A GIS GRASS-embedded decision support framework for flood

Proceedings of the Open source GIS - GRASS users conference 2002 - Trento, Italy, 11-13 September 2002

A GIS GRASS-embedded decision support framework for flood forecasting

Sandra G. García*

* Department of Thermic Engineering and Fluids. Hydraulic Engineering Area, Technical University of

Cartagena, Paseo de Alfonso XIII, 52. 30203 Cartagena. Murcia, tel. ++34968325935, fax ++34968325435, e-mail . [email protected]

1 Abstract In this study spatial analysis tools are presented which allow the simulation and prediction of flash floods in semiarid basins to be carried out. Different applications are analysed of the Shyska operational system in real basins, developed with functions embedded in a Geographical Information System (GIS), which combines information from latest generation data acquisition technologies in real time. By means of a Digital Elevation Model (DEM) the topographical parameterisation of integrated hydrological models is carried out. Shyska is a support tool for the taking of decisions in real time, based on spatial analysis, when faced with a flash flood event. The results of applying these methodologies to basins in the south east of Spain are presented. 2 Introduction In this work, firstly an introduction is given as to the subjects of the study, their interrelation and importance. Different hydrological applications of the Geographical Information System (GIS) and Digital Elevation Model (DEM) in several basins in the south east of Spain which allow the concepts introduced to be understood better are discussed. Different methodologies and automatic processing tools for raster DEMs for the topographical parameterisation of hydrological models are presented together with examples of their application. For the development of this work, a computational tool known as Shyska [4], whose characteristics are analysed, has been used. Finally, the results are presented of models of spatially distributed rainfall-runoff transformation, where inputs come partially from a flood warning system in real-time.

2.1 SAIH and hydrological simulation and forecasting models The Hydrological Information Automatic Systems (SAIH) are real-time hydrometeorological information collection networks. Their main aim is to warn of flash flood situations. The coordinated integration of data provided by the SAIH with other sources of real-time information like those which are supplied by meteorological institutions (weather forecasts, radar and radar-satellite data, etc.), civil protection (areas at greatest risk), etc., supply more support instruments for the taking of decisions with the aim of lessening the effects of flooding. In this case, efficient integration tools and methodologies and the presentation of information with a high degree of space-time resolution are necessary. In the Mediterranean basins, where flash flooding is usual, it is not enough merely to be supplied with real-time information integration and coordination between the responsible bodies. Because of the short length of time, which exists between the hydrograph peak and the associated rainfall, the warning of this type of event should be based on forecasting, both meteorological (quantitative rainfall forecasting) and hydrological. In

A GIS GRASS-embedded decision support framework for flood forecasting 2

addition to the information integration techniques, forecasting models are needed which are able to exploit the space-time data available, presenting precise forecasts early enough.

2.2 Geographical Information Systems in Hydrology and Water Resources The GIS are computational tools to store, recover, process and visualize spatial data. A very interesting concept, presented by [9] defines the GIS as an information integration technologies which may include aspects of cartography, remote sensing, demography, economy, landscape attributes, computational science, etc. The applications of the GIS in Hydrology and Water Resources can be classified in two categories, as [21] suggests: management and analysis. The uses referred to with management include data storage, recovery and visualization. For example, certain attributes may be stored within the GIS, such as the sitting of wells, the hydrographical network, reservoirs, etc. The managers and planners of this resource can use this information for taking decisions to do with water resources and land planning. The analytical applications of the GIS refer to modelling. The real potential of the GIS lies in its analytical capacity, which allows, among other possibilities, the generation of new layers of information. The most common strategies of integrating environmental models and GIS can be summed up in three types, according to [19]: loose coupling, tight coupling and embedded.

2.3 Digital Elevation Models (DEM) in Hydrology The DEM represent the spatial distribution of ground heights with respect to an arbitrary reference level. The DEM may present depressions and flat areas which make the drainage pattern difficult to define. The depressions in the DEMs constitute one of the main problems in hydrological applications, since they impede flow propagation [7]. These areas contain cells surrounded by others of a similar or greater height. While the depressions and flat areas may be real attributes of the landscape, many times they are errors which may be caused by data entry, interpolation processes and/or the limited horizontal or vertical resolution of the DEM, as [12] suggest. Whatever their origin, they need a treatment which allows the definition of the complete drainage network. The paths which the flow follows may be used to carry out the flow propagation all over the basin, the delineation of basins and the estimation of upstream contribution area, and indirectly to estimate the wetness indexes, as [13] suggests. Some hydrological applications of the DEMs are presented in greater detail further on. The problem of depressions in the DEMs has been widely written on in the literature. Recently, several authors ([12]; [16]) have presented new algorithms for the treatment of depressions and flat areas. This problem constitutes a line of research at present. The DEMs are being used increasingly for the automatic extraction of drainage networks, so the problem of defining the scale of the initiation of channels takes on considerable importance in the effort to obtain networks similar to real ones. To identify the drainage network from the DEM two methods may be used, a constant threshold area or a threshold area depending on the slope able to reproduce the spatial variation of drainage density. Both ways of thinking, together with objective methodologies which allow the most suitable range of threshold areas to be discerned, have been integrated into the Shyska system.

