Entwine Point Tiles for fast 3D visualization and analysis of massive point cloud data

NASA and ESA are increasingly utilizing the cloud to store and distribute Earth data, with a particular urgency around ambitious new missions that produce massive data. ICESat-2 is one such mission. ICESat-2 mission data products have been highly anticipated, having applications for everything from climate change to wildlife conservation, but the data volumes are massive at up to 1 terabyte per day. Unlocking the scientific potential of this data requires new approaches to organizing the data and new tools to process and visualize this data.

The Multi-Mission Algorithm and Analysis Platform (MAAP) is a joint ESA and NASA open science platform for global biomass modeling. In this post, we share the MAAP approach of using the AODS technology Entwine Point Tiles (EPT) for the ICESat-2 Land and Vegetation Height product (ATL08). We also talk about Analytics Optimized Data Stores (AODS) more broadly. If you are at AGU we hope you will check out our session on AODS formats and approaches (requires AGU login).

ICESat-2: Satellite Laser Altimetry Data Supports Varied Science Disciplines

The Advanced Topographic Laser Altimeter System (ATLAS) instrument aboard the ICESat-2 mission generates height profiles of the Earth’s surface. The instrument produces canopy height profiles by measuring the time between sending and receiving a laser pulse to the Earth’s surface.

AGU’s Fall Meeting 2020 includes several sessions highlighting the utility of ICESat-2 products across science disciplines. The ICESat-2 Applications Program engages scientists across hydrology, ecology and the Navy to share how they are using ICESat-2 data.Early adopters shared their work with ICESat-2 data at AGU’s ICESat-2 Town Hall in fields from wildfire to wildlife.

Birgit Peterson showed one great example of using ICESat-2 data for wildfire research. She uses ICESat-2 photon data to differentiate between burned and unburned vegetation. This enables “burn severity mapping” which supports better post-fire decision making and is used as an input to other risk measurements.

The following scientists featured in the town hall also showcased their work using ICESat-2 data.

• Christopher Parish of Oregon State University is fusing ICESat-2 with Landsat 8 Operational Land Imager (OLI) to create contiguous coastal and nearshore maps.
• The Naval Research Lab is using ICESat-2 to measure sea ice thickness, crucial for understanding the risk of storms to coastal communities.
• Michael Wethington (Stony Brook University) and Valerie Casasanto (University of Maryland) are using ICESat-2 to study how sea ice dynamics impact seabird populations.

Using ICESat-2 for Mapping Trees

Development Seed is working with NASA and ESA to develop a cloud-based platform for scientists to collaborate on creating global biomass models and maps. This platform, known as the Multi-Mission Algorithm and Analysis Platform, or MAAP, supplies users with analytics-optimized data stores (AODS) for handling large data volumes.

The ATL08 Land and Vegetation Height product includes tree canopy height measurements. The biomass research community is using ATL08 to scale the global spatial and temporal extent of their models¹. However, the volume of files for doing this work is prohibitively large for this task.

Key Challenge: Handling Large Data Volumes

ATLAS is one of a new class of recent space-based LiDAR sensors that produce a large volume of data with sparse geographic density. ATLAS, which came online in October 2018, produces at least 1 TB per day (Blumenfeld, 2019), and will collect data for at least 3 years. The current ATL08 version 3 subset data published on the MAAP platform is roughly 3.3 Terabytes in size. File-based access of the full extent of this data is not easy on a single-user machine.

Additionally, the biomass modeling task faces these challenges:

• Estimating global carbon stocks from proxies like canopy height requires a modeler to work with large volumes of data from the entire globe over a longer temporal range (Albinet, 2019).
• Visualizing the dataset requires identifying and then reading thousands of files for a given bounding box. HDF5 is not optimized for use via web requests. It is just not possible to run an interactive visualization on data spread across hundreds of HDF5 files.
• Like visualization, most meaningful analysis involves selecting a contiguous area of interest which can come from many granules (scenes) of HDF5 files. Each data record has X,Y, Z coordinates and additional dimensions (variables), filtering before downloading greatly reduces the size of data transfer.

