by Neural Networks. Paper presented at the Stochastic and Statistical Methods in Hydrology and Environmental Engineering, University of Waterloo, Waterloo, Ontario, Canada.

Zhu, M., ; Fujita, M. (1993, June 1993). Application of Neural Networks to Runoff Forecast. Paper presented at the Stochastic and Statistical Methods in Hydrology and Environmental Engineering, University of Waterloo, Waterloo, Ontario, Canada.