Sandra G. García 3

3 The integrated tool: Shyska environment

3.1 Objectives The Shyska computational system was conceived as a support tool for the taking of decisions in real-time, which helps to plan strategies on a basin scale in the face of a flash flood event. The Shyska environment presents a structure based on a GIS oriented to the forecasting/simulation of flows. It is made up of a series of modules which allow the management and processing of hydrometeorological episodes in real-time which come from SAIH systems. But the reasons which underlie the Shyska system go beyond the simulation and forecasting of flash flooding in real-time. It is powerful tool, which by making use of the latest technology for the treatment of information with a high spatiotemporal resolution, enables studies to be carried out of the hydrological characterization of basins, erosion/sedimentation models, the study of the hydrological effects of changes in land use, etc.

3.2 Platform of development The system core was developed with the Tcl/Tk language [14], using the Tix language [8] the most sophisticated effects have been reached, and the C language has been used for the management and processing of time series, and computationally intensive mathematical algorithms. Commands belonging to the GRASS (Geographic Resources Analysis Support System) GIS have been integrated, along with new spatial commands based on GIS libraries and encoded in C language. Figure 1 shows an example of the system working under the Linux operative system.

Figure 1: Environment of work. Linux

3.3 Characteristics and structure The Shyska computational system with respect to its relationship with the GIS belongs to what has been defined as embedded integration. It has been developed with the same language and libraries as the GRASS GIS.

A GIS GRASS-embedded decision support framework for flood forecasting 4

Shyska presents a modular structure. These modules are oriented to (Figure 2): • Data Management. • Terrain Analysis. • Hydrological Modelling.

Figure 2: Shyska environment. main menu and first level of menus

The following possibilities are presented by Shyska:

(a) Combines information from last generation technologies of data acquisition, both supplied by SAIH systems and from remote sensing (rainfall field products of radar-satellite technology).

(b) Integrates spatially distributed and hybrid hydrologic models, topographically based.

(c) Automatically extracts from the DEM the relevant parameters for formulating the hydrologic models used.

(d) Integrates meteorological forecasting in real-time. 4 Hydrological applications of GIS: Study cases Three GIS application typologies can be found in hydrological modelling:

• The use of GIS for storage, processing and generation of data necessary to parameterise simulation models. The parameterisation of models using techniques based on GIS is the most common use in Hydrology.

• GIS assistance in automatic processes to derive morphometric characteristics of the drainage basin. These morphometric characteristics constitute or make up a part of the parameters of the model.

• The development of spatially distributed models with techniques embedded in the GIS.

4.1 Model parameterisation In accordance with [11] opinion, it is considered that in hydrological modelling, the most limiting factor does not correspond to the ability to characterize the hydrological processes but to the ability to determine the suitable parameters of the models. The GISs present digital data analysis techniques which greatly facilitate this task. The simulation models, both those with lumped or distributed parameters, need data as to land types and uses, slope, basin limits, etc. The GIS is used to analyse the different layers of information and to generate either a lumped parameter for the whole basin (or study area), or the spatial distribution of the parameter. A clear example of this application is the automatic generation of the Curve Number (CN) parameter of the Soil Conservation Service (SCS) runoff generation method, using the hydrological group

Sandra G. García 5

maps for soils uses, types of soil, practice and management, and hydrological condition. The result constitutes the spatial distribution of the parameter for the study area. A model, which applies the methodology presented by [20] modified, has been integrated into the Shyska system to estimate the CN value for each cell. Other examples of model parameterisations are based on attributes extracted from the DEM, either indexes or relations valid for the whole basin or spatial distributions. These subjects are discussed in greater detail in the following section.

