Strategies for improving the communication of satellite-derived InSAR data for geo-hazards through the analysis of Twitter and online data portals

Satellite-based earth observation sensors are increasingly able to monitor geophysical signals related to natural hazards, and many groups are working on rapid data acquisition, processing, and dissemination to data users with a wide range of expertise 15 and goals. A particular challenge in the meaningful dissemination of Interferometric Synthetic Aperture Radar (InSAR) data to non-expert users is its unique differential data structure and sometimes low signal to noise ratio. In this study, we evaluate the online dissemination of ground deformation measurements from InSAR through Twitter, alongside the provision of open access InSAR data from the Centre for Observation and Modelling of Earthquakes, Volcanoes and Tectonics (COMET) Looking Into Continents from Space with Synthetic Aperture Radar (LiCSAR) processing system. Our aim is to evaluate (1) 20 who interacts with disseminated InSAR data, (2) how the data are used, and (3) to discuss strategies for meaningful communication and dissemination of open InSAR data. We found that the InSAR Twitter community was primarily composed of non-scientists (62%), although this grouping included Earth Observation experts in applications such as commercial industries. Twitter activity was primarily associated with natural hazard response, specifically following earthquakes and volcanic activity, where users disseminated InSAR measurements of ground deformation, often using wrapped and unwrapped


Capabilities and communication of SAR and InSAR data
The availability of open access Sentinel-1 data over the last ~8 years, in addition to other emerging SAR satellites, creates an opportunity for making SAR and InSAR data accessible and useable to non-experts.However, there is a contrast between the intuitive accessibility and processing requirements of 'visual' optical satellite imagery compared to the active microwave sensing of synthetic aperture radar (SAR) (Figure 2).Similarly, limited availability of open access to SAR satellites has slowed the operational and scientific uptake until the Copernicus Sentinel-1 programme started in 2014.
Optical imagery is intuitive to interpret by the human eye and is therefore interpretable by non-specialists (Voigt et al., 2007), even considering non-visible spectral information such as near-or short-wave infrared, which can reveal non-visible characteristics such as the presence of chlorophyll and therefore the health of vegetation (Figure 1a) (Pettorelli et al., 2005).
By comparison, SAR data, representing transmitted radio wave returns from a side-looking antenna, produce an all-weather view of the earth characterising the scattering properties of the reflecting surface or object (Rosen et al., 2000) but are less readily interpretable.The image brightness or intensity is influenced by the reflecting surface characteristics including texture, shape, water content, and satellite viewing angle (Gens and Van Genderen, 1996) (Figure 1b).The viewing angle can also cause shadowing and layover issues for steep terrain or for tall objects (Hagberg and Ulander, 1993).The SAR acquisition mode, wavelength, and polarisation also add to the complexity of data that is acquired due to variable radar penetration depths and interaction with surface features.The recent expansion in commercial SAR constellations (e.g.ICEYE and Capella), which often focus on providing derived products and results rather than raw data, also influences the future interpretability requirements of SAR data.
Interferometric Synthetic Aperture Radar (InSAR) uses pairs of radar images to measure radar wave phase differences, which corresponds to ground displacement after applying geometric, topographic and atmospheric corrections (Simons and Rosen, 2007).InSAR measurements of ground displacements are used to study natural hazards and processes such as earthquakes (Wright et al., 2004;Massonnet et al., 1993), landslides (Lauknes et al., 2010;Calò et al., 2012), volcanoes (Hooper et al., 2004;Ebmeier et al., 2018), and ice flow (Goldstein et al., 1993), in addition to anthropogenic signals such as ground water extraction and subsidence (Colesanti et al., 2003).For large and complex deformation, displacements derived from optical satellite imagery are a valuable alternative, particularly using daily revisit constellations such as Planet cubesats (Kääb et al., 2017).Over the last decade, new commercial radar satellites and open access Sentinel-1 data have transformed InSAR into a routine analytical and monitoring technique (Biggs and Wright, 2020).Common products generated using InSAR include wrapped and unwrapped interferograms, coherence maps, and displacement timeseries (Figure 2).Wrapped interferograms show colour cycle fringes representing λ (wavelength)/2, equivalent to 2.8 cm of motion per cycle for Sentinel-1 towards or away from the satellite, which are unwrapped to show a line-of-sight displacement in an unwrapped interferogram (Figure 2).Coherence is low if the scattering properties of the reflecting surface change between SAR acquisitions, which can inhibit displacement measurements (e.g.close to a fault rupture following an earthquake but more commonly associated with changes in vegetation or snow cover).However, changes in coherence also reveal insights into other Earth surface processes such as glacier movement (Atwood et al., 2010), volcanic eruption (Dualeh et al., 2021) and landsliding (Burrows et al., 2019), or urban damage following disaster events (Pilger et al., 2021).
Routine or on-demand production of InSAR products is available through several platforms including LiCSAR (Lazecký et al., 2020), the Alaska Satellite Facility (Kennedy et al., 2021;Meyer et al., 2017), and the GeoHazards Exploitation Platform (Galve et al., 2017).The LiCSAR processing system is detailed by Lazecký et al. (2020).Briefly, Sentinel-1 SAR data are automatically processed over active tectonic and volcanic regions to primarily produce coherence maps, wrapped and unwrapped interferograms at 0.001° resolution in WGS-84 coordinate system (111 m at the equator), which can be used to derive displacement time series using LiCSBAS software (Morishita et al., 2020).Open access datasets and online analytical tools and geographic information services can broaden the user base of satellite data by reducing the requirement for complex processing expertise.This is important when developing products and systems that integrate earth observation data, for example in-country monitoring capabilities for natural hazards to reduce disaster risk, in addition to using the data for disaster response (Joyce et al., 2009;Wang et al., 2012;Poblet et al., 2014).