AODS technologies provide access to large volumes of data without users having to download a single file. One AODS technology which has garnered a lot of attention for managing massive point clouds like the ICESat-2 data products is Entwine Point Tiles (EPT).

What are Entwine Point Tiles?

The EPT format is a cloud-optimized point cloud data format which re-organizes points into a cloud friendly spatially indexed data structure. MAAP uses AWS S3 to store an ATL08 EPT store and serve the data over OGC specified APIs: 3DTiles for visualization and Features for querying. These APIs allow for interactive 3D visualizations in a web browser, including notebook environments and facilitates on the fly subsetting for interactive data exploration, all of which can be applied to other similar sensors.

How did we do it?

The MAAP data team used AWS Step Functions to generate an ATL08 EPT store for a subset of variables. 101,088 source HDF5 files were transformed into intermediary LAS files using the Point Data Abstraction LIbrary (PDAL). The Entwine library used these LAS files to index over 639 million individual points, with global coverage starting in October 2018 through mid-July 2020².

Description of Entwine Point Tile (EPT) for ATL08:

• A 3D octree spatial index reorganizes data by 3D geography to optimize spatial queries. (Mosa et al., 2012)
• Over 100,000 input files are combined into an indexed directory structure that can be queried as a single data source. In this case we used online cloud storage: S3 on AWS.

The data store normalizes data to common LiDAR dimensions:

• X: Longitude
• Y: Latitude
• Z: DEM height
• ElevationLow: segment terrain height best fit (h_te_best_fit)
• HeightAboveGround: canopy height (h_canopy)
• OriginId: a reference to an origin file for each data point
• GpsTime: standardize timestamps to GpsTime for cross dataset querying.

Here’s how to use it!

See the source code or launch an example jupyter (python) notebook for exploring the data store with Binder.

Are you attending AGU?!

We are presenting this work in a poster session at the 2020 American Geophysical Sciences (AGU) Fall Meeting in a poster session “Lessons Learned on Supporting Analysis Ready Data (ARD) with Analytics Optimized Data Stores/Services (AODS) in Collaborative Analysis Platforms” (requires AGU login).

Other posters in this session:

• Data store alternatives for the Multi-Mission Algorithm and Analysis Platform (MAAP), Author: Dai Hai Ton That, NASA IMPACT
• Sentinel-2 Cloud-Optimized GeoTIFF Public Dataset, Author: Matt Hanson, Element84
• Using TileDB and Pangeo to Provide Access to Thousands of NetCDF Files as Analysis-Ready Data, Author: Peter Killick, MET Office Informatics Lab
• Pangeo-Forge: Crowdsourcing Analysis-Ready, Cloud Optimized Data, Author: Ryan Abernathy, Lamont Doherty Earth Observatory at Columbia University
• Cloudy Oceanography using Analysis Ready Datasets, Author: Chelle Gentemann, Fallaron Institute
• Analysis Ready SST Data for the Oceans, Author: Edward Armstrong, NASA JPL
• Hybrid Serverless Cloud and Supercomputing Workflow to Support Methane Plume Detection and Regional Analysis, Joseph Jacob, NASA JPL

We look forward to discussing:

• What are the shared lessons learned from building and using collaborative science platforms?
• What is the difference between Analysis Ready Data and Analytics Optimized Data Stores and why does it matter?
• How are analytics optimizations being used in practice?

Footnotes

1. In addition to ICESat-2, the MAAP provides scientists with other field, airborne and satellite optical, SAR, and LiDAR data to use for their biomass models. Two other new missions this community is excited about are the BIOMASS (SAR) and GEDI (LiDAR) satellite missions to scale their biomass modeling globally.
2. The MAAP team expects to update the Entwine Point Tile store with the most recent ICESat-2 ATL08 product in early 2021.

