Geo for Good 2021 Poster Gallery
Abstract
Agricultural Land Cover
The availability of satellite imagery and Artificial Intelligence has been recognized by Statistics Indonesia as one of effective data collection tools and methodology. In Statistical Mapping Division, we develop and implement methods using satellite imagery Sentinel-2 and machine learning algorithms, CatBoost (gradient boosting tree algorithms), for identifying the potential agricultural lands in order to support Census of Agriculture 2023. The algorithm able to classify the land cover not only the national level but with enhanced within country disaggregation.
Goals
Methods
Result
Conclusion
Lessons Learned
The output, the model able to acquire the average accuracy over 93.6% using F-1 score as evaluation metrics. However, the accuracy below 80% occurred in particular class such as mixed plantation class due to the high heterogeneity of pixel characteristics.
Indonesian Agricultural Land Cover
for Census of Agriculture 2023 in Statistics Indonesia BPS
Achmad Fauzi Bagus Firmansyah, achm.firmansyah@gmail.com
Wida Widiastuti, nenkwida@gmail.com
Data Acquisition
Data Labelling
Model Development
Data Production
Start
Divide Indonesia into 20x20 km grid
Acquire Sentinel 2 yearly composites and topographical data using GEE
Preprocessing and vegetation indices calculation
Grouping Indonesian grid utilizing ecoregions
Choose sampling grid 5% for each ecoregion
Create Data Training through visual interpretation and additional data
Removing Outlier for each label
Split Training 80% and Testing 20% with label stratification
Recursive Feature Elimination and Hyperparameter Tuning using Optuna
3-fold cross validation for model building
Record the performance
Post Classification using window filtering and calbirating with other national ministry maps
Classifying data using developed model
Finish
Next Step
Class | F1-score | Class | F1-score |
Rice Fields (1) | 0.9 | Wetlands (6) | 0.9 |
Non-irrigated crop fields (2) | 0.86 | Shrub (7) | 0.87 |
Palm plantations (31) | 0.89 | Pastures (8) | 0.86 |
Sugarcane plantations (32) | 0.86 | Artificial Surfaces (9) | 0.91 |
Tea plantations (33) | 0.87 | Pond (10) | 0..85 |
Other annual plantations (34) | 0.86 | Water Surface (11) | 0.96 |
Mixed plantations (4) | 0.78 | Others (12) | 0.87 |
Forest (5) | 0.95 | | |
Average F1-score for each class nationally
Result for Java Island
Sentinel 2
Composite 2020
Agricultural Land Cover
Research Background
The Yawuru People are the Traditional Owners of the land and sea Country around Broome, Western Australia. Managing ecologically and culturally sensitive Country alongside a working cattle station requires Yawuru managers to balance competing priorities to protect Country and Culture. Nyamba Buru Yawuru (NBY) have developed a nation-leading geospatial mapping program¹ to assist in negotiations and decision making.
In a novel application of the SEEA-Ecosystem Accounting², we are harnessing the accessibility and compute power of Google Earth Engine to build strategic intelligence about biophysical and cultural values across Yawuru Country. We have developed an interactive dashboard to monitor biophysical change over time for vegetation, surface water and cultural burning.
Harnessing Technology
Geospatial Intelligence for Indigenous Land Management
Anna Normyle, Bruce Doran, Dean Mathews and Julie Melbourne
anna.normyle@anu.edu.au
We acknowledge the Yawuru people, the Traditional Custodians of the lands and waters in and around Rubibi
(the town of Broome). We acknowledge and respect their continuing culture and contribution to this research.
Results
ٰ¹ Australian Human Rights Commission, 2012. Native Title Report. Sydney.
² Normyle et al., 2021. Including Indigenous Perspectives in SEEA-EA in Theory and Practice. London Group on Environmental Accounting 27th Meeting.
SEEA: System of Environmental-Economic Accounting
What do conservation & resource managers need?
Data to fill monitoring gaps...
Conservation Districts assist with ecological monitoring of agricultural areas in the USA. They report on the status of ecosystems embedded in mostly privately-owned lands. Easier access to remotely sensed data could improve the extent and effectiveness of their monitoring programs.
What is needed?
Bringing Earth Engine into ecosystem monitoring
Identifying relevant & feasible GEE analysis
Tracking landscape change
We developed scripts in the Earth Engine code editor using Sentinel-2 satellite imagery to:
Transferring capability to a broader user group
Current work and next steps
Adding Earth Engine to the conservation toolbox
Amanda T. Stahl, atstahl@wsu.edu
| |
| |
References
Stahl, A.T., Fremier, A.K., Heinse, L., 2021. Cloud-Based Environmental Monitoring to Streamline Remote Sensing Analysis for Biologists. BioScience. https://doi.org/10.1093/biosci/biab100.
User Library (under development): http://labs.wsu.edu/ecology/research-projects/cbem-user-library/
Developing a workflow with GEE and monitoring programs
WSU partnered with two local Conservation Districts to co-develop an analysis workflow for monitoring. We worked with Sentinel-2 image classification in Google Earth Engine. The products of this initial workflow have been used by three counties for reporting to the state. We summarized our collaborative process in a series of 5 steps that could be used in any setting.
Introduction
Creating a Carbon Budget
Exploring the effect of Scale
Methodology
Mapping land cover was one of the first satellite mission and the GLASS (The Global Land Surface Satellite) CDRs (Climate Data Records) project is a good example of collating LandSat AVHRR data from 1982 to 2015, but for a city carbon budget this resolution is not sufficient.
Last month the latest Land Cover product was announced, the ESA 10m world cover, with 11 distinct classes.
Besides these two, there are many more products, each with an unique combination of resolution and classification. In this study 6 different Land Cover Classification products are compared, and assessed on usability at local level, to produce the most accurate carbon budget
Process
Results
Conclusion
Carbon Budgeting
Outlook
References
Why Scale Matters - Creating a Carbon Budget
Gijs van den Dool - gvdd@everimpact.com - Connect on LinkedIn
Ankur A. Shah - ankur@mycelium.ngo - Connect on LinkedIn
Land Cover Change by spatial resolution
ESA World Cover (10m) | |
Land Cover | CO2 Amount |
Tree cover | (0.68) |
Shrubland | (0.00) |
Grassland | (0.71) |
Cropland | (0.35) |
Built-up | 2.81 |
Bare | 0.53 |
Permanent Water | 0.00 |
Herbaceous Wetland | (0.00) |
Total (over 7200 sq km) | 1.59 |
Mt (million tons)/year | |
”All models are wrong, but some are useful” George E. P. Box
Nottingham Wards - Carbon Budget
Estimated CO2 Amount in Mt (million tons) / year
Introduction
Floods
Floods are natural phenomena that have always been occurring, in some regions vital for the functioning of ecosystems and human activities such as fishing and agriculture. However, floods are also a devastating natural hazard, causing damage, displacement and loss of life, exacerbated by human-induced land use changes and climate change.
Flood Mapping
Flood maps (forecasting, near real-time or historical) can be derived from models and/or earth observation (EO) data. We will focus on those derived from EO. Regardless of the method, they can provide valuable insights into the geographic extent and progression of a flood. There are currently many providers of EO-derived flood mapping products (too many to list here), which on its own poses a problem for many end-users (e.g. governments, NGOs, humanitarian agencies), as they cannot be expected to understand the strength and limitations of each.
Actionable information needs
However, even a perfect flood map does not provide actionable information on its own, as this requires:
This is often lacking or handled in a too simplistic manner for it to provide information on which important decisions can be based.
Method
We use the HYDrologic Remote sensing Analysis for Floods (HYDRAFloods) tool to create flood maps, which utilizes Google Earth Engine [1,2]. This is developed within the SERVIR-Mekong program, a partnership between USAID and NASA. The tool is operational over parts of Southeast Asia and has also been used ad-hoc in Central America. Research projects have been carried out in several other countries (e.g. Myanmar, Ethiopia).
In order to provide context to these maps we consider flood drivers (e.g. precipitation, topography) and historical water presence, through existing sources (e.g. JRC GSW, locally available data) and/or by creating custom maps. Issues or uncertainties can arise from:
Impact is often handled by overlaying exposure datasets (e.g. land use, population, infrastructure), but this also introduces many uncertainties. This is related to the same topics already mentioned above, but it also often misses information on flood depths and velocity, which are critical when looking at impacts (but notoriously hard to obtain from EO data).
Results
We show that the inclusion of reference data provides more detailed information and provides context, which helps end-users to better understand the flood maps.
We have also shown that impacts can vary widely based on the used exposure data, even without considering flood depths or velocity. Using a single data source can provide a false sense of security, but the uncertainty that becomes visible when using multiple sources also poses problems, as it can make it difficult to understand for end-users.
Conclusion
Context
This requires knowledge of the area, particularly hydrological expertise, preferably at the regional or local level.
So you have a flood map...now what?
Arjen Haag, arjen.haag@deltares.nl
References
Impact
This requires further research into best practices, new exposure datasets and EO-derived flood information.
Floods in Myanmar, late July 2018. Photo: MOI Myanmar
SIKU: the Indigenous Knowledge Social Network, is a powerful platform developed by Inuit to support Indigenous self-determination, safety, food security and knowledge transfer.
The SIKU app makes it easy to create posts with photos, tag
Inuktut sea ice terminology, and support knowledge and
language transfer in communities! Learn more
Method
C) Ice Watch Challenge
D) Next Up: Machine Learning and Sea Ice Safety
When the computer and the harpoon work together: Using Inuit Knowledge, machine learning, Earth Engine, and satellite radar to support sea ice safety through SIKU: The Indigenous Knowledge Social Network
People from across Inuit Nunangat took part in our first annual Ice Watch Challenge, making 295 ice posts across 32 communities. The posts documented dangerous ice, shared Inuktut terminology, and will be used to improve the SIKU Ice Map. Learn more
B) SIKU Ice Map: Inuit Knowledge, Sea Ice Charts, Satellite Radar
Our top priority is to use machine learning to find and display ikiraasak (polynyas - open water surrounded by sea ice) located within the landfast ice, followed by other ice features important for safe travel like apputainaq (open water hidden by snow) and siqummaq (mobile cracks). This project is just beginning, and will use two main data sources: RADARSAT Constellation Mission radar imagery and observations of polynyas. Interested in learning more or contributing? Contact us!
The SIKU Ice Map (top) highlights differences between a) landfast ice / mobile ice, b) growing (thinner) ice / thicker ice, and c) rough ice / smooth ice.