Natural hazards communications
Both optical and radar satellite sensors, often in combination, are used to monitor and respond to natural hazards owing to their large spatial coverage, increasingly short revisit periods, and the diverse scope of analytical capabilities (Elliott et al., 2016;Kirschbaum et al., 2019;Joyce et al., 2009;Tronin, 2010;Voigt et al., 2007).For emergency response, data are analysed to produce mapping products (e.g. of building damage, floods, and fires), which are communicated to relevant in-country organisations to support response coordination, such as through the Copernicus Emergency Management Service (Directorate Space, Security and Migration, European Commission Joint Research Centre (EC JRC), 2020).Satellite data that are usually restricted or expensive are also provided for research and hazard response through international initiatives such as the CEOS Working Group on Disasters Pilots and Demonstrators (Pritchard et al., 2018;Kirschbaum et al., 2017).Recently, social media sites have been used by scientists and citizen scientists to investigate disasters in real-time using satellite data and eyewitness multimedia and accounts to collaboratively build knowledge in an open forum (Lacassin et al., 2020;Shugar et al., 2021;Ruan et al., 2022;Hicks, 2019;Earle et al., 2011).Twitter in particular has emerged as a platform used by scientists to create, discuss, and share research outputs (Van Noorden, 2014;Bruneau et al., 2021).National agencies such as the United States Geological Survey or volcano monitoring observatories also use dedicated social media accounts to provide updates on a range of natural hazards.These Twitter communications are open, therefore allowing direct public interaction, and are also increasingly directly embedded in news media reporting (Oschatz et al., 2021;Broersma and Graham, 2013).This highlights the important ethical considerations regarding the release of data, with or without interpretation or communication of uncertainties, by users with differing expertise.
Effective communication and knowledge co-production between scientists and citizens is essential to creating effective risk communication and disaster risk reduction strategies (Hicks and Barclay, 2018); however, it requires negotiating complex professional conduct considerations including: leadership, communications, cultural differences, community requirements and expectations, access to information, and misinformation (Newhall, 1999;Pallister et al., 2019;West et al., 2021).Provision of open-source satellite data and analysis tools offers one mechanism to break down geographical and social barriers to knowledge co-production (Turner et al., 2015).However, the ethics of scientific conduct and data dissemination are complex, particularly during extended disaster events such as volcanic crises, for example as described by Newhall et al. (1999).We do not evaluate data use for specific disaster events in this study; however, these considerations provide an important context to the dissemination and discussion of data products for individuals or teams that are remote to the event, for example: (1) when and what data to release and with what level of interpretation, considering interpretation differences and underlying uncertainties; (2) considering whether releasing data will distract from official advice, or undermine local teams; (3) whether releasing the data can contribute to public safety, even if results are only preliminary.

Study aim: LiCSAR and InSAR communication
In this study, we aimed to evaluate how InSAR data are communicated and engaged with through Twitter and the automatic InSAR processing system LiCSAR.Specifically, we address what we can learn from Twitter communication to improve the utility of this processing system.Through COMET, open-access LiCSAR data supports research into tectonic, volcanic, and other earth processes, in addition to responding to natural hazard events and aiming to support the integration of earth observation data into disaster risk reduction strategies.
Noting the less intuitive visual outputs and applications of InSAR to non-specialists (Figure 2), we explore whether communications are predominantly in-reach (to other academic scientists) or outreach (e.g. to industry experts, the media, and decision-makers), and the types of data that are communicated in each case.We incorporate user feedback from LiCSAR to identify how the current provision of data products aligns with those communicated using Twitter.We also consider engagement with the LiCSAR Earthquake InSAR Data Provider (EIDP) that automatically processes and disseminates InSAR data for major earthquakes.We then identify future directions to improve the communication and application of InSAR products to natural hazard disaster risk reduction.

Data and methodology
To investigate the prevalence and types of InSAR data communication over Twitter, we first collected and analysed a database of InSAR themed Tweets, which were also used to derive the community of InSAR Twitter users by analysing user connections and key words in profile descriptions.Engagement with open-access LiCSAR InSAR data was then evaluated using website analytics from the LiCSAR portal, combined with a user feedback survey.We then used the Earthquake InSAR Data Provider (EIDP) and the COMET Volcano Deformation Portal as example use cases of LiCSAR data to determine how these data align with those communicated using Twitter, and to identify directions to improve these InSAR data dissemination.

