Less is more: Optimizing vegetation mapping in peatlands using unmanned aerial vehicles (UAVs)

Abstract

1. Northern peatlands are inaccessible wetlands that serve important ecosystem services to humans, including climate regulation by storing and sequestering carbon. Unmanned aerial vehicles or drones are ideal to map vegetation and associated functions in these ecosystems, but standardized methods to optimize efficiency (highest accuracy with lowest processing time) are lacking. 2. We collected high-resolution drone imagery at three different altitudes (20 m, 60 m, and 120 m) of two Irish peatlands contrasting in pattern complexity and evaluated to what extent classification accuracy of vegetation patterns (microforms and plant functional types) changed using different flight altitudes, minimum segment size and training/testing sample size. We also analysed the processing time of all classifications to find the most efficient combination of parameters. 3. Classification accuracy was consistently high (>90 %) and estimated areas of both patterns were uniform among all flight altitudes, independent of pattern complexity. Minimum segment size and training/testing sample size were also important parameters affecting the efficiency of classifications. Total processing time from imagery capture to final map was 19–22 times faster with drone imagery at 120 m altitude than at 20 m, and seven times faster than at 60 m. 4. Our findings suggest that flying at the maximum legal altitude of 120 m is the most efficient approach for landscape-scale mapping of vegetation in peatlands or other ecosystems with similar short vegetation structure. We conclude that flying higher is always more efficient as long as the pixel size of drone imagery remains under the pixel size of the pattern under investigation.

Keywords

• Peatlands
• Vegetation patterns
• unmanned aerial vehicles (UAVs)
• Heterogeneous landscapes
• Remote sensing
• Machine learning