The SIKU Ice Map is designed to bring together the best information to help northern communities access information about ice. It currently uses SIKU Ice Posts, ice charts from the Canadian Ice Service and radar imagery from the Sentinel-1 satellites, which are combined using Earth Engine. We are working to build more Inuit Knowledge into the SIKU Ice Map by using SIKU Ice Posts made by Inuit sea ice experts to train machine learning models! Learn more
Introduction
Taking soil moisture to the next level
Soil Moisture is a key parameter in many environmental disciplines as well as it is crucial for practical decisions in agriculture. However, recent soil moisture data either lack spatial or temporal resolution, their application requires an advanced technical background or they need an extensive computing infrastructure. In this context, I am developing a soil moisture product within the Google Earth Engine that is both high resolution and easy to use, making practical agricultural decisions possible and overcoming the aforementioned problems.
Objectives:
Method
Results
The high resolution soil moisture product based on the Sentinel-1 satellite mission gives reasonable results compared to in situ measured data. The error between estimated and in situ soil moisture was evaluated over the Rur catchment and ranges between 2.6 % and 7.0 %, depending on the vegetation coverage. Using the Earth Engine, both high resolution products for field-scale use (10 m) as well as mid-resolution products for catchment-scales (200 m) can be computed fast and on demand.
Conclusion
The Next Steps
References
High Resolution Soil Moisture Mapping with Earth Observation
David Mengen, mengendavid@gmail.com
The method is based on the alpha approximation method proposed by Balenzano et al. [1], assuming that between two consecutive Synthetic Aperture Radar (SAR) recordings, a change in backscattering signal is caused mostly by a change in soil moisture, as it has the fastest rate in temporal change compared to other surface parameters. The algorithm is embedded into an automated workflow, using the cloud-processing platform Google Earth Engine. As input data, the Sentinel-1 A/B SAR image collection is used. To reduce the time between consecutive observations, all available orbits as well as both ascending and descending acquisition modes can be used. Looking a bit ahead, the method is also developed for the upcoming L-band SAR satellite mission ROSE-L from the European Space Agency.
Step-by-step
Introduction
Watershed prioritization is the classification of a catchment’s various sub-watersheds according to the order in which soil management and conservation steps must be taken.
Need of proper soil and water management treatments to overcome continuous soil loss by runoff and other ecological problems in Kosi watershed because of lack of capability to withstand it. Also, speedy land use modification as it undergoes issues of abrupt bank erosion and high sedimentation, stream displacement, soil fertility degradation occurs throughout the basin has raised the risk of soil erosion.
Kosi watershed situated in Almora and Nainital (central-eastern mountainous regions of Kumaon area of Uttarakhand) was preferred for this study.
OBJECTIVES
Materials and Methods
Study area description:
Geographic location – 29°21’N to 29°52’N Latitude and 79°3’E to 79°52’E Longitude.
Total area - 1842.56 sq.km.
Altitude -363 m to 2732 m a.m.s.l.
Topography - Mountainous regions
Step-by-step
Results and Discussion
In the VIKOR model, sub-watershed 17, 13, and 19 with the highest results (0.982, 0.967 and 0.649) are most susceptible to erodibility, and sub-watersheds 6, 20, and 11 divulging lowest score (0.011, 0.019 and 0.088) have shown the least erosion susceptible.
Sub-watersheds 17, 13, and 19 with the maximum value (0.811, 0.803 and 0.784) of rank 1 to 3 are most vulnerable to erodibility, according to the outcome of prioritization of sub-watersheds in terms of their vulnerability to erosion using the TOPSIS process. Contrarily sub-watersheds 6, 20, and 11 have demonstrated the least vulnerability to erosion, with ratings of 0.241, 0.249, and 0.387, respectively.
Conclusion and Recommendation
References
VIKOR and TOPSIS based Prioritization of Kosi Watershed using Geomatics
Purabi Sarkar, Pankaj Kumar, A. K. Sharma and S Sharma spurabi08@gmail.com
Healthy vs. Eutrophic Ecosystem
Introduction
An iterative app for coastal water managers to detects symptoms of coastal eutrophication solely using satellite-derived chlorophyll-a (CHL) concentration. The app can detect areas with severe symptoms of eutrophication and those with improving water quality and is relevant in tracking the progress against Sustainable Development Goals.
What was the motivation?
How the assessment is done in the app
Results
Conclusion
Lessons Learned
References
Globally consistent assessment of coastal eutrophication
Elígio de Raús Maúre1✉, Genki Terauchi1, Joji Ishizaka2, Nicholas Clinton2 & Michael DeWitt3
1Department of Research and Study, Northwest Pacific Region Environmental Cooperation Center (NPEC), Toyama, Japan. 2Institute for Space-Earth Environmental Research, Nagoya University, Nagoya, Japan. 3Google LLC, 1600 Amphitheater Parkway, Mountain View, CA, USA. ✉email: eligiomaure@gmail.com
Trends in annual CHL max based on Sen’s slope method (Sen, 1968) at 90% significance level. Polar regions with a few observations (< 70% of the study period) were masked.
Baltic Sea
Gulf of Mexico
Comparative eutrophication assessment in the Bohai Sea: a (1998-2015) and b (1998-2019)
Eutrophication assessment in the Baltic Sea (c) and Gulf of Mexico (f) based on the global eutrophication watch .
Spatial distributions of bottom hypoxia and anoxia over time (Carstensen et al. 2014)
A satellite image during summer 2006 showing high concentrations of phytoplankton in reds and greens (g) (Miller & Spoolman, 2014)
e
d
c
f
g
h
The world’s third largest oxygen depleted zone (after the Baltic Sea and the northwestern Black Sea). Oxygen < 2 mg L−1 (h) (Miller & Spoolman, 2014)
Conceptual diagram comparing a healthy system to unhealthy system exhibiting symptoms of eutrophication (Bricker et al. 2007)
The global eutrophication watch app on the left and the animation of eutrophication assessment steps in the Northwest Pacific region on the right
Introduction
COVID-19 Breakthrough Hospitalizations & Deaths
Data on COVID-19 breakthrough infections, when fully inoculated people contract the virus, is relatively new. There is some evidence that these cases are becoming more common as the more transmissible COVID-19 Delta variant surges. The US Centers for Disease Control and Prevention estimated that fully vaccinated people accounted for less than 3% of coronavirus hospitalizations nationwide and less than 1% of virus deaths. Looking at how many hospitalizations and deaths have involved fully vaccinated people is a common, but crude measure of how well the vaccines are working.[1]
Rationale
Methodology
The New York Times asked all 50 states and Washington, D.C. to provide data on breakthrough infections roughly spanning back to the first months of the COVID-19 vaccination campaign at the end of 2020 and beginning of 2021. Florida and Arkansas, while in the Southeastern US, did not provide data. [1]
Data from State health departments (breakthrough hospitalizations, breakthrough deaths and some total deaths), U.S. Department of Health and Human Services (hospitalizations), New York Times database of cases and deaths (some deaths) was used. [1] A Google sheet (shown below) was created highlighting the number of breakthrough COVID-19 hospitalizations & deaths from the NY Times data combined with publicly available US Census Bureau GIS state data. [3]
This combined information was then uploaded and styled accordingly for display using Google My Maps.
Results
Key: (State, # of Breakthrough COVID Deaths)
States are shaded based on the percentage of Breakthrough COVID-19 Deaths as compared to all COVID-19 Deaths. Darker shades indicate greater COVID-19 mortality (death) rates.
These results show us that the percentage of COVID-19 breakthrough deaths in select states in the Southeastern US are relatively low in the population of people who have been vaccinated against COVID-19.
View the full size version of the map here.
Conclusion
Recommendations
References
Breakthrough COVID-19 Mortality
in Select Southeastern US States, 2021
Elizabeth Carter, elizabeth@publichealthmaps.org & Edward Paul Vallejo, edward@publichealthmaps.org
Limitations
For more information on COVID-19 and COVID-19 vaccines, please visit the US Centers for Disease Control and Prevention’s COVID-19 Vaccine Page and the US Department of Health and Human Services’ “We Can Do This” COVID-19 Public Education Campaign
Introduction
Sexual Violence is a pandemic
Method
Results
Conclusion
References
Safecity - a crowdmap for sexual and gender based violence
ElsaMarie DSilva, elsamarieds@gmail.com
Non verbal
Verbal
Physical
The data can be used by:
Police
Added insights into specific issues and locations
School and college administration
New data set which they may not have
Prevention of sexual violence on campus
Elected representatives
Focussed spending on budgets
Legal advocacy
Influencing policy, amending legislation or advocacy for new legislation where required
INDIVIDUALS
- can make better choices regarding their own safety
COMMUNITIES
INSTITUTIONS
35000
reports
10 country chapters
1 Million
Citizens
Engaged
5 Police
Partners
40000
People trained
2000 Youth leaders
Achievements
Introduction
What is Accessible Life?
Accessible Life in Google Earth
is a visual search system that allows you to explore the earth to discover accessible outdoor locations.
The search - completely visual - takes place crossing successive layers until you reach the area you are looking for.
Accessible Life is a collective Map, built by contributors like us who want to help others.
Our Goals
How I Did it
I started by creating a standard structure in Google Earth, replicating it to create several interconnected layers.
In the next step I prepared the standard structure for Google My Maps, so I worked to have a coherent set of instructions to give to contributors to ensure a standard and consistent result. When everything was ready, I contacted some friends Local Guides, to see if they were able to create a map, applying the required standard, without further instructions.
Our tools
Results
Conclusion
Where we are, Where we want to go
Ermes Tuon Poster: Accessible Life in Google Earth
Ermes Tuon, ertuon@gmail.com
We have currently mapped Accessible Places across four continents (Asia, Europe, North America, South America), and the map is constantly growing with new contributions.
The interaction with the contributors also allows to evolve and refine the system, adding new details to increase the ease of use by users.
Introduction
Introduction
What inspires me: The successful project from Ermes Tuon (ertuon@gmail.com) working on Google Earth inspires me to learn, to open to a new perspective in different ways, to challenge, and to participate.
My Goals
We have a beautiful State, Michigan USA, with so much to offer. If I can add more info about accessibility via Google Earth to make it just a click away I will. Let’s have the individual with the disabilities to determine if the location or parks is accessible for him/herself.
How I Did it
How I did it?
The tools to use:
- Google My Maps: is the tool to use to create
the map
- Google Earth: is the search engine where my
maps are linked to.
