Swiss Re AG · SIX:SREN
Swiss Re — GainForest — Technical Presentation (2021)
39 slides from Swiss Re's technical presentation, 2021
Slide 1: Using Artificial Intelligence to help restore the Natural World
Using Artificial Intelligence to help restore the Natural World David Dao ETH Zurich Photo: © Sebastian Unrau on Unsplash 1
Slide 2: 2
2
Slide 3: Masungi Georeserve
Masungi Georeserve Ann Dumaliang, Managing Trustee of Masungi Georeserve 3
Slide 4: Kayapo People $1.2M
Kayapo People $1.2 M Needed by Kayapo Project to support guard posts (2021) 4
Slide 5: The climate and biodiversity crisis
The climate and biodiversity crisis “The natural world is deteriorating in rates unparalleled in human history” The Intergovernmental Science-Policy Platform on Biodiversity and Ecosystem Services 5
Slide 6: Deforestation rates are increasing
Deforestation rates are increasing Largest deforestation in × 14 years (Size of HK) 20% Global emissions Source: BBC with MODIS imagery 6
Slide 7: A Humanitarian Crisis
A Humanitarian > Miners Kill Indigenous Leader in Brazil During Invasion of Protected Land
Slide 8: Difficult Monitoring = Slow Payouts
Difficult Monitoring = Slow Payouts $25bn Public funds reserved for But overlapping land claims and results-based payments in slow performance measurement forestry make funds inaccessible Source: REDD+
Slide 9: Difficult Monitoring = Slow Payouts for Landowners
Difficult Monitoring = Slow Payouts for Landowners 300 USD/Ha for forest monitoring, verification and reporting 9 Source: FAO
Slide 10: We can't value what we can't measure
We can’t value what we can’t measure* *we need hundreds of billions of dollars to reverse deforestation, but the cost of failing is in the trillions 10
Slide 11: Can algorithms help us monitor nature?
Can algorithms help us monitor nature? 11
Slide 12: Training generator...
Training generator... GeneratorModel : Step 290 Global Loss : 233.069 PSNR Error : 22.8307 Sharpdiff Error: 11.8287 Training discriminator... Training generator... Training discriminator... Training generator... Training discriminator... Training generator... Training discriminator... Training gener
Slide 13: Deforestation Forecasting
Deforestation Forecasting Ground Truth Deforestation Forecast 13 D. Dao et al., 2019
Slide 14: Flood Forecasting
Flood Forecasting Explore what we built at trillium.tech/eie Earth Intelligence Engine by Natalia Diaz Rodriguez, Björn Lütjens, Océane Boulais, Brandon, Leshchinskiy, Farrukh Chishtie, Christian , Requena-Mesa, Chedy Raissi, Alex Lavin 14
Slide 15: AI on satellite imagery not enough
AI on satellite imagery not enough Public Low-Resolution Forest analytics models on satellite Imagery Deforestation forecasting Landcover classification Deforestation driver classification Landowner Automated Forest Stakeholders Monitoring and Valuation Payment for Ecosystem Services (PES)
Slide 16: We need multi-view data fusion
We need multi-view data fusion Public Low-Resolution Forest analytics models on multi-view data Imagery Tree tracking Species classification Carbon estimation Landowner Collected High-Resolution Automated Forest Stakeholders Data Monitoring and Valuation Payment for Ecosystem Services (PES) 16
Slide 17: Multi-View Data
Multi-View Data Satellite Auditability Monitoring Drone Monitoring Field peoatianteg Resolution
Slide 18: Drone-based Carbon Estimation
Drone-based Carbon Estimation Tree Height Biomass Estimation Patches of aerial imagery Lütjens et al., 2019 https://arxiv.org/abs/1912.07850 Tree Species Classification and *the values of the depicted tree have illustrative Segmentation 18 purpose and are not correct
Slide 19: Field Monitoring: Citizen Science-Based Tree Labeling App
Field Monitoring: Citizen Science-Based Tree Labeling App 1. Tree Species Prediction Predicts a tree species from its bark 2. Tree Width Estimation Predicts tree width from a single image Development lead by research team at ETH Zurich DS3Lab and Crowtherlab 19
Slide 20: Merging Drone Data and Field Data
Merging Drone Data and Field Data
Slide 21: Can we trust citizen-science data?
Can we trust citizen-science data? Public Low-Resolution Imagery Landowner / Reported High-Resolution Automated Forest Stakeholders Data collector Imagery Monitoring and Valuation Payment for Ecosystem Services (PES)
Slide 22: Challenge
Challenge • Automated Forest Validation opens up possibility of untruthfully reported imagery • Attack vectors Reported Land-Use true time wrong time true time modified image true location true location wrong location Detected Forest Cover medium high high high [1] Simona Santamaria*, David Dao*, Bj
Slide 23: TrueBranch reference
[1] Simona Santamaria*, David Dao*, Björn Lütjens*, TrueBranch: Metric Learning-Based Verification of Forest Conservation Projects, 2020 ICML Workshop (CCML), *equal contribution, https://arxiv.org/abs/2004.09725 23
Slide 24: Real-world impact
Real-world impact 24
Slide 25: Forest AI Benchmark
Forest AI Benchmark Collaboration with WWF Ecuador [1]: 100k trees of ETH + Waldlabor [1]: >3k trees of >30 >50 species from 6 project site species from 1 site [1] Dave Thau, Isabell Hillman, B. Luetjens et al. (in [2] S. Rusco, D. Dao, L. Caprez, et al. (in collection) 25 collection)
Slide 26: Empowering Conservationists with Technology
Empowering Conservationists with Technology Collaboration We are working with Crowther Lab and the Restor foundation to provide × Sharing of ecological data × Monitoring of species and progress × Transparency for donors 26
Slide 27: Climate Change AI
Climate Change AI Increasing trust for AI in decision-making and climate finance We will host a special forest event with policy makers at UN COP26 in Glasgow 27
Slide 28: From measure to value
From measure to value 28
Slide 29: 29
29
Slide 30: Developed by GainForest
Developed by GainForest Nonprofit legally based in Zurich 9 members from 5 countries
Slide 31: Machine Learning as Proof-of-Care
Machine Learning as Proof-of-Care 31
Slide 32: Crypto for good: Proof-of-care to create new digital assets
Crypto for good: Proof-of-care to create new digital assets Whitepaper release at GainForest.net 32
Slide 33: GainForest NFTrees
GainForest NFTrees NFTrees are cryptocollectible that is sold in an auction. NFTree holders pay a regular stewardship fee. 33
Slide 34: GainForest Proof-of-Care
GainForest Proof-of-Care Co-designed milestones with indigenous communities 34
Slide 35: Project Sites
Project Sites Currently in private sale Masungi Georeserve, La Cotinga, Costa Rica Phillippines Forest Foundation, Kayapo Project, Brazil Namibia 35
Slide 36: Crypto for Good
Crypto for Good Payments in Brazilian Real are paid via a network of local supermarkets 36
Slide 37: What we can do
What we can do NATIONAL ‘GEOGRAPHIC EXP LOR ERS FESTIVAL €? MASUNGI — GEORESER VE —
Slide 38: Climate + AI DS3Lab initiative
Climate + AI DS3Lab initiative Catherine Simona Marc Watine Santamaria Gyri Reiersen Cang Iveta Rott Ming Zhang Kenza Amara 38
Slide 39: Twitter: @dwddao IG: @gainforestofficial
Twitter: @dwddao IG: @gainforestofficial 39