4.2 Extraction of topographical attributes of hydrological interest from the

DEM 4.2.1 Study basins Different examples of the application of the DEM in the topographical parameterisation of hydrological models are presented in the two basins chosen which belong to the Segura River Basin in the south east of Spain (Figure 3). The basins studied belong to the Mula River Basin (169 km2), which is regulated by the La Cierva Reservoir and the Rambla Salada Basin (112 km2) which has no regulating structures. Although some examples are also presented for other basins of the Segura River. A mean annual temperature of 14-16 ºC, mean annual precipitation of 300-500 mm, and a mean annual evapotranspiration of approximately 850 mm characterize the study region. The Rambla Salada Basin is an ephemeral channel, while the Mula River Basin presents a stational regime with low flows in summer. Both basins present sensors belong to SAIH system.

Figure 3: Location of study area and SAIH raingauges 4.2.2 Primary topographical attributes From an idealized DEM, a series of topographical attributes may be extracted which are vital in order to obtain others which make up the parameters of certain models. Between the primary topographical attributes with hydrological interest, are the drainage flow accumulation, drainage direction, slope, and aspect. From the primary topographical attributes others known as secondary may be obtained, such as for example the topographical index defined as ln(a/tangβ), where a represents

A GIS GRASS-embedded decision support framework for flood forecasting 6

the drainage accumulation and β the slope. This index is used in the formulation of the TOPMODEL model [1], and has been widely applied by various authors ([2]; [15]). 4.2.3. Automatic extraction of basin from DEM From the drainage direction maps, the contributing basins for the stream gauges stations of the SAIH system were defined automatically. The contrasts between the limits of the basin traced manually from the printed (black line) topographical maps and the watershed extracted from the DEM (solid) have been satisfactory (Figure 4).

(a)

(b)

Figure 4: Contrast between basins automatically extracted from DEM 50 m: a) Rambla

Salada Basin, and b) Mula River Basin 4.2.4 Estimation of the spatial distribution of parameters For the basin defined automatically from the DEM, the spatial distributions of the flow velocity in each cell, the flow path length (Figure 5) and the corresponding flow travel time were estimated. The flow velocity field is estimated invariant with time. The methodology put forward by [11], is applied. The flow travel time map is estimated as the sum of the partial flow times through the cells that make up the flowpath. From this map, the hystograph time-area is obtained.

Figure 5: Spatial distribution of the flowpath lenght, (m) 4.2.5 Derivation of basin morphometric properties and hypsometric curve

Sandra G. García 7

Morphometric characteristics of the drainage basin are estimated for the basin automatically extracted from the DEM (Figure 6). Between the characteristics estimated below Shyska are: geometric indexes (area, perimeter of the basin, etc.), and of relief (maximum and minimum height, etc.), as well as hypsometric curve.

Figure 6: Estimation of morphometric indexes. Shyska module and example of hypsometric curve estimated from DEM 70m

4.2.6. Automatic extraction of the drainage network and estimation of scale properties of the channel network The estimation of indexes and scale properties for the hydrographical network from a DEM can be summed up as a four-step process:

1. Estimation of primary topographical attributes. 2. Automatic extraction of the drainage network. 3. Automatic encoding of the network by means of Strahler´s flow ordination

scheme. 4. Estimation of indexes and scale properties.

Two positions are to be found as to the extraction of the drainage network from a DEM: to adopt a constant threshold area or a threshold area which depends on the slope. In both cases, what ought to be made clear is that the choice of threshold area should not be arbitrary, as it affects the morphological and scale properties derived from the network, as various authors have shown ([6]; [4]). Strahler´s flow orders are assigned to each section of the channel corresponding to the total drainage network (Figure 7). The geomorphological relationships of the basin can be used as predictors of the flash flood properties. As examples, the mean flow of a river can be related to the basin area or to the drainage density of the hydrographical network, as well as the length of the main channel can also be used in equations which define the basin concentration time. Finally, Horton´s geomorphological relationships come into the parameter definitions for hydrological models, as for example the Geomorphological Instantaneous Unit Hydrograph (GIUH). The GIUH model was originally presented by [17]. From the encoded network according to Strahler´s scheme, a series of topological indexes may be obtained (drainage density, channel frequency, etc.)

A GIS GRASS-embedded decision support framework for flood forecasting 8

For each channel order the number of channels, channel length, drainage area, mean length and mean drainage area are defined. Once the Hortonian drainage network analyses have been carried out, Horton´s relationships of area, bifurcation and length can be estimated, by means of regression analysis. In Figure 7, the module developed to carry out this analysis can be seen. In addition, this module integrates different formulations to estimate the basin concentration time, even applying the formulations at cell scale. From the Horton’s relationships, fractals dimensions could be estimated in order to describe [19]: the length of main channel, the tree structure of the channel network and the drainage structure as a whole.

