Hej everyone, my name is Melanie Stammler and I have recently worked with Trimble eCognition when semi-automatically mapping Northern Swedish aeolian sand dunes for my MSc. thesis. Now, it is my honor to contribute my experience to this blog as a guest author.
Indeed, I would be very surprised if aeolian sand dunes represented your first association with Arctic Northern Sweden, more precisely the area where Norway, Sweden and Finland meet. However, parabolic croissant-shaped aeolian sand dunes represent wonderful features in this area and have probably been in existence since the end of the last ice age. Putting the surprise aside, it is important to acknowledge their important value in understanding landscape development of this highly sensitive Arctic region. While their locations reflect (previous) sand sources and suggest reasons for sand availability, their orientation indicates wind direction during genesis and illustrates potential contrast to today’s wind patterns. Their geomorphometry allows us to derive additional hypotheses, for instance about their state of activity. In short, Northern Swedish aeolian sand dunes are what you are looking for when gauging the impact of climate change on an area that is today experiencing some of the fastest rates of change globally.
In my MSc. thesis, I substantiate dune characteristics relevant for landscape evolution analysis by evaluating geomorphological concepts and having a look through the system science lens. Based thereon, I deduce a substantial need to systematically and spatially analyze aeolian sand dunes in Northern Sweden. Focus lies on accurately mapping these features to determine their location, orientation and geomorphometric characteristics. Geomorphological expert-decision based mapping has a long tradition in the discipline. While it has many advantages, there are costs, such as time intensity, limited reproducibility as well as reduced transparency of the process. As many of us will be aware, geographic object-based image analysis allows for creating reproducible mapping boundaries while treating the target forms as objects. Further, the transferability of rule-sets that make increasing spatial scale mainly a question of processing power represents yet another benefit of the technique. Thus, I concluded GEOBIA to likely be a better approach for my endeavor of mapping aeolian sand dunes in Arctic Northern Sweden.
The real world is continuous. Isolated structures are therefore subjective and artificial portions of reality, and the biggest initial problem is the identification and separation of meaningful sections of the real world” Chorley and Kennedy (1971 : 1)
As mentioned in the introduction, the study area I chose for my analysis is located far above the Arctic circle. It comprises a comparatively flat landscape that gains in topography towards the Norwegian border. Due to its glacial history, it is characterized by drumlins, eskers and outwash plains. These features are well-represented in a high resolution digital elevation model (DEM) made available by the Swedish mapping agency, Lantmäteriet. Further, Sentinel-2 imagery and high resolution orthophotographs provide a good overview of the landscape.
Let’s get to the method insight: The powerful high resolution DEM unlocks the potential to use a residual-relief separated DEM along with derived slope and curvature data as input to the eCognition Developer 9.5 multi-resolution segmentation. Additionally, I tested the suitability of aspect data and the above-mentioned optical imagery as input. In a rule-based approach, the image segments are circled through an alternation of classification and refinement. Discrimination by mean slope and curvature values is followed by merging the respective image objects to make sure no features are split into parts. I then make use of the shape parameters available in eCognition Developer 9.5 to discriminate further. The parameter settings are determined using point datasets published for a selection of aeolian sand dunes in Northern Sweden.
The classification scheme accepts on average 2.5 % of the segmented image objects as potential dune sites. Subsequent expert-decision in 17 test areas confirms on average 25 % of the classified image objects as identified sand dune locations. Regarding input data, the residual-relief separated DEM along with slope and curvature datasets produces best results. This might yield yet another surprise but can be easily explained by vegetation cover strongly reducing the dunes spectral characteristics’ significance. When opposing the expert-accepted and the classified image objects to all segmented image objects, the slope dataset shows most distinct contrast. As described in many papers, I also face a challenge due to implementing fixed boundaries: The rule-based classification provides best results when targeting a smaller area in comparison to the entire study area as the fixed boundaries cannot adequately mimic the variability within the dune characteristics.
Now, I want to take you to the geomorphology side (we may have cookies, too) and present the potential inherent in interpreting the polygon data created in this thesis. Also, being a geographer writing this blogpost without including a map would have gone against my nature.
Most importantly, you can perceive the GEOBIA derived dune polygons in dark blue. In red, you see arrows indicating the dune’s orientation. Among others, glaciofluvial deposits such as eskers are visualized in green, (drumlinised) till in light blue and former glaciofluvial channels with black arrows. Having gained knowledge on the dunes’ location and orientation I continue with investigating sand sources (spoiler alert: esker systems) and reasons for sand availability (potentially the disturbance by glaciofluvial river systems). The polygon dataset further allows determination of a predominantly parabolic shape, a predominant south-east pointing orientation indicating northwesterly winds during formation and a coexistence of single with coalesced dunes. Analyzing the geomorphometry of the dune polygons reveals a bimodal slope distribution representing windward and lee slopes. Differences in fragmentation, also perceivable from the polygon data, represent variability in the features state of erosion. Not only for dating experts: if these differences represent age variability, they exemplify one benefit object-based image analysis can yield to an increased understanding of landscape development.
Without going into to much detail about the geomorphological interpretation of aeolian sand dunes in Northern Sweden, I want to highlight the door-opening role of GEOBIA to this kind of important analysis. The transferability of developed rulesets, the transparent and reproducible mapping boundaries, the opportunity to fuse different sources of data in eCognition Developer and the applicability over larger areas are from my point of view the most crucial benefits that helped me to conduct my study. They allowed me to answer my research question with ‘yes, adequately mapped aeolian sand dunes provide insight in landscape development and contribute to a better understanding of long-term landscape (in)stability in Northern Sweden’.
Thank you for virtually letting me take you to the Arctic! I hope that with this blogpost I could showcase how using Trimble eCognition software helps unravel the location, orientation, and geomorphometry of aeolian sand dunes above the Arctic circle. Exploring GEOBIA on sand dunes was quite an adventure to me, and I am more than happy to hear about your thoughts and associations with this blogpost, or even my MSc. thesis in general.
Further, I want to thank Keith Peterson for offering this blogpost opportunity, the entire eCognition team for supporting me throughout my thesis and my MSc. thesis supervisor Thomas Stevens at Uppsala University for his tremendous support and for helping me navigate this spectacular journey.
Source for the quote: Chorley, R.J., Kennedy, B.A., 1971. Physical Geography: a systems approach. Prentice Hall, London