InSAR tweets
Twitter was used to derive InSAR communications (tweets), including how widely the tweets were circulated (retweeted) through a network of users, each with profile descriptions that often indicate professions and interests.We used the Twitter application programming interface (API) and Rtweet package v0.7.0 (Kearney, 2019; R Core Team, 2021) to retrieve public tweets (excluding retweets) 18 th August 2020 to 9 th February 2022 mentioning "InSAR".We excluded retweets to avoid considering the same tweet text multiple times, though we do consider the retweet count for each tweet.Retweets reveal how and what information is spread through Twitter, so could provide useful context on the spread of misinformation in relation to natural hazards.We processed tweet text to remove stop words, mentions, uniform resource locators, emojis, numbers, and punctuation (e.g.Ruan et al., 2022).Finally, we removed a list of other words, tweets, and user IDs that were not relevant to our study, including "InSAR" tweets relating to the International Society for Autism Research (Supplementary information).
We used word counts from the remaining tweet text to identify the most commonly occurring words or hashtags in the InSAR tweets to identify the context of the InSAR tweets.
To identify the most common and most popular InSAR Twitter communications, we examined the user networks for the top ten InSAR tweeters, (1) ranked by number of tweets, and (2) the tweeters of the top ten most retweeted tweets.For each of the top ten user IDs (a static ID number which unlike the Twitter user handle, cannot be changed) in ( 1) and (2), we used Rtweet to find the network of users that were followed by each account, therefore unravelling active components of the Twitter InSAR network.User connections were summed to find commonly connected users within the InSAR network.For example, a user ID followed by five of the top ten tweeters would feature a count of five.We also extracted all user IDs that followed the top ten accounts in (1) and (2).These followers represent the users that would receive tweets on their timeline from the top ten InSAR tweeters in ( 1) and ( 2).The profile description of these users were classified using key word search strings from Côté et al. (2018) into 'scientist' and 'non-scientific' groups using the R package 'stringr' (Wickham, 2017) to determine different applications of InSAR and whether InSAR communications were predominantly amongst scientists or outreach.The keyword strings we used are shown in the Supplementary information.Examples of the 'scientist' class included: 'professor', 'phd', 'post doc', 'doctoral', and 'research scientist' and could be within academia (universities) or applied and research-focussed industry or government agencies.Profiles that did not match the scientist classification were assigned 'non-scientist'.'Nonscientists' are expected to include decision-makers such as government agencies; media and journalists; outreach organisations such as museums and science educators, and other members of the public (Côté et al., 2018).These classifications were used to disaggregate different users of InSAR data; however, we acknowledge the distinction between 'scientist' and 'non-scientist' is ambiguous in some cases.We validated the classification by manually inspecting a random sample of 500 users from each of the scientist and non-scientific groups.We did not disaggregate the non-scientific group into additional categories such as 'media' or 'decision makers' due to the high misclassification rates observed by Côté et al. (2018).

Website Analytics
To evaluate how open access InSAR data are used and by whom, we used statistics from the COMET LiCSAR project, combined with our analysis of InSAR communications using Twitter.
Website analytics for the LiCSAR portal were obtained from Google Analytics from January 2017 to December 2019, and WordPress.comstarting in January 2020.Both platforms use a page 'views' metric, which represents instances of a user viewing the portal web page; however, the two analytics are potentially not directly comparable.User feedback on the LiCSAR Portal was collected using a Google Forms survey containing 15 questions (Figure S1), which were designed to collect information on who was using the portal, their usage and experience with LiCSAR data, and suggestions for future developments.Responses were provided by visitors to the LiCSAR portal that clicked the 'Give User Feedback' link on the portal homepage.Responses were also informally solicited by COMET members through peer networks and existing partnerships by sending a link to the feedback form.COMET membership was asked as a survey question to distinguish these responses.In this study, we used anonymised responses obtained from the initiation of the survey in February 2020 to February 2022 (n=80); however, a link to the survey remains on the LiCSAR Portal for ongoing feedback.

LiCSAR Earthquake InSAR Data Provider (EIDP)
LiCSAR data are produced following specific earthquake events and consolidated through an 'Earthquakes' tab on the LiCSAR Portal.First, earthquake events that meet a threshold magnitude and depth, trigger the activation of LiCSAR frames for data processing (Lazecký et al., 2020).These events are plotted on an interactive Leaflet map along with key metadata from the United States Geological Survey (USGS).The map interfaces with a data table that allows users to search and download data for each event from an online event page.These event pages are automatically generated for each earthquake event and are regularly updated in the following weeks to include new coseismic and post-seismic data.Details of the earthquake events with a link to the event page are automatically tweeted by @COMET_database, in addition to the first processed interferogram overlapping the earthquake epicentre.We used page views logged using Google Analytics to evaluate user engagement with the EIDP system and we report statistics for the same reporting period as the Twitter data (18 th August 2020 to 9 th February 2022).

COMET Volcano Deformation Portal
The COMET Volcanic and Magmatic Deformation Portal (https://comet.nerc.ac.uk/comet-volcano-portal/) allows observatory scientists to critically interrogate automatically generated Sentinel-1 interferograms and identify deformation around active volcanoes (Rigby et al., 2021).The volcano portal provides access to interferograms and time series (Lazecký et al., 2020;Morishita et al., 2020) for an initial subset of volcanoes where processing is most complete, with the aim of eventually including all subaerial Holocene volcanoes.The portal allows users to browse through images of interferograms using a 'slider' tool and to plot timeseries (including choice of reference area) and profiles for comparison with topography interactively.
Wrapped and unwrapped interferograms can be compared, as well as time series with and without GACOS atmospheric correction (Yu et al., 2018) and spatiotemporal filtering (Morishita et al., 2020).The aim is to provide users with tools that support interrogation of the data, with a particular focus on distinguishing between atmospheric features and volcanic deformation (e.g.Pinel et al., 2011;Ebmeier et al., 2018;Biggs et al., 2021).We present new Sentinel-1 InSAR results in the context of historical observations of volcano deformation (as described in Ebmeier et al., (2018)) organised according to volcano numbers from the Smithsonian list of Holocene volcanoes (Global Volcanism Program., 2022).Machine Learning based tools (Biggs et al., 2022) that flag high magnitude displacements also provide users with a tool for quickly finding major deformation events in a time series.We deliberately do not provide authoritative interpretations of the interferograms, but provide tools for an engaged user to assess the automatically processed data for themselves, facilitating comparisons to both ground-based measurements and independent processing of the InSAR data.The spatial resolution of images used for the on-the-fly time series analyses is low, being based on boxes centred on volcanoes (or sites of active magmatic deformation) with a modal length and width of 50 km (this would be sufficient to capture ~90% of historical magmatic and volcanic displacements, although in a few cases signals do exceed ~100s km in diameter (Ebmeier et al., 2018;Henderson and Pritchard, 2013)).These tools are designed to allow on-the-fly analysis, and also allow the user to download either the time series results shown on screen (currently in csv format) or the original interferograms.
Our communication strategy for volcanic signals is in contrast to the more responsive role of the EIDP and reflects the challenges of forecasting risk associated with volcanic hazards that evolve dynamically on temporal scales that can range from hours to years.While InSAR measurements have the potential to capture displacements significant for understanding and even forecasting volcanic hazards, there are specific obstacles to the exploitation of satellite radar for volcanology.Development of the volcano InSAR portal tools has also been informed by consultation with volcano observatory colleagues, including colleagues at Addis Ababa University, Ethiopia and Instituto Geofisico Escuala Politecnica Nacional, Ecuador.

