How Google is changing the weather forecasting game
How Google is changing the weather forecasting game
PLUS: a new toolbox to detect anomalies, measuring building height from street-view images, how the digital economy is reducing emissions, and more.
Hey guys, welcome to this week’s edition of the Spatial Edge — a weekly round-up of geospatial news that you can digest in less than 5 minutes. We’re disrupting your inbox the same way What3Words disrupted the coordinate system — gently and quietly and without much success.
In today’s newsletter:
Google AI Predicts Climate Trends: combining traditional methods with ML.
Digital Economy and CO₂: how the digital economy reduces emissions.
New Toolbox for Extreme Events: AIDE uses AI to detect and assess extremes.
Estimating Building Heights: Pano2Geo model uses street-view images for height estimation.
Geospatial Datasets: UCDP releases new armed conflict data for 2024.
Research you should know about
1. Google AI can now predict long-term climate trends and weather in just minutes
Google has developed a new AI-powered weather forecasting model called NeuralGCM, which combines traditional physics-based prediction methods with machine learning.
Traditional climate projections use General Circulation Models (GCMs), which simulate the atmosphere with physics-based equations. I did a deep dive on GCMs that you can check out here.
Lately, machine learning models have been better than GCMs at short-term weather predictions, but they haven’t been able to keep up when it comes to long-term forecasts.
Google’s NeuralGCM is different. It can nail short-term weather forecasts (one to ten days) and also medium-term ones (up to fifteen days), as well as traditional models. On top of that, if you feed it sea surface temperature data, it can predict climate patterns for decades.
For long-term forecasts, it’s on par with the European Centre for Medium-Range Weather Forecast’s ensemble model (ECMWF-ENS), which is basically the gold standard.
So, why should we care about this?
Accurate weather and climate forecasts are helpful for planning and decision-making in a bunch of areas:
agriculture
disaster preparedness
managing water resources
Better short-term forecasts can help us prepare for storms and extreme weather, potentially saving lives and reducing damage.
On the long-term side, predicting climate patterns for decades can help with tackling climate change. If we can see what's coming, we can develop better strategies to mitigate the impacts. NeuralGCM is game-changing because it gives us the best of both worlds—more accurate predictions in the short term and reliable long-term forecasts. Plus, it's more efficient, which means we can get these insights faster, cheaper, and with less hardware.
The code for the project is available here: https://github.com/google-research/neuralgcm
2. Impact of the digital economy on carbon dioxide emissions
A new study in Scientific Reports looks at how the digital economy affects CO₂ emissions in cities that rely heavily on natural resources. They found that as the digital economy grows, it can actually help cut down CO₂ emissions, especially in the early stages.
Here’s how it works:
Raising Awareness: The digital economy makes people more aware of environmental issues, because ‘digital information’ can be spread quickly and more widely. This public concern pushes businesses and governments to go green and adopt more sustainable practices.
Boosting Efficiency: Technology like the Internet of Things (IoT), big data, and AI help make everything more energy-efficient. E.g. smart sensors that cut down energy waste in factories.
Stricter Regulations: In places with strong environmental regulations, the impact is even bigger. These rules encourage the use of digital tech to save energy and use cleaner energy sources.
The study examined 107 resource-based cities in China from 2011-2021. They used data on CO2 and the digital economy (the China Digital Inclusive Finance Index) from the Chinese statistical yearbooks.
Interestingly, the biggest reductions in CO₂ emissions happen early on as the digital economy develops, and then it starts to level off.
This is a big deal for cities that depend on natural resources and are trying to tackle both pollution and economic challenges. Basically, embracing digital technology could be a game-changer for making these cities more sustainable and reducing their carbon footprints.
3. A new toolbox for identifying extreme events
Researchers from the University of Valencia have developed a new toolbox on Artificial Intelligence for Disentangling Extremes (AIDE). AIDE essentially provides a full pipeline for the detection, characterization and impact assessment of extreme events using ML and computer vision tools.
The pipeline consists of:
Data loading and pre-processing
ML model selection and training
Evaluation and visualization of results
So how does this all work?
Users can provide temporal (i.e. time series), spatial (across space), or spatio-temporal (both time and space) datasets. This can come from a bunch of different sources like satellite images, climate models, or historical climate data.
The toolbox can be used to clean and transform the data to make it useful for analysis. This might include normalising the data, removing noise, etc. The toolbox then offers a number of different machine learning and deep learning models.
Users can then select the model based on their use case (e.g. anomaly detection or event prediction). The trained model can identify anomalies or extreme events in the data. It can then be used to classify events (e.g. droughts or floods) to assess their impacts.
The toolbox also involves explainable AI, so you can understand which features (e.g. temperature, soil moisture, etc.) are important for detecting anomalies.
You can access the toolbox here: https://github.com/IPL-UV/AIDE
4. Improving agricultural maps using geospatial methods
A publication by the Asian Development Bank describes how geospatial technology can be used to more easily and accurately map agricultural land.
Technology like GPS devices, satellite imagery, and drones are becoming increasingly affordable. These tools can change the way agricultural statistics are measured since they unlock more precise and frequent monitoring of farming activities.
We’re now seeing national statistics offices and agricultural ministries starting to use these tools to get more detailed info about where farms are and how big they are, making the whole process a lot more efficient.
The traditional way of gathering farm data can be pretty costly, done infrequently, and inaccurate because it relies on farmers' estimates. Plus, many national statistics offices had previously struggled with new tech because they needed skilled staff and better IT systems to manage all the data. These issues make it tough to get the accurate, reliable agricultural stats that are crucial for good policy-making and resource allocation.