References

• Albinet, C., Whitehurst, A.S., Jewell, L.A. et al. A Joint ESA-NASA Multi-mission Algorithm and Analysis Platform (MAAP) for Biomass, NISAR, and GEDI. Surv Geophys 40, 1017–1027 (2019). https://doi.org/10.1007/s10712-019-09541-z
• Blumenfeld, J., 2019. ICESat-2 Data Usher in a New Age of Exploration | Earthdata [WWW Document]. URL https://earthdata.nasa.gov/learn/articles/tools-and-technology-articles/icesat-2-data/ (accessed 11.17.20).
• Mosa, A.S.M., Schön, B., Bertolotto, M., Laefer, D.F., 2012. Evaluating the Benefits of Octree-based Indexing for Lidar Data. Photogramm. Eng. Remote Sens. 78, 927–934. https://doi.org/10.14358/PERS.78.9.927

Creating Analytics Optimized ICESat-2 Data for Biomass Mapping was originally published in Development Seed on Medium, where people are continuing the conversation by highlighting and responding to this story.

Build this…

… using this.

Flyovers and 3D maps provide context and perspective, especially when terrain is part of the story. By following a GPX trace, bring users along for an Iditarod, a dangerous border crossing, a hike along the PCT, or their afternoon run. In the next few weeks, a GPX trace plugin is coming to Mapbox GL Director. Sign up for updates.

Terrain Plus Data

Overlaying data unlocks powerful new insights: landslide risk, microclimate zones, wildlife habitat range, solar mini-grid proposals, and more. In the coming weeks, there will be new features to overlay, combine, and style data directly within Mapbox GL Director. Sign up for updates.

Build better together

The code behind Mapbox GL director is open source under an MIT license. You can contribute to the project or file bugs and feature requests on GitHub.

Add flyover to any app with Mapbox GL Director was originally published in Development Seed on Medium, where people are continuing the conversation by highlighting and responding to this story.

I’m excited to welcome Jennifer to Development Seed! Jennifer is a Cloud Engineer who is passionate about accessibility and meaningfully expressing systems and concepts in code.

Alongside the Earthdata team, she will be working to build core infrastructure for Cumulus, which helps to better leverage cloud computing for data processing, storage, and retrieval of NASA’s Earth observation data.

Before joining Development Seed, Jennifer worked as a Junior Software Engineer at Acorns where she helped build the backend of the Acorns Spend product.

When not contributing to open source work, Jennifer enjoys meditating and trying new recipes.

Join me in welcoming Jennifer to the team! 👋

Welcome Jennifer Tran! was originally published in Development Seed on Medium, where people are continuing the conversation by highlighting and responding to this story.

I’m excited to welcome Tânia to Development Seed! Tânia is an experienced UI/UX designer who infuses digital products with positivity and meaning. At Development Seed, she will think deeply about how people interact with geospatial data in the browser and designing products that are highly polished and delightful to use.

Before joining Development Seed, Tânia worked as a Digital Product Designer at Major, a design studio in Lisbon. She worked on several projects, from creating and expanding a marketplace to a complete redesign of Portugal’s national postal service. Throughout, she oversaw projects from the initial discovery phase to user testing and analytics.

Outside of work, Tânia enjoys reading books, watching movies, and spending time at Lisbon’s gardens and miradouros. As a new mom, most of her spare time is spent playing with her daughter and raising her to be a feminist 🌈

Say hi 👋 to Tânia on Twitter and welcome her to the team!

Tânia joins the team in Lisbon! was originally published in Development Seed on Medium, where people are continuing the conversation by highlighting and responding to this story.

I am so pleased to welcome Omnia to the DevSeed family! She joins our operations team in a vital capacity that will increase our ability to execute on day-to-day operations, culture, and financial administration. Omnia brings extensive experience running nonprofit operations and will help create and implement processes around office management, admin and logistics, and financial reporting. She serves in a critical role that not only ensures seamless execution of operations tasks, but will help build out our diversity and inclusion program, and provide onboarding support to new team members. She is a rock star and will be essential to ensuring DevSeed is a people-centric workplace!

Omnia embodies the role of activist both in her personal and professional life. She developed a deep passion for working with groups focused on providing meaningful and impactful change domestically and globally while studying government and politics / women’s studies in college. In DC, she worked to scale operations at both the macro and micro level to combat the ever present homeless crisis for an organization dedicated to implementing strategic housing initiatives for women in the region. She also worked for a group dedicated to eradicating poverty globally, and in both capacities refined her operational expertise to working with globally distributed teams.