InSAR Twitter
We collected a total of 16,621 tweets from 18 th August 2020 to 9 th February 2022 mentioning "InSAR", which was reduced to 5,349 after removing retweets, and down to 4,312 after cleaning the dataset (Section 2.1).These cleaned tweets originated from 1,935 unique Twitter users.We observed a mean of 8±7 tweets per day and spikes related to notable events including scientific conferences, news media reports, earthquakes, and popular Twitter threads (Figure 3).A data gap was present in the timeseries (29/10/2020-06/12/2020); however, the source was unknown and could be related to an issue with the collection of tweets.Specific earthquake events related to increased activity included those in Croatia (USGS ID: us6000d3zh.Date: 2020-12-29), Greece (USGS ID: us7000df40, Date: 2021-03-03), China (USGS ID: us7000e54r, Date: 2021-05-21), and Haiti (USGS ID: us6000f65h, Date: 2021-08-14).Typically, these events included multiple other earthquake events (foreshocks and aftershocks) at each location.Other peaks in tweets were related to InSAR sessions at the American Geophysical Union (AGU) and European Geosciences Union (EGU) conferences in December 2020 and April 2021, respectively.News media reports were associated with tweet peaks in June 2021 related to the upcoming Venus Emissivity, Radio Science, InSAR, Topography, and Spectroscopy (VERITAS) mission to map Venus, and to reports of subsidence in Mexico City.Finally, two peaks in tweet activity were related to specific Twitter threads written in Spanish on subsidence in Mexico City, and earthquake seasonality in Mexico (Figure 3).

255
Word count analysis of the text of InSAR tweets revealed that mentions of 'data', including Sentinel-1 SAR, and applications of InSAR, such as to natural hazards, occurred most frequently (Figure 4).'Data' was the most frequently occurring word (n= 676) followed by 'Earthquake' (n= 312) and 'Deformation' (n= 269) (Figure 4a).The most popular hashtags used in InSAR tweets were 'Sentinel1' (n=323), 'Earthquake' (n=259), and 'SAR' (n=143) (Figure S2).The most popular words and hashtags 260 in the top 50 most retweeted InSAR tweets were 'Deformation' (n=11), 'Earthquake' (n=10), and 'Dike' (n=9) (Figure 4b).Specific data sources identified in the tweet word analysis included 'Sentinel-(1)', 'VERITAS', 'NISAR' (NASA-ISRO-SAR), and 'ALOS' (Advanced Land Observing Satellite) (Figure 4, Table S1).are coloured by their count (node_connections) where a user is followed by two or more of the top ten tweeters (Figure 5a), or a tweeter of a top ten most retweeted tweet (Figure 5b).The black lines with transparency represent other Twitter users followed by a labelled (black bold text) user.Two users were present in both groups: 'ericfielding' and 'mr_insar' (Figure 5a,   b).The top ten tweeters followed 7,334 users of which 916 had two or more node connections (i.e. were followed by two or more of the top ten tweeters) (Figure 5a).By comparison, the top ten most retweeted tweeters followed 5,056 users, of which 687 had two or more node connections (Figure 5b).Users with high node connections (out of a potential maximum of ten), in some cases featured in both network plots (example users in blue text in Figure 5a, b) despite not being a top ten tweeter or retweeted tweeter themselves.S3).The count shows the number of times each pair of words occurred.
The followers (n=31,680) of users in the top ten tweeter groups (black text in Figure 5a, b), were classified into scientist and non-scientific audiences to determine the in-reach and outreach status of the InSAR tweets.We excluded the space reporter 'thesheetztweetz' since their retweet was about the NASA VERITAS, which is a Venus mission we exclude in our study of Earth Science InSAR.We found that the networks were predominantly outreach, since 62% of the followers (n=19,778) were non-scientists and 38% (n=11,902) were scientists.The profile description word network for the scientist group was more connected than non-scientists (Figure 5c, d).Commonly occurring paired words for scientist profiles that were indicative of user interests included 'machine learning' (n=113), 'natural hazards (n= 80), 'structural geology' (n=80), and 'active tectonics' (n=60) (Table S2).Non-scientist paired words were more diverse with lower occurrence counts, including 'civil engineer' and 'animal lover' (n=21), and 'gis specialist' (n=18) (Table S3).Commonalities between both communities included 'remote sensing', which was most frequent in both lists and suggests the non-scientist group included a large community of satellite data users, likely including industry Earth Observation experts applying InSAR analyses.
Of the ten most retweeted InSAR tweets (Figure 6), two were plain text (a, e); one contained a video of volcanic deformation (b); one contained a photograph and description of subsidence (c); three contained wrapped interferograms of volcanic deformation of Nyiragongo, Democratic Republic of the Congo (d), Iceland (g), and Galapagos (h); one contained new opensource SAR code (f); and two contained line of sight displacement measurements, one of volcanic deformation in Iceland (i) and one of an earthquake in Haiti (j).Four of the tweets presented Sentinel-1 data (d, g, h, i) and one presented ALOS-2 data (j).Notably, these tweets mostly featured data annotations and text explanations of the associated images and were retweeted a total of 1,487 times.