So, to address this, ADB has combined geospatial data with traditional survey methods. For example, countries like China, Korea, and Sri Lanka have seen better agricultural stats by combining farm lists with geographical data. In the Cook Islands, using high-res satellite images to map farm areas showed how effective this tech can be for getting accurate measurements.
This ultimately helps governments with planning for food security, agricultural programs, and managing environmental impacts.
5. Pano2Geo: estimating building heights from street-view photos
A new paper from the ISPRS Journal of Photogrammetry and Remote Sensing introduces the Pano2Geo model, which takes street-view panoramic (SVP) images and converts them to geospatial coordinates to estimate building heights.
A 360° street-view image is taken from a camera (mounted on a car) — similar to the Google Street View cars. Each pixel in this image is mapped onto a 3D space. Using the direction the car is facing, the authors adjust the 3D points to match real-world locations. They use geometry to convert the angles and distances from the camera to each building into real-world coordinates (i.e. latitude, longitude, and height).
Following this mapping to real-world coordinates, the authors segment buildings (i.e. to find the outlines of buildings). They then detect the roof and base of each building and calculate the vertical distance between the two. The final product is building height information.
This can be pretty useful for urban planning. Instead of sending surveyors with expensive equipment, you can use the Pano2Geo model. You can collect street-view images, process them through the model, and get accurate building heights quickly and efficiently. This approach saves time, costs less, and provides reliable building height estimates at scale.
P.S. The source dataset and code are available via https://github.com/Giser317/Pano2Geo.git
Geospatial Datasets
1. UCDP Candidate Events Datasets
Uppsala Conflict Data Program (UCDP) has released its candidate events data for the first six months of 2024. UCDP is my go-to resource for armed conflict-related data.
This new release includes disaggregated datasets on georeferenced events and candidate events. There are also yearly datasets on armed conflict, one-sided violence, non-state conflict, and more.
2. Level 4C Footprint Level Waveform Structural Complexity Index (WSCI) Dataset
WSCI is important for understanding habitat quality, species diversity, and how ecosystems work. The data, collected through detailed airborne laser scanning, includes uncertainty estimates and quality flags, covering April 2019 to March 2023.
Future updates will improve data resolution and geolocation accuracy, making it an even more valuable tool for ecological research and conservation.
3. Hyperspectral Albedo Maps Dataset
A new dataset on hyperspectral albedo maps gives a detailed look at how different surfaces on Earth reflect sunlight across various wavelengths.
These maps help precisely identify and analyze surface materials like vegetation, soil, water, and urban areas. They help us to understand the Earth's energy balance and climate processes.
The maps can be pretty useful for monitoring environmental changes, assessing vegetation health, and improving climate models.
Other useful bits
ESA’s EarthCARE satellite, which launched into orbit on 29 May, has begun transmitting images. Its first images provide info on different clouds and cloud temperatures across the globe.
ESA and UKSA announced a joint funding opportunity: InCubed - Innovation in Public Services with Satellite Earth Observation. This includes projects on infrastructure monitoring, solar and wind energy, precipitation modelling, and nowcasting.
The UK Space Agency is investing £33 million in space sector innovation through their National Space Innovation Programme. This investment is split between eight projects, all aligning with the Agency’s goals of supporting innovative technology, providing opportunities across the UK, and bringing the benefits of space exploration to people on Earth.
Overture Maps—which Microsoft, AWS, and Meta back—launched its first open map datasets, which included open address data for 14 countries for just over 200 million address records. This map pools a bunch of data sources, including open datasets from projects like OpenStreetMap, government data, their own proprietary info, and even data from Google.
ESA’s soon-to-be-released Φsat-2 will also have an application that aims to convert satellite images into street maps to identify roads that remain accessible during emergencies. The main idea behind the app is to provide quick, useful information to ground teams during emergencies like floods.
NASA created a new tutorial on navigating the Earth's Surface Mineral Dust Source Investigation (EMIT) L2B Estimated Mineral Identification and Band Depth and Uncertainty data. EMIT L2B provides estimated mineral identification and band depths in a spatially raw, non-orthocorrected format. You can learn more about it here.
Copernicus published a situational report on the oil spill in Manila Bay, Philippines. The oil spill was observed before the tanker even capsized, but it was difficult to contain following the aftermath of the recent typhoon in the country.
Geoawesome details how the maps in Game of Thrones were created, including those for the hit spin-off series House of the Dragon. They talk about some of the surface visualisation techniques used to provide detailed fictional landscapes.
This is a very cool video of a suborbital rocket launch (the OS-X1, to be exact) seen from a satellite perspective. You don’t see rocket launches from this perspective that often (or even at all). Learn more about the OS-X1 here.
Jobs
World Economic Forum is looking for a Specialist in Space Technology. The specialist will be under its Centre for the Fourth Industrial Revolution (C4IR).
An ESA-funded project is looking for a Postdoc in Machine Learning for Earth Observation.
Our World in Data is looking for a Research and Data Project Lead in Economics.
Global Facility for Disaster Reduction and Recovery (GFDRR) is looking for a young Geospatial Data Scientist who will join the World Bank’s Junior Professional Associates (JPA) Program.
Coastal Carbon is looking to fill a number of vacancies: AI Scientists, AI Engineers, MLOps Engineers, and Chiefs of Staff.
GAF Munich is looking for a Project Manager in Earth Observation and Copernicus Land Cover and Land Use (LCLU) downstream Services.
Development Seed is recruiting a Geospatial Services Engineer to develop geospatial data services and advance open-source geospatial technologies.
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.