Results
Conclusion
Conclusions:
This will be a wonderful way to access location with Google Earth and I strongly believe that in 5 years or more to the future, Google Earth will be the hot spot in the technology industry. A lot of works but fun to do!
Erna LaBeau Poster: Accessible Life in Google Earth
Everyone who has the link will be able to see the beauty of the parks in Google Earth. Every Point of Interest will be provided with photos and explanations. This way the viewer can imagine what to expect if they are to visit the location.
Introduction
Climate change awareness
We wanted to educate and support our communities on the growing challenge of climate change and urban resilience. This include increased vulnerabilities to flooding and rising temperature. In 2019, we developed a Nature-based Climate Adaptation Programme for the urban areas of Penang Island (first in Malaysia). We supported the programme through data-driven approach in our climate research and demonstrative projects which include our study of Urban Heat Island effects in Malaysian cities.
Our goals
How We Did It
We chose five highly developed and populated cities in Malaysia for the study to assess the change in their surface temperatures between two different periods. Using GIS and Google data platforms, we were able acquire open-source satellite data captured by the United States Geological Survey (USGS).
Impacts and Implications
Conclusion
Lessons learned
References
Cities are becoming hotter
Think City study: Malaysian cities got hotter over the decades; KL had lowest increase, Ipoh the most
FRIDAY 5 MARCH 2021
Perak moves to reduce urban heat
MONDAY 8 MARCH 2021
Adaptation Fund will help Penang reach its green development goals
SUNDAY 14 MARCH 2021
Malaysia’s getting hotter. Can its leaders rise to the climate challenge?
THURSDAY 18 MARCH 2021
Step-by-step
Introduction
ABMI’s Mission
The Alberta Biodiversity Monitoring Institute’s mission is to track changes in Alberta’s wildlife and their habitats and provide ongoing, relevant, scientifically credible information for decision makers, land stewards, and all Albertans. Part of this mission is to produce open source and openly available geospatial landcover products. The Alberta wetland inventory is the culmination of three major regional wetland mapping initiatives and was published in March 2021.
Objectives of Alberta wetland mapping
Method
The wetland classification was based on open source Sentinel-1 and Sentinel-2 imagery compiled in Google Earth Engine, ALOS DEM 30 m topographic data, and ABMI’s “photoplots” : eight hundred 3 by 7 km photointerpretation blocks spaced across Alberta which were used to train the classification models. The processing was done in Google Earth Engine, SAGA GIS, ArcGIS, R, and Python. Different object oriented modelling approaches were used for each of the three major regions.
Classification Models
Results
Conclusion
References
The Making of the Alberta Wetland Inventory
Fiona Gregory, Evan DeLancey, John Simms, Agatha Czekajalo, Jahan Kariyeva fgregory@ualberta.ca
Please replace these image placeholders with images related your work or project, or remove completely if you want to use this space for text.
ABMI’s Photoplot Interpretation Data
Image Processing in Google Earth Engine
SNIC segmentation for isolated wetlands in the Rocky Mountains
SNIC segmentation in southern Alberta prairie pothole region
Convolution Neural Network (CNN) model for the Boreal Forest region
Collaborators
Map by Evan DeLancey, ABMI
Alberta, Canada
Rationale
Growing A World Wonder
“Sustainably managed and restored trees and forests in Africa’s drylands have the potential to reverse land degradation and therefore contribute to poverty reduction, food security, biodiversity and climate change mitigation and adaptation.” (Sacande et al., 2020) The Great Green Wall of Africa is a transcontinental afforestation project spanning 8,000km long and 15km wide. It was launched in 2007, lead by the African Union in collaboration with other partnering institutions such as the United Nations Convention to Combat Desertification. The project aims to plant trees across the Sahelian region (the route passes through 11 countries) to reverse desertification processes and combat Climate Change.
Methods
Due to the transcontinental scale of the project, Earth observation satellite data was necessary for this study. Remote sensing methods provided the tools to assess vegetation dynamics across such a large study site, and at the time no monitoring and evaluation research on this project using remote sensing data had been published.
Results
Key Findings
Conclusion
References
A Novel Open-Source Remote Sensing Tool To Improve Monitoring and Evaluation of the Great Green Wall of Africa
Gemma Newbold, gemma@justdiggit.org
Challenges limiting progress towards the project’s ambitious targets
Based on a recent assessment, many challenges (financial, technological, biophysical, governance, and monitoring) to the implementation of the Great Green Wall of Africa have meant the project is not on track to meet its ambitious 2030 targets, which include:
Therefore, there was a research gap for a monitoring tool to provide temporal and spatial insights into the success of the project so far, and where more funding or technical expertise needs to be directed.
Benefits of the project
Dataset
Vegetation Index
MSAVI
The Modified Soil Adjustment Vegetation Index, which was chosen for the analysis, performs better than NDVI in sparser canopies, as it corrects for soil background reflectance.
MODIS Terra
Methodology
A ‘greened’ pixel is defined, for the purpose of this analysis, as a pixel which has a positive difference in MSAVI value when subtracting its 2007-2009 average from its 2018-2020 average. This greening could have occurred from natural causes or human intervention.
Background
The demographics of trees
Many studies have explored the relationship between forest and tree cover (Riley and Gardiner 2020) and green spaces (Dai 2011) to different demographics such as race (Lu et al., 2021) and income (Holt and Borsuk 2020). In this study, I looked at the relationship between forest cover, race and income in my home state of Virginia.
Typically, studies conclude that people who are white, asian, and/or higher income (Pearsall and Eller 2020) reside in areas with higher percentages of trees and green spaces. As of 2020, Virginia has a population of over 8.6 million people; of which 67% are White, 19% are Black, and 6% are Asian [1]. In 2019, the overall median household income for the state was $74,222 [2].
Objectives
I wanted to look at how forest cover relates to race and income in Virginia. Do the relationships that past papers have shown hold up here? If not, how much do these relationships differ?
My objective is:
How to
To determine the relationship between tree cover, race, and income, the following datasets were used:
The Forest Cover raster data was summarized by census tracts to create a total forest cover % by census tract. This way I could see the total % forest cover, race, and income for each census block.
The relationships between these three variables were done by running correlations between the three variables using the ggcorrplot [4] and ggpubr [5] libraries in R.
The correlation of these variables tells me the type of relationship (positive or negative) that race and income has against forest cover for each census tract.
Results
Conclusion and next steps
Final thoughts and where to go from here:
There are only a few of the ways I would like to improve upon this study and am very interested in hearing your perspective!
References
[1] Virginia population and race percentages: https://worldpopulationreview.com/states/virginia-population
[2] Virginia overall median household income: https://www.census.gov/quickfacts/fact/table/VA/AFN120212
[3] Hansen Global Forest Cover rasters: https://earthenginepartners.appspot.com/science-2013-global-forest/download_v1.7.html
[4] Ggcorrplot: Alboukadel Kassambara (2019). ggcorrplot: Visualization of a Correlation Matrix using 'ggplot2'. R package
version 0.1.3. https://CRAN.R-project.org/package=ggcorrplot
[5] Ggpubr: Alboukadel Kassambara (2020). ggpubr: 'ggplot2' Based Publication Ready Plots. R package version 0.4.0.
The Demographics of Trees in Virginia
This flowchart shows the general process I used to create my final map of race vs forest cover and income vs forest cover.
Mean forest cover had the highest correlation with White identifying populations.
% Mean Forest Cover by Census Tract
Population by Census Tract
Median Household Income by Census Tract
Introduction
Why need to create a landslide imagery database
Landslides are common natural disasters around the globe. Understanding the accurate spatial distribution of landslides is essential for landslide analysis, prediction, and hazard mitigation. So far, many techniques have been used for landslide mapping to establish landslide inventories. However, these techniques either have a low automation level (e.g., visual interpretation-based methods) or a low generalization ability (e.g., pixel-based or object-based approaches); and improvements are required for landslide mapping.
What was our goal
Method
We used Google earth engine as backend to enable users to search on NAIP imagery on Google map and discover/predict the landslide by drawing the bounding box in the map interface.
Our tool pipeline
Results
Our web portal is developed based on the google cloud platform. It provides a user-friendly interface to improve landslide mapping efficentcy. The deep learning model guides users’ attention to landslide- suspicious area in the image. Users can review and correct machine labeled landslides using the labeling tool.
Conclusion
Utilizing an interactive AI-empowered web portal for landslide labeling
Guanlin He (gbh5146@psu.edu) ,Te Pei, Savinay Nagendra, Srikanth Banagere Manjunatha, Daniel Kifer, Tong Qiu, and Chaopeng Shen
Department of Civil and Environmental Engineering, Pennsylvania State University
Department of Computer Science, Pennsylvania State University
Please replace these image placeholders with images related your work or project, or remove completely if you want to use this space for text.
Please replace these image placeholders with images related your work or project, or remove completely if you want to use this space for text.
.
Introduction
Mapping Habitat Loss
Agricultural expansion is a major driver of habitat loss and threat to biodiversity in sub-Saharan Africa (Tilman et al., 2017). Accurate land use maps help conservation planners monitor ecosystem health and manage protected areas. In Angola, however, thousands of landmines have made mapping agricultural expansion on foot impossible. And a dynamic savanna ecosystem, structured by annual fires, means that well-vetted techniques used to map habitat loss in forests do not transfer. Google Earth Engine’s global imagery database and machine learning algorithms provided a unique opportunity to create a new, more accurate habitat loss map for the Angolan miombo woodlands ecosystem.
What was our goal?
Method
Methods
Results
Figure: Distribution of woodland (dark green), grassland (light green), agricultural areas (pink) and urban areas (red) in the Angolan miombo woodlands, mapped using GEE with 86% accuracy.
Agriculture has expanded significantly (by 5%) since 2013, driving loss of savanna woodland and grassland habitat. Rapid expansion is occurring (1) along major roads (2) around regional cities and (3) along an agricultural frontier advancing northeastward from the Angolan highlands—the large agricultural belt in the region’s west.
Conclusion
Key Points and Opportunities for Future Study
References
Mapping Post-Conflict Habitat Loss in Angola’s Woodlands
Imma Oliveras Ph.D., Ty Loft ty.loft@bnc.ox.ac.uk
Please replace these image placeholders with images related your work or project, or remove completely if you want to use this space for text.
A chameleon crossing the road in the miombo woodlands. Photo credit: Ty Loft
Introduction
A cash crop amongst many...