Outside of work, Omnia can be found running after her two cats and reading anything from blogs on home renovations to books soon to be made into movies and TV shows.

Join me in welcoming Omnia to the team! 👋🏼

Omnia joins the operations team! was originally published in Development Seed on Medium, where people are continuing the conversation by highlighting and responding to this story.

I’m pleased to welcome Ciaran to Development Seed! He is a cloud engineer who brings deep expertise around developing serverless tools and environments to use with global sized datasets to the team. In particular, he has extensive experience with Sentinel data and deep scientific knowledge of satellite imagery and processing. He will join our team working to ingest, catalog and distribute new commercial datasets with groups like NASA and other data providers. In this capacity he’ll extract satellite orbit data to display satellite tracks on the frontend and focus on improving the overall architecture for this capability.

Ciaran has developed rigorous coding standards due to the nature of his background. As a data engineer at the UK Hydrographic Office (UKHO) he produced serverless data pipelines that used deep learning models to perform inferences at a global scale. He’s built cartography products that people’s lives depended on, including for his dissertation project, where he created a spatial dashboard for first responders and disaster relief planners.

Outside of work, Ciaran often daydreams how he can achieve his childhood dream of becoming an astronaut 👩‍🚀, but for now has settled on working with satellite imagery.

Say hi 👋🏼 to Ciaran on Twitter and welcome him to the team!

Welcome Ciaran, cloud engineer, to the team! was originally published in Development Seed on Medium, where people are continuing the conversation by highlighting and responding to this story.

Managing models, predictions and infrastructure

ML Enabler makes it incredibly easy to spin up infrastructure to run your model along with all necessary resources. Through the new user interface, you can upload new models, spin up AWS resources, generate and preview predictions. Previously, there was a minimal CLI tool to upload models and fetch predictions.

Behind the scenes, ML Enabler uses AWS Cloudformation and will work with any AWS account. A few key infrastructure choices like instance count, and concurrency can be made directly from the ML Enabler interface. ML Enabler uses lambda functions for downloading base64 images for inference from the specified Tiled Map Service (TMS) endpoint and writing inference outputs into the database.

You can monitor the tile prediction queues right from the UI. When the processing is complete, predictions are automatically displayed in the map tab. It’s easy to toggle between different classes in your model, and filter predictions based on confidence threshold. Over each tile, the model’s raw output and the confidence score is displayed. This makes it really convenient to explore spatial patterns within the inferences.

On-demand predictions

ML Enabler generates and visualizes predictions from models that are compatible with Tensorflow’s TF Serving, on-demand. All you need is to drag and drop a zip with trained classification or object-detection model, provide a TMS end point, and an AOI for the inference. ML Enabler will spin up the required AWS resources and runs inference to generate predictions. Running a classification model inference over a medium sized city, which is divided into approximately 4,000 zoom 18 tiles takes approximately 2 minutes.

The prediction tiles are indexed using quadkeys for easy spatial search. To help facilitate these on-demand predictions ML Enabler has integrated in many of the components of Development Seed’s Chip-n-Scale project.

Support for classification and object detection models

Currently, ML Enabler supports two common machine learning model formats — classification, and object detection. ML Enabler works with binary as well as multi-label classification models. The infrastructure setup and prediction visualization adapts automatically based on the model format. For object detection models, ML Enabler converts coordinates in the pixel space to geographic space for every prediction along with bounding box and confidence score.

Additionally, for classification models, ML Enabler supports a custom lambda function to create supertiles. Supertiles are incredibly useful to overcome objects that may lie across tile edges. For example, zoom 18 tiles offers a higher resolution than zoom 17, but one draw-back is that sometimes buildings get split between multiple tiles. Supertiles allows for the aggregation of the four zoom 18 tiles within the zoom 17 footprint to create a (512, 512, 3) training chip, instead of the typical (256, 265, 3) training image chip.

