Academic
- Author: eartharxiv.org
- Category: article
- Document Tags: paper
- URL: eartharxiv.org/repositor…
- Author: MapScaping
- Category: podcast
- Document Tags: audio
- URL: podcasts.apple.com/gb/podcas…
- Author: Aravind
- Category: rss
- URL: newsletter.terrawatchspace.com/why-scien…
- Author: Population Division of the United Nations Department of Economic and Social Affairs (UN DESA)
- Category: pdf
- URL: population.un.org/wup/asset…
-
The world has become increasingly urban; more people live in cities today than in towns or rural areas.
-
The number of “megacities” (10 million inhabitants or more) continues to grow; over half are in Asia.
-
More people live in small and medium-sized cities than in megacities; many of these smaller settlements are among the fastest growing, especially in Africa and Asia.
-
Growth of the world’s city population between now and 2050 will be concentrated in seven countries.
-
City population growth is uneven; most cities are growing, but thousands have shrinking populations.
-
Towns are home to more than a third of humanity and are critical for sustainable development.
-
As the world’s rural population approaches its peak size, it faces unprecedented challenges.
-
The expansion of built-up areas is outpacing population growth worldwide.
-
The Degree of Urbanization methodology reveals the world is more urbanized than national statistics suggest.
-
Sustainable development requires integrated planning that treats cities, towns and rural areas as interconnected and interdependent.
- Embeddings provided by Google Satellite Embeddings V1
- Aerial images are provided by Esri World Imagery
- Liverpool boundary used to filter pixels by ONS
- Author: Jonathan Reades, Yingjie Hu, Emmanouil Tranos, Elizabeth Delmelle
- Category: pdf
- Document Tags: paper
- URL: www.nature.com/articles/…
- Author: Krzysztof Janowicz, Gengchen Mai, Weiming Huang, Rui Zhu, Ni Lao & Ling Cai
- Category: article
- Document Tags: paper
- URL: www.tandfonline.com/doi/full/…
- Author: Nathan Zimmerman
- Category: article
- URL: element84.com/machine-l…
Yesterday, we (the Imago team) gave a workshop on satellite embeddings for a room with about 50 people from different corners of the UK Government. It was fantastic in every respect (with the possible exception of Github bringing down our website because it thought it was receiving a DDoS attack lol). It’s so exciting to see these ideas and technology move at the speed of light from state of the art research to Government. Anything we can do at Imago to facilitate that transfer, we’re here for that!
If you’re curious, materials are available at:
imago-sdruk.github.io/EMBED2Soc…
And if this looks interesting to your and/or your organisation, do get in touch!
🔗 Earth Embeddings: Towards AI-centric Representations of our Planet
Very very timely paper that captures the current zeitgeist in EO and AI. If nothing else, it serves as a fantastic introduction to one of the technologies that I think(/hope) will help the most bring imagery to the masses in the coming years.
Metadata
Highlights
Earth embedding vectors emb are produced by a family of embedding functions E that map continuous location inputs (i.e., longitude, latitude with optionally elevation, and time) into a d-dimensional vector space:
Figure 2: Earth embeddings provide different functions: (1) They compress high-dimensional data into a lower-dimensional vector format. (2) They fuse together different geospatial data modalities, from different types of images to text and tabular data. (3) They can interpolate to unseen spatiotemporal locations, where raw data is missing. (4) They are interoperable with other AI foundation models, such as LLMs, through aligned embedding spaces.
as explicit models, extracting embeddings from raw data (e.g. satellite imagery) associated with a location (emb ∼ Eexplicit(datalocation))
implicit models, returning embeddings from only location inputs (emb ∼ Eimplicit(location)).
Earth embeddings map places and times that share similar properties closer together in embedding space.
GeoFMs are large-scale modeling and learning frameworks, whereas Earth embeddings constitute the interoperable, location-indexed data outputs that can be stored, shared, or queried independently of the model that created them.
We posit that Earth embeddings will emerge as the dominant format of geospatial data in the AI age
ways in which users can employ Earth embeddings for prediction, conditioning, simulation, and search
Call to action: Advancing analyses and applications with Earth embeddings.
• Evaluating and benchmarking Earth embeddings
• Explainable and interpretable Earth embeddings
• Learning planetary processes with Earth embeddings
Earth Embedding Models: Explicit Feature Extraction versus Implicit Neural Representation
Challenges and opportunities for improving Earth embeddings.
• Model capacity
• Spatio-temporal heterogeneity
• Data curation and scaling
• Learning objective
The research agenda we outline is fundamentally interdisciplinary: Earth embeddings will rely on feedback from domain scientists, e.g. in ecological, geological, oceanographic, and atmospheric sciences, that incorporate Earth embeddings into their analyses and from data practitioners apply- ing Earth embeddings in their workflows and products.