West Africa has lost 90% of its original moist forest and what remains is heavily fragmented and degraded (Leach and James, 2000). Cocoa farms, amongst others, such as oil palm, rubber and coconut are the major drivers of deforestation in Ghana and Côte d’Ivoire (Ruf, 2015). Ghana has lost more than 2.5 million hectares (Mha) (33.7%) of its forest since the early 1990s (Fuglie et al., 2013). The statistics for Côte d’Ivoire are even worse, as it has experienced rapid deforestation since the mid-1950s (Barraclough and Krishna, 2000) and this is partially due to the increase in cocoa demand (Nalley et al., 2014). It was recorded that between 2000 and 2013, cocoa plantations in West Africa were responsible for 57% of annual global expansion, and in 2013, 6.3 Mha was allocated to cocoa cultivation in the region (Ordway et al., 2017).
Protected areas (PAs) constitute a major conservation tool for protecting forests and the biodiversity they shelter. In tropical Africa, management effectiveness of several PAs is deficient, and that PAs continue to be exposed to threats such as wildlife hunting, logging and agriculture.
Cocoa farming is one of the agricultural activities threatening PAs in tropical Africa.
What was my goal?
How I Did It
An approximate 20 degree belt either side of the Equator with humid tropical climates with regular rain and a short dry season is favourable to the growth of cocoa (Kyei et al., 2011). As shown in this Figure, climatically the most suitable cocoa areas in Ghana (GH) are mainly in the Eastern, Central, Ashanti, Western and southern Brong-Ahafo regions, while in Côte d'Ivoire (CIV) they are mainly in Sud-Comoé (Comoé since 2011), Agnéby (Lagunes since 2011), Moyen Comoé (Comoé since 2011) and Sud-Bandama (Bas-Sassandra since 2011) districts (Läderach et al., 2013). Accordingly, we focused our mapping efforts on these regions and districts and neighbouring regions.
Result & Discussion
The area of cocoa recorded for cote d’Ivoire was 3.69 Mha and 2.15 Mha for Ghana. The cocoa plantations detected in close proximity to Protected Areas (PAs) may correspond to the cocoa agroforestry systems which have been suggested as a management strategy to reduce the influence of surrounding land-use activity on PA biodiversity. Although our study does not enable the direct investigation of how effective this strategy is in mitigating the effects of land-use activities on PA biodiversity, we clearly show that this strategy is ineffective in preventing the encroachment of cocoa plantations into PAs. The PAs housing cocoa plantations are concentrated in the southern parts of Côte d’Ivoire and Ghana, corresponding to the cocoa belt, and 105 PAs have cocoa plantations exceeding 35% of their area.
Conclusion
Detecting cocoa plantations in Côte d’Ivoire and Ghana and their implications on protected areas
Itohan-Osa Abu, Zoltan Szantoi, Andreas Brink, Marine Robuchon, Michael Thiel
itohan-osa.abu@uni-wuerzburg.de
For more information please scan the barcode
The dataset consists of median composites of S-1 (VV and VH) and S-2 (13 bands) for the year 2019 (Figure 2). The size of the median window was set to the twelve-month period lasting from 1/1/2019 to 31/12/2019 because it was the minimum window size that could produce a cloud-free composite for S-2 in the study area. After initiating these steps, the S-1 composite was created for deriving S-1 derivatives and layer stacking with S-2 composite’s derivatives for the classification. Based on the S-2 composite, the NDVI calculation and Tasseled Cap Analysis (TCA) has been performed using functions written in GEE. Oil palm plantations (Descals et al., 2019), settlements (JRC Global Human Settlement Layer) and water bodies (JRC Global Surface Water Explorer) were masked out from the S-1 and S-2 composite and the derivatives.
Hansen, Matthew C., et al. (2013)
The climatically most suitable cocoa regions in Ghana and districts in Côte d’Ivoire (Läderach et al., 2013).
Cocoa plantations map for Côte d’Ivoire and Ghana representing the year 2019.
Fig. 8. Estimated area of cocoa plantations (%) within protected areas. NP = National Park, FR = Forest Reserve, RR = Resource Reserve, NR = Not Reported. 1. Marahoué NP, 2. Niégré FR, 3. Dassieko FR, 4. Monogaga FR, 5.Mabi FR/Yaya FR, 6. Séguéla FR, 7. Rapide Grah FR, 8. Kani-Bandaman NR, 9. Azagny NP, 10. Sassandra FR, 11. Haut De, NR, 12. Bolo FR, 13. Bossematie FR, 14. Mont Peko NP, 15. Moyenne Marahoué NR, 16. Koba FR, 17. Port Gautier FR, 18. Bouafle FR, 19. De FR, 20. Ngadan Ngadan FR, 21. Banco NP, 22. Ile Ehotilé NP, 23. Bia Tawya FR, 24. Tai NP, 25. Anhia NR, 26. Bia NP, 27. Tano Ofin FR, 28. Asenanyo FR, 29. Bia North FR, 30. Manzan FR, 31. Sukusuki FR, 32. Ankasa RR.
The Why
Alberta’s Dynamic Landscapes
The landscapes of Alberta, Canada, are ever-changing. In addition to natural disturbances (e.g., wildfire, insect defoliation), human land uses like resource extraction, agriculture, and urban expansion lead to continuous changes. Up-to-date mapping and monitoring are key to effective land use planning, and sustainable resource management.
Existing and comprehensive data on human footprint (HF) in the form of province-wide digital Human Footprint Inventories [1] are a valuable source of information (see map to right). They treat all features equally, however, regardless of the temporary nature of some features (e.g., forest harvest areas).
Our goal is to enhance current HF data for the province of Alberta with remote sensing-based information on vegetation recovery.
How We Did It
Our Results
Key Takeaways
References
Mapping post-disturbance recovery on human footprint features in Alberta, Canada
Jen Hird, jennifer.hird@ucalgary.ca, @JNHird
Earth Engine & the Landsat Archive
We leveraged the long-running Landsat archive using Google’s Earth Engine platform to extract spectral recovery trajectories for several types of polygonal HF features in Alberta. In short, we:
Forest Harvest Areas (Alberta)
Well Sites (Oil Sands)
Reclaimed Mines (Oil Sands)
The majority of harvest areas in our provincial dataset (n = 57.8K; harvested 1989-2013) have recovered their pre-harvest NBR spectral signals in 2018.
Reclaimed mine areas in our Alberta Oil Sands dataset (n = 377) show highly variable levels of spectral recovery in 2018.
Active well sites in our Alberta Oil Sands dataset (n = 545) show some variety in levels of spectral recovery in 2018, but a peak around 65-86% spectrally recovered.
Abandoned well sites in our Alberta Oil Sands dataset (n = 3427) show less variability in spectral recovery in 2018, with an even higher peak between 65-86% spectrally recovered.
𝜇 = 72.8%
𝜎 = 22.8%
𝜇 = 105.2%
𝜎 = 20.9%
𝜇 = 76.3%
𝜎 = 25.3%
𝜇 = 66.7%
𝜎 = 24.7%
We Learned That:
Spectral recovery varies for different HF features, but shows some expected patterns (i.e., lower for active wellsites, higher for regenerating harvest areas).
Main Contributions:
What’s Next? Analyses of environmental driving and limiting factors, and comparisons with ground observations to better understand our recovery metrics
Introduction
Coastal wetlands
The long-term fate of the mud dominated coastline of Suriname, part of the Guianas coastal system stretching from the Amazon river to the Orinoco delta, is influenced by migrating subtidal mudbanks. Their alongshore movement causes a cyclic instability of alternating erosion and progradation phases. For climate resilient management strategies, that need to to account for changes that might occur in the coming decades, it is critical to incorporate spatial and temporal variability of coastline dynamics that reflect the complex interplay between controlling factors such as mudbank migration.
Mudbank cyclicity
Method
Remote Sensing
We use a remote sensing approach (de Vries et al., 2021) to quantify the influence of mudbank migration on local coastline dynamics, along the entire coast of Suriname between 1985 and 2020 from medium resolution Landsat images that are available in Google Earth Engine.
For subtidal features, resuspension of mud and migration processes are responsible for the spatiotemporal variability of their footprints. Abrupt changes in mud abundance are thus considered as fuzzy transitions that correspond to mudbank boundaries. Accordingly, image analysis techniques that involve semi-automatic unmixing of cover fractions can be used for the analysis of these gradients. Simultaneously, the NDWI index is used to detect the interface between land and water with an Otsu thresholding approach (Donchyts et al., 2016).
Results
Conclusion
References
Coastline Dynamics Controlled by Migrating Subtidal Mudbanks from Remote Sensing Images in Google Earth Engine
Job de Vries
Unsupervised Decision Tree (UDT)
Faculty of Geosciences
Departhment of Physical Geography
j.devries4@uu.nl
https://www.uu.nl/staff/JdeVries5
Acknowledgements
The MangroMud project is funded by NWO-WOTRO (W 07.3030.106) and Utrecht University.
Figure 3.The presence and absence of mudbanks in relation to observed coastline changes in Suriname between 1986 and 2020.
Figure 1.The coastal area of Suriname with the different coastal sections separated by the two major rivers.
Figure 2. The applied workflow to determine the coastline position and mudbank presence.
Figure 4. Distribution of the rates of coastline change between 1985 and 2020 for all positions along the coast of Suriname that are either fronted by a mudbank or not.
Introduction
UNI Tallgrass Prairie Center Programs
The University of Northern Iowa (UNI) Tallgrass Prairie Center (TPC) is a leading native prairie proponent in the midwestern United States and has a demonstrated 30-year commitment to prairie reconstruction, restoration, management, and advocacy. The primary programs of the Center are Research and Restoration, Iowa Roadside Management (IRM), Plant Materials, and Prairie on Farms (POF).
Overview
Method
In collaboration with the TPC, the UNI GeoTREE Center planned collection of 360° photos at a variety of Prairie on Farms and Iowa Roadside Management locations. GeoTREE Center student research assistants collaborated with TPC Americorps members to leverage those photos and Google Earth technologies to build a suite of virtual resources.
Methods
Results
There were 191 photos published to Google Street View while multiple virtual resources were developed for both the Iowa Roadside Management and Prairie on Farms programs. These virtual resources include POF and IRM flyovers and IRM Google Earth Projects. The GeoTREE Center also developed a custom web mapping application using Google Maps JavaScript API and open source 360 photo viewer.
Conclusion
Accomplishments and Acknowledgements
Google Earth Technologies to Communicate Iowa Prairie Stories
John DeGroote, john.degroote@uni.edu
Introduction
Grasslands are important for global biodiversity, food security, and climate change analyses. Yet, grassland vegetation monitoring at spatial and temporal resolution relevant to land management (e.g., ca. 30-m, and at least annually over long time periods) is challenging due to, among others, often limited data availability.