LiCSAR Portal
The LiCSAR Portal disseminates automatically processed Sentinel-1 InSAR data for active tectonic and volcanic regions, thereby focussing on natural hazards, including a dedicated earthquake processing system known as EIDP.User feedback provided on LiCSAR and usage statistics for LiCSAR and EIDP were used to evaluate engagement with the system and how this related to InSAR communications over Twitter.Views to the LiCSAR portal homepage February 2021 to February 2022 (period of the user feedback survey) totalled 17,403, with an additional 3,538 views to the EIDP homepage (Table 1).Annual views increased by 4,833 (2020-2021) and averaged 1,148 views monthly (Figure 7).Evaluation of the LiCSAR system was therefore expected to capture a range of users across the InSAR community.

User feedback
User feedback on the LiCSAR Portal February 2020 to February 2022 revealed the profession of LiCSAR users and their views on current and future InSAR products (Figure 8).We report statistics using all responses (n=80); however, in some cases not all questions were answered by respondents.Most users (83%, n=64) that provided feedback were not associated with COMET so were independent of the LiCSAR system.Respondents were from 29 countries and used LiCSAR data to study a 335 total of 39 countries.Most respondents (77%, n=72) were academic (scientific) users.Geological/geophysics survey (7%) and public sector (6%) were the next highest groups (Figure 8a); however, the InSAR applications in these groups (scientific vs non-scientific) were not known.Users predominately used unwrapped and wrapped interferograms and coherence products (Figure 8b).We observed a broad spread in the number of interferograms used, with the majority using less than 100 interferograms and some using over 400 (Figure 8d).Similarly, most used fewer than five data frames (individual frame IDs 340 on the LICSAR Portal), though one used over 250 (Figure 8e).Displacement time series, mean velocities, and subsidence/uplift rates were the most desired future products (Figure 8c).Overall ease of access to data through the portal averaged 3.8±1.1 on a scale of difficult (1) to easy ( 5) and most respondents identified a benefit to future integration of the data in a cloud-based platform such as Google Earth Engine, scoring 3.6±1.2on a scale of no benefit (1) to strong benefit (5).

Earthquake InSAR Data Provider (EIDP)
The EIDP component of the LiCSAR Portal included 430 earthquake event activations between April 2019 and February 2022.
The earthquake event pages and first wrapped interferogram generated over the epicentre (Figure 9) were automatically tweeted by @COMET_database, which featured in the top ten InSAR tweeting accounts (Figure 5a) and accounts for the high instance of 'LiCSAR' word occurrence (n=190) in the InSAR tweets (Figure 4a).The five most retweeted EIDP interferograms all displayed a clear earthquake signal, shown by the colour fringe cycles and black arrow (Figure 10a).Three of these interferograms corresponded to earthquakes that were associated with peaks in tweet activity identified in Figure 3    We evaluated the engagement with EIDP using page views logged through Google Analytics for events that occurred during our Twitter data collection (18 th August 2020 to 9 th February 2022).Users accessed EIDP webpages directly (n=1,232), through social media (n=736), through referrals (linked from another website) (n=263), and organic searches (n=76).Twitter accounted for 98% of social media access to EIDP with the remainder from Facebook.The page view time series for the ten most viewed events revealed a peak in page views associated with Sentinel-1 InSAR data tweets from @COMET_database for events with a deformation signal (n=6) (Figure 11).Three events did not have an associated data tweet and one had no deformation signal.
The mean time between an earthquake and subsequent tweeting of InSAR data was 4.7±2.8days, with the shortest gap being one day for the 2021 Mw 6.3 earthquake 'us7000df40' in Týrnavos, Greece (Table 2).The five most retweeted interferograms (Figure 10a) were amongst the most viewed EIDP event pages, of which the Mw 7.3 earthquake in Southern Qinghai, China (21/05/2021, us7000e54r) had the most page views (n=1,231) (Figure 11, Table 2).

Engagement with the COMET Volcano Deformation Portal 390
Although we did not have usage statistics for COMET Volcano Deformation Portal in this study, we considered barriers to the use of InSAR data by volcano observatories from ten respondents to a short questionnaire summarised in Table S4.All agreed that ready-processed and analysed data to be beneficial to their operations, improving their monitoring capacity.They would use such data to inform their deployment of ground-based monitoring instrumentation and to alert the necessary authorities of unrest.The responses identified that automated tools must be accessible without the need for high-speed internet or advanced computing facilities and that data need to be delivered in a timely manner and support would be required to ensure proper understanding.Though ready-analysed data is desirable, the responders also stated that they would like training to accompany these resources, so they have a good understanding of the outputs and the processes leading to these outputs.

Discussion
Following the open access availability of SAR data acquired by the Sentinel-1 satellites (adopted for the Copernicus programme by the EU) and a transition towards open-source processing techniques, we aimed to evaluate the applications and engagement with InSAR products that are communicated online outside of traditional scientific publication outputs.First, we analysed InSAR geoscience communication through Twitter to determine the types of outputs communicated (e.g.text, images, videos), their data sources (e.g.Sentinel-1), geoscience applications (e.g.earthquakes, volcanoes), and level of user engagement.We then used our experience with the LiCSAR Sentinel-1 InSAR processing system to determine the extent to which this open access data provision aligned with Twitter InSAR communications.