Figure 7: Example of Hortonian analysis. Threshold accumulation= 15 cells. Channel network automatically extracted. Application Strahler’s scheme. Corneros River Basin.

Segura River Basin. 4.2.7 Derivation of Geomorphological Instantaneous Unit Hydrograph (GIUH) and Characteristic Indexes Within the Shyska system, a module has been developed which allows the automatic estimation of the ordinates and characteristics (flow peak and time at peak) of the Instantaneous Unit Hydrograph model proposed by [18] which is really a combination of the GIUH shape and Nash´s Model.

Sandra G. García 9

Figure 8: Example of estimation of GIUH characteristics

4.3 Spatially distributed hydrological modelling: Examples of application Spatially distributed methodologies oriented to episode simulation have been integrated into the Shyska system, part of their parameters having been obtained from the DEM. The flow is routed at the basin outlet using a Unit Hydrograph that could be present a lineal translation component (pure translation model) and of storage (translation-storage model). Both models are based in the spatial distribution of flow velocity, proposed by [11]. The models developed have been applied to one of the study basins (Rambla Salada) for various episodes registered by SAIH system, in the period 1997-1999 [4]. The models parameters have been calibrated and validated. The results obtained are considered satisfactory. Using the same parameters, the velocity fields have been estimated for the both study basins obtaining good results in the hydrographs timing. In the Rambla Salada Basin, the pure translation model (T.P.) have been applied. While the translation-storage model (T.A.) have been adjusted for the Mula River Basin. An example of Shyska session is presented in the following Figure 9 for the episode 0997. This figure shows the main menu of Shyska, module for the selection of time period of simulation, and the simulation results (numerical and graphical results).

A GIS GRASS-embedded decision support framework for flood forecasting 10

Figure 9: Example of Shyska session. Episode 0997. Rambla Salada Basin The Figure 10 presents the results of simulation (hydrographs and hyetographs) in the Mula River Basin, for the episode registered in the period 30/09/1997 05:00 hs. – 30/09/1997 21:00 hs.

0

50

100

150

200

250

1 3 5 7 9 11 13 15 17 19 21 23 25 27 29 31 33Intervalos temporales (30min)

Escu

rrim

ient

o (m

3 /s)

0

5

10

15

20

25

30

Prec

ipita

ción

(mm

)PrecQobsQsim

Modelo T.A.

Figure 10: Observed and simulated flows. Mula River Basin. Episode 0997 (30/09/1997 05:00 hs. – 30/09/1997 21:00 hs.)

Sandra G. García 11

4 Conclusions The advances in computational sciences and in GIS have allowed the development of spatially distributed hydrological models, operative in real-time, whose functional relationships are applied at the cell scale into which the basin is discretized. The use of raster structures allows an efficient integration of radar-satellite and/or radar rainfall fields and simulation models and real-time forecasting, as has been shown in different applications ([3]; [5]). Every day the demand for high resolution spatial information is greater, both for management and modelling activities. The development of techniques for the integration of space-time information is important. In hydrometeorological warning situations in the Mediterranean basins, characterized by flash flood type events, decisions have to be taken in short periods of time. It is absolutely vital to be able to make use of systems which act as a support when taking decisions in real-time. This type of system should combine the latest information treatment technologies with the most efficient and precise simulation and parameterisation techniques. As mentioned above, more important than the modelling as such is precision in the specification of the model parameters. In this article, the application of a computational system, Shyska, which facilitates the real-time management of a flash flood event, has been presented. It is presented as a support tool for decision taking by local and regional authorities, when broadcasting warnings as to the possibility of flash flooding, in order to lessen the effects of the flood. 5 Acknowledgments The author thanks to the Confederación Hidrográfica del Segura for providing data for this study. References [1] Beven, K. and Kirkby, M. J., 1979. A physically based, variable contributing area

model of basin hydrology. Hydrol. Sci. Bull., 24, pages 43-69.

[2] Chairat, S. and Delleur, J.W., 1993. Effects of the Topographic Index Distribution on Predicted Runoff Using Grass. Water Resources Bulletin. American Water Resources Association, 29, pages 1029-1034.

[3] García, S.G., 2001. A Real-Time Flood Forecasting System based on GIS and DEM.

Remote Sensing and Hydrology 2000 (Proc. symposium held at Santa Fe, New Mexico, USA, April 2000), (eds. Owe, M., Brubaker, K., Richtie, J. & Rango, A.), IAHS Publ. no. 267, pages 439–445.