Our new approach bridges this gap, allowing to map short- and long-term changes in grasslands at 30-m resolution based on the entire Landsat record available in Google Earth Engine,
We tested the approach in the grasslands in the Caucasus ecoregion.
Method
Results
Presented methodology and results are part of Lewińska et al. (2020) and Lewińska et al., [in revision].
Conclusion
References
Changes in the grasslands of the Caucasus based on
Cumulative Endmember Fractions from the full Landsat record
Katarzyna Ewa Lewińska*, Johanna Buchner, Benjamin Bleyhl, Patrick Hostert, He Yin, Tobias Kuemmerle, and Volker C. Radeloff
* corresponding author lewinska@wisc.edu | @kelewinska | http://silvis.forest.wisc.edu
Results
The approach is based on the Cumulative Endmember Fractions (CEF), i.e., annual sums of monthly ground cover fractions of: soil, green vegetation, non-photosynthetic vegetation and shade (Lewińska et al. 2020).
Processing consists of the following steps:
Changes are classified into negative change pathways:
and positive change pathways:
Grassland vegetation in the Caucasus was highly dynamic during 1987-2019.
Short-term negative change pathways were most abundant in 1990s and 2010s, while positive short-term change pathways were the most common before 1997.
Changes in grasslands were not controlled by weather conditions and the effect of livestock varied among regions.
Our approach:
Vegetation change maps are available at https://kelewinska.users.earthengine.app/view/caucasusgrasslands
Why you should care about Hong Kong’s forests
Hong Kong’s Natural Beauty
Goals:
Combining Freely Available Datasets
Pixel- versus Object-based Classification
Looking Forward
Harnessing the Power of Google Earth Engine to Map Vegetation in Hong Kong
Katherine Strattman katstratt1@gmail.com, Jed Kaplan jkaplan@hku.hk
revise
First Classification attempt:
Second Classification attempt:
Pixel-based Classification
Object-based
Classification
Sentinel-2 MSI
Image Stack
Pixel-based Random Forest Model
LiDAR-derived
CHM and DEM
Sampled Training Points
Categorical Land Use/Land Cover Points
Sampled Testing Points
Validation and Prediction
Mask all Non-vegetated Areas
Sampled Training Points
Sampled Testing Points
Object-based Random Forest Model
Low Quality Classification
Sentinel-2 MSI
Image Stack
Validation and Prediction
High Quality Classification
LiDAR-derived
CHM and DEM
98% Accuracy
84% Accuracy
Introduction
Drought in the Navajo Nation
The Navajo Nation is the largest sovereign tribal nation in
the U.S. in total land area, with boundaries in Arizona,
New Mexico, and Utah.
On the Navajo Nation, they have observed increased drought, exacerbated by warming temperatures and precipitation deficits, which is adversely affecting their lives, including their farmland (>212,000 acres) that produce $92 million in agricultural products annually.
The land size and the region’s diversity, ranging from deserts to snowy forests, makes it challenging to identify who was most impacted.
Objective
Generate spatially resolved drought information that supports distribution of emergency funds to regions that are more severely affected by drought
Method
This project is a collaboration between the Navajo Nation, the Desert Research Institute (DRI) and NASA’s Western Water Applications Office. In order to meet the objective, the Drought Severity Evaluation Tool (DSET) was created as a web-based drought reporting tool that:
Results
The Drought Severity Evaluation Tool (right) shows the Navajo Nation boundaries and the 85 rain-gauge locations. This map displays the CHIRPS-generated Standardized Precipitation Index (SPI) values from June to August 2018. The data underscores the variability of the SPI across the Nation, particularly during the monsoon season.
DSET allows virtual calculations of historical or current precipitation levels for any area in the region, without requiring someone to manually check a rain gauge on the ground. It combines the most accurate and longest monitoring rain gauges with the remote sensing data from satellites to provide enhanced accuracy of measurements.
Conclusion
References
The Drought Severity Evaluation Tool for the Navajo Nation
Keiko Nomura, keiko@climateengine.com
Image: NASA’s Western Water Applications Office
Satellite and climate datasets in DSET
Image: Amber Jean McCullum, Rachel Green, Carlee McClellan
Tutorials are available https://wwao.jpl.nasa.gov/water-portfolio/water-projects/nasa-navajo-drought-severity-tool-user-guide/
The progression of drought and climate conditions in the Navajo Nation
Introduction
Mangroves play a vital role in carbon sequestration and contribute significantly to coastal protection in Guyana by reducing wave action and decreasing wave energy, capturing sediments, and stabilizing shoreline substrates. Present monitoring operations depend on field observations; however, more effective methods for monitoring the full extent of Guyana's mangrove ecosystems are necessary to better monitor restoration sites, identify crucial areas of mangrove decline, and facilitate protection, among other things. The mangrove monitoring system will ultimately measure extent, biomass and health of the mangroves on an annual basis. Mangroves have distinct surface spectral reflectance properties that allows them to be easily identified from other land and marine areas as well as other vegetation. Both optical and radar images were used to classify mangrove and non-mangrove features using Google Earth Engine.
What are my Objectives?
Method
Step by Step:
Results
Conclusion
References
A National Mangrove Monitoring System for Guyana
Kim Chan: kbagot19@gmail.com
Figure 1: Pilot area Pomeroon-Supenaam (Region 2 Guyana)
Case Study
The North-Eastern Coast of Pomeroon-Supenaam (Region 2) Guyana was used as the pilot area to develop the spatial mangrove monitoring system. It is located at latitudes 6º 55’ 00” to 7º41’0” North, and longitudes 58º28’0” to 58º52’0” West. A combination of anthropogenic activities and natural forces has significantly altered this fragile mangrove ecosystem.
Figure 2: Ground-Truthing Map showing Target Sites
Figure 3: 2020 Mangrove Extent for Pilot Area using GEE. Preliminary overall accuracy is 0.99.
Mangrove Classification using GEE
Ground-Truthing
Figure 4&5: Preliminary analysis revealed several areas of intense change in mangrove coverage. Field validation confirmed significant anthropogenic changes and minor natural changes to the mangrove environment from 2018 baseline. Mangroves were removed to facilitate large coconut farms and other agricultural activities, also for construction of buildings for both residential and commercial purposes.
Google Earth Engine was shown as a useful tool in classifying mangroves along the coast of Pomeroon Supenaam, with an overall accuracy of 0.99.
Ground truthing activity further validated the model and highlighted key areas of mangrove losses between 2018 and 2020. These changes include anthropogenic influences (agricultural activities) and natural influences (erosion).
These results will be used by NAREI and other Government Agencies to monitor existing restoration sites, identify potential areas for new restoration sites and to develop policy framework to protect the mangroves as they protect us from the sea.
Next Steps:
Acknowledgement:
The author would like to acknowledge the follow persons for their assistance on the completion of this activity:
Dr. Marc Simard, Dr. Temitope Oyedotun, Ms. Abigail Barenblitt, Dr. Lola Fatoyinbo, Ms. Kelsey Herndon, Ms. Kene Moseley, Mr. Brian Zutta, Ms. Zola Narine
Damage inspection too tedious
Results
Looking for two awesome ML engineers? Hire us!
DamageMap: A post-wildfire damaged-buildings identifier
Krishna Rao (kkrao@stanford.edu), Marios Galanis (margalan@stanford.edu),
Xinle Yao, Yi-Lin Tsai, Jonathan Ventura, G. Andrew Fricker
Damage detection using post-fire images: a new approach
Unlike existing technologies which use both pre- and post-fire images, we use post-fire images only o decrease image sourcing challenges (see publication.)
Extensively validated
Damage assessment for high-profile fires available on a GEE App (link)
source: abcnews
Introduction
Abstract:
The coverage of mobile networks in Tanzania has embarked a lot of mobile Apps use within local communities which is enhanced due to the availability of Swahili language in Android phones. WWF-Tanzania is piloting android mobile apps use for spatial data collections, monitoring, information disseminations and awareness and advocacy which is bridging the gap between different management level data visualizations, analysis and decision making by reducing paperwork and time in data collection and increasing accuracy and real-time monitoring by putting spatial data use on both online and offline mode. The mobile apps are in three different categories based on thematic areas which are terrestrial, marine and freshwater. The communities use the Apps in Village Land Forest Reserves, Beach Management Units and Water User Associations as GPS, Data Collections, Compass Awareness, Sending spatial SMS and providing real time satellite processed data like fire and weather plus administrative and landuse as base layers. Information sharing between different projects implementing partners has been efficient (ie: easy, fast and timely) and increased transparency compared to old cumbersome manual methods. Also, the apps have increased general public awareness on communities’ sustainable conservation and responsible utilization projects.
Data & Method
Most of the data are processed on Google Earth Engine using the Python API and archived onto WWF-Tanzania Monitoring, Evaluation and Learning (M&EL) Platform. The apps contains Tanzania Terrestrial, Marine and Freshwater Priority Areas spatial data which includes the Administrative boundaries & locations, National Parks, Nature & Forest Reserves, Game Reserves and Community Based Natural Resources Management(CBNRM) Areas (ie: Wildlife Management Areas - WMA, Village Land Forests Reserves - VLFR, Community Fisheries Management Areas - CFMA & Water User Associations Areas - WUA), SAGCOT, and Economic Exclusive Zone(EEZ) area.
Results
The Apps have shown to be effective tools for landuse implementations monitoring, environmental monitoring and as a data collections/communications of current/recorded/analysed spatial data/information in the WWF-Tanzania Priority Areas. These apps are potential tools to the field personnel especially the Community Based Organizations (CBO) by applying mapping in their day-to-day monitoring, evaluation and learning activities of natural resources sustainable conservations and responsible utilizations. In additions, the tools have proven to increase transparency by disseminating digital advocacy and awareness information's to the CBOs and partners/stakeholders in Swahili. (Vitini) SMS.
Conclusion
The CBOs are managing part of the conserved areas set aside by village landuse plans and under the close supervision from Village Natural Resources Committee (VNRC), both will use the apps to participate in spatial-temporal tracking, recording, reporting, awareness, advocacy and information disseminations/sharing with/to their community as part of Citizen Science in the WWF-Tanzania priority areas by enhancing community environment degradations monitoring early warning and increases the transparency, governance and accountability from all levels of government administrations (ie: Village, District, Regional and at the Ministry).