InSAR Twitter
InSAR Twitter content was primarily aligned with natural hazards and disaster event response.However, peaks in Twitter activity were also associated with academic conferences (American Geophysical Union (AGU) and European Geosciences Union (EGU)), four earthquake events, news articles, and Twitter threads (one on subsidence and one on earthquake seasonality) (Figure 3).Analysis of the most commonly occurring words in the InSAR tweets revealed that InSAR was most associated with earthquake deformation (Figure 4a, b), likely reflecting the newsworthy status of these events drawing user engagement (e.g.Zawacki et al., 2022).In addition to earthquakes, applications of InSAR to volcanic deformation was also prevalent in our analysis, particularly in the most retweeted InSAR tweets.Words such as 'dike', 'eruption', 'volcano', and 'magma' were present alongside specific events such as the volcano-tectonic events that occurred around the Icelandic Fagradalsfjall volcano on the 'Reykjanes' peninsula, and the 'Nyiragongo' eruption in Democratic Republic of the Congo (Figure 4b).Wrapped interferograms of these events were in the top ten most retweeted tweets (Figure 6).
Our analysis of InSAR tweets reflect a prevalence of Sentinel-1 data usage as both the most commonly occurring hashtag (Figure S2) and use as the data source in four of the top ten most retweeted InSAR tweets (Figure 6).The Sentinel-1 series comprised a pair of SAR satellites (Sentinel-1A launched in 2014 and Sentinel-1B in 2016, the latter which malfunctioned in December 2021), which provided a step change in provision and development of processing tools for open access radar data (Geudtner et al., 2014;Elliott, 2020).For LiCSAR data, we observed greater retweeting of interferograms displaying an earthquake deformation signal compared to those without (Figure 10), suggesting a degree of collaborative information filtering to amplify the presence of a positive detection.This aligns with other studies that found Twitter can facilitate rapid knowledge co-production that receives transparent, though unconventional, peer review through public Twitter discussions (Lacassin et al., 2020;Mendoza et al., 2010).Similarly, we found that scientists accounted for 38% of users within our defined Twitter InSAR network, though many of the non-scientist grouping likely represented Earth Observation experts in other applications, including commercial industries.Whilst there is no guarantee that this network of users acts to positively filter InSAR data, for example, to mitigate common challenges such as misinterpretations of atmospheric effects as a deformation signal (for which we show a tweeted example of relating to Agung Volcano in Figure S3), other studies have shown that Twitter has been an useful platform for geoscience discussions and collating information on natural hazard events (Lacassin et al., 2020;Shugar et al., 2021;Ruan et al., 2022).

LiCSAR data dissemination
The availability of open access Sentinel-1 data has led to the development of semi-or fully-automated processing systems to produce InSAR information from the vast amount of data acquired each day.Scientific systems are largely focused on geohazards (e.g.Lazecký et al., 2020;Kennedy et al., 2021;Galve et al., 2017;Piter et al., 2021), whereas commercial systems cover a diverse range of applications including monitoring subsidence, mining activities, engineering projects, and geohazards.Two commercial systems using Sentinel-1 and other data sources (@TRE_ALTAMIRA and @SatSenseLtd) were identified in the top tweeters InSAR network (Figure 5a).Open access LiCSAR data disseminated through an online portal and @COMET_database was also a top ten tweeter of InSAR data, specifically through automatic tweets of coseismic earthquake interferograms (Figure 10).LiCSAR InSAR data has been used in studies of earthquakes (Ganas et al., 2018), ground deformation (Morishita et al., 2020;Weiss et al., 2020;Chaabani et al., 2020), volcano deformation including machine learning of deformation patterns (Moore et al., 2019;Anantrasirichai et al., 2018;Lloyd et al., 2018;Gaddes et al., 2018;Albino et al., 2019), and landslides (Burrows et al., 2019;Tsironi et al., 2022).The primary link to data dissemination through Twitter was from the EIDP system that generates interactive web pages, processes interferograms, and automatically tweets the outputs (Figures 9,10,11), thereby providing a resource for preliminary interpretation and data download.Although LiCSAR data did not feature in the most retweeted InSAR content (Figure 6), the underlying data products in these tweets (wrapped and unwrapped interferograms) were the same as those generated by LiCSAR and these automatic tweets were associated with directing views to earthquake event pages (Figure 11).The most retweeted tweets generally featured supplementary data annotations and text explanations of the observed deformation signals, which are not a feature of the automated LiCSAR data dissemination.Similarly, 30% (n=24) of the LiCSAR survey respondents identified uncertainties as a useful future product (Figure 8c), which would ideally be disseminated in parallel to other outputs to reduce misinterpretation of interferogram fringes for example, that are often related to atmospheric or topographic effects (Zhou et al., 2009).

Application of tailored online toolsthe COMET Volcano Portal
Despite the prevalence of volcanic InSAR tweets (e.g. Figure 6), no responses were recorded from volcano observatories in the LiCSAR user survey (Figure 8a).This may be in part because presentation of the COMET Volcano Portal thus far has been focused on targeted workshops (e.g., 'Online International Workshop on Volcano Monitoring Infrastructure on the Ground and In Space ', Feb 18, 19, 22 2021, and the annual COMET InSAR Training Workshop) rather than social media.It may also reflect a preference of volcano-focussed users to engage with the volcano portal tools rather than downloading LiCSAR imagery.
We considered potential barriers to uptake of InSAR data for volcanic applications to inform the portal development.These barriers can be grouped into two broad classes, the first relating to the awareness of data availability and confidence in interpreting the data, the second relating to logistical challenges such as the human and computing resources required to process satellite data and challenges associated with internet bandwidth (Table S4).Similar natural hazard information accessibility and usability issues were summarised by Gill et al. (2021).The COMET Volcano Portal minimises logistical issues by providing access to ready-processed InSAR.Data are downsampled and clipped to the area of interest for individual volcanoes such that the timeseries data are served over a standard web browser in small files sizes (typically several megabytes).Dealing with the second barrier of data awareness and confidence requires an approach incorporating a communication strategy.This could include Tweeting examples of portal content and promoting the portal at conferences and through local observatory networks, also considering other forms of social-media communication.We consider regular communication of volcano portal updates to be a better strategy than automatic direct social media responses to eruptions or unrest, due to the potential for eclipsing or distracting from the official communications from local volcano observatories and civil protection.We also consider further direct engagement with volcano observatory scientists to be critical, both for making the portal as applicable and transparent in its operation as possible, and to provide comparisons between the InSAR processing and ground truthing measures.These are also aided by engagement with international initiatives such as the CEOS Volcano Pilot/Demonstrator (Pritchard et al., 2018) or the Global Volcano Monitoring Infrastructure Database (GVMID).Portal developments will focus on mitigation of 'fading signal' in long term time series and establishing near real-time processing that functions globally.The provision of Machine Learning tools will also be developed to incorporate a variety of complementary methods.

