Download to read offline
![PARSEC WP2: decisions via a CNN*
•
convolutions
Average or max [+ sub-sampling]
Non linear function
(= activation function)
normalisation
*Convolutional Neural Network](https://image.slidesharecdn.com/parsec191119slideshare-191204225231/85/Parsec-191119-slideshare-36-320.jpg)
![PARSEC WP2: decisions via a CNN*
•
convolutions
Average or max [+ sub-sampling]
Non linear function
(= activation function)
normalisation
*Convolutional Neural Network](https://image.slidesharecdn.com/parsec191119slideshare-191204225231/75/Parsec-191119-slideshare-36-2048.jpg)
Uploaded byAlison Specht
Parsec 191119 slideshare
The PARSEC project utilizes satellite imagery and artificial intelligence to analyze the impact of protected areas on socioeconomic outcomes, bridging the gap between scientific research and digital resources. It aims to enhance environmental change research through interdisciplinary collaboration while tackling issues like species extinction and ecosystem degradation. The project emphasizes the importance of properly citing datasets and sharing data to improve its discoverability and reuse, providing guidance and best practices for researchers and stakeholders.
More Related Content
Similar to Parsec 191119 slideshare
Parsec 191119 slideshare
- 1. Using satellite imageryand AI to discern the effects of protected areas in your backyard, while improving the interface between scientists and the digital world (the PARSEC project). Alison Specht, SEES, University of Queensland With contributions from Shelley Stall, AGU, David Mouillot, U. Montpellier, Nicolas Mouquet, CNRS, FRB-CESAB, Laurence Mabile, U Toulouse, Marc Chaumont & Gérard Subsol, LIRMM.
- 2. The Belmont Forumis a global partnership of funding organizations, whose operations aim to encourage: International transdisciplinary research providing knowledge for understanding, mitigating and adapting to global environmental change. The Science-driven e-Infrastructure Innovation (SEI) aims to: • Enhance the impact of environmental change research by supporting technological innovation that would accelerate discovery, inform policy, and support decision making. • Enable teams of computer and information scientists and technologists to work together with natural and social scientists and related stakeholders in transnational projects. • These teams would integrate data streams and analysis systems, amalgamate best practices from public and private sectors, and foster open data and open access. Belmont Forum « www.belmontforum.org »
- 3. The Belmont Forumis a global partnership of funding organizations, whose operations aim to encourage: International transdisciplinary research providing knowledge for understanding, mitigating and adapting to global environmental change. The Science-driven e-Infrastructure Innovation (SEI) aims to: • Enhance the impact of environmental change research by supporting technological innovation that would accelerate discovery, inform policy, and support decision making. • Enable teams of computer and information scientists and technologists to work together with natural and social scientists and related stakeholders in transnational projects. • These teams would integrate data streams and analysis systems, amalgamate best practices from public and private sectors, and foster open data and open access. Belmont Forum « www.belmontforum.org »
- 4. The Belmont Forumis a global partnership of funding organizations, whose operations aim to encourage: International transdisciplinary research providing knowledge for understanding, mitigating and adapting to global environmental change. The Science-driven e-Infrastructure Innovation (SEI) aims to: • Enhance the impact of environmental change research by supporting technological innovation that would accelerate discovery, inform policy, and support decision making. • Enable teams of computer and information scientists and technologists to work together with natural and social scientists and related stakeholders in transnational projects. • These teams would integrate data streams and analysis systems, amalgamate best practices from public and private sectors, and foster open data and open access. Belmont Forum « www.belmontforum.org »
- 5.