Collecting feedback about predictions and retraining

Another exciting feature we added to ML Enabler is the ability to collect feedback about predictions from within the interface. Users can tag a tile as valid or invalid. Predictions tagged as valid switch to green, predictions tagged as invalid switch to white, and predictions that haven’t been manually validated stay red.

ML Enabler can then convert these validated predictions back into labeled training data matched up with imagery to allow users to easily re-train a new model with the validated model predictions.

Future

We think that it can make the integration between mapping tools and model infrastructure easy and less intimidating. There are an increasing number of internal and partner projects that rely on ML Enabler and we will continue developing and maintaining.

Some of our immediate plans include ability to run inference over imagery sources other TMS, automate re-training workflows, and the ability to add more detailed model metadata to the registry. We hope you get the chance to experiment ML Enabler for yourself. Please reach out with any questions or comments on Github or Twitter!

On-demand machine learning predictions for mapping tools was originally published in Development Seed on Medium, where people are continuing the conversation by highlighting and responding to this story.

I’m delighted that Katy Money is joining Development Seed as our Head of Business Development to help us grow and better serve our partners and clients.

Our Partners are what makes working at Development Seed the most rewarding and exciting work of our lives. We are extremely fortunate to work closely with the some of the most innovative and driven teams at massive data players like NASA, European Space Agency, UNICEF, USGS, the World Bank, and others. Not to mention our partnerships with exciting entrepreneurial change-makers like OpenAQ and Humanitarian OpenStreetMap Team, and dozens of mission aligned nonprofits, innovative tech companies, and municipal governments that inspire and invigorate us every day. I’m so very grateful that these organizations trust us to help them build their dreams; to build the change they want to see in the world. We are smarter and better for these partnerships and for contributing to their extremely important, world changing initiatives.

For every group that we work with today, there are a dozen others that could benefit from the same open software, open algorithms, open data, and open knowledge that we are generating in our existing work. Katy joins Development Seed to help us grow our impact through new and deeper partnerships to better serve our communities. She will help us to reimagine the ways we can serve our existing partners and ways that we can package our work to make it more useful and accessible to a broader set of users. Katy is going to scale our impact-per-person, allowing us make a bigger dent in the universe while remaining a lean and hungry team.

Katy is no stranger to this challenge. Katy led growth and partnerships at Ushahidi, an organization that I deeply respect for its ability to engage the global development, humanitarian, and mapping communities. She’s developed a keen insight for the value of data-driven decision making for social good in her work at GeoPoll, Chemonics International, and AidData.

Outside of work (and in pre-pandemic times) Katy loves getting outside, traveling to new countries, trying new restaurants, and playing golf.

Reach out to Katy and welcome her to the team or find a time to connect about the work you’re doing and how DevSeed could help:

Twitter: @katymny

Github: @katymoney27

LinkedIn: https://www.linkedin.com/in/katy-money-4861733b/

Email: katy@developmentseed.org

Katy Money joins DevSeed to lead growth and business strategy was originally published in Development Seed on Medium, where people are continuing the conversation by highlighting and responding to this story.

As of July 14, India has more than 900,000 confirmed cases of COVID-19. Only the US and Brazil have more confirmed cases. India took an aggressive approach to lock down around the country in March after only a few dozen confirmed cases. Since lifting most movement restrictions, cases have rapidly increased in India, particularly in large cities with significant urban slums.

Mapbox Movement Data provides a look at how these restrictions unfolded in different communities across India. When combined with population data on vulnerability and COVID disease dynamics, a compelling story emerges around the impact of national and local policies through the pandemic. We explored mobility trends pre/post the lockdown announcement, and looked at the high concentration of containment zones in slums and vulnerable areas.

Explore the interactive story →

Movement in cities has a high correlation to the spread of the virus. Reopening plans must consider highly localized movement patterns to design policies that are appropriate for very different communities.

As governments begin to feel pressure to re-open their economies, our aim is to help them do so safely and more efficiently by collecting, publishing, and better understanding geospatial and location data surrounding COVID. Correlations between movement and social policies reveal that in both urban and rural areas, no two response efforts yield the same results. To learn more about our COVID mobility work, check out our latest piece on diverging patterns as states in the U.S. begin to open up again and the impact on case numbers.

