2.3. Mapping of wetland vegetation types and wildfires

Fanourios-Nikolaos Sakellarakis

Tour du Valat

The mapping of vegetation types, or habitats, is necessary for the management and applied conservation of ecosystems, including wetlands. Reedbeds and wet meadows are two key elements of the wetland vegetation of the Lesser and Great Prespa lakes, forming very important habitats that support a wide range of ecosystem services. Thus, the distribution and composition of the lakeshore vegetation around Lesser Prespa was assessed as part of the Prespa LIFE Waterbirds project, aiming to guide vegetation management.

In order to map the current state of the vegetation – looking mainly at the distribution of vegetation types and the dominant species in the reedbeds – drone-derived orthophoto mosaics of fine-scale resolution, as well as Landsat satellite images, were used. These datasets were processed using traditional digitising methodologies, and then analysed using classification tree analysis. The classification of Landsat images was 89.5% accurate in the validation subset, and when the classification results (Figure 1, Map B) were further evaluated using ground-truthed data, the overall accuracy was found to be 92.2%. For the creation of the orthophoto mosaic, an unmanned aerial vehicle (UAV) with both RGB and NIR cameras and an original pixel size of 0.1 m x 0.1 m was used, allowing a precise view of the vegetation. Orthophoto mosaic classification tree analysis had an overall accuracy of 75.9% and 87% on the training and ground-truthed subsets respectively (Figure 1, Map A). By overlaying the two resulting maps (orthophoto and Landsat) we were able to identify and locate any misclassified areas (Figure 1, Map C).

Figure 1: Methodology for the creation of the final map: A) Landsat derived classification, B) Orthophoto mosaic derived classification, C) Comparison of Landsat and Orthophoto analysis with the misclassified areas depicted in red, and D) final map of the study area.

Modifications were made, based on expert judgment, fieldwork data and the fine-scale resolution of the mosaic, in order to produce a 100% accurate final map (Figure 1, Map D). As a result, the total surface area in the study area of stands dominated by Lesser bulrush (T. angustifolia) and Common reed (P. australis), the two main species comprising the reedbed around Lesser Prespa, was calculated to be 188.5 ha and 128.8 ha respectively.

In addition, wildfires are also mapped in the Lesser Prespa Lake reedbeds on an annual basis, as they are a key factor in reedbed disturbance. By analysing Sentinel-2 satellite images with remote sensing methodologies based on the DeltaBAIS2 index (Wilm et al., 2020), a workflow was created, ensuring quick and accurate mapping of the total surface area of burnt reedbed around the lake on a yearly basis (e.g. Map 1).

To summarise, drone-derived orthophoto mosaics have proved extremely useful for mapping the current state of reedbed vegetation in Prespa, and can serve as a base map for the monitoring of vegetation dynamics in the future. Furthermore, the use of satellite images in the context of wildfire monitoring is a powerful tool, allowing a better understanding of post-wildfire vegetation development.