- 6. IPBES 2019 GlobalAssessment Report Demographic and sociocultural Economic and technological Institutions and governance Conflicts and epidemics Natural ecosystems have declined by 47% on average, relative to their earliest estimated states. Approximately 25% of species are already threatened with extinction in most animal and plant groups studied. Biotic integrity-the abundance of naturally-present species-has declined by 23% on average in terrestrial communities, *Since prehistory Terrestrial INDIRECT DRIVERS ECOSYSTEM EXTENT AND CONDITION SPECIES EXTINCTION RISK ECOLOGICAL COMMUNITIES BIOMASS AND SPECIES ABUNDANCE NATURE FOR INDIGENOUS PEOPLES AND LOCAL COMMUNITIES Freshwater Marine Direct exploitation Land/sea use exchange Climate change Pollution Invasive alien species Others The global biomass of wild mammals has fallen by 82%. Indicators of vertebrate abundance have declined rapidly since 1970. 72% of indicators developed by indigenous people sand local communities show ongoing deterioration of elements of nature important to them. 47% 25% 23% 82% 72% Valuesandbehaviours
- 8. Synthesis strand: To combineremote sensing, artificial intelligence and socioeconomic data to assess change in socioeconomic conditions Data strand: To increase the number of properly cited data sets, provide credit and attribution, and accurately track data and code reuse 2018 2022Four years PARSEC project in a nutshell
- 9. Synthesis strand: To combineremote sensing, artificial intelligence and socioeconomic data to assess change in socioeconomic conditions Data strand: To increase the number of properly cited data sets, provide credit and attribution, and accurately track data and code reuse Determine the influence of protected areas on socioeconomic outcomes 2018 2022Four years PARSEC project in a nutshell
- 10. Synthesis strand: To combineremote sensing, artificial intelligence and socioeconomic data to assess change in socioeconomic conditions Data strand: To increase the number of properly cited data sets, provide credit and attribution, and accurately track data and code reuse Determine the influence of protected areas on socioeconomic outcomes Improve practices and tools for interdisciplinary projects worldwide 2018 2022Four years PARSEC project in a nutshell
- 11. PARSEC – atransnational team BRAZIL: EPUSP : P Pizzigati Corrêa NISR : J-P Ometto São Paulo U : KMPMB Ferraz SCIELO : S Santos USA: AGU: S Stall Southern Oregon U: JE Trammell U California: M O’Brien The Nature Conservancy: S Reddy, J Evans UK : H Glaves : BGS FRANCE: N Mouquet : CESAB-FRB, CNRS A Cambon-Thomsen, L Mabile, M Thomsen: Toulouse U M Chaumont, G Subsol: LIRMM J Claudet, L Thiault : CNRS L Durieux, F Sèyler : IRD D Mouillot, L Velez : Montpellier U O Hologne, R David : INRA JAPAN: Y Murayama, K Imai: NICT Y Kondo, : RIHN T Osawa: Tokyo Met U AUSTRALIA: A Specht: U of Queensland L Wyborn: NCI Associates:
- 12. PARSEC data strand Leader:Shelley Stall, Senior Director, Data Leadership, American Geophysical Union
- 13. • incentives forresearchers… • a sturdy credit and attribution infrastructure that benefits researchers… • recommendations and best practices that work for researchers to encourage data sharing and data reuse… PARSEC data strand: objectives Increase the number of properly cited data sets, provide credit and attribution, and accurately track data reuse. For this we need:
- 14. • incentives forresearchers… • a sturdy credit and attribution infrastructure that benefits researchers… • recommendations and best practices that work for researchers to encourage data sharing and data reuse… PARSEC data strand: objectives Increase the number of properly cited data sets, provide credit and attribution, and accurately track data reuse. For this we need:
- 15. • incentives forresearchers… • a sturdy credit and attribution infrastructure that benefits researchers… • recommendations and best practices that work for researchers to encourage data sharing and data reuse… PARSEC data strand: objectives Increase the number of properly cited data sets, provide credit and attribution, and accurately track data reuse. For this we need:
- 16. A. Robustly connectidentifiers across papers, people, and repositories. Currently, even if these identifiers are included, the necessary linking to allow tracking is not fully implemented. B. Conduct outreach and adoption campaigns on the importance of persistent identifiers and their infrastructure to all relevant stakeholders—these include the data repositories, publishers, researchers, and the key groups that set standards for publishers for reference tagging. C. Promote and extend data usage metrics generated by RDA’s Data Usage Metrics Working Group and data citations generated by RDA’s Scholix Working Group. By requiring that data be cited from a trusted, community-accepted repository, the value of repositories can be better measured. Data sharing through citation increases the likelihood of data discovery and reuse. D. Provide guidance to our own science-synthesis team and the selected project teams from our partners to optimize data reuse as well as data deposition of generated data for possible reuse. Demonstrate that when researchers follow the FAIR Data Principles, data are better prepared for others to understand, reuse increases, and discovery is improved. E. Promote the work of integrated guidance that will address recommendations (generic and specific to the ecological/biodiversity community) to improve each step of the process of data sharing, reuse, credit and reward for researchers and repositories. PARSEC data strand: work program
