Biodiv Sci ›› 2018, Vol. 26 ›› Issue (8): 892-904.  DOI: 10.17520/biods.2018039

Previous Articles     Next Articles

An analysis of lightweight-drone-assisted mapping accuracy in tropical forest plot

Deng Yun1,3,4,*(), Wang Bin2, Li Qiang2, Zhang Zhiming2, Deng Xiaobao1,3, Cao Min1, Lin Luxiang1,3   

  1. 1 CAS Key Laboratory of Tropical Forest Ecology, Xishuangbanna Tropical Botanical Garden, Chinese Academy of Sciences, Mengla, Yunnan 666303
    2 School of Ecology and Environmental Sciences, Yunnan University, Kunming 650091
    3 National Forest Ecosystem Research Station at Xishuangbanna, Xishuangbanna Tropical Botanical Garden, Chinese Academy of Sciences, Mengla, Yunnan 666303
    4 University of Chinese Academy of Sciences, Beijing 100049
  • Received:2018-02-06 Accepted:2018-08-05 Online:2018-08-20 Published:2018-09-27
  • Contact: Deng Yun

Abstract:

Accurate coordinate position is a prerequisite for combining drone-assisted remotely sensed data and ground survey data. However, in the practice of surveying forests, many factors prevent accurate measurement of coordinate position and inaccurate coordinates may lead to incorrect conclusions. Therefore, researchers must pay attention to factors effecting accuracy of position. In this study, we compared location error of ground control points (GCPs), model error of photogrammetric point cloud (estimated by Photoscan software) and reprojection error of camera exposure position. First, we found that real time kinematic (RTK) global navigation satellite system (GNSS) cannot locate position in tropical forest with high accuracy. The root mean square error (RMSE) of GCPs in canopy gaps were 0.167 ± 0.158 m and 0.297 ± 0.170 m in the horizontal and vertical axes respectively. In comparison, RMSE of GCPs within forests were 0.392 ± 0.368 m and 0.657 ± 0.412 m respectively for horizontal and vertical axes. Second, the number and measurement accuracy of GCPs influenced model error of photogrammetric point cloud. Third, reprojection error of camera exposure position (18.434 ± 5.252 m and 34.042 ± 6.920 m in horizontal and vertical axes respectively) was much greater than location error of GCPs when the drone acquired position with a single-station GPS system. Fourth, standard deviation of difference between estimated digital terrain model (DTMestimated) and measured digital terrain model (DTMmeasured) was positively correlated with mean canopy height (r = 0.713, P < 0.05). DTMestimated was better estimated at 20 ha scale than at 1 ha scale. Based on these results, we suggest that uniform distribution and sufficient numbers of GCPs can improve drone-assisted mapping accuracy. Lightweight-drone-based photogrammetry has an advantage in requiring fewer equipment and enabling creation of accurate DSM (digital surface model), but remains incapable of estimating ground elevation. Researchers should consider these factors related to accuracy before using drones for surveys.

Key words: lightweight drone, global navigation satellite system, location accuracy, forest plot