IEEE Geoscience and Remote Sensing Magazine - June 2021 - 119

TECHNICAL COMMITTEES
MANIL MASKEY
Advancing Machine Learning
Tools for Earth Science Workshop
T
he IEEE Geoscience and Remote Sensing Society
(GRSS) Earth Science Informatics (ESI) Technical
Committee (TC) cosponsored a workshop,
" Advancing Machine Learning Tools for Earth Science, "
in 2020. The objective was to bridge the gap between
academics, government, and industry and discuss
how to best utilize artificial intelligence (AI) and
machine learning (ML) techniques with Earth observation
data to address environmental challenges. One of
the main functions of the ESI TC is to advance applications
of informatics to Earth science. The NASA Earth
Science Data Systems (ESDS) program shares similar
goals for maximizing the use of Earth science data in
disruptive technologies, such as AI and ML, for societal
benefits. Thus, ESDS and the ESI TC organized this
workshop with shared interests and aims.
DATA SYSTEM PERSPECTIVE
The ESDS program manages the complete lifecycle of
Earth science data, including a generic and repeatable
process of flowing information through several stages,
from acquisition to distribution and archiving
(Figure 1). There are opportunities to leverage AI
within each stage of the data lifecycle, and they can
be summarized, at a high level, into five categories:
1) improving data utilization, 2) automating operations,
3) improving service levels, 4) providing innovative
services, and 5) utilizing existing information
stewardship skills to curate labeled data. To
take advantage of these opportunities, well-curated,
large-scale training data sets need to be constructed.
Additionally, the Earth science community will reap
the full benefits of AI when training data and models
are benchmarked, standardized, reusable, interoperable,
and accessible. Investments are needed
Digital Object Identifier 10.1109/MGRS.2021.3059525
Date of current version: 17 June 2021
JUNE 2021 IEEE GEOSCIENCE AND REMOTE SENSING MAGAZINE
to make Earth science information more tractable for
AI algorithms.
A lack of such training data continues to be the
main bottleneck within Earth science [1]. ML requires
large amounts of human effort to label the training
data necessary for supervised learning. Creating
training data sets and developing new ML models
is laborious and time-consuming. Guided activities
to build and share training data sets at a scale for ML
algorithms are needed. These large training data sets
need to be open, discoverable, and shareable to reduce
redundant efforts and encourage data reuse. In
addition, training data sets should conform to accepted
standards for interoperability. Similar standards
are needed for AI models to promote knowledge sharing
among users.
WORKSHOP STRUCTURE
The workshop organizers identified two major goals.
The first was to define scientific requirements for training
data catalogs, with associated topics that included
the following:
◗ defining and measuring uncertainty in training data
◗ treating sparsity in training data
119
IEEE Geoscience and Remote Sensing Magazine - June 2021

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