Reopening a country of 1.3B people was originally published in Development Seed on Medium, where people are continuing the conversation by highlighting and responding to this story.

Mobility data is a powerful tool for understanding and anticipating COVID infection rates. Mobility data from cell phone has been used to track the initial spread of the disease, to inform social distancing rules, and to evaluate the impact of recent protests for racial justice on disease spread. But not all movement is the same. As we work with governments and decision-makers to use mobility data for COVID planning, we have noticed a surprising trend. The relationship between movement and COVID infection rates is evolving along different paths in different locations.

Using movement data from our good friends at Mapbox, we analyzed movement patterns across 1129 cities and counties from pre-COVID through the current spike of cases. From this data, we created a unique “Movement-COVID Fingerprint” for these areas. These fingerprints show the very different paths that communities have taken through COVID response and recovery.

Our preliminary conclusions from our work with mobility data are:

• At the beginning of the crisis, there was a strong correlation between decreased movement and a subsequent reduction of infection rates a few weeks later.
• Over the past few months, some cities have been able to increase movement and maintain decreasing infection rates, while others have not.
• Of a range of social and demographic indicators, lower social vulnerability and percentage of African American residents are most correlate with locations that maintain decreasing R(t) (transmission rate) while increasing movement.
• Voting patterns are also correlated, but less so. Areas that vote primarily democratic are slightly more likely to have a favorable Movement-R(t) Fingerprint. This is consistent with other research that suggests that people who identify as Democrats are also more likely to wear a mask and maintain safe distances while moving.

Mobility data for COVID response

Movement restrictions are a critical tool for governments to combat the COVID-19 pandemic. However, these restrictions are extremely painful. Understandably, governments want to relax restrictions to the minimum level necessary to keep the virus under control. We’ve seen the best results from governments that have taken a highly data-driven approach to determining when and how to relax restrictions. Mobility data is extremely valuable for evaluating actual social distancing practice and fine-tuning policies to reopen the economy.

We use Mapbox Mobility Data with our partners. We like the Mapbox data because it is:

• very high spatial resolution (to the city block)
• frequent and fast (daily data delivered by next day)
• privacy protecting (no personal data collected from the phone)

To support data-driven COVID response, Mapbox has created a new mobility dataset to measure social distancing practices under COVID. Mapbox Movement Index (MMI) is a mobility dataset to monitor population movement relative to pre-COVID movement rates, down to the city block. This data is incredibly useful when combined with data on infection rates, hospitalization rates and data on social vulnerability.

In the very beginning of the crisis, through the initial lockdown we saw a significant correlation between movement [MMI] at one point of time and infection rates [R(t)] roughly two weeks later.

However over the course of reopening, we’ve seen this relationship diverge. Some locations have been able to increasing movement without a corresponding increase in R(t).

Movement-R(t) Fingerprints

An analysis tool that our partners have found particularly valuable, is a simple plot of the evolving relationship between movement and R(t) at the city level.

Epidemiologists and policymakers track the average ratio of new infections spread per person in a region, or R(t), to understand how the virus spread. Government and public health officials have all honed in on a goal of an R(t) of less than 1. An R(t) of 1 means that you’ve flattened the curve. Sustained rates above 1 mean that you are on a path to overwhelming your hospital system. The higher the number, the more quickly you will overwhelm the hospital system and the disease will be even more lethal. You can read more detailed explanations here.

A standard time-series line chart can show both MMI and R(t) on the same chart.

But this view of the data can slightly obscures the relationship between the two variables, and how they move with a time delay. We created an alternative “four quad” view. This chart plots the same MMI and R(t) values for each date as a scatter plot, connected by a line to show progress through time. We then annotate these scatter plots with policy decisions and thresholds to indicate broader patterns.

Looking at the relationship between movement data and R(t) in this fashion provides a fine gauge on how movement and social distancing policies impact actual behavior and movement levels, and in turn how these translate into slowing COVID infection rates.

