Elsevier

Journal of Hydrology

Volume 331, Issues 1–2, 30 November 2006, Pages 58-72
Journal of Hydrology

Modelling of the flooding in the Okavango Delta, Botswana, using a hybrid reservoir-GIS model

https://doi.org/10.1016/j.jhydrol.2006.04.040Get rights and content

Summary

The Okavango Delta is dominated by annual flood events from the Okavango River. During such events the inundated area increases from about 5000 km2 to 6000–12,000 km2. Several models of a conceptual character were developed previously to represent hydrological processes in that system. Although essentially successful in their applications, the models have been criticised for their conceptual simplicity and the arbitrary way of representing long-term variation in outflows. All the existing models required the use of correction factors to address the apparent non-homogeneity of the time series. This paper presents a new model that has overcome these shortcomings by introducing more physical knowledge of the Okavango Delta system into the model. In view of the spatial complexity of the system and lack of data to support spatially distributed parameterisation of hydrological processes, the semi-distributed semi-conceptual approach, based on large units, has been retained. The major improvements of the model are: a better representation of surface water–groundwater interactions and the use of measurement-based rather than model calibrated parameterisation of topographic controls of floodplain water storage. These enabled a successful representation of 34 years of observed outflows and 15 years of observed inundation area in a conceptually sound way. Additionally, a GIS model has been developed for determination of spatial distribution of the simulated floods. In this model, the within-unit flood distribution is obtained from remote sensing-derived flood maps. In this way, in spite of the semi-distributed nature of the hydrological model, flood distribution maps and ecologically important flooding characteristics can be determined for simulated and predicted floods with a spatial resolution much higher than that of the computational units. The hydrological model developed forms the basis for subsequent analyses of ecosystem response to hydrological change of both floodplain and dryland ecosystems.

Introduction

In the past, wetland management practices were dominated by negative perceptions of wetland functions, resulting in flood regulation measures aiming at reclamation of fertile wetland soil for agriculture, protection from water-borne diseases and reduction of wetlands “losses” to evaporation in order to increase available surface water resources (Mitsch and Gosselink, 2000). Recently, however, wetlands are increasingly recognised as important ecosystems that are not only wildlife habitats but which also provide a range of benefits to human communities (e.g. Williams, 1990, Mitsch and Gosselink, 2000). Thus, the current management imperative aims at maintaining and improving existing wetlands and restoration of those that are lost or degraded. These activities are implemented either voluntarily, by accepting obligations of the Ramsar convention, or imposed by international law (e.g. EU Water Framework Directive). Hydrological processes determine the physical, chemical and biological characteristics of wetlands (Mitsch and Gosselink, 2000), and thus understanding of hydrological processes and the linkage between hydrology and ecology is a prerequisite to the success of wetland management. Hydrological models are increasingly accepted as tools for quantitative description of hydrological processes in wetlands, as they have the capacity of simulating (predicting) hydrological effects of various management alternatives. Wetland modelling approaches differ in complexity, physical basis and in data requirements, and the choice of a model depends on the specific purpose of modelling exercise. Wetland models vary from simple lumped water balance models through semi-distributed conceptual models to 1-D hydraulic models (e.g. generic models MIKE11, HEC) and distributed models (e.g. USGS MODFLOW with wetland’s package, or MIKE SHE). Although the distributed physically based models are considered the ultimate modelling tools and are increasingly used (e.g. Thompson et al., 2004), they need large amounts of data, use effective parameters that do not correspond to point measurements and are not easily identifiable from model calibration due to the equifinality problem (Pappenberger et al., 2005, Beven, 2006). Simpler, semi-distributed or lumped models of conceptual character, although deficient in terms of representation of spatial variability, are much less data demanding, and the (usually few) effective parameters are easier identifiable through calibration. Wetland models of this type are usually based on a lumped simulation of wetland storage in a spatial unit, and determination of inundated area using a volume–area curve obtained from topographical data. Such models have been applied to many small and large wetlands: e.g. floodplains in Australia (Costelloe et al., 2003), Hadejia wetlands in Nigeria (Thompson, 1995), dambos in Zambia and Zimbabwe (Wolski, 1999), Esteros del Ibera wetlands in Argentina (Ferrati and Canziani, 2005), Sudd in Sudan and Inner Niger Delta (Sutcliffe and Parks, 1989, Mohamed et al., 2004), Tonle Sap in Vietnam (Kite, 2001), Murray River in Australia (Whigham and Young, 2001). The conceptual models are shown to be useful in determination of dominant hydrological processes in a wetland and facilitate management decision. Importantly, lumped and semi-lumped models, in spite of their simplicity, can be directly linked to ecological responses, as, when merged with a DEM, they are able to provide information on inundation area, depth and duration, which are important ecological characteristics (e.g. Whigham and Young, 2001).