[4] García, S. G., 2002. A real-time flood forecasting and simulation system based on GIS

and DEM: Analysis of sensitivity to scale factors (Spanish text). Doctoral Thesis. ProQuest Information and Learning. UMI Division. ISBN 0-493-47612-1. Michigan, USA, pages 498.

[5] García, S. G. 2002a. The modelling of flash flood with SAIH and radar-RAINSAT

data combination. Fifth International Symposium on Hydrological Applications of Weather Radar – Radar Hydrology- (Proc. Symposium held in Kyoto, Japan, Nov. 2001) , pages 401-406 (in revision for Hydrological Processes. Special Issue.).

A GIS GRASS-embedded decision support framework for flood forecasting 12

[6] Helmlinger, K.T, Kumar, P. and Foufoula-Georgiou, E., 1993. On the Use of Digital Elevation Model Data for Hortonian and Fractal Analyses of Channel Networks. Wat. Resour. Res., 29(8), pages 2599-2613.

[7] Jenson, S. K. and Domingue, J. O., 1988. Extracting Topographic Structure from

Digital Elevation Data for Geographic Information Systems Analysis. Photogrammetric Engineering and Remote Sensing, 54 (11), pages 1593-1600.

[8] Lam, I. K., 1995. Tix Reference Manual - Draft. Computer Graphics Laboratory,

University of Pennsylvania. Online Report. [9] Loague, K. and Corwin, D. L., 1999. Regional-scale assesment of non-point source

groundwater contamination. In: Hydrol. Applications of GIS. (Eds. Gurnell, A. M., and Montgomery, D. R.). Advances in Hydrological Processes. John Wiley & Sons. Great Britain, pages 137-145.

[10] Maidment, D. R., 1993. GIS and hydrologic modeling. In: Environmental Modeling

with GIS. (Eds. Goodchild, M. F., Parks, B. O., and Steyaert, L. T.). Oxford University Press, Oxford, pages 147-167.

[11] Maidment, D. R., Olivera, F., Calver, A., Eatherall, A. and Fraczeck, W., 1996. Unit

hydrograph derived from a spatially distributed velocity field, Hydrol. Proc., 10, pages 831-844.

[12] Martz, L. W. and Garbrecht, J., 1999. The treatment of flat areas and depressions in

automated drainage analysis of raster digital elevation models. In: Hydrol. Applications of GIS. (Eds. Gurnell, A. M., y Montgomery, D. R.). Advances in Hydrol. Proc. John Wiley & Sons. Great Britain, pages 23-35.

[13] O'Loughlin, E.M., 1986. Prediction of Surface Saturation Zones in Natural

Catchments by Topographic Analysis. Wat. Resour. Res., 22(5), pages 794-804. [14] Ousterhout, J. K., 1994. Tcl and the Tk Toolkit. Addison Wesley (ISBN 0-201-

63337-X). Online Manual. [15] Quinn, P., Beven, K., Chevallier, P. and Planchon, O., 1993. The Prediction of

Hillslope Flow Paths for Distributed Hydrological Modelling Using Digital Terrain Models. In: Terrain Analysis and Distributed Modelling in Hydrology. (Eds. Beven, K.J. and Moore, I.D.), Advances in Hydrol. Proc., pages 63-83.

[16] Rieger, W., 1999. A phenomenon-based approach to upslope contributing area and

depressions in DEMs. In: Hydrol. Applications of GIS. (Eds. Gurnell, A. M., and Montgomery, D. R.). Advances in Hydrol. Proc. John Wiley & Sons. Great Britain, pages 37-52.

[17] Rodriguez-Iturbe, I. and Valdés, J. B., 1979. The Geomorphologic Structure of the

Hydrologic Response. Wat. Resour. Res., 15(6), pages 1409-1420. [18] Rosso, R., 1984. Nash model relation to Horton order ratios. Wat. Resour. Res., 20

(7), pages 914-920. [19] Roth, G., La Barbera, P. and Greco, M. (1996). On the description of the basin

effective drainage structure. J. Hydrol., 187, pages 119-135.

Sandra G. García 13

[20] Srinivasan, R. and Engel, B. A. (1991) r.cn. Agricultural Engineering Department, Purdue University. http://www.geog.uni-hannover.de/grass/gdp/html_grass5/html/r.cn.html

[21] Thompson, S. A., 1999. Hydrology for Water Management. A.A. Balkema.

Rotterdam, Netherlands.