Modelling water-related ecosystem services with InVEST - developing guidance on how to select appropriate land cover input data

Abstract

Modelling and mapping water-related ecosystem services (ES) are getting more important in the scientific community and decision making contexts. Especially straightforward, easy to operate ES model suites that are based on Land Use / Land Cover (LULC) data have gained popularity in the scientific realm, including but not limited to the InVEST Model Suits. Model sensitivity to input factors has been widely assessed. However, little attention has been given to the effect on modelled ES distribution by user decisions such as the selection of LULC dataset. These crucial input data influence model results and hence, validity and credibility of model outputs and maps. Yet, many model applications aim to support policy and decision making, without properly specifying uncertainties of their modelling and underlying data. Therefore, we investigated how to select appropriate and representative LULC data for model application. To test the effects of input LULC data on modelling results, we modelled the three regulating ecosystem services of water erosion control, water quality and water flow retention using InVEST. Different input LULC datasets were used to analyse how these datasets affect the modelling and mapping of ES supply. Taking a case study on Terceira Island, the Azores (Portugal), 3 LULC datasets were applied: (1) the EU-wide CORINE LULC (2018), (2) the Azores Region official LULC map (COS.A 2018) and (3) a remote sensing-based vegetation map using Sentinel-2 satellite imagery (2018). Output maps were compared by statistical analysis of class for distribution and similarity and visualized in similarity maps, showing the spatial variability between the three input LULC model results. Model results show significant differences in distribution of water-related ES based on the different input LULCs. For the ES erosion control (Sediment Delivery Module), spatial distribution of modelled output maps differed greatly. Large homogenous agricultural areas in LULC datasets, in combination with steep slopes, present a skewed picture of erosion rates, simplifying the small patch structure with hedges and stone rows found on Terceira island. The modelling of Water quality, based on Nutrient Export Module, and flow retention, based on the Seasonal Water Yield Module, showed a more balanced and similar, yet significantly different spatial pattern of ES supply. Therefore, we developed a guiding scheme to help researchers and practitioners select appropriate input LULC data for their ES modelling. Hereby, the availability of different LULC data is a first criterion. Factors such as LULC classes, especially linked to aquatic, riparian and agricultural land uses determine the level of detail of water-related ES modelling. Also, the scale of the assessment should be reflected in the average feature size and Minimum Mapping Units of the LULC dataset. Especially for local model applications, availability of high resolution LULC data, including structural elements, is preferred to obtain precise results.