- 17. A. Robustly connectidentifiers across papers, people, and repositories. Currently, even if these identifiers are included, the necessary linking to allow tracking is not fully implemented. B. Conduct outreach and adoption campaigns on the importance of persistent identifiers and their infrastructure to all relevant stakeholders—these include the data repositories, publishers, researchers, and the key groups that set standards for publishers for reference tagging. C. Promote and extend data usage metrics generated by RDA’s Data Usage Metrics Working Group and data citations generated by RDA’s Scholix Working Group. By requiring that data be cited from a trusted, community-accepted repository, the value of repositories can be better measured. Data sharing through citation increases the likelihood of data discovery and reuse. D. Provide guidance to our own science-synthesis team and the selected project teams from our partners to optimize data reuse as well as data deposition of generated data for possible reuse. Demonstrate that when researchers follow the FAIR Data Principles, data are better prepared for others to understand, reuse increases, and discovery is improved. E. Promote the work of integrated guidance that will address recommendations (generic and specific to the ecological/biodiversity community) to improve each step of the process of data sharing, reuse, credit and reward for researchers and repositories. PARSEC data strand: work program
- 18. A. Robustly connectidentifiers across papers, people, and repositories. Currently, even if these identifiers are included, the necessary linking to allow tracking is not fully implemented. B. Conduct outreach and adoption campaigns on the importance of persistent identifiers and their infrastructure to all relevant stakeholders—these include the data repositories, publishers, researchers, and the key groups that set standards for publishers for reference tagging. C. Promote and extend data usage metrics generated by RDA’s Data Usage Metrics Working Group and data citations generated by RDA’s Scholix Working Group. By requiring that data be cited from a trusted, community-accepted repository, the value of repositories can be better measured. Data sharing through citation increases the likelihood of data discovery and reuse. D. Provide guidance to our own science-synthesis team and the selected project teams from our partners to optimize data reuse as well as data deposition of generated data for possible reuse. Demonstrate that when researchers follow the FAIR Data Principles, data are better prepared for others to understand, reuse increases, and discovery is improved. E. Promote the work of integrated guidance that will address recommendations (generic and specific to the ecological/biodiversity community) to improve each step of the process of data sharing, reuse, credit and reward for researchers and repositories. PARSEC data strand: work program
- 19. A. Robustly connectidentifiers across papers, people, and repositories. Currently, even if these identifiers are included, the necessary linking to allow tracking is not fully implemented. B. Conduct outreach and adoption campaigns on the importance of persistent identifiers and their infrastructure to all relevant stakeholders—these include the data repositories, publishers, researchers, and the key groups that set standards for publishers for reference tagging. C. Promote and extend data usage metrics generated by RDA’s Data Usage Metrics Working Group and data citations generated by RDA’s Scholix Working Group. By requiring that data be cited from a trusted, community-accepted repository, the value of repositories can be better measured. Data sharing through citation increases the likelihood of data discovery and reuse. D. Provide guidance to our own science-synthesis team and the selected project teams from our partners to optimize data reuse as well as data deposition of generated data for possible reuse. Demonstrate that when researchers follow the FAIR Data Principles, data are better prepared for others to understand, reuse increases, and discovery is improved. E. Promote the work of integrated guidance that will address recommendations (generic and specific to the ecological/biodiversity community) to improve each step of the process of data sharing, reuse, credit and reward for researchers and repositories. PARSEC data strand: work program
