IEEE Geoscience and Remote Sensing Magazine - June 2022 - 190

different scientific (and nonscientific) communities are often
a challenge for an effective interdisciplinary approach.
FUTURE OPPORTUNITIES
In this article, we recognize the importance of the availability
of data and computational facilities for the success
of DL models. Developments in this direction are offering
new opportunities to the EO community. For half a century,
Earth has been under continuous observation by satellites
to monitor and understand environmental processes.
However, historically, RS data were foremost available to
those governmental agencies, research institutes, and commercial
companies that had direct access to the corresponding
sensors. A mixture of different political, organizational,
and legal reasons made a free distribution of acquired data
difficult to impossible, resulting in a limited number of
mostly small data sets [278]. The traditional approach to
developing and testing new methods on small and local
data still prevails today. In particular, in the context of DL,
this is problematic, as approaches are evaluated on data sets
(often consisting of only a single small image as, e.g., the
HSI data set Indian Pines) that do not provide a sufficient
amount of independent test samples.
BIG AND OPEN GEODATA
Fortunately, during the last years, large parts of the community
have been moving away from closed data and embraced
open science principles such as findable, accessible,
interoperable, and reusable [279] and free and open source
software [280]. These developments enable transparent
and reproducible scientific research, allow the distribution
and reuse of data and methods, and lead to the more efficient
creation of new data products. This led to open code
libraries [such as Open RS (http://openremotesensing.
net) and the IEEE RS Code Library (https://rscl-grss.org/),
public evaluation servers [such as the IEEE Geoscience
and Remote Sensing Society Data and Algorithm Standard
Evaluation website (http://dase.grss-ieee.org)], and modern
benchmark data sets for a multitude of combinations of EO
sensors and tasks, which are rapidly replacing older smallscale
data sets. These are more in line with the actual situation
in RS, as current EO data have passed the petabyte
scale and pose common big data challenges regarding volume
(i.e., the amount of data), velocity (that is, the temporal
pace at which new data are acquired), and variety (the
heterogeneity regarding image acquisition, such as sensors
types and modes as well as environmental factors) [281].
CLOUD COMPUTING INFRASTRUCTURES
New opportunities are also arising from the availability of
cloud computing infrastructures that allow the visualization
and analysis of large-scale data [e.g., Landsat [282]
and Sentinel data (https://scihub.copernicus.eu/)] directly
in the cloud without the need for local downloading, storing,
and processing. Examples include Digital Earth Australia
[283], Earth System Data Lab [284], Swiss Data Cube
190
[285], Copernicus Data and Information Access Services,
and Google Earth Engine [239]. The availability of large
and open data sets in combination with powerful computing
infrastructures sets the premise for researchers to work
more cohesively on addressing the environmental and societal
challenges of our time.
A GLOBAL PICTURE OF WORLDWIDE PHENOMENA
Finally, we want to remark that the combination of DL and
EO offers the opportunity to obtain a truly global picture of
environmental and societal phenomena that go beyond national
boundaries as opposed to the data typically collected
by national statistical agencies. As discussed in [240], the
national-level reporting structure of the SDGs limits the ability
to capture environmental phenomena that cross national
borders. Moreover, differences in the data collection practices
of the national offices often result in inconsistent data. EO
can provide spatially and temporally consistent data, while
DL offers the tools to extract semantic information in an objective
and reproducible manner. We, thus, advocate for the
use of DL and EO to monitor progress toward the SDGs and
encourage the geoscience and RS communities to play active
roles in discussions with stakeholders and policy makers.
CONCLUSION
We have reviewed the latest developments in the context
of DL for EO and a large number of applications that contribute
to the UN agenda for sustainable development.
The combination of DL and EO appears to be a strategic
asset that can play an essential role in addressing many
of the challenges raised by the UN agenda and, beyond
that, some of the most urgent demands of human societies.
Understanding the role of DL in EO for extracting
nationwide geospatial statistical data has far-reaching
societal implications for policy development and decision
making. Going beyond the SDG agenda, DL and EO
can play significant roles in other international agendas,
such as the New Urban Agenda (https://habitat3.org/
the-new-urban-agenda) or Sendai framework for disaster
risk reduction.
ACKNOWLEDGMENTS
GCV was partly supported by the European Research
Council Synergy Grant " Understanding and Modelling the
Earth System with Machine Learning (USMILE) " under the
Horizon 2020 research and innovation programme (grant
855187), and by the EU H2020 2.1.6.3. project " DeepCube:
Explainable AI pipelines for big Copernicus data " (grant
101004188). We thank Dr. Manuel Campos-Taberner from
Universitat of València for reproducing Figure 4 and discussions
about the role and opportunities of deep learning for
land cover classification.
AUTHOR INFORMATION
Claudio Persello (c.persello@utwente.nl) is an associate
professor with the Faculty of Geo-Information Science and
IEEE GEOSCIENCE AND REMOTE SENSING MAGAZINE JUNE 2022
http://openremotesensing.net http://openremotesensing.net https://www.rscl-grss.org/ https://www.habitat3.org/the-new-urban-agenda http://dase.grss-ieee.org https://www.habitat3.org/the-new-urban-agenda https://scihub.copernicus.eu/
IEEE Geoscience and Remote Sensing Magazine - June 2022

Table of Contents for the Digital Edition of IEEE Geoscience and Remote Sensing Magazine - June 2022