InSAR communications and disaster risk reduction
Overall, InSAR Twitter was orientated towards natural hazards and specifically disaster or event response, rather than preparedness or disaster risk reduction, which reflects the current affairs nature of the Twitter platform (Petrovic et al., 2013;Murthy, 2011;Acar and Muraki, 2011).Here, InSAR contributes top-down situational awareness to these natural hazard events, usually through providing satellite observations of ground deformation (e.g. Figure 6, 10a), in some cases on interactive maps that could be used alongside 'on the ground' mapping (Figure 9).By comparison, bottom-up aggregation and analysis of 'on the ground' crisis-communication tweets are shown to be effective at identifying the immediate aftermath of events such as earthquakes, and providing the first accounts of impacts (Vieweg et al., 2010;Acar and Muraki, 2011;Earle et al., 2011).We found that peaks in InSAR tweets corresponded earthquakes (Figure 3) and ongoing volcanic events (Figure 4b), where InSAR products were used to reveal the spatial patterns of deformation.For the automated @COMET_database tweets, data were acquired, processed, and tweeted within 4.7±2.8days of the earthquake event, which may limit applicability of the data in the immediate response effort.It would therefore be useful to investigate the links between data dissemination, knowledge building over Twitter, and local event response to determine the specific applications and benefits to disaster management (Figure 12).This is particularly relevant when considering the coupling between open access data, derived products, related uncertainties, and quality interpretation, which are all necessary for effective communication and relates to a broader discussion on the engagement of natural hazard scientists in disaster risk reduction (e.g.Gill et al., 2021).In addition to event response, we observed tweets related to longer timescale events such as ground subsidence (e.g. Figure 6c).Explanatory Twitter threads are effective means of knowledge building and dissemination (Lacassin et al., 2020) and could therefore be used to incorporate InSAR analysis into the broader and longer-term disaster management cycle including reduction, readiness, response, and recovery (Elliott, 2020;Joyce et al., 2009) (Figure 12).Twitter threads unrelated to disaster events, for example on long-term measurements of active faults, offer a mechanism to balance the current focus on postdisaster response applications and media reporting, which increasingly embed Twitter threads within their news articles and therefore increase public exposure (Broersma and Graham, 2013).
Pre-processed and open access InSAR data such as produced by LiCSAR could support disaster risk reduction strategies, for example through long-term strain rate mapping over active faults to improve seismic hazard assessments (Lazecký et al., 2020;Stevens and Avouac, 2021), or monitoring regions prone to landslides (Novellino et al., 2021;Rott and Nagler, 2006) or volcanoes (Ebmeier et al., 2018).Similarly, InSAR can reveal urban ground deformation (Morishita, 2021) due to groundwater extraction or the weight of a city itself (Parsons, 2021).Therefore, several opportunities exist to integrate InSAR into disaster management strategies and broader UN Sustainable Development Goals to develop sustainable cities with reduced risk from natural hazards (UNISDR, 2015; UN, 2020).Increased uptake and dissemination of InSAR data has followed the breakdown of barriers to data accessibility, including open access data and processing chains; however, improved metadata detailing processing methods and uncertainties are still required to maximise applications of InSAR data and minimise misinterpretations due to radar data complexities.For example, evaluating apparent deformation signals alongside weather models is often essential at volcanoes to avoid incorrectly inferring a deformation signal (e.g. Figure S3).Integration of InSAR datasets into cloud-based analysis platforms and portal such as Google Earth Engine or the European Space Agency's Geohazard Exploitation Platform could further facilitate analysis with other data sources, and offers cloud computing capabilities for users lacking local resources (Gorelick et al., 2017), which is potentially advantageous for embedding InSAR data into disaster management agencies and volcano observatories (Table S4).

Limitations and other considerations
Although widely used for geoscience communications, InSAR communications over Twitter represent one social-media platform and other platforms and dissemination routes that are more relevant to specific stakeholders could be considered.
Additionally, identifying types of InSAR users was limited to profile description classification in our study, producing scientist vs non-scientists groups that were used to infer whether communications were in-reach or outreach.However, these groupings are not exclusive and preclude a more comprehensive evaluation of how InSAR data are communicated and used by scientists, industry experts, the media, and decision-makers for example.Self-reporting in feedback surveys could help address this.
Finally, ethical considerations exists despite the public nature of Twitter data and prior usage consent being given through acceptance of Twitter's terms of service (Sinnenberg et al., 2016;Gold, 2020).In our study, we collected tweets using the Twitter API, and report anonymised and aggregated statistics, other than reporting on the top tweeters of InSAR data, where we believe reporting a subset of Twitter usernames does not create a risk of harm, provides recognition to active members of the Twitter InSAR community, and adds context (e.g.Lacassin et al., 2020).

