The Crowd4Sat Project: Augmenting Satellite Observations with Crowdsourcing

Transcription

1 The Crowd4Sat Project: Augmenting Satellite Observations with Crowdsourcing Dr. Suvodeep Mazumdar Department of Computer Science The University of Sheffield, UK In collaboration with the Crowd4Sat consortium ESA Project: AO/1-8068/14/F/MOS

2 Crowd4Sat project overview European Space Agency (ESA) funded 170,000 / 14 months recently completed (June 2016) Explore new ways and methodologies to use CS Earth Observation data validation / exploitation Demonstrate value of CS for science, applications, education and citizen engagement 4 use cases demonstration projects targeting key scientific issues across space domains Develop strategy for better exploitation of CS ESA data exploitation educational activities

3 Two main task SotA: Analysis and roadmapping existing crowdsourcing projects and communities challenges and needs raised by CS initiatives technological and community trends ways to capitalise on such opportunities for ESA and the wider industry 4 Demonstration Projects (DPs) / case studies explore opportunities for crowdsourcing and OS quality, reliability and usability of crowdsourced data differing types of CS, application domain, OS products and data Key complex technical and societal problems

4 DP1: Snow coverage Snow Covered Area (SCA) estimation valuable Accessibility and safety of transport routes and settlements Leisure activities (skiing, hiking, etc.) Avalanche prediction, etc. Snow melt is key parameter for management of water resources runoff modelling Sentinel-1A will improve SCA accuracy and revisit time but Mountainous terrains remain problematic: slant-range distortion effects foreshortening, layover, shadowing

5 DP1: Findings Engage with stakeholders, citizens and special interest groups (e.g. hikers) Shared interest Domain-specific information Interest in providing and receiving (hiking maps) information Initial stages Lot of interest from hiking communities, received maps Dissemination in public interest increased However, observations were not recorded Repeated contacts More information vs data overload? Deviation from traditional sources of information Providing information vs performing activity External factors weather?

6 DP2: Traffic and pollution Vehicles are the source of 50% emissions and 90% of the health impacts within the atmosphere. 3.4 million deaths each year costing the economy $3.5 trillion satellites only sees the full column (c.f., atmospheric inversions) CS data (via telematics) can augment OS data to improve: pollution / emissions / exposure models pollution mapping traffic management and city planning

7 DP2: Findings There are clear benefits in augmenting OS and CS data, as validated by experts (e.g. road pollution estimation) Augmenting ground based observations (e.g. road pollution data) at larger scales such as regions of cities are more helpful Smaller sections (e.g. roundabouts) are designed by planners to have minimal elevation gradients. However, over larger scale, elevation gradients are enhanced and can provide better estimates

8 DP3: Flood emergency mapping Flooding is most recurrent natural disaster causes significant damages and losses. next 70 years will see doubling in: number of people affected by flooding each year (to million) annual damages (increasing to billion) OS data and imagery used for flood mapping Request to acquisition can be 24 hours too slow for rapidly changing situation CS can bridge the gap and augment OS-derived flood mapping

9 DP3: Findings Opportunistic Crowdsourced Social Media Quick means of collecting large volumes of data Observations on the ground : practical considerations Where? Less than 1% of Tweets have geolocated information Automatic Geolocation can result in identifying information from out of the area of interest Differing contextual information Twitter vs Facebook vs Pinterest Some sources more useful than others e.g. Pinterest, Youtube Public organisations provide more relevant information compared to citizens Need to deal with duplication of information Requires post-observation analysis (time)

10 DP4: Land use Land use is key parameter in the management of water resources and the wider environment. Land cover and land cover change needed by decision-makers in the implementation of Water Management plans (2000/60/EC) Flood Risk Management plans (2007/60/EC) CORINE Land Cover (CLC) main resource 44 different classes only refreshed every 5 years 2 year delay between image acquisition and derived results CS data can improve accuracy and timeliness of the land cover information improvement of models

11 DP4: Findings Multiple levels of granularity and points of view CS observations inform situations on the ground OS inform situations from a high level Validation tasks (esp. when requiring volunteer thinking and observations) need to understand human factors of observations Can the CS participant see areas in the same scale as OS data (CORINE data)? Observe behind obstructions? Access areas of difficult terrain? (as a result impair observations) Visual inspection prone to be misclassified if categories is not distinct enough, open to interpretation and confusion even trained users

12 Lessons Learned from Demonstration Projects Taxonomy Increasing involvement DP1 DP2 Passive Volunteer Computing Volunteer Thinking DP3 Environmental, Ecological obs. DP4 Participatory Civic, comm Science Scale DP1 Increasing scale Local, city DP4 DP2 DP3 Regional Country International

13 Citizens as Sensors Worst possible sensors Often unavailable (e.g. sleep) Imprecise and temperamental Believes in rumours Politically, ideologically, personally motivated (at times) lack of common sense Processing Information, not data - Unstructured, unformalised information source Best possible sensors Large availability in time and space Highly reliable (esp. as a group) Lots of common sense Processing Information, not data understand variety of contexts

14 Engaging with Communities Practical experience Participatory crowdsourcing can be Challenging Unpredictable Influenced by external events Develop understanding of Communities Field-work together with technology development Interaction with participants community based focus groups, inperson interviews, dissemination by citizens within communities Co-Design and Iterative user-centered process Understand motivations, limitations and concerns Combine with other forms of Crowdsourcing passive, opportunistic

15 Questions & (some) Answers 1. Use of data/information from heterogeneous sources DPs showed various success / advantages 2. User needs for sharing of data/information Engagement <-> user needs 3. User needs for INSPIRE compliance and the benefits of INSPIRE??? 4. Integrate Copernicus data/information in existing business/working processes??? 5. Application in a completely different sector CS has wide ranging general application (mobility <-> football)

16 Questions? Dr Suvodeep Mazumdar Department of Computer Science The University of Sheffield, In collaboration with the Crowd4Sat consortium ESA Project: AO/1-8068/14/F/MOS