LLMs can’t jump, a position paper by Tom Zahavy at Google DeepMind, argues that AI has become good at finding patterns in data and at deriving consequences from assumptions, while the step between the two, inventing the assumption, remains out of reach. The argument is about AI. The consequence I keep returning to is about researchers, and what we are for once the rest of the process runs itself.

The difference between a statistical model and a machine learning model is not that one is interpretable and the other is not. A better framing is this: traditional statistical models usually start with an explicit theory about the relationship between variables, while machine learning models usually start with a prediction goal and let the algorithm learn complex patterns from data. The real contrast lies in purpose, assumptions, model structure, and interpretation.

Scope. This review covers ~25 papers published between 2023 and 2026 at the intersection of geospatial reasoning, large language models, and multi-agent systems. The focus is on GIS-domain journals (IJGIS, IJDE, JAG, TGIS, Annals of GIS, Big Earth Data, GeoInformatica), machine learning venues (ICLR, ICML, NeurIPS, ACM SIGSPATIAL), and relevant preprints. The goal is to map what the community is actively pursuing and, more importantly, to identify gaps that deserve attention.

Names like S2, H3, and A5 - these are not branding choices or GIS shortcuts. They come directly from mathematics and encode how geometry and symmetry are used to model the Earth in a rigorous and globally consistent way.

This post summarizes the panel discussion held as part of the OGC DGGS × AI pilot project. The conversation brought together pilot contributors and geospatial AI researchers to reflect on the current state and future potential of combining Discrete Global Grid Systems (DGGS) with artificial intelligence.

Richard Sutton’s essay The Bitter Lesson argued that general methods leveraging computation eventually outperform methods built on human knowledge. In his recent interview with Dwarkesh Patel, Sutton extends this argument into a broader claim about intelligence itself: real intelligence is not about imitating human language, but about having goals, acting in the world, learning from experience, and continually improving through feedback.

This post contains my personal notes from the course Introduction to Generative AI (Spring 2024), taught by Prof. Hung-yi Lee at National Taiwan University. The course provides a comprehensive overview of generative models.

One of the powerful applications of Discrete Global Grid Systems (DGGS) is spatial binning of point data using their equal-area cells as analysis units. This enables efficient aggregation, monitoring, and statistical analysis for a wide range of geospatial applications such as population census, natural disturbance tracking, and environmental mapping.

The Open Geospatial Consortium (OGC) defines a set of core operations that a Discrete Global Grid System (DGGS) must support. These include quantization, spatial relations, and interoperability. Together, they form the foundation for integrating DGGS into spatial data infrastructures and enabling robust geospatial analysis.

A Discrete Global Grid System (DGGS) can be understood through several complementary lenses. These perspectives highlight its versatility in organizing, analyzing, and sharing geospatial data across disciplines and applications.

In 1994, Michael Goodchild proposed 14 criteria for evaluating DGGS designs. These were refined by Kimerling et al. in 1999. Recognizing that no DGGS can satisfy all criteria perfectly, the Open Geospatial Consortium (OGC) released a formal DGGS standard in 2017 that identifies key design requirements considered achievable and essential. These are summarized below, based on the DGGS Core Conceptual Data Model and Abstract Specification.

Discrete Global Grid Systems (DGGS) are expanding in both theory and application. This post explