Error propagation of open global land cover maps on automatic fluvial ecosystem services extraction.
MetadataShow full item record
Fluvial ecosystem services (FES) as a subset of all ecosystem services (ES), in their healthy state provide range of goods and services for people and environment (Brian et al., 2003). The interaction of natural factors, such as flow pattern, sediment and organic matter inputs, temperature and light characteristics, chemical and nutrient conditions, and the plant and animal assemblage, in space and time define the dynamic nature of freshwater ecosystem services (Baron et al., 2003). However, the treat to freshwater ecosystems by humans and extensive use of their services made water ecosystems the most affected of habitats by historical changes (MEA, 2005). For better river management ecosystem services need to be monitored , and their value implemented into decisions of environmental managers and planners (Daily et al., 1997). To reach this, a mapping method must be developed. The best option for the method is to use globally available public data on land cover with global coverage. With the use of these open data, the method can be applicable everywhere and especially in the areas with national data scarcity. However, for fluvial FES extraction there is lack of automated method that use globally available land cover data. The method can be developed, but how far will the error in classification in land cover map affect the FES extraction results? And this is the main objective of this study. To reach the main objective three steps were executed: to assess the error of OSM and GlobeLand30 in the field, to develop an automated model for fluvial features and FES extraction and to propagate the error in the landcover map towards longitudinal profiles of FES. The error of OpenStreetMap (OSM) and GlobeLand30 was assessed in the field by collecting the observation points and comparing them to the same location on both maps. The model for fluvial features extraction and their FES scores was semi-automatic. For the model river characteristics and land cover are required and were derived from the public data. The basics of the mapping method, such as the linking cascade of fluvial features to ecosystem services and scoring rules were retrieved from the earlier study by Large and Gilvear (2015). The study area to apply the method on, a 105 km long section of the Desna River in Ukraine, was selected for its naturalness and because no detailed data are available for this area. The input data consisted of two 30 m resolution global datasets, the SRTM 1-Arc Second digital elevation model(DEM) and GlobeLand30 land cover map, and the OSM vector data. Using map algebra on combinations of the input data, 18 features were extracted that were subsequently converted to scores on 13 separate FES and to their main categories: Provisioning, Supporting and Regulating. To focus on the fluvial ecosystem services of the Desna River, the spatial resolution of the FES scores was 5 km sections in longitudinal direction. Laterally the sections were constrained by the height above the main drainage. To error propagation consisted of two steps. First, was to apply the error matrix of GlobeLand30, to generate new land cover maps with the new accuracy. Thereafter, the developed model was run with new input maps creating new outputs. The accuracy of GlobeLand30 was 52.3%, which is low, but higher than the accuracy of 39.6% for OSM map. Generally the results in longitudinal profiles and score graphs showed high scores for features to deliver an ecosystem service. The highest scores of FES were in the area with the highest sinuosity of the river combined with large area of forest. The first effect of error, when error matrix was used to generate new maps, was on spatial distribution of the classified units of the original GlobeLand30 maps and on the FES scores internal variation. The decreased accuracy of the input land cover maps, and the resulting loss of spatial autocorrelation of the features led to decrease in the FES scores compared to the original input map with an average average maximum score of 4.0. However, the same score distribution pattern as the original input result was preserved. Based on the statistical results, the answer on the main objective was that the effect of the error of almost 50% in GlobeLand30 maps is only slightly affecting the FES scores. Thus, the proposed method is a good foundation for further research on fluvial ecosystem services. However, there are limitations in the model due to its dependence on public data quality and its preserved simplicity. But, virtual globes are growing and public data quality is improving what will lead to more accurate results of the model in the future. And manually executed adjustments (the simplifications for study area) in the model can be reduced, making it more automatic. The improvement of the mapping method by future users, by making the model more automated or applicable on historical maps and different rivers combined with its public accessibility, will implement more knowledge for people, environmental managers and planners about the ecosystems and help them to make more informed decisions.