- 20. A. Robustly connectidentifiers across papers, people, and repositories. Currently, even if these identifiers are included, the necessary linking to allow tracking is not fully implemented. B. Conduct outreach and adoption campaigns on the importance of persistent identifiers and their infrastructure to all relevant stakeholders—these include the data repositories, publishers, researchers, and the key groups that set standards for publishers for reference tagging. C. Promote and extend data usage metrics generated by RDA’s Data Usage Metrics Working Group and data citations generated by RDA’s Scholix Working Group. By requiring that data be cited from a trusted, community-accepted repository, the value of repositories can be better measured. Data sharing through citation increases the likelihood of data discovery and reuse. D. Provide guidance to our own science-synthesis team and the selected project teams from our partners to optimize data reuse as well as data deposition of generated data for possible reuse. Demonstrate that when researchers follow the FAIR Data Principles, data are better prepared for others to understand, reuse increases, and discovery is improved. E. Promote the work of integrated guidance that will address recommendations (generic and specific to the ecological/biodiversity community) to improve each step of the process of data sharing, reuse, credit and reward for researchers and repositories. PARSEC data strand: work program
- 21. PARSEC data strand:credit and reward landscape
- 22. PARSEC data strand:partnership projects Slide Credit: Martin Fenner, DataCite, PARSEC meeting at RDA P14, 21 Oct 2019
- 23. PARSEC data strand– synthesis strand linkage
- 24. PARSEC data strand– synthesis strand linkage
- 25. PARSEC data strand– synthesis strand linkage
- 26. PARSEC data strand– synthesis strand linkage
- 27. PARSEC synthesis strand Leader:David Mouillot, University of Montpellier, France
- 28. PARSEC synthesis strand:work program • WP1: Stratified sampling of 200 rural communities close to and far from natural protected areas (PAs) using matching algorithms. • WP2: Estimate socioeconomic conditions in the selected rural communities using remote sensing and artificial intelligence. • WP3: Using paired comparison tests determine whether proximity to a PA can improve socioeconomic outcomes. Identify contributing factors.
- 29. PARSEC WP1: stratifiedsampling Step 1: Identification of suitable data for the project Selection of socio-economic systems close to a Protected Area (PA) • PA: IUCN category 1-5; >10km2; creation date 2000- 2015 • Town/village: <5000 inhabitants < 20km from PA, >100km from large city Step 2: Acquiring Data – for example • PA: Dipperu NP, IUCN category 1a, 112.05 km2, created 2014 • Town: Nebo, 840 inhabitants, circa 20km from PA, >100km from Mackay
- 30. PARSEC WP1: stratifiedsampling Step 1: Identification of suitable data for the project Selection of socio-economic systems close to a Protected Area (PA) • PA: IUCN category 1-5; >10km2; creation date 2000- 2015 • Town/village: <5000 inhabitants < 20km from PA, >100km from large city Step 2: Acquiring Data – for example • PA: Dipperu NP, IUCN category 1a, 112.05 km2, created 2014 • Town: Nebo, 840 inhabitants, circa 20km from PA, >100km from Mackay
- 31. PARSEC WP1: stratifiedsampling Further criteria for selection (a) Socio-economic status of the town so information can be derived about (for example): • Gross Domestic Product (GDP) • Human Development Index (HDI) • Child Growth Failure (CGF) • Consumption Expenditure (CE) • Asset Health (AH) Sources: UNESCO MICS surveys, World Bank studies, information that allows comparison across the countries, and beyond. Local surveys OK, but more difficult to use for comparison. (b) Image analysis across time: before and after the creation of the PA. Are the images available? In what format (Landsat, SPOT, QuickBird etc)? Assessment of socio-economic status using AI (c) What is the possibility for mirror sites?
- 32. PARSEC WP1: stratifiedsampling Further criteria for selection (a) Socio-economic status of the town so information can be derived about (for example): • Gross Domestic Product (GDP) • Human Development Index (HDI) • Child Growth Failure (CGF) • Consumption Expenditure (CE) • Asset Health (AH) Sources: UNESCO MICS surveys, World Bank studies, information that allows comparison across the countries, and beyond. Local surveys OK, but more difficult to use for comparison. (b) Image analysis across time: before and after the creation of the PA. Are the images available? In what format (Landsat, SPOT, QuickBird etc)? Assessment of socio-economic status using AI (c) What is the possibility for mirror sites?
- 33. PARSEC WP1: stratifiedsampling Further criteria for selection (a) Socio-economic status of the town so information can be derived about (for example): • Gross Domestic Product (GDP) • Human Development Index (HDI) • Child Growth Failure (CGF) • Consumption Expenditure (CE) • Asset Health (AH) Sources: UNESCO MICS surveys, World Bank studies, information that allows comparison across the countries, and beyond. Local surveys OK, but more difficult to use for comparison. (b) Image analysis across time: before and after the creation of the PA. Are the images available? In what format (Landsat, SPOT, QuickBird etc)? Assessment of socio-economic status using AI (c) What is the possibility for mirror sites?
