IEEE Geoscience and Remote Sensing Magazine - December 2022 - 167

possible contribution to the final scenicness score. These
weights allow the model to find land use-related subtle
variations of scenicness around the mean score of the dataset.
From the weights one can see that, on average, urban
fabrics contribute negatively to landscape aesthetics, while
open spaces, nonagricultural vegetated areas, and inland
wetlands tend to contribute positively.
Summarizing this application, social media and crowdsourcing
tell us something that is generally not available
in traditional land prediction tasks: information about the
subjective reaction of the users of space. This information
is contained in social media data and can be used to learn
about perceptions about space and spatial preferences, and
can potentially lead to better location- (and customer-)
based analytics.
CRISIS RESPONSE
During a crisis situation, quickly gaining as much information
as possible about the progression of events is of crucial
importance. Gathering information is necessary for
developing situational awareness and can mean the difference
between life and death. Social media is one source of
such information that has started garnering interest in recent
years. Twitter users write about disaster preparations,
developments, recovery, and many other topics [158].
Retrieving this information can lead to improvements in
disaster management strategies. In contrast to many other
information sources, social media posts appear nearly immediately
whenever there is a new occurrence (as long as
telecommunication infrastructure is still intact) and can
therefore deliver information very quickly. Such messages
can also provide new perspectives that would not be available
any other way at this speed, e.g., ground photos. In
addition to factual information, social media can offer
personal insights into the occurrences as well as a backchannel
to users for relief providers, government agencies,
and other official institutions as well as the media. In a
2010 Red Cross study, 69% of Americans said that emergency
response agencies should respond to calls for help
sent through social media channels [183]. A highly comprehensive
overview of social media usage in crisis situations
is given in [184].
The difficulty lies in the retrieval and classification of
such messages. As described in the " Social Media Data
Management " section, the incoming volume of Twitter
messages in the livestream is huge. In any given event, the
majority of these posts will not be relevant to the event or
useful to service providers. The task of finding social media
posts in a crisis may appear clearly defined at first but
quickly becomes more convoluted when attempting an
exact definition. Existing publications have defined this
problem statement in a variety of ways. As described in
[185], research generally focuses on three qualities of messages:
whether they relate to the crisis event at all, whether
they are relevant, or whether they are informative. All
of these questions, especially the last two in particular,
DECEMBER 2022 IEEE GEOSCIENCE AND REMOTE SENSING MAGAZINE
depend highly on the users of the system, be they affected
citizens, family members, the government, news media,
or others. In many cases, users are assumed to be relief
organizations. In addition, each of these users may be interested
in a different use case of the system. For instance,
humanitarian and governmental emergency management
organizations are interested in " understanding the big picture, "
whereas local police forces and firefighters desire to
find " implicit and explicit requests related to emergency
needs that should be fulfilled or serviced as soon as possible "
[186]. Moreover, some of these use cases may require
a high precision of the detected tweets while possibly
missing some important information; others may be more
accepting of false alarms while focusing on a high recall.
To date, several detection approaches have been presented
in the literature and fall into three categories: filtering
by characteristics, crowdsourcing, and ML based. The
most obvious strategy is the filtering of tweets by various
surface characteristics, as shown, for example,
in [187].
Keywords and hashtags are used most frequently for this
task and often serve as a useful prefilter. Olteanu et al.
[159] developed a lexicon called CrisisLex for this purpose.
However, this approach easily misses tweets that do not
mention the keywords specified in advance, particularly
when changes occur or the attention focus shifts during
the event, or it may retrieve unrelated data that contain the
same keywords [188]. Geolocation is another frequently
employed feature that can be useful for retrieving tweets
from an area affected by a disaster. However, this approach
misses important information that could be coming from
a source outside the area, such as help providers or news
sources. Additionally, as described in the " Social Media
Data Availability " section, only a small fraction of tweets
is geotagged at all, leading to a large quantity of missed
tweets from an area [72].
To resolve these problems, crowdsourcing strategies
were developed, i.e., asking human volunteers to manually
label the data. Established communities of such volunteers
can be activated quickly in a disaster event, e.g., the
Standby Task Force (https://www.standbytaskforce.org/). To
facilitate their work, platforms have been developed over
the years. One of the most well-known systems is Ushahidi
(https://www.ushahidi.com). This platform allows people
to share situational information in various media, e.g.,
by text message, e-mail, and of course, Twitter. Messages
can then be tagged with categories relevant to the event.
Other examples include Artificial Intelligence for Disaster
Response [188], which contains automatic analysis tools,
or CrisisTracker [168]. In CrisisTracker, tweets are also collected
in real time and clustered by topics so that volunteers
can analyze them jointly.
In recent years, approaches based on deep learning
techniques have come to the forefront of research. Caragea
et al. [160] first employed CNNs for the classification of
tweets into those related to flood events and those unrelated.
In many of the following approaches, a type of CNN
167
https://www.standbytaskforce.org/ https://www.ushahidi.com
IEEE Geoscience and Remote Sensing Magazine - December 2022

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