Archive

Archive for May, 2021

Faked Satellite Imagery: Another opportunity to be critical of the data

As we have written about frequently in this blog, all geospatial data should be viewed critically. The user needs to carefully assess the attributes, resolution, date, source, and other characteristics before deciding whether that data is fit for use. The same is true with satellite imagery, for reasons we have described here (Be critical of the data–imagery too!) and here (Imagery–It is what it is. Well, not always).

But a new and disturbing reason for critical thinking has appeared more recently, and that is faked imagery. One of a growing number of articles about this issue is entitled A Growing Problem of ‘deepfake geography’: How AI (Artificial Intelligence) Falsifies Satellite Images. In the research article referred to here, entitled Deep fake geography? When geospatial data encounter Artificial Intelligence, by Bo Zhao, Shaozeng Zhang, Chunxue Xu, Yifan Sun, and Chengbin Deng in Cartography and Geographic Information Science, the authors describes their study. The goal of the study was not to show that geospatial data can be falsified, but rather, “the authors hoped to learn how to detect fake images so that geographers can begin to develop the data literacy tools, similar to today’s fact-checking services, for public benefit.” They suggest timely detections of deep fakes in geospatial data and proper coping strategies when necessary, with a goal to cultivate critical geospatial data literacy and “understand the multi-faceted impacts of deep fake geography on individuals and human society.”

Fake satellite images of a neighborhood in Tacoma with landscape features of other cities. (a) The original CartoDB basemap
tile; (b) the corresponding satellite image tile. The fake satellite image in the visual patterns of (c) Seattle and (d) Beijing, from Zhao et al. article in Cartography and Geographic Information Science.

Situating the issue of images that have been purposefully falsified in a broader context is this very useful article by Pierre Markuse, who advocates that a user needs to differentiate between three different ways an image could be understood (or really debunked) as being a fake: 1. Perceived as fake but in fact just a different representation of the data, 2. Perceived as fake but just a misrepresentation of facts, and 3. Actually faked satellite images. Pierre provides excellent illustrations of each of these three ways, including a supposed fire in Central Park in New York City and a “pollution plume” spilling from a river into a sea. Pierre very helpful concluding section on how to determine if an image is faked or out of context focuses on the themes of this blog–providing practical advice on what questions to ask as you examine and work with images. I highly recommend both of these articles for students, instructors, and researchers.

Along these lines, I would also advocate any user of GIS or remote sensing software to pay close attention to the defaults when images are brought into your software and displayed and rendered. These defaults are not nefarious, to be sure, but they are created to encompass the needs of a wide variety of users. Your needs might very well be different, so make sure you understand what the defaults are and how to change them, so that you are not misunderstanding your data or inadvertently leading others into misunderstanding.

These developments are not unexpected, and while the deliberately faked images are unfortunate, they provide more opportunity to assist students and colleagues around us to always be vigilant and critical of the data–including and perhaps especially geospatial data.

Joseph Kerski

Categories: Public Domain Data

Coupling data with scholarly research, Part 2

May 10, 2021 1 comment

Recently we wrote about coupling data with scholarly research as a means to enable researchers and practitioners to avoid “starting over” when they wish to tackle a problem with GIS or any other set of tools. This is part of an important and much wider discussion, and any blog essay by its very nature will not do it sufficient justice. But it is worth expanding that discussion at least with one further essay at this juncture, and opening the topic up to the wider community via the comments that you, the reader, can make, below.

I spoke about this with our Esri Chief Scientist, Dr Dawn Wright, who said that in her view, “there are presently two separate but related discussions: (1) the publishing of data either with papers on its own vs. (2) the publishing of software code, or workflows/methods within software, either with papers or on its own. This in turn begs the question of how to properly cite that data or software once it is published on its own or along with a paper. These are long-standing issues as tackled by the Earth Science Information Partners (ESIP; e.g., https://www.esipfed.org/esip-endorsed ); the NSF-funded EarthCube initiative (for example, https://www.esipfed.org/data-help-desk), and the many working and interest groups of the Research Data Alliance (RDA), for example: https://rd-alliance.org/groups/data-citation-wg.html.” In addition, Dr Wright also mentioned her essay that takes these discussions to the practical level, Making story maps citable; for example, with Digital Object Identifiers, focusing on Making Story Maps Citable (e.g., with Digital Object Identifiers) but with implications beyond story maps.

Along these lines, another colleague of mine here at Esri, Dr Kevin Butler, reminded me that the Nature Research journal Scientific Data is a ‘peer-reviewed, open-access journal for descriptions of scientifically valuable datasets, and research that advances the sharing and reuse of scientific data.’  Kevin believes it represents a nice balance between just a bucket full of data (no paper) that only a few will understand vs. taking space from an analysis based paper to describe the data and processing in detail. I agree. Costs are reasonable for publication, and they also host the data, too.  And MDPI has a similar type of journal that fits into this discussion, as does the Data Science Journal (e.g., https://datascience.codata.org/about/research-integrity/ ), and Elementa.

Coupling data with scholarly research, Part 2. Photograph by Joseph Kerski.

–Joseph Kerski

Categories: Public Domain Data