A bit late to post about (though not to publish!), we have a new episode of the #GLaDpodcast out. If nothing else, be enticed by the title (the oldest profession in Geography?!?!?!); if something else, delight in Anthony Robinson’s views on maps, AI, and microwave ovens!
🎧 From Data Dump to Data Product
So many common points and arguments that really resonate here and make me more hopeful for Imago. The discussion of data as infrastructure, invisibility as success, and thinking really hard about how to make sure “it”’s not only here now, but tomorrow and the day after are points that’ll stay with me. And it’s also great to find more people who’re thinking creatively (not only from the tech side of things) to ensure the world has more collective-ness around data. Most recommended listen.
Metadata
Highlights
The way I frame it is like, the game is figuring out how to lower the cost of asking questions.
🔗 Why "Science-as-a-Service" Doesn't Work for Earth Science
Very important, if sobering, piece on the TerraWatchSpace newsletter on why “handing off to industry” is not a great idea for Earth Observation (EO) for basic science. In some ways, I see parallels with the discussion in the social sciences around how traditional sources (think decadal censuses, but also large surveys, etc.) could potentially be replaced by new sources such as mobility from phones or, for that matter, modern uses of Earth Observation. Don’t get me wrong, I am more excited than most about the potential of new data in the social sciences (imagery in particular!). We do need more data than a drop every ten years to not fly blind through everything that happens between release points (which is a lot). The bit that makes me very uneasy here is the replace, rather than complement. Without the census, satellites and phones are fairly close to useless for social scientists, and the reasons are very similar to why commercial EO needs large, public, and free programmes like Sentinel and Landsat.
Metadata
Highlights
Jared Isaacman, President Trump’s nominee for NASA Administrator has articulated a compelling vision: “NASA needs to constantly be recalibrating to do the near impossible, what no one else is doing - and the things they figured out, they hand off to industry.”
Earth Science Data Is Infrastructure, Not a Service
Infrastructure requires institutional commitment that transcends market cycles and political administrations. It requires transparency, neutrality, and guaranteed long-term access. It requires optimization for societal benefit rather than profit margins.
Just in time for a cozy listen this coming break, a new #GLaDpodcast episode! This time, we welcome the one and only Serge Rey to talk all things open, open source, and academia! Come for the code, stay for the stories of lives changed!
With all the rush last month, I forgot we released a new #GLaDpodcast episode. We finally gave in to the GeoAI craze and went straight to the source. Join us for a conversation with Krzysztof Janowicz, who’s been laying the grounds for what today is called “GeoAI” for longer than you can think of!
🔗 World Urbanization Prospects 2025
Cities are (still) a pretty cool thing…
Metadata
Highlights
Entry for #30DayMapChallenge
I have an entry today at the GDSL’s #30DayMapChallenge with an attempt at “Dimensions”. It’s a story of 64 dimensions, mapped in two, with a bit of an extra dimension.
Official description:
This “map” is a projection to 2D of the 64 dimensions provided by Google’s Satellite Embedding V1 for all pixels of 100m x 100m in Liverpool for 2017 and 2024. Each dot is thus one area at one point in time. Because embeddings encode semantic similarity, nearby dots represent similar areas. Three areas (A, B, C) with substantive change are highlighted, and this change can be seen in the “journey” across the map from their 2017 location to that in 2024. Suggestive aerial images are included also to help understand the nature of the change.
Credits:
🔗 The City As Text
Neat paper by a great gang.
Metadata
Highlights
This Review seeks to ground this opportunity in an introduction to the kinds of text and tools available to researchers, providing examples of the state of the art in urban research while contextualizing these applications in the broader framework within which this interest in textual data evolved.
(Satellite imagery) embeddings for the rest of us
My five(ish) minute contribution to the Leeds workshop on Geo Foundation Models.
🎧New episode of the #GLaDpodcast 🎧 We liked talking to Wenfei Xu so much that we invited her back! Come for the fancy data, stay for engaging conversation about hopping between academia and industry, traversing disciplines, and… an Exhibitions entry in your CV.
🔗 GeoFM: how will geo-foundation models reshape spatial data science and GeoAI?
This was a much more insightful read than I anticipated. The first part is a fantastic introduction to the idea of state of the art foundation models today, in particular in the Geo space. The second is more prospective, and thus a little more speculative. Either way, very good food for thought.
Metadata
Highlights
Given these three motivating factors, GeoFM can be defined as follows: Geo-foundational models are foundation models specifically trained on heterogeneous spatiotemporal data, capable of reliably performing advanced spatiotemporal reasoning, and designed to incorporate spatial, temporal, and other contextual factors into their output to support a wide range of (geo)spatial downstream tasks in geography and neighboring disciplines that benefit from a spatial or geographic perspective.