The Okavango Delta (Fig. 1) is a large wetland formed on an alluvial fan of the Okavango River (Kgathi et al., Kniveton and Todd , both this issue). During the annual flood event, the inundated area increases from about 5000 km2 to 6000–12,000 km2, depending on the size of the flood. The Okavango Delta is located in semi-arid NW Botswana, where highly seasonal rainfall (November–March) is of the order of 300–500 mm a−1 and potential evaporation (corrected pan A) amounts to 2100 mm a−1. The distance of more than 600 km from the headwaters of the Okavango River and the low topographic gradient of the alluvial fan (1:3500) cause a delay in the annual flood of the system. As a result, flooding occurs in the distal part of the Okavango Delta only during the late dry season (August–October). Such a setting creates an environment with a very specific ecology. As explained by (Murray-Hudson et al., this issue), the frequency and duration of annual floods are the principal drivers of the ecosystem. Both the feeding river and the Okavango Delta itself have a high potential for water resource development (Kgathi et al., this issue, Andersson et al., this issue).

In this paper a hybrid reservoir-GIS hydrological model of the Okavango Delta is presented, where fluxes and water storage are modelled for large lumped units, and distributed inundation maps are obtained by GIS analyses. The model was created to provide a tool to analyse the impacts on that system of upstream abstractions and development that modify the hydrograph of the Okavango River, as well as possible impacts from climatic change. Model outputs are suitable for analyses of ecological responses of the system to hydrological change (Murray-Hudson et al., this volume). In addition, the model enhances understanding of the hydrology of the Okavango Delta and provides insight into the possible effects of long-term changes to the system.

Section snippets

Characteristics of the hydrological system of the Okavango Delta

The hydrology of the Okavango Delta has been described in detail in the literature (e.g. Gumbricht et al., 2004b, McCarthy et al., 1998, Gieske, 1997; Kgathi et al., this issue) and only the main elements are presented here. Flooding in the Okavango Delta is primarily caused by the annual flood wave arriving from the Angolan part of the Okavango River basin (see Hughes et al., this volume). At Mohembo, the Okavango River enters a 10–15 km wide, 150 km long, flat-bottomed valley called the

Previous modelling

Several hydrological models of the Okavango Delta were previously developed (Dinçer et al., 1987, SMEC, 1990, Scudder et al., 1993, Gieske, 1997, WTC, 1997). These were reservoir models that simulated either a part or the entire system. These were calibrated against observed outflows in one of the many distributaries, the Boro (Fig. 1). Table 1 summarizes main features of the previous models. The old models have two major drawbacks that prevent them from being applicable for the purpose of this

Adopted modelling concept

Since 15-year series of NOAA-AVHRR derived flood maps are available (McCarthy et al., 2004), we developed a tool that integrates reservoir modelling and GIS-modelling. The reservoir model provides information on the inundation area in large units (>500 km2) and the fluxes between them. Subsequently, the GIS model allows for the translation of the unit flood size into a flood distribution map, and hence spatially distributed flood characteristics, which can be used to determine eco-system

Outflows

Model performance criteria are summarized in Table 2. Outflows from the Boro are well simulated (Fig. 6) with the correlation coefficient of the observed and simulated monthly flows of 0.91 and a root mean squared error (RMSE) of 11.8 M m3/month. These results are better than those obtained by the previous models (Dinçer et al., 1987, SMEC, 1990, Scudder et al., 1993), and similar to those obtained by Gieske (1997). In terms of Boro flow duration and flood size exceedance, the model performs very

Summary and conclusions

The Okavango Delta system can be adequately described by a conceptual model consisting of only 9 units that represent the major distributaries of the Okavango Delta. The conceptual model is straightforward with a minimum number of parameters. Many of the parameters are physically based and can be obtained through independent measurements. The GIS model that translates the inundation area in the hydrological model to distributed inundation maps is a tool that can be applied to determine

Acknowledgements

Work presented here was funded by EU in the framework of a project “Water and Environmental Resources for Regional Development”, coded ICA-4-CT-2001-10040. We thank the Department of Water Affairs, Republic of Botswana for providing hydrometric data.

References (40)

  • F. Pappenberger et al.

    Uncertainty in the calibration of effective roughness parameters in HEC-RAS using inundation and downstream level observations

    J. Hydrol.

    (2005)
  • J.R. Thompson et al.

    Application of the coupled MIKE SHE/MIKE 11 modelling system to a lowland wet grassland in southeast England

    J. Hydrol.

    (2004)
  • P.A. Whigham et al.

    Modelling river and floodplain interactions for ecological response

    Math. Comput. Model.

    (2001)
  • P. Wolski et al.

    Dynamics of surface and groundwater interactions in the floodplain system of the Okavango Delta, Botswana

    J. Hydrol.

    (2006)
  • Allen, R.G., Pereira, L.S., Raes, D., Smith, M., 1998. Crop evapotranspiration. Guidelines for computing crop water...
  • Andersson, L., Wilk, J., Todd, M., Hughes, D., Earle, A., Kniveton, D., Layberry, R., Savenije, H., this issue. Impact...
  • Bauer, P., 2004. Flooding and salt transport in the Okavango Delta, Botswana: key issues for sustainable wetland...
  • Dinçer, T., Heemstra, H.H., Kraatz, D.B., 1976. The study of hydrological conditions in an experimental area in the...
  • T. Gumbricht et al.

    The micro-topography of the wetlands of the Okavango Delta, Botswana

    Earth Surf. Process. Landforms

    (2004)
  • Hughes, D., Andersson, L., Wilk, J., Savenije, H., this volume. Regional calibration of the Pitman model for the...
  • Cited by (0)

    View full text