Derivation of forest inventory parameters from high-resolution satellite imagery for the Thunkel area, Northern Mongolia. A comparative study on various satellite sensors and data analysis techniques.
by Holger Vogt
Date of Examination:2021-05-10
Date of issue:2021-06-03
Advisor:Prof. Dr. Martin Kappas
Referee:Prof. Dr. Martin Kappas
Referee:Dr. Dominik Seidel
Files in this item
EnglishWith the demise of the Soviet Union and the transition to a market economy starting in the 1990s, Mongolia has been experiencing dramatic changes resulting in social and economic disparities and an increasing strain on its natural resources. The situation is exacerbated by a changing climate, the erosion of forestry related administrative structures, and a lack of law enforcement activities. Mongolia’s forests have been afflicted with a dramatic increase in degradation due to human and natural impacts such as overexploitation and wildfire occurrences. In addition, forest management practices are far from being sustainable. In order to provide useful information on how to viably and effectively utilise the forest resources in the future, the gathering and analysis of forest related data is pivotal. Although a National Forest Inventory was conducted in 2016, very little reliable and scientifically substantiated information exists related to a regional or even local level. This lack of detailed information warranted a study performed in the Thunkel taiga area in 2017 in cooperation with the GIZ. In this context, we hypothesise that (i) tree species and composition can be identified utilising the aerial imagery, (ii) tree height can be extracted from the resulting canopy height model with accuracies commensurate with field survey measurements, and (iii) high-resolution satellite imagery is suitable for the extraction of tree species, the number of trees, and the upscaling of timber volume and basal area based on the spectral properties. The outcomes of this study illustrate quite clearly the potential of employing UAV imagery for tree height extraction (R2 of 0.9) as well as for species and crown diameter determination. However, in a few instances, the visual interpretation of the aerial photographs were determined to be superior to the computer-aided automatic extraction of forest attributes. In addition, imagery from various satellite sensors (e.g. Sentinel-2, RapidEye, WorldView-2) proved to be excellently suited for the delineation of burned areas and the assessment of tree vigour. Furthermore, recently developed sophisticated classifying approaches such as Support Vector Machines and Random Forest appear to be tailored for tree species discrimination (Overall Accuracy of 89%). Object-based classification approaches convey the impression to be highly suitable for very high-resolution imagery, however, at medium scale, pixel-based classifiers outperformed the former. It is also suggested that high radiometric resolution bears the potential to easily compensate for the lack of spatial detectability in the imagery. Quite surprising was the occurrence of dark taiga species in the riparian areas being beyond their natural habitat range. The presented results matrix and the interpretation key have been devised as a decision tool and/or a vademecum for practitioners. In consideration of future projects and to facilitate the improvement of the forest inventory database, the establishment of permanent sampling plots in the Mongolian taigas is strongly advised.
Keywords: remote sensing; Mongolia; forest inventory; Unmanned Aerial Vehicle