Peter van Horssen
Department of Environmental Studies
University of Utrecht, Utrecht, The Netherlands
Abstract
An existing non-spatial logistic regression model (Barendregt, 1993) has been modified and implemented in a GIS as a step towards landscape ecological modeling of a wetland area. The model can calculate responses for 102 plant species in marsh ecosystems to spatially variable environmental conditions. Input variables derive from existing maps or from interpolation between point measurements. The output is a map of ecological groups. These maps are validated by records on absence and presence of species of these ecological groups.
Introduction
In the Netherlands interventions in the surface water and ground water system have affected water quality and water budget, to the detriment of the plant species composition of marsh land ecosystems The main activities are ground water extraction for industrial and drinking water supply and lowering of the surface water levels for agriculture in winter and spring. Water shortages in summer and late spring are compensated by intake of nutrient rich surface water from a nearby river system. This water infiltrates and causes ground water quality to change to nutrient rich water types (Wassen 1990, Barendregt 1993, Schot, 1991).
The project aims to select scenarios, by means of a landscape ecological model, for sustainable use of land and water including surviving wetland ecosystems. An integrated landscape ecological has been build to study the combined effect of interferences in this aquatic and phreatic system on the quality of marsh ecosystems. The model integrates the biotic and abiotic component and the spatial and temporal aspects of the system and focuses on water and nutrient transfers by process-oriented modeling of ground water and surface water flows. Environmental site conditions are computed and taken as input variables for the vegetation response model. Because the landscape ecological model must work in an spatial manner, all input and output variables are processed in a GIS but before this research the vegetation response model was not spatially dimensioned. This study describes the implementation of this model (Barendregt, 1994) in a GIS (PCRaster )(Wesseling en van Deursen 1994 ) and show some results and applications. The study area was chosen in the Vecht streek on its availability of relevant data from earlier research.
Area description
The investigated area (Fig 1) in the central part of the Netherlands contains a polder area with a soil of clay and peat and adjacent a sandy glacial ridge. The polders lie between +1.0 and -3.5 above mean sea level (a.m.s.l.). Surface water levels are completely controlled on fixed winter levels and lower summer levels. The land use is predominately agricultural, mainly dairy cattle breeding, nevertheless substantial parts are natural reserves with lakes, broads, marshes and peat land. The glacial ridge ( +1 to + 30 a.m.s.l.) has coarse sandy soils and gravels. The dominant landuse is forest (both decidious and coniferous), heathland and towns (population 50.000 - 75.000). Several groundwater pumping stations are situated here.
Data
The vegetation response model was developed on the basis of 906 record of vegetation and environmental conditions in the investigation area (Barendregt & Wassen ,1989). The model is based on gaussian logistic regression technique and predicts a response value related to the values of 18 different environmental variables per plant species. This particular model calculates the response values of 102 plant species and uses information on soil texture, management, phreatic conditions and ground water quality. Data are used from the period 1985 - 1990. The ground water quality data (Fig 2 and (Fig 3) are processed into maps by means of spatial interpolation resulting in input maps (Fig 4 and Fig 5) for ground water quality with estimated block mean averages of 500 x 500 x 10 meter blocks. Data for validation are taken from independent environmental inventories. The model has been validated using a resolution of 1 x 1 km.
Results
The model yields species maps (Fig 6) which show the response value per plant species to the environmental conditions. For the presence or absence of a species a cut-off level of 0.5 * the maximum calculated response value is taken. Is a response below this level, a plant species is considered to be not present in the cell, if above the species is considered to be present in the cell. Species are classified according to the ecological groups of (Den Held c.s.1992). The three distinctive properties are : (Appendix 1)
The groups of species compositions are ordered in decreasing chemical loads concentrations. So the A B C E group species are generally related to nutrient rich brackisch/groundwater like water types, while J K L group species are related to nutrient poor rainwaterlike water types. The species of the D F G species group are composed of species of both intermediate nutrient and mixed rainwater - groundwater like water type. Species of group j, f and c are respectively the most found species groups. (Fig 9)
Validation
All valid estimates, that is the found versus calculated and the not found versus not calculated cases, are summed in a valid map ( Fig 10 ) while all the non valid cases are summed in a non valid map (Fig 11). The pattern of the valid map shows a small part of the area in the southwest and south with valid predictions > 80 cases. Most of the area shows valid predictions between 40 and 60 cases,while an area with relatively few valid predictions ( < 20 ) is found in the west central part of the area. The map with the non valid cases shows (logically) the complementary pattern of the map with valid cases.
Discussion & Conclusions
A vegetation response model has been integrated in a GIS and the results in coherent and valid patterns at the landscape scale. The validity of the model was tested with input variables which are estimated from measurement by means of interpolation. Errors caused by interpolation of the input data and in the actual regression model affect the results. In the final model, where the input data are generated by process models, the error in the results is expected to be lower (this remains to be investigated). This constructed spatial vegetation response model can be used as a tool to asses the impact on wetland ecosystem.
Classification of plant species (Den Held,1992)
Literature
Barendregt A. (1993) Hydro-ecology of the Dutch polder landscape. thesis, Faculty Geographical Studies, University of Utrecht, Utrecht, pp. 200.
Barendregt, A. en Wassen, M.J. (1989) Het hydro-ecologisch model ichors (versie 2.0 en 3.0). report Department of Environmental Studies, University of Utrecht, pp. 72.
Den Held,A.J.,Schmitz, M. & Wirdum van G. (1992) Types of terestrializing fen vegetation in the Netherlands in: Verhoeven, J.T.A. (eds) Fens and Bogs in the Netherlands Vegetation dynamics, history, nutrientdynamics and conservation, pp 237 - 321. Kluwer Academic Press. The Netherlands.
Schot, P.P(1991) Solute transport by groundwater flow to wetland ecosystems. thesis, Faculty of Geographical Studies, University of Utrecht, Utrecht. pp 134.
Wassen, M. (1990) Water flow as a major landscape ecological factor in fen development. thesis, Faculty Geographical Studies, University of Utrecht, Utrecht, pp. 197.
Wesseling, C. and van Deursen, W. (1993) The PCRaster package. Department of Physical Geography, University of Utrecht.
Peter van Horssen
Department of Environmental Studies
University of Utrecht, Utrecht
The Netherlands