- 34. The machine learningcomponent PARSEC WP2: estimation of socio-economic conditions using remote sensing and artificial intelligence
- 35. PARSEC WP2: machinelearning protocol • STEP 1. We « show » the « network » examples and counter- examples • STEP 2. We use the network Deep learning fish other fish other THELEARNING
- 36. PARSEC WP2: decisionsvia a CNN* • convolutions Average or max [+ sub-sampling] Non linear function (= activation function) normalisation *Convolutional Neural Network
- 37. PARSEC WP2: predictionof poverty with a CNN? Predicted poverty in Uganda Image from Xie et al. (2016) Using 400x400 pixel images (1 km x 1km) the CNN should predict a poverty value (scalar) (0 = Low ; 100 = high) Google Static Maps API, for the image 400 × 400 pixels at zoom level 16 (poverty annual consumption level of households) Predicted poverty probabilities at a fine-grained 10 x 10km block level
- 38. PARSEC WP2: Partsof the image that “react” Original daytime satellite images from Google Static Maps filter activation maps Overlay of activation maps onto original images urban areas nonurban areas water roads Jean et al. 2016
- 39. PARSEC WP2: butis this already done? From Figure 3, Xie et al. (2016) * World Resources Institute, 2009
- 40. PARSEC WP2: butis this already done? From Figure 3, Xie et al. (2016) * World Resources Institute, 2009
- 41. PARSEC WP2: butis this already done? From Figure 3, Xie et al. (2016) * World Resources Institute, 2009
- 42. PARSEC WP2: butis this already done? From Figure 3, Xie et al. (2016) Only 70% correlated to the ground truth evidence * World Resources Institute, 2009
- 43. PARSEC WP2: howto improve the prediction? • More images with more ground truthing • Work on methods using a small number of images. • Work with temporal sequences • Add diversity in the learning database: • Really poor to very rich • Various places in the world (should we create only one unique CNN or …) • Challenges • poverty data accessibility (within time-period, frequency, type) • Integration of multi-resolution, multi-source, sparsity, time integration, incomplete data, etc ..
- 44. And there ismore!
- 45. WP3: Using pairedcomparison tests determine whether proximity to a PA can improve socioeconomic outcomes. Identify contributing factors.
- 46. But that’s for2022 Thankyou!
- 47.
- 48. IPBES (2019) Summaryfor policymakers of the global assessment report on biodiversity and ecosystem services of the Intergovernmental Science-Policy Platform on Biodiversity and Ecosystem Services. Eds S. Díaz, J. Settele, E. S. Brondizio E.S., H. T. Ngo, M. Guèze, J. Agard, A. Arneth, P. Balvanera, K. A. Brauman, S. H. M. Butchart, K. M. A. Chan, L. A. Garibaldi, K. Ichii, J. Liu, S. M. Subramanian, G. F. Midgley, P. Miloslavich, Z. Molnár, D. Obura, A. Pfaff, S. Polasky, A. Purvis, J. Razzaque, B. Reyers, R. Roy Chowdhury, Y. J. Shin, I. J. Visseren-Hamakers, K. J. Willis, and C. N. Zayas. IPBES secretariat, Bonn, Germany. 39 pages. Jean M., Burke M., Xie M., Davis W. M., Lobell D. B. , Ermon S. (2016) Combining satellite imagery and machine learning to predict poverty. Science 353(6301): 790-794. World Resources Institute (2009) Mapping a better future: how spatial analysis can benefit wetlands and reduce poverty in Uganda. 39p. Washington, D.C. (USA): WRI. Xie M., Jean N., Burke M., Lobell D., Ermon S. (2016) Transfer learning from deep features for remote sensing and poverty mapping, pp 3929-3935. Proc. 13th AAAI Conference (AAAI-16). PARSEC: cited references
Editor's Notes
- #6 As you all known, advances in science, both today and in the future, will depend on the openness, accessibility and reusability of data, software, samples, and data products. This a a collective responsibility but also an opportunity for us to push forward our global intelligence to face modern challenges we are confronted to.