The availability of Swahili language in the Android mobile operating system has enabled the community to participate on natural resources sustainable conservations and responsible utilizations by using the smartphones equipment also the self learning is enhanced by including all training materials in digital formats so that all reference materials are available at their fingertips for reference and knowledge sharing.
References:
ENVIRONMENTAL DEGRADATION NOTIFICATIONS
FOR CBOs AND CITIZEN SCIENTIST
Langen R. Mathew - lmathew@wwftz.org
Please replace these image placeholders with images related your work or project, or remove completely if you want to use this space for text.
The Java Programming language was used to create the Android based mobile apps which utilizes the hardware sensors of the smartphones (ie: GPS,). The prepared data was converted into GeoJSON to allow compatibility with LeafletJS as static spatial information in the apps even if there is no network coverage. The apps were designed to allow data collections. especially waypoints and tracks, the recorded data can be exported into Keyhole Makeup Language (KML) and shared. With strong network coverage (ie: Internet Speed), the apps can retrieve near-real time NASA satellite modelled data like Fire, WIND Directions, Temperatures, etc.
Figure on right: Participants from Northern Tanzania WMAs (SOKNOT-Unganisha Transboundary Landscape) who attended the MIOMBO App training-workshop at Uhuru Hostel, Moshi, Tanzania funded by Germany -BMZ-1 Project through WWF-Tanzania.
Figure above: Webmaps for providing access to spatial data retrieval, manipulation, analysis and visualization
Above Figure: One of the Dashboard which provides visualization to CBOs and Citizen Scientist for better decision making.
Below Figures: Android app Screenshots which shows the data collection, position data/Spatial SMS sending option, Compass/Search, Alert subscription and tutorials/brochure.
The CBOs (Key Users) are:
The Android Applications screenshots for CBOs/Citizen Scientist to Collect, Visualize, Search and Dissemination (Alert/Earlywarning)
Mobile Applications on
Google Play Store:
Free and Open Source
Software (FOSS) in use:
Introduction
Method
Results
Conclusions
References
Poggio, L., de Sousa, L, Batjes, N.H., Heuvelink, G.B.M., Kempen, B., Ribeiro, E., Rossiter, D. SoilGrids 2.0: producing quality-assessed soil information for the globe. SOIL, 2021.
SoilGrids 2.0: Soil Information for the Globe
Laura Poggio on behalf of SoilGrids for ISRIC, soilgrids@isric.org
Soil information is fundamental for a large range of global applications, including assessments of soil and land degradation, sustainable land management, and environmental conservation.
SoilGrids is a global product that fulfils two main goals:
1) it is a source of consistent soil information to support global modelling, and
2) it provides complementary information to support regional and national soil information products in data-poor areas.
The SoilGrids uses remote sensing to derive environmental covariates to model soil properties at the global scale. The main covariates considered are derived from a mix of optical and radar sensors:
The workflow incorporates Digital Soil mapping (DSM) practices and adapt them to the challenges of global DSM with legacy data. It builds on previous global soil properties maps, integrating up-to-date machine learning methods, the increased availability of standardised soil profile data for the world and environmental covariates . In particular:
SoilGrids requires an intensive computational workflow, with numerous steps integrating different software. SoilGrids is entirely based on open source software, in particular: SLURM for job management, GRASS GIS for data and tiles management, and R statistical software for model fitting and statistical analysis. Predictions were computed in a high-performance computing cluster. A dynamic geographic tiling system was developed with GRASS GIS to maximise the use of memory for each job.
The maps were obtained at 250m spatial resolution using machine learning methods. Soil observations from about 240 000 locations worldwide and over 400 global environmental covariates were used as inputs.
The quantitative evaluation showed metrics in line with previous global, continental and large regions studies. The qualitative evaluation showed that coarse scale patterns are well reproduced.
Introduction
Overview
Over the years, the utilization satellite-derived land surface temperature has been increasingly recognized as a significant parameter in understanding the temperature changes on the land surface of the earth and gained importance for various climate applications.
This presentation presents the relationship between satellite derived LST and air temperature measurements from a weather station located in Port Area, Manila City Philippines.
The researchers used available air temperature data from 2014 to 2018, provided meteorological agency in the Philippines, and analyze its data pattern and variability. In addition, LST was derived from MODIS satellites data through Google Earth Engine (GEE).
Problem (What we want to solve)
Method
Results
Conclusion
Conclusion
References
Temporal Analysis between Air Temperature and Land Surface Temperature in Port Area, Manila City, Philippines
Mark Angelo C. Purio, purio.mark-angelo894@mail.kyutech.jp
Problem (What we want to solve)
Land Surface Temperature (LST)
Data Source: MOD11A1 V6 product known as the Terra Land Surface Temperature and Emissivity
Google Earth Engine (GEE): used to retrieve the LST values from the specified point where Manila Port synoptic weather station is located
�
Air Temperature Data
Data Source: MOD11A1 V6 product known as the Terra Land Surface Temperature and Emissivity
Google Earth Engine (GEE): used to retrieve the LST values from the specified point where Manila Port synoptic weather station is located
LST & Air Temperature Relationship Analysis
The collected data were examined and cleaned before using it as input parameters for the correlation analysis.
The respective data were observed by looking at their trend and establishing the warm and cold recorded temperatures within the year
Trend Analysis of Air Temperature and LST
Relationship between Air Temperature and LST
Future Work
Introduction
Measurement of on-road air-pollution in Bengaluru, India
Meenakshi Kushwaha1, Pratyush Agrawal1, Adithi R. Upadhya1, Jonathan Gingrich2, Jai Asundi3, V. Sreekanth3, Julian D. Marshall4, Joshua S. Apte2
1 ILK Labs, Bengaluru 560046, India
2 Civil and Environmental Engineering, University of California, Berkeley, CA 94720
3 Center for Study of Science, Technology and Policy, Bengaluru 560094, India
4 Civil and Environmental Engineering, University of Washington, Seattle, WA 98195
Materials and Methods
Results
BC
Specific study roads are in green.
Study neighborhoods are color-filled shapes.
Challenges
BC(μg/m3)
UFPs concentration on different road types
BC concentration on different road types
Monte-Carlo analysis
Spatial map of BC and UFPs concentration in a residential neighbourhood
UFPs (#/cm3)
Instruments used on the mobile platform
Who
•Undergraduate students taking General Education Geography Courses including Physical Geography and Will the World Provide?, a sustainability course
•~ 80 students per year
•Typical student majors include Sports Management, Education, Exercise Science, and Geography
How
What
•Google Tours to explore different environments
•Google Tours to explore different surface features
•Create their own Google Tours as a final project
•Timelapse to explore changes to climate and environment
•Timelapse to explore human-driven changes to Earth
Why
Social Impact: Undergraduate Education
Melinda Shimizu, Ph.D. melinda.shimizu@cortland.edu
•Dynamic views and photos in the Google Earth interface are far better at capturing and communicating Earth features than static photos
•It is interactive and students thrive when presented with interactive opportunities to learn
•The Timelapse feature is so helpful to showing changes to environments in an accessible way
Why mangrove phenology is important?
Estimating phenological metrics
So far..
Next..
References
Elucidating mangrove forest phenology in the Sundarbans
Mohammad Shamim Hasan Mandal and others
Postdoctoral Researcher, Graduate School for Advanced Science and Engineering Hiroshima University
mshmandal@gmail.com | mandal@hiroshima-u.ac.jp
Phenological metrics::
Pre-processing and
Time series in
Google Earth Engine (GEE)
Cloud masking and filtering
Phenological parameters
Spectrum using FFT
Smoothing
MODIS (MOD13Q1) 16-day
Remote Sensing Data
NDVI, EVI
Time series: four forest types
Pneumatophore Vivipary
Sundarbans mangrove forest
But,
F1 - Heritiera fomes dominated,
F2 - Excoecaria agallocha dominated,
F3 - Ceriops decandra dominated,
F4 - mixed forests of F1, F2, and F3
A clear annual seasonality in canopy greenness
Phenological calendar
Future study
Lesson learned
Introduction
Visualize the Environmental impact of Project Finance
My aim is to demonstrate the environmental impact in the implementation of new projects, through the use of Google Earth, in order to allow the various authorisation levels to make decisions that include both saving costs and the environment concern.
Bullet List
Method
I created a project in Google Earth, detailed with pictures and plans of the project including the necessary information to describe the issue.
In order to represent the situation in the best possible way, and to allow the boards to understand the problem, I shared the project so that each individual could analyse the issue in the best possible way.�
Bullet List
Results
Changes to the initial plan to create gas lines affecting the turtle reserve were accepted and a new project was considered.
the real difficulty was to instil within an old-style managerial class the possibility of assessing the feasibility of new projects through the use of open source resources such as Google Earth
Conclusion
Bullet List
References
Visualize the Environmental impact of Project Finance
Nicolò Spelgatti, nic.spelgatti@gmail.com
Logo
Please replace these image placeholders with images related your work or project, or remove completely if you want to use this space for text.
Please replace these image placeholders with images related your work or project, or remove completely if you want to use this space for text.
Please replace these image placeholders with images related your work or project, or remove completely if you want to use this space for text.
Please replace these image placeholders with images related your work or project, or remove completely if you want to use this space for text.
Introduction
Spatiotemporal Assessment of Forest Cover Change
Understanding changes of forest cover is important for identifying hot-spot areas and drivers of change to form a basis for formulating effective conservation and management strategies. However, study using a combined method of phenology-based threshold classification (PBTC) and Google Earth Engine (GEE) is still unavailable in Sri Lanka and in South Asia. Here, we the GEE along with the Landsat TM and Landsat OLI collections and applied PBTC method to map the spatial distribution and analyze temporal changes of forest in Vavuniya District, Northern Province, Sri Lanka since 2001.
Our study aims:
Methodology
We acquired and processed 234 collections of Landsat 5 TM and Landsat 8 OLI during dry-phenology season (May to September) to map the land covers for 2001, 2006, 2010, 2016, and 2020 to obtain the images for analysis (Figure 2).
We applied the minimum and maximum EVI threshold values for dry monsoon forest and open forest.
EVI thresholds as listed below:
Accordingly, land cover classification and spatiotemporal mapping is performed by executing PBTC method in GEE. Our study yielded five PBTC land cover maps.
We assessed the rates of annual changes for whole district’s forest and its four Divisional Secretariats divisions.
