Monitoring the Population Development of Indicator Plants in High Nature Value Grassland Using Machine Learning and Drone Data

Abstract

The temporal monitoring of indicator plant species in high nature value grassland is crucial for nature conservation. However, traditional monitoring approaches are resource-intensive, straining limited funds and personnel. In this study, we demonstrate the capabilities of a repeated drone-based plant count for monitoring the population development of an indicator plant species (Dactylorhiza majalis (DM)) to address such challenges. We utilized multispectral very high-spatial-resolution drone data from two consecutive flowering seasons for exploiting a Random Forest- and a Neural Network-based remote sensing plant count (RSPC) approach. In comparison to in situ data, Random Forest-based RSPC achieved a better performance than Neural Network-based RSPC. We observed an R² of 0.8 and 0.63 and a RMSE of 8.5 and 11.4 DM individuals/m², respectively. The accuracies indicate a comparable performance to conventional plant count surveys. In a change detection setup, we assessed the population development of DM and observed an overall decline in DM individuals in the study site. Regions with an increasing DM count were small and the increase relatively low in magnitude. Additionally, we documented the success of a manual seed transfer of DM to a previously uninhabited area within our study site. We conclude that repeated drone surveys are indeed suitable to monitor the population development of indicator plant species with a spectrally prominent flower color. They provide a unique spatio-temporal perspective to aid practical nature conservation and document conservation efforts.