Future design and dissemination of LiCSAR and InSAR products
Our analysis of InSAR communications has shown that Twitter is a valuable tool for disseminating InSAR data to scientific and other Twitter users (Figure 5) and that current communications are predominantly focussed on natural hazard response (Figure 3, 4).However, LiCSAR products are not produced with a real-time disaster response mandate.Therefore, improving the dissemination of LiCSAR products for use in disaster response requires links with local or national advisory agencies, for example the British Geological Survey in the UK.The type of disaster (e.g.earthquake or volcanic eruption) and event-specific sensitivities require consideration to avoid hindering response efforts and to ensure effective and ethical communication (Newhall, 1999).Since ongoing hazard events have specific contexts and require local considerations beyond the scope of COMET LiCSAR, data dissemination is best channelled to organisations that can effectively use the InSAR data and link with civil protection agencies.This could involve additional signposting of data portals (e.g.EIDP Figure 9) and provision of guidance text and flowchart diagrams for data interpretation.Longer term integration of InSAR data into disaster mitigation (Figure 12) could also benefit from expert led Twitter threads discussing and demonstrating the capabilities and limitations of InSAR for specific applications.
Communication of uncertainties in LiCSAR InSAR data are currently lacking and were noted as a desired future product in a user survey (Figure 8c).Data processed with higher levels of corrections (e.g.atmospheric) particularly using a single methodology, do not always improve the end result and could cause misinterpretations.Additionally, other spatially variable (e.g.depending on land cover) biases can accumulate in time series analysis of interferograms (Maghsoudi et al., 2022;De Zan et al., 2015).Therefore, effective communication of processing steps is required to avoid instilling a false sense of confidence in derived outputs.Higher level processing steps require waiting for the necessary correction data and further processing time, which may delay use of the data.Therefore, there is a trade-off between data timeliness and appropriate level of processing, which is variable between hazard types and individual events.For example, there is merit in producing interferograms of earthquake events with a large deformation signal before atmospheric corrections (Figure 10a), whereas the signal for smaller or deeper events could be obscured.
We identify the following recommendations as a pathway for LiCSAR development and InSAR communications: context of an urbanising global population.We presented an example of using volcano observatory user feedback to develop the COMET Volcano Deformation Portal, by considering solutions to barriers to uptake of satellite data; however, in this case we drew on a small number (n=10) of respondents and continued feedback should be sought.We summarised future development directions using five recommendations, which included both LiCSAR development and broader InSAR communications.For example, effectively communicating uncertainties, such as a spatially distributed displacement uncertainty, or uncertainties that accumulate through processing steps including atmospheric correction, are critical to increase accessibility given the often complex interpretability of InSAR products.
the development of the COMET Volcano Deformation Portal.We thank Ailsa Naismith and an anonymous reviewer for comments that helped improve this study.

Figure 1 :
Figure 1: Comparing optical and SAR data.(a) Example of a 10 m resolution Sentinel-2 optical image (27 th June 2021) over the city of Bishkek and mountains to the south.Left panel shows a true-colour (red, green, and blue bands) image and the right panel shows a false colour composite of (shortwave infrared, near infrared, and blue bands).(b) Sentinel-1 radar multi-looked intensity image (26 th June 2021) from the Looking Into Continents from Space with Synthetic Aperture Radar (LiCSAR) processing system (Lazecký et al., 2020) covering the same area of Bishkek in (a) at 50 m resolution.

Figure 3 :
Figure 3: Timeline of InSAR tweets from September 2020 to February 2022.Dashed red line shows the mean of eight tweets per day (excluding data gap).Vertical lines and annotation are shown for events exceeding 22 tweets per day (two standard deviations from the mean).Dates shown for the earthquake events in bold represent the date of peak Twitter activity.The peak in March 2021 was omitted since the post and account were removed by the user.

Figure 4 :
Figure 4: (a) Top 50 words occurring in the text of InSAR tweets.(b) Combined top 50 words and hashtags occurring in the 50 most retweeted InSAR tweets.Wordcloud representations of the same data are shown in Figure S2.We analysed the InSAR Twitter social network for (1) the top ten tweeters of InSAR content, and (2) the top ten most retweeted tweeters of InSAR content.Therefore, capturing (1) the greatest tweeters of InSAR related tweets, and (2) the tweeters of content with the highest level of engagement.The network plot (Figure 5a, b) circle markers show common connections and

Figure 5 :
Figure 5: (a) User network showing who the top ten InSAR tweeters follow.Black lines with transparency represent individual users 280

Figure 8 :
Figure 8: Results from select questions in the LiCSAR data user feedback survey (February 2020 to February 2022, n=80).Additional survey results are discussed in the text.
(China 21/05/2021, Croatia 29/12/2020, and Greece 03/03/2021).EIDP tweets coseismic interferograms regardless of the presence of a deformation signal and five examples with no earthquake signal are shown in Figure 10b, all of which were only retweeted once.

Figure 9 :
Figure 9: Earthquake Event Data showing USGS earthquake parameters and an interactive map showing activated frames, the first 360

Figure 10 :
Figure 10: Examples of wrapped interferograms tweeted for earthquake events (magnitude, date, and USGS ID are annotated) by @COMET_database, including the five most retweeted interferograms (a) with the earthquake signal annotated with the arrow, and five interferograms without an easily-identifiable earthquake signal (b).Note that the 100 km scale bar is applicable to all panels except the Mongolia earthquake (us6000d7ix).Colour cycle fringes represent 2.8 cm of motion per cycle towards or away from the satellite.

Figure 11 :
Figure 11: Event page views for the ten most viewed earthquakes (Table 4).Vertical line shows the date that LiCSAR InSAR data 385

Figure 12 :
Figure 12: InSAR intersection with examples of natural hazard disaster management.