Conclusion
Assessment of Forest Cover Change using the Phenology-based Threshold Classification and Google Earth Engine
Sharaniya Vijitharan (st121566@ait.ac.th), Manjunatha Venkatappa, Nophea Sasaki
Figure 1. Visualization of forest cover types
Naturally-regenerating forest types
Study Site
Methods
Figure 3. PBTC in GEE
Figure 2. Methodological framework
Results
Conclusion
Our research achieved the average overall accuracy of 87% and mean Kappa coefficient of 0.83.
Forest area expanded between 2001 and 2010, while a reduction measured during 2010-2020.
We assessed:
In 2020, 57.5% of district’s area was covered by forest.
Forest cover changes in DS divisions:
Vavuniya – increased until 2020 and then declined
Vavuniya South -- slightly increased
Vengalacheddikulam – gradually expanded
Vavuniya North – continuously declined by 19.2% (2001-2020)
Figure 4. Overall trend of forest cover
Figure 5. Forest cover change in DS divisions
Forest cover changes between 2001 and 2020:
From 2010 to 2020, total forest cover reduced by 6.04%. Forest area expanded between 2001 and 2010, while a reduction observed during 2010-2020. Dry monsoon forest cover declined continuously for 19 years. Overall open forest cover was slightly increased.
Highest forest cover was in Vavuniyabut nut it declined from 83.8% in 2001 to 67.7% in 2020.
Classification of open forest is challenging due to its smaller extent mixed with croplands, and soil reflectivity during dry season.
Without GEE, it would not have been possible to access data since 2001 as such data are rarely available in Sri Lanka.
Figure 6. Land cover maps of Vavuniya District, Sri Lanka
Main Points
Introduction
Climate change and human activities have had a significant impact on the forest in recent decades. Forest fires are caused by climate and human usage, resulting in infrastructural damage and loss of life. Several studies learn more about the adverse effects of forest fires on flora, biomass, and soil characteristics (Daz-Delgado et al., 2002; Verma et al., 2019). Forest fires in Nepal's forests are quickly rising each year. According to studies, more than 90% of vegetation fires in Nepal are likely caused by humans, with forest fires affecting 3 to 5 million hectares yearly (Durbar & Kathmandu, 2018; Yasmi et al., 2019). Since 2005, Nepal has lost an estimated 200,000 hectares of forest cover per year due to forest fires, according to the report (Chaudhary et al., 2016; Paudel, n.d.). The forest fire trigger by a prolonged drought and lack of rain. Its smoke pollutes the air and forms a thick layer in the atmosphere. As a result, forest fire identification and monitoring research are necessary to determine its kind, magnitude, and impact and develop a risk map and early warning system to reduce its repercussions.
Objectives
Method
In Gandaki Province's trans-Himalayan region, Manang has dense pine forest and approximately 25% of the Annapurna protection area (Bhattarai & Conway, 2021; Huddart & Stott, 2020). Currently, there is no road access in the district.A wildfire raged for two months in Thanchok and Kote's woods in January, destroying approximately 700 hectares of forest cover (Wildfire Burns 700 Hectares of Forest Forest Cover in Manang, n.d.; Wildfires Destroy Hundreds of Hectares of Forest in Nepal - Xinhua | English.News.Cn, n.d.). It contains a highly flammable thick pine forest, which may have been ignited by accident or on purpose.
The research area is estimated using the Landsat-8 satellite image indexes, object-based categorization, fire area, and burning intensity. It uses Google Earth Engine to generate a potential forest fire map of Manang generated using different sources for the forest fire area’s burn severity.
Results
The differenced Normalized Difference Vegetation Index (dNDVI) and differenced Normalized Burn Ratio (dNBR) map of the study area is the difference between the pre-fire and post-fire NDVI and NBR obtained from the satellite images to estimate the burn severity. dNDVI index obtained by subtracting the NDVI index after the fire from the NDVI index before the fire. Higher values indicate dense vegetation cover and vice versa. A higher value of dNBR indicates more severe damage, while areas with negative dNBR values may indicate regrowth following a fire. The differenced Green Normalized Difference Vegetation Index (dGNDVI) map for distinguishing stressful vegetation and the Relative differenced Normalized Burn Ratio (RdNBR) map measures the spectral variation.
Conclusion
References
Detecting Forest Fire using Landsat Images: A Case Study of Manang District, Nepal
What and why?
Watch my Wilderness
The Wilderness Society (Australia) is an ENGO that campaigns to protect Australia’s iconic wild places and biodiversity. We empower the community to influence the decisions about the places they love. One of the tools we are developing is a community powered deforestation monitoring platform. The Watch my Wilderness (WmW) platform provides easy access to Sentinel-2 data to compare images to detect land clearing/deforestation. The users can trace suspected clearing and submit it to our database for verification. We aim to contribute to open source training data for change detection models that are being built to assist with biodiversity conservation, vegetation management/compliance and carbon related projects.
WmW aims to:
How?
Watch my Wilderness uses Google Earth Engine, the Google Cloud platform, and Google Maps to:
Did it work?
We have released a restricted access version of Watch my Wilderness to test user needs for monitoring Australia. Sentinel-2 EOM provides high frequency, medium resolution imagery and Google makes it easily and freely available for grass roots platforms like Watch my Wilderness. The users found the platform to be very useful for remotely monitoring land without having to go into the field. Users also found it useful to have access to GIS professionals to assist in image interpretation and to verify clearing. All users wanted more advanced auto detection to guide the monitoring, which the WIlderness Society is currently working on. The platform is used by a diverse group of people, mostly non-GIS professionals, and has a verified clearing database of over 400 clearing events from 10 trial users in Queensland alone.
What next?
References
Land clearing and deforestation monitoring in Australia
Rachel Fletcher, rachel.fletcher@wilderness.org.au
Citizen science:
Australia has a high rate of deforestation and land clearing across a large area. The Wilderness Society is harnessing the power of our community of volunteer citizen scientists to detect and record clearing events across Australia.
Introduction
Sustainable Supply Chain
Developing a sustainable supply chain from seed to delivery and everything in between is something I’m passionate about. I’m meeting with people responsible for these initiative at my clients who mostly reside in the SLC industry.
Method
Results
Conclusion
References
Sustainable Supply Chain & Logistics
Scott Hitchcock, scotthitchcock@google.com
Introduction
Google Earth Election Maps
Leverage the power of open data and free Google apps to educate the public about elections.
Visualization of election data provides a clearer interpretation of election outcomes and uncovers patterns which may be used to measure performance.
Oblige Google Maps Election Project is an online workshop and toolkit that demonstrates how to transform election data into useful and understandable maps that improve understanding of elections. Oblige demonstrates how to easily leverage free Google Apps for election visualization, analysis, strategy, and public education.
Dissipation of local news outlets has created an aperture in the media marketplace for the visualization of local election data. News media data visualization resources are strained across markets. The election mapping market is dominated by proprietary mapping professionals and hobbyists using softwares with steep costs and learning curves. Very few resources provide interactive maps for free public consumption or iterations.
Oblige Google Earth Election Maps Mission
Approach
Clean data may be mapped in Google Earth manually and methodically, creating interoperable maps that can be shared and further analyzed. Building upon the principles of the Open Election Data Initiative developed by Google, NDI, and the USAID; this project seeks to demonstrate how to visualize that data for better comprehension and wider consumption.
Method
Insights
The 2020 Presidential Turnout Map to the right demonstrates an in depth analysis of election results. Election and voter data was gathered and cleaned using Google Sheets and Drive. Calculations were mapped in Google Earth and saved and shared to Google Maps and disseminated. The results shown in the map present a clear pattern that economically depressed areas had lower turnout in the election.
The ability of this data to be easily read and shared interactively on mobile devices in the field allows political campaigns and organizations to allocate resources nimbly and strategically to effectuate greater election participation.
How may we Oblige?
Fostering a better understanding of community through elections data visualization using Google Earth to engender more trust and understanding of democratic elections.
Leverage free tools to educate and improve performance outcomes and identify disparities.
Oblige: Google Election Maps
Seth Platt, Sethplatt@oblige.org
Barriers are access to election data/clean data under non-transparent or unorganized governments.
Resources needed to map election results: Human, GE processing power, data.
The map above delineates Democrats and Republicans elected as precinct committee people per voter district in Broward County, Florida. This map identifies areas of strength and weaknesses in the organization's’ County wide human resources.
Publicly available addresses of elected members were uploaded to Google Earth from Sheets and demarcated manually in Google Earth according to how many resided in each voter precinct. This map define a resource allocation strategies.
States with FAW presence
Introduction
Fall armyworm (FAW) arrival in India
Goal
Method
Results
Conclusion
References
Wu, Q.-L.; He, L.-M.; Shen, X.-J.; Jiang, Y.-Y.; Liu, J.; Hu, G.; Wu, K.-M. Estimation of the Potential Infestation Area of Newly-invaded Fall Armyworm Spodoptera Frugiperda in the Yangtze River Valley of China. Insects 2019, 10, 298.
Forecasting Fall armyworm using multi-sensor data for maize farmers in India
Sonal Bakiwala, Sravani Gunda, Parmita Ghosh, Raman Babu, Anu Swatantran
Farmer
Agronomist
IBM Weather
Data
AWS
Database
ML Model
FAW Risk Probability
Prediction
Satellite Data
Engineered Features
Data Digitization from Farmer location
Step by Step
Haversine’s law
Distance and Bearing
Sentinel-1 SAR
Catboost Model Test Accuracy 88%
Total | High | Medium | Low | Healthy |
8197 | 1532 | 2994 | 2896 | 775 |
Ground Truth data for model building and validations
Ground truth data for FAW incidence ground survey
Predicted FAW Severities from multi sensor model(14 days in advance)
Misclassified
severity
Challenge
We did not have information on farmers if they applied any chemistry to manage FAW. Its possible predictions are correct, but farmers may have applied solutions and they don’t witness heavy damage due to FAW
FAW has been a menace for the entire country since 2019. Extensive research and studies have been done to manage FAW better through multi-disciplinary sciences. We have produced a strategy to exploit multi-sensor data and integrate them together to forecast FAW risk better. We leverage information and data from farmers directly. We understand that its very important to help the community at large especially small farmer holders. Using high resolution satellite data and weather data we can deliver early alerts 14 days in advance for FAW risk and help farmers (especially small holders) to mitigate the FAW risk and not suffer yield loss because of FAW damage. More validations across country and cross seasonal validations are required.
