Literature Responses
Davies et. al, 2019
Open source geospatial technologies provide an important avenue that is accessible for peoples around the world to actively engage and interact with their changing landscape in an age of increased climatic variability. The article An Open-Source Mobile Geospatial Platform for Promoting Climate-Smart Livelihood-Landscape Systems in Fiji and Tonga (Davies et. al, 2019) outlines a project that, as you might expect with such a title, develops a highly accessible mobile platform of geospatial information in order to better provide largely self-subsistence farmers within regions of Fiji and Tonga with tools helpful in the interaction with a climatically vulnerable landscape. Rather than develop an entirely new technological platform to achieve this goal, this project funded by the Australian government screened numerous open-source geospatial platforms with specific criteria in order to determine which present options best fit their needs. This resulted in the determination of QField, an Android accessible mobile version of QGIS, as the base platform for users to digitize geospatial information which can then be extracted as maps in the full version of QGIS. Additionally, the project team relied upon PostGIS for reference spatial datasets and Google Earth Engine for dynamic land cover and climate layers. After users collect geospatial data through the QField plugin of QGIS, the new information is synchronized into the PostGIS reference datasets.
The results of this project are not yet known, as field-testing only began in June and July in 2019, though the authors emphasize that value will be found in a system of platforms which is capable of continual exploration, refinement, and development of spatial information that is directly linked to the users’ need for this information. I must agree that the value in this set-up is self-evident, and such a project can only be properly developed through the use of open-source geospatial platforms. As such, the people who are impacted by changing conditions of a landscape due to increased climatic variability can first-hand upload and share information which will then assist neighboring users.
DOI Link: https://doi.org/10.5194/isprs-archives-XLII-4-W14-31-2019
Allen et. al, 2016
Geospatial information from social media sites, such as Twitter, have the potential to be an important source of data to perform a variety of geographic analyses. In the article Applying GIS and Machine Learning Methods to Twitter Data for Multiscale Surveillance of Influenza (Allen et al, 2016), Twitter data from the 2012-2013 winter season is used to build a model that is able to gather and filter tweets pertaining to influenza. Then, with this model, the paper examines if Twitter is a reliable source of disease outbreak monitoring.
To collect data from Twitter, the paper employs linguistic query parameters of ‘flu’ and ‘influenza’ geographically focused on 31 cities throughout the country, listed in full in the paper. A variable geographic search radius is used for each city, though the length of the search radii for each city is not specified. The objective of the automated model that is developed from this accumulation of tweet data is to be able to determine which tweets are positively correlated with an actual case of influenza and which are not. To train the binary model, the paper randomly selected 1,500 tweets among the queried tweets and manually determined each one as either positively or negatively correlated with a real case of influenza. After that, the statistical method inverse document frequency is used to screen for shared words among all of the tweets manually selected as either representative or not representative of an actual influenza case. These indicators of positive or negative influenza correlation were then weighted. Then, with words positively and negatively weighted, a cumulative value can be automatically assigned to each tweet in the model. Once a certain threshold value is passed, a tweet is deemed as positively correlated with a real case of influenza.
Among the explanation of the model’s formation, there is a lack of explanation around the assignment of weighted values which is necessary for the paper to be both reproducible and replicable. It is not made clear which words the statistical model determined as key words associated with both the tweets positively correlated with influenza cases and those negatively correlated, except that general words such as ‘influenza’ did not belong to either category. Further, the weighted value assigned to these positive and negative indicators are not defined and the value threshold set for tweets in order to be determined as a positively correlated tweet is not reported. Without this information, it is not possible to reproduce the process employed to construct the model. However, the model could still be tested for replicability with differently assigned values on words gleaned from the stated statistical method. If the model is indeed robust, the temporal or spatial scale used for the acquisition of tweet data is not important, so long as there is a sufficient quantity of data.
After the construction of the model based on the 2012-2013 tweet data, the model was run for 2013-2014 tweet data with the same query parameters. The outputs were then compared against the official government-measured Influenza-like illness (ILI) rates at national, regional, and municipal scales to measure the precision and recall score of the model. Through the course of this inductive research, the paper determined that despite regional variability, there is a high rate of correlation between the twitter model and official ILI rates, demonstrating the potential reliability of Twitter as a source of geospatial illness monitoring.