With this visualization, you hope for a C-shaped graph. At the start of the crisis, people are moving normally and the virus is spreading at its natural rate (R(t) of around 2). Cities and states then enact restrictions on gathering and movement. First, movement levels fall dramatically and within a few weeks these restrictions result in lower infection rates (movement from quadrants A→B→C). A well managed county can then gradually and carefully reopen. In these cases, movement increases without a subsequent increase in R(t) (A→B→C→D). If a county opens too early or too aggressively, infection rates could go back up (A→B→[C→D]→A). These patterns, especially when overlaid with policy decisions, can show us how effectively individual counties have responded to the pandemic.

New York, NY

New York City starts with high-movement and high-R(t). As social distancing policies are enacted, movement decreases, but it takes a few weeks for R(t) to catch up. Eventually once R(t) is below 1.0, movement restrictions can be eased while the resulting impact on R(t) is carefully tracked.

Washington, D.C.

Washington, D.C. moves in a similar pattern. More restrictions imposed seemed to have slowed down movement, though not to the extent of NYC. Interestingly, the Stay At Home order seems to have been instituted after movement had already significantly decreased.

Harris County, TX

Harris County in Texas (home to Houston), represents a county where the curve hasn’t followed the same path. Movement only dropped to about half of the baseline, as opposed to 20–35% in DC in New York. The infection rate in Harris County continues to rise. A mask mandate was ordered on July 2nd.

Miami-Dade County, FL

Miami-Dade County’s curve looks similar to Washington, DC for the first month. As movement starts increasing, there is a sharp rise in R(t). It subsequently falls for a two week period before rising again.

Divergent Reopening experiences

In April and May, fingerprints followed predictable patterns. A significant decrease in mobility would eventually be followed by a steady decline in R(t). Cities that did not reduce movement would churn in a high R(t) cycle.

In June and July we’ve seen a greater divergence in experience. Overall movement has increased in most cities and counties. However some counties were able to increase movement while keeping R(t) low. Others were not.

This suggests that not all “re-opening” is the same. How people move about and interact matters. In addition to where they are going, what activities they engage in, and critically what precautions they take to safely go back into public can have significant effects on resulting transmission rates. Do people wear masks in public? Do they keep a safe distance? Do they avoid crowded indoor spaces?

What factors lead to “good reopening”

This divergence presents a critically important research question for safely reopening an economy. What practices allow a community to maintain low infection rates even as you increase the movement and economic activity? It isn’t enough to study locations that have low infection rates vs. high infection rates. It is more interesting to study communities that have been able to increase movement while maintaining low R(t), to understand what allows them to achieve this. Essentially, what are the characteristics that allow an area to break the relationship between movement and disease rates. This provides hope for developing safer ways to reopen the economy.

We looked at a range of demographic, economic, and social indicators to determine which best correlate with places that have managed to break this relationship across 1129 cities and counties in the US.

Our preliminary findings suggest that:

• Areas with vulnerable populations are harder to reopen safely. A county’s Social Vulnerability Index (SVI) score was the most predictive factor on whether increased movement would lead to increased transmission rates.
• Numerous studies suggest that Black and Brown people are disproportionately paying the cost of COVID. Our analysis is consistent with that finding, in that areas with a higher percentage of African Americans were less likely to escape the movement/COVID relationship.
• Areas that tend to vote for Democrats were slightly more likely to be areas that fared better as movement increased. This is consistent with other research that finds that people who identify as Democrats are more likely to trust public health officials and are more likely to engage in practices like wearing masks.
• Restaurants probably deserve scrutiny. For the small subset of cities where restaurants reservation data is available, restaurant reservation trends were highly correlated with disease trends. That doesn’t necessarily mean that restaurants are worse than other locations. We didn’t have any data that would allow us to look at attendance at gyms, bars, tattoo parlors, libraries, or grocery stores.

There is more research to do. We will continue to investigate how communities reopen safely and would love to team up with other researcher and policymakers. Please contact us if you’d like to collaborate.

R(t) data from the CovidActNow API

Mobility and COVID Infection Rates was originally published in Development Seed on Medium, where people are continuing the conversation by highlighting and responding to this story.