Conclusion
High Severity
Medium Severity
Low Severity
Corn areas
Ground truth
Cross validation
Accuracy – 92.1%
Corn area Prediction using Sentinel-1
---Internal Use---
Introduction
Increase the resilience of cities
Nowadays our cities are flooding with increased frequency, due to more severe weather events but also to increasing anthropogenic pressure (e.g. soil sealing, land subsidence, uncontrolled development, etc.).
It is therefore necessary to increase the resilience of our cities and facilitate their adaptation to current and future extreme flooding events
What about the challenges?
We want to equip communities with the appropriate flood risk intelligence for supporting preparedness and resilience:
Method
SaferPLACES is a cloud-based climate service for assessing pluvial, fluvial, and coastal flood hazards and risks under current and future climate scenarios. A DIGITAL TWIN of the urban environment can be generated in few steps in any city of interest, to provide accurate and actionable insights to flood risk assessment and mitigation.
The algorithms
Results
SaferPLACES platform provides a cost-effective and user-friendly cloud-based solution for flood hazard and risk mapping. Moreover SaferPLACES supports multiple stakeholders in designing and assessing multiple mitigation measures such as flood barriers, water tanks, green-blue based solutions and building specific damage mitigation actions.
Conclusion
SaferPLACES capabilities
References
SaferPLACES Global Platform: A Digital Twin Solution for
Flood Risk Intelligence in Urban Areas
Stefano Bagli, stefano.bagli@gecosistema.com
Introduction
Satellite data to monitor montane forest recovery
Method
Results
Recovery trajectories vary after different kinds of disturbance. Large areas of burned forests show no recovery at all, while other areas recover. Disrupted recovery and no recovery areas can be targeted for assisted and active restoration interventions.
Conclusion
Challenges
Assessing recovery trajectories in disturbed tropical montane forests through multi-temporal satellite data
Tina Christmann and Dr Imma Oliveras, tina.christmann@worc.ox.ac.uk
A promising tool for restoration planning
Satellite-based recovery classifications can guide planning and decision making on ecosystem restoration in inaccessible ecosystems such as tropical mountains. For instance, by identifying areas of successful and unsuccessful recovery, restoration priority hotspots can be scoped and suitable restoration methods chosen. Such methods can als be transferred to monitor existing large scale active restoration projects (e.g. reforestation projects) and improve transparency and accountability of ecosystem restoration.
Fig. 1: A tropical montane forest degraded by fire in the Brazilian Atlantic Forest (own photograph)
Fig. 2: Workflow of degradation and recovery classification
Fig. 4: Recovery trajectories in different degradation areas
Fig. 3: Spatially explicit classification map for four recovery classes based on trend analysis
Degradation and recovery classification
Introduction
Story begins
Having previous work experience with poverty mapping and being locked at home during pandemics for months, I have decided to spend my time taking care of my newborn baby and experimenting with new AI methods. Not only to implement, however also to to tune it all and possibly check for freely available input data. After analysis, I focused on two research papers. “Classifier” and “Detector”.
Goooooaaaaal
How it was done
In Google Earth, I manually located polygons of the target country, where very high resolution (0.3m/px) is available and downloaded them. Due to resolution and HW restrictions, only cca 200 out of 1500 administration regions were used in this project.
In Earth Engine, Sentinel 2, VIIRS nightlight imageries and Worldpop datasets were aggregated and downloaded for AOI.
Virtual Machine with GPU from Google Colab was used for the training and inference phases of the modelling.
Fastai2 and Pytorch provided machine learning frameworks for the creation of classifiers(arch ResNet34) and detectors(arch Yolov5).
Linear models for final poverty mapping and validation were implemented based on the scikit-learn library.
Results
Creating “classifier” and “detector” separately, I was able to reach comparable results with the original papers. Furthermore, the resulting combined ensemble model gained high poverty prediction accuracy (r_squared=0.82) by exploiting existing methods, different model architectures and satellite image resolutions. For more detailed description please see published paper.
Conclusion
References
Poverty mapping from space
Yolov5 framework from Ultralytics with sample detections, Google Earth image (2016 CNES/Airbus)
VIIRS Nighttime Composites Version 1
Composites derived from Landsat-ETM+ 30m (left) and WorldView2 2m resolution data (right).
Introduction
NSF South Big Data Hub’s
3D Wetlands Project
Produced Datasets
Datasets in GEE
DEM & Habitat Classification Maps are now on GEE
github.com/luislizcano/3D-wetlands-app
Next Steps
3D Wetlands : 2m Habitat Class + Elevation in the SE USA
Luis Lizcano-Sandoval {luislizcanos@usf.edu}, Tylar Murray {mail@tylar.info}
Target area WV2 coverage
DataSet 1 : Digital Elevation Model
DataSet 2 : Land Cover
The Problem
Science informs management
The Nature Conservancy (TNC) and partners are working to reveal and understand coral health at different scales to answer a range of vital questions for resource managers, governments, and conservationists. Scientists are collecting data using different technologies and platforms, each of which provides an important layer of information to inform marine conservation and management initiatives. Airborne imagery from the Arizona State University's Global Airborne Observatory (GAO) can be used to derive spatial data such as live coral cover and habitat complexity, or rugosity. These maps will help coral managers identify specific locations to outplant coral to maximize survival rates.
In the past, site selection for coral outplanting was based on local knowledge and observations, with little quantitative information involved in this process. However, in November 2019, The Nature Conservancy and Arizona State University’s Global Airborne Observatory took a step forward by collecting data and developing a new tool to inform site selection. Coupled with local knowledge of ideal in-water conditions, this data allowed for selection of sites that improved coral survival.
The Data
Arizona State University's Global Airborne Observatory (GAO) is an airborne laboratory that collects high resolution (1m) coral health and habitat information to inform effective coral outplanting site selection with increased survivorship. The GAO plane flies at an altitude of 1000m and collects 427 samples of light per pixel that correspond to the chemistry of corals, generating 324 GB/hour of data. Data is processed on a cluster machine matching spectrometer information with high-precision positioning data at an accuracy of 7-10cm.
Hyperspectral imagery was collected by ASU's Global Airborne Observatory (GAO) for The Nature Conservancy in Saint Croix, U.S. Virgin Islands and the Dominican Republic. ASU derived high-resolution maps from this aerial data representing: bathymetry, live coral cover, fine habitat complexity, algal cover, seagrass cover, and sand cover. TNC scientists used water surface drone imagery to collect high-precision GPS data and to predict percent live coral to train the machine learning model. These maps are made available here for use by resource managers and conservation scientists.
The GEE App
On the Ground Results
The Coral Outplanting Siting Guide app allows on-the-ground coral managers to find sites that meet suitable criteria ranges of different variables such as percent algal cover, habitat complexity (rugosity), live coral cover, and bathymetry. This scientific method for selecting sites can guide conservation and management efforts when coupled with local knowledge. This app is available at CaribbeanMarineMaps.tnc.org and a similar app is currently under development for the Hawaiian islands by The Nature Conservancy and partners.
Reference:
Coral Outplanting Siting Guide
Valerie Pietsch McNulty, valerie.mcnulty@tnc.org
© Paul Selvaggio
© Paul Selvaggio
© Paul Selvaggio
This tool was successfully implemented during a coral mania organized by Fundación Grupo Punta Cana in Cabeza de Toro, Bavaro. This event brought together dozens of volunteers belonging to NGOs, Dominican environmental authorities, dive operators and other local stakeholders aimed to outplant 1,711 Acropora cervicornis fragments into different sites. By October 2020 (11 months after outplanting), survivorship remained above 76%. These results demonstrate higher than average success rates for coral outplant survival for this species.
Introduction
Globally, about 930 million hectares of forests are classified as degraded forests, while another 10 Mha are deforested annually. Foreseeing this danger, the New York Declaration on Forests and the UN Decade on Ecosystem Restoration set the target to restore 350-500 Mha of the degraded ecosystems by 2030. At COP26 of the UNFCCC, world leaders agreed to completely stop deforestation by 2030. To implement these agreements at scale, identification of the appropriate degraded and/or deforested lands is prerequisite. However, little study exists on this important identification. Such identification at scale needs scalable technology along with high computing power.
What was our aim?
Method
Results
We were able to prioritize the areas for forest restoration according to the following degraded levels: critically degraded (~96,693 ha), highly degraded (48,878 ha), moderately degraded (46,487 ha), and slightly degraded forests (75,567 ha).
If all these degraded forests were restored, 193.73 TgCO2 could be sequestered in Siem Reap between 2021 and 2030 (Fig. 6).
Depending on carbon prices, US$0.8 – 26.6 billion could be generated.
Conclusion
Main points
References
Identification of the Degraded Lands for Forest Restoration on Automation
Manjunatha Venkatappa, Nophea Sasaki manjunathagis@gmail.com | www.leetintel.com | www.ait.ac.th
Figure 1. UN Ecosystem Restoration by 2021-2030
Figure 2. Tropical forest degradation levels
Figure 3. Level of forest degradation
Figure 4. Forest transition and level of forest degradation
Figure 5. Forest cover change between 1990 and 2018
Figure 6. Potential carbon sequestration (million tons) between 2021 and 2030 and benefits
Figure 7. Strategies for the New York Declaration on Forests Restoration
Note: ANR = assisted naturally regenerated restoration, EP = enrichment planting, PLR = preventing logging reentries and RIL = reducing impact logging
Introduction
Early Active Fire Detection
In recent years, wildfires have become major devastating hazards that affect both public safety and the environment. Thus,agile detection of the wildfires is desirable to suppress the wildfires in the early stage. Owing to the high temporal resolution, GOES-R satellites offer capabilities to rapidly obtainimages every 15 minutes. By utilizing the GOES-R data offered by Google Earth engine, I would like to provide a way to accurately and instantly detect active fires.
Goal
Method
Due to the reason that GOES-R has 2km spatial resolution, to achieve accurate purpose, the image is resampled to 375m spatial resolution. For each pixel, a window of 11 by 11 is selected to form the time-series and to be fed into the GRU-Network to generate the output. As a compromise to accuracy requirements because of the interference of the smoke and cloud, the max-aggregated image is used in inference.
Steps:
Results
The output of the GRU-Network is overlaid on a Sentinel-2 images and the progression of the active wildfire is clearly visualized. Also the detection of the wildfire is earlier than VIIRS 375m active fire product for our study areas.
The Accuracy for the proposed method is also higher than the baseline GOES-R active fire product.
Conclusion
Conclusions
References
EARLY DETECTION OF WILDFIRES WITH GOES-R TIME-SERIES AND DEEP GRUNETWORK
Yu Zhao, zhaoyutim@gmail.com