IEEE Geoscience and Remote Sensing Magazine - June 2021 - 88

PERSPECTIVES
DEVIS TUIA, RIBANA ROSCHER, JAN DIRK WEGNER, NATHAN JACOBS,
XIAO XIANG ZHU, AND GUSTAU CAMPS-VALLS
Toward a Collective Agenda on AI
for Earth Science Data Analysis
I
n past years, we have witnessed the fields of geosciences
and remote sensing and artificial intelligence
(AI) become closer. Thanks to the massive availability
of observational data, improved simulations, and algorithmic
advances, these disciplines have found common
objectives and challenges to help advance the modeling
and understanding of the Earth system. Despite such
great opportunities, we have also observed a worrisome
tendency to remain in disciplinary comfort zones, applying
recent advances from AI on well-resolved remote
sensing problems. Here, we take a position on the research
directions for which we think the interface between
these fields will have the most significant impact
and become potential game changers. In our declared
agenda for AI in Earth sciences, we aim to inspire researchers,
especially the younger generations, to tackle
these challenges for a real advance of remote sensing
and the geosciences.
DATA-DRIVEN APPROACHES IN THE
GEOSCIENCES, THEIR PITFALLS AND
OPPORTUNITIES
AI promises to change the way we do science. Today
it is widely accepted, and almost a mantra, that
data, along with faster computers and advanced machine
learning (ML) algorithms, can solve any data
science problem. Approaches from ML, computer vision,
applied mathematics, or big data,
in general,
are undoubtedly revolutionizing the way we tackle
challenges in remote sensing and geoscience. This is
particularly visible as deep learning has entered the
arena [1]: the promise of a technology that can process
large amounts of data and learn the complex
structures of environmental processes, thus leading
to improved modeling, is making ML an unavoidable
Digital Object Identifier 10.1109/MGRS.2020.3043504
Date of current version: 17 June 2021
88
approach. For a multidisciplinary overview, see the
work of Camps-Valls et al. [2].
We are convinced that the rise of data-driven approaches
in the geosciences is beneficial and will lead to
important discoveries. However, we want to raise awareness
of pitfalls and the fundamental questions that are
currently understudied by the community.
The first risk is that of seeing everything as an
opportunity to apply ML and then deploy massive
technology regardless of whether such technology is
necessary and adapted to the problem at hand. Were
that to become common practice, one would miss the
opportunity to use (and improve) such new technology
to tackle new challenges that could not be solved
without it. For example, remote sensing is a data-hungry
discipline that embraced ML very early on: after
an exploratory phase [3]-[5], we now see the need for
an impulse to embrace this technology to unlock new,
difficult problems that will, in turn, create value for
these geospatial data. Some of the concepts presented
in this article illustrate these directions (see Table 1).
First is the question of introducing reasoning in the algorithms
as a way to mimic cognitive processes about
space and time (direction 1 in Table 1). Second is the
need for exploring unconventional data modalities to
capture the complexity of the visual world (direction 2
in Table 1). And third are the new methods of human
interaction with remote sensing models, for instance,
via question answering as a way to retrieve image content
on demand (direction 3 in Table 1).
A second pitfall is the blind faith in data science.
Data science has, regrettably, been a misleading term.
Should it be replaced with " data for science, " as some
researchers have suggested? Science is about contrasting
hypotheses, understanding physical phenomena,
and validating causal and explanatory models. If those
objectives were to be achieved through data analysis, a
new science would be born.
IEEE GEOSCIENCE AND REMOTE SENSING MAGAZINE JUNE 2021
IEEE Geoscience and Remote Sensing Magazine - June 2021

Table of Contents for the Digital Edition of IEEE Geoscience and Remote Sensing Magazine - June 2021
