IEEE Geoscience and Remote Sensing Magazine - March 2023 - 113

scale, there lies a typical representation of a semantic
segmentation task in which the learner approximates
the given annotation (the second sample in Figure 6).
While doing so, the learner struggles with the borders
of the given wilderness area, which can be explained by
1) the edge-effect [31], which complicates the segregation
of ecological units from the previous perspective,
and 2) the imperfect annotations used when training
the learner.
Later, the learner has been further investigated by employing
an interpretable-by-design architecture to study
the patterns in the decision-making process. The high-resolution
sensitivity maps in Figure 9 make it evident that
the learner holds a deeper understanding of wilderness
that disentangles wilderness from human influence. The
maps produced in this experiment provide pixel-level sensitivity
information that could be utilized in the process of
inferring new scientific insights.
Although the behavior of the learner can often be explained
away with some confidence, in the light of the
experimental results explained previously, it is still unclear
when and why the learner behaves in certain ways.
Motivated by this observation, we urge environmental
science, conservation, computer science, and RS researchers
to study the ambiguity in the ill-defined elements of
nature to better monitor, understand, and protect nature,
our home.
SUMMARY AND CONCLUSION
With this article, we 1) introduce a novel task of wilderness
mapping and 2) publish MapInWild, a large-scale benchmark
dataset curated for the task of wilderness mapping.
MapInWild is a multimodal dataset and comprises various
geodata acquired and formed from a diverse set of RS
sensors. The dataset consists of 8,144 images with a shape
of 1,920 × 1,920 pixels and is approximately 350 GB in
size. The images are weakly annotated with three classes
derived from the WDPA: strict nature reserves, wilderness
areas, and national parks. With MapInWild, for the
purpose of deepening our understanding of what makes
nature wild, we embark on the complications induced by
the ambiguity of the term wilderness and study the vagueness
in nature and propose our dataset as a test bed for
ML research concerning environmental RS. We are convinced
that getting closer to understanding the concept
of wilderness is of great value to the community to further
bridge the gap between DL applied to environmental
RS and conservation. Both the MapInWild dataset and
the code are publicly available at https://dataverse. harvard.
edu/dataverse/mapinwild and https://github.com/burakekim/
MapInWild.
ACKNOWLEDGMENT
This work was supported by the German Research Foundation
(DFG project MapInWild, Grant RO 4839/5-1/SCHM
3322/4-1). The authors acknowledge the computing time
MARCH 2023 IEEE GEOSCIENCE AND REMOTE SENSING MAGAZINE
granted by the Institute for Distributed Intelligent Systems
and provided on the GPU cluster Monacum One at the University
of the Bundeswehr Munich.
AUTHOR INFORMATION
Burak Ekim (burak.ekim@unibw.de) is with the Institute of
Space Technology and Space Applications, Department of
Aerospace Engineering, University of the Bundeswehr Munich,
85577 Neubiberg, Germany.
Timo T. Stomberg (timo.stomberg@uni-bonn.de) is with
the Institute of Geodesy and Geoinformation, University of
Bonn, 53115 Bonn, Germany.
Ribana Roscher (ribana.roscher@uni-bonn.de) is with
the Institute of Geodesy and Geoinformation, University of
Bonn, 53115 Bonn, Germany.
Michael Schmitt (michael.schmitt@unibw.de) is with
the Institute of Space Technology and Space Applications,
Department of Aerospace Engineering, University of the
Bundeswehr Munich, 85577 Neubiberg, Germany.
REFERENCES
[1] X. X. Zhu et al., " Deep learning in remote sensing: A comprehensive
review and list of resources, " IEEE Geosci. Remote
Sens. Mag., vol. 5, no. 4, pp. 8-36, Dec. 2017, doi: 10.1109/
MGRS.2017.2762307.
[2] M. Schmitt, S. A. Ahmadi, and R. Hänsch, " There is no data
like more data - Current status of machine learning datasets
in remote sensing, " in Proc. IEEE Int. Geosci. Remote Sens.
Symp.
igarss47720.2021.9555129.
[3] G. Sumbul et al., " BigEarthNet-MM: A large-scale, multimodal,
multilabel benchmark archive for remote sensing image classification
and retrieval [Software and Data Sets], " IEEE Geosci.
Remote Sens. Mag., vol. 9, no. 3, pp. 174-180, Sep. 2021, doi:
10.1109/MGRS.2021.3089174.
[4] M. Schmitt, L. H. Hughes, C. Qiu, and X. X. Zhu, " SEN12MS:
A curated dataset of georeferenced multi-spectral sentinel-1/2
imagery for deep learning and data fusion, " in Proc. ISPRS
Ann. Photogrammetry, Remote Sens. Spatial Inf. Sci., Sep. 2019,
vol. IV-2/W7, pp. 153-160, doi: 10.5194/isprs-annals-IV-2
-W7-153-2019.
[5] X. X. Zhu et al., " So2Sat LCZ42: A benchmark data set for the
classification of global local climate zones [Software and Data
Sets], " IEEE Geosci. Remote Sens. Mag., vol. 8, no. 3, pp. 76-89,
Sep. 2020, doi: 10.1109/MGRS.2020.2964708.
[6] X. Sun et al., " FAIR1M: A benchmark dataset for fine-grained
object recognition in high-resolution remote sensing imagery, "
ISPRS J. Photogrammetry Remote Sens., vol. 184, pp. 116-130, Feb.
2022, doi: 10.1016/j.isprsjprs.2021.12.004.
[7] D. Silvestro, S. Goria, T. Sterner, and A. Antonelli, " Improving
biodiversity protection through artificial intelligence, " Nature
Sustain., vol. 5, no. 5, pp. 415-424, Mar. 24, 2022, doi: 10.1038/
s41893-022-00851-6.
[8] C. Persello et al., " Deep learning and earth observation to support
the sustainable development goals: Current approaches,
open challenges, and future opportunities, " IEEE Geosci. Remote
113
IGARSS, Jul. 11, 2021, pp. 1206-1209, doi: 10.1109/
https://dataverse.harvard.edu/dataverse/mapinwild https://dataverse.harvard.edu/dataverse/mapinwild https://github.com/burakekim/MapInWild https://github.com/burakekim/MapInWild
IEEE Geoscience and Remote Sensing Magazine - March 2023

Table of Contents for the Digital Edition of IEEE Geoscience and Remote Sensing Magazine - March 2023
