Inequalities in flood adaptation measures
PLUS: assessing the accuracy of rainfall estimates in real-time, measuring field boundaries under clouds, global poverty updates and more.
Hey guys, here’s this week’s edition of the Spatial Edge — a weekly round-up of geospatial news that you can digest in less than 5 minutes. After last week’s edition, I’m happy to report we now have a real testimonial from Professor Neil Lee:
This means there’s only one more testimonial left on my bucket list. So Barack, if you’re reading this…
In today’s newsletter:
Inequalities in Flood Adaptation: Low-income areas receive less support
Real-Time Rainfall Accuracy: NOAA's NPreciSe tool validates satellite data
Field Boundaries in Clouds: PTAViT3D detects fields despite cloud cover
Converting SAR to Optical Images: Dual-GAN enhances radar visuals
EU Crop Map 2022: Detailed crop types across Europe
Research you should know about
1. Inequalities in flood adaptation measures
A new study in Nature Communications looks at inequalities in flood adaptation policies in the US.
The study assessed the impact of the National Flood Insurance Program's Community Rating System (CRS) using a FEMA dataset of ~2.5 million flood insurance claims. They used a method called CausalFlow, which is a deep learning-based causal inference methodology they developed. This is used to figure out how flood adaptation measures impact communities based on factors like income, race, and flood risk.
The results showed that flood adaptation can save communities between $5,000 and $15,000 per household, but these savings aren’t shared equally. Low-income communities, especially those at higher risk of flooding, get much less help compared to wealthier areas.
P.S. You can access the data here.
2. Assessing the accuracy of rainfall estimates in real-time
A new study from Scientific Data introduces NPreciSe, a tool from NOAA that checks how accurate satellite rainfall estimates are in almost real-time.
The tool compares satellite data against reliable ground observations from the Multi-Radar/Multi-Sensor (MRMS) network and NOAA's Stage IV data. It essentially helps us determine which satellite-based datasets are the most accurate. This in turn helps us assess the various biases in satellite-based estimates.
The study found that NPreciSe effectively standardises the validation process, making it easier to understand the strengths and weaknesses of various datasets. As I discussed in last week’s edition, I find it pretty annoying that different satellite-based datasets provide different results when measuring the same thing. My hope is that this type of feedback can help us improve satellite-based estimates going forward.
P.S. You can access the code here.
3. Measuring field boundaries amongst the clouds
New research from CSIRO and the University of Strasbourg provides new insights on how to detect agricultural field boundaries that are located in cloudy areas.
The team introduced a 3D Vision Transformer architecture, with the catchy name of ‘PTAViT3D’, to process time series data from Sentinel-2 and Sentinel-1 satellites. They developed two models: PTAViT3D, which uses either Sentinel-2 or Sentinel-1 data, and PTAViT3D-CA, which combines both datasets using a cross-attention mechanism. These models are essentially able to handle images with partial or dense cloud cover without the need for manual cloud filtering.
The results show that PTAViT3D can pinpoint field boundaries just as well as traditional methods that only use cloud-free optical images.
P.S. You can access the code here.
4. Using generative AI to convert SAR data into optical-like images
Another week and another attempt to translate SAR data into optical-like imagery. This time, a new study explores a method of doing this using Generative Adversarial Networks (GANs).
They compared a bunch of different GAN models—including Pix2Pix, CycleGAN, and S-CycleGAN—and introduced a novel dual-generator GAN that uses partial convolutions and transformers. They trained these models on paired SAR and optical images from Sentinel-1 and Sentinel-2 satellites.
Their findings showed that the dual-generator GAN produced optical images with improved ‘visual fidelity’ and feature preservation compared to existing models.
However, that said, we need to be careful of using GAN-based results for analysis. I go into more details about these concerns here. But of course, if the aim is to provide something that looks nice and visually appealing, then that’s totally fine.
Geospatial datasets
1. EU crop map dataset
A new Scientific Data study presents a detailed map showing the types of crops grown across the European Union and Ukraine in 2022. It combined satellite images from Sentinel-1 and Sentinel-2 with ground data from over 134,000 locations in the LUCAS survey.
P.S. You can access the replication code here.
2. Forest regrowth wildfires dataset
Another study from Scientific Data provides a global map of forest recovery after wildfires from 2000 to 2020. It includes details on forest regrowth like height, biomass, leaf area, and light absorption, all at a 30-meter resolution.
P.S. You can access the replication code here.
3. Gross primary production (GPP) dataset
The Hi-GLASS GPP v1 provides high-resolution data on gross primary production (GPP) across China from 2016 to 2020. It uses 30m Landsat data and a light use efficiency model to improve estimates of GPP.
4. Individual canopy dataset
The Winniepeg individual canopy data provides info on over 1.4 million tree canopies in the city. This includes data on canopy height, crown area, and density.
Other useful bits
Methane and carbon dioxide plumes have been successfully detected for the Carbon Mapper satellite mission. The instrument was designed by NASA’s Jet Propulsion Lab, and is able to pinpoint emissions from oil and gas facilities.
The World Bank's September 2024 Global Poverty Update brings some encouraging news: global extreme poverty has largely returned to pre-COVID levels. However, many low-income countries are still struggling to recover due to inflation and other challenges. This update includes new survey data and refined methods, providing a clearer picture of poverty trends across 170 economies.
The second edition of Geocomputation with R is close to completion and features some big updates: it now includes faster tools like the terra package and new features for handling spherical geometry. Plus, there are new sections on cloud services and geographic metadata. The book remains open-source, and the full release is coming in January 2025.
Google Earth Engine has launched its Earth Engine v1 Python library, making it easier to use geospatial analysis tools in Python. This simplifies access to satellite imagery and environmental data and allows us to better integrate it with Pandas and NumPy.
The World’s most innovative country is Switzerland, according to WIPO. ‘Innovation’ here is measured in terms of patents, scientific publications, R&D expenditure and so on. It’s also calculated using a bunch of big data, such as the number of GitHub commits per country.
Jobs
Orbital Insight/Privateer is looking to fill several vacancies, ranging from Director roles to Data Scientists and Engineers.
The European Environment Agency is looking for an Expert to lead the implementation of the Copernicus Contribution Agreement.
The World Bank is looking for a Data Scientist to join their Data Lab and Development Data Partnership team.
Hummingbirds is looking for a Remote Sensing Intern based in Paris, France who will support their Nature-based Solutions portfolio development.
ESA is looking for an Earth Observation Digital Innovation Engineer based in Frascati, Italy who will work within their ESA ɸ-lab.
Just for fun
ESA’s Euclid mission has released its first map of the Universe, featuring millions of stars and galaxies in a massive 208-gigapixel mosaic.
That’s it for this week.
I’m always keen to hear from you, so please let me know if you have:
new geospatial datasets
newly published papers
geospatial job opportunities
and I’ll do my best to showcase them here.
Yohan