Just as LLMs encode the syntax, semantics, and pragmatics of human language, GeoFM could encode the language of space, i.e., the place-agnostic properties that define geography – spatial dependence and heterogeneity (Anselin,Citation1988) and its related concepts such as scale, adjacency, spatial and temporal scopes, and so on.
why do we need geo-foundation models (GeoFM) at all, and what exactly are they or will they be? First, foundation models can only generalize within the scope of their training data.
Second, many geospatial tasks are highly specific
Third, geography is inherently local/regional or contextual.
GeoAI advances along two major dimensions: (1) it applies novel methods and technologies from the broader AI and machine learning community to geographic and geospatial research questions and (2) it feeds its own, novel theoretical and methodological contributions back to the broader AI community
location embeddings can be trained separately and concatenated with the embeddings representing learned building footprints, land classes, and so on (Mac Aodha et al. Citation2019; Yan et al. Citation2019; Mai et al. Citation2020). Note: This is an idea I’ve had for a while and would be good to check some of these references to see how they approach it. Although modern foundation models were not yet on the horizon in the early 2010s, it was already clear that the era of custom, single-purpose models was slowly giving way to workflows developed around reuse and transferability. This shift raises a key question for GeoAI research: how can we distinguish progress driven by GeoAI-specific innovation from improvements mostly gained through the application of transfer learning (and related methods) from general-purpose models?
The successful combination of few-shot, prompt engineering, and transfer-learning methods on top of powerful general-purpose models raises the old question again: is spatial really special?
we can roughly classify the existing GeoFM-related research into the following categories: 1) adapting existing FMs on geospatial tasks via prompt engineering and task-specific fine-tuning; 2) developing advanced LLM agent frameworks for geospatial tasks; and 3) developing novel geo-foundation models via geo-aware model training and fine-tuning.
we further classify the current GeoFMs in four categories based on the data modalities they support and their application scenarios: geospatial language foundation models, geospatial vision foundation models, geospatial graph foundation models, and geospatial multimodal foundation models.
three major ways of realizing GeoFM or using generalist FM
For now, it is unclear whether one of the paths is preferred to approach the vision of generally capable GeoFM so that the research community could consolidate our efforts, or if this is task-dependent, and, hence, varying paths should be taken for different types of tasks.
Designing architectures that can jointly process such heterogeneous data, scale to large datasets, and accomplish effective cross-modality alignment remains a major open challenge.
A fundamental question is whether those subjective and complex human experiences should become part of GeoFM.
this raises concerns about GeoFM misrepresenting geography, be it by introducing bias or by learning representations that do not align with those of groups or societies.
spatial priors should ideally be incorporated into the pre-training of GeoFM […] those priors change across scale, resolution, modality, and so forth, and it is presently not clear how to best handle those. For instance, should they be explicitly engineered or implicitly learned?
Without co-evolving our data and benchmarks, the true potential of GeoFMs will remain constrained.
most present work on AI alignment does not account for regional, e.g., cultural, differences. However, as geographers, we know that the aforementioned societal goals, values, and norms vary greatly across geographic space and time – without any being inherently superior to others.
skills that help us better interact with such agents, critically think about their outputs, align AI with societal goals, and so on, will increase in importance.
🎧New episode of the #GLaDpodcast 🎧 Everything you ever wanted to know about mobile phone data in social science and never dared to ask, answered neatly in an entertaining™️ conversation. Check it out here, or wherever you get your podcasts:
🔗 Mission Critical -- Satellite Data is a Distinct Modality in Machine Learning
Some of the arguments made here would ring obvious to traditional spatial analysts, others to traditional remote sensers and, I suppose, many might seem “bread and butter” for the ML crowd. But it is not each individual argument that is the point here; it is putting them together, now, and in a contemporary and fresh framework that makes this paper worth a read. Well worth it indeed.
Metadata
Authors: Esther Rolf, Konstantin Klemmer, Caleb Robinson, Hannah Kerner Category: article URL: arxiv.org/abs/2402….
🔗 Why We’re Talking About a Centralized Vector Embeddings Catalog Now
The white paper mentioned below is well worth a read. It puts in much more eloquent words many of the reasons why I’m very excited about the new generation of satellite foundation models and the potential embeddings have to make satellite data more useful, useable, and used! A lot of food for thought and great argumentation for why we need to think about satellite images more and more like abstract tables than like images of pixels.
Metadata
Highlights
our team published a detailed white paper in which we make the case for how Earth Observation (EO) data providers such as NASA can dramatically improve access to their data by creating a centralized vector embeddings catalog
We gave in to the hype and did one #GLaDpodcast episode on how we use AI, for Geography, Life, Geography Life, and data…
New PhD, a collab between IBM Research, SDR-UK’s own Geographic and Imagery Data Services to explore using vision foundation models & high resolution imagery of the Liverpool City Region. If this sounds like you (or a friend!), all info here!
Getting excited for ESA’s Living Planet Symposium, to happen in Vienna in just over a week! I’ll be there Sunday to Thursday, who else is coming? hit me up if you’d like to talk all things Imago or just catch up!