Towards standardised large-scale monitoring of peatland habitats through fine-scale drone-derived vegetation mapping

Abstract

Northern peatlands provide key climate regulating services by sequestering and storing atmospheric carbon as peat, but also provide habitat for specialised plant and animal species. Ecosystem-wide monitoring of the functions associated with these services is necessary to better inform policy and management and facilitate carbon financing schemes. Mapping peatland vegetation as an ecological indicator of their functions using drones has proven promising herein, yet the absence of standardised methods limits implementation.We developed drone-derived vegetation maps and compared them with two types of field-based ground-reference maps: 1) habitat distribution (ecotopes) and 2) habitat quality (status) for five Irish peatlands (35–124 ha). We also explored spatial transferability of our mapping approach across peatlands. First, orthomosaics and digital terrain models (DTM) were derived from drone imagery, after which plant functional types and microforms were separately classified. Second, ecotope and status maps were classified using the proportions of the fine-scale vegetation and the range in DTM within 20x20m grid cells as input predictor variables.Drone-derived ecotope and status maps captured the overall vegetation zonation of the conventional maps well, with the least mismatches for the peatlands displaying clear concentric zonation. Classification performance ranged between 88% for status and 72% for ecotope maps between peatlands. The lower classification performance for ecotopes was partly an artifact from gridding the conventional polygon-shaped ground-reference maps. Further classification errors resulted from artificial landscape features, variable plant phenology, and inaccuracies in the detrended DTM data at peatland-scale. Spatial transferability of the mapping approach was limited. Particularly, using pooled ground-reference data for classification decreased model performance with 5% for status and 10% for ecotope maps, largely because microform and plant functional type proportions associated with peatland habitat classes in the conventional maps varied between peatlands.Our findings highlight that both fine-scale vegetation patterns and habitats can be classified consistently on the peatland-scale using drone-derived imagery products and machine learning classifications. Yet, status is currently mapped notably more accurately than ecotopes. Also, peatland-specific ground-reference data is required until the conventional vegetation classes are more standardised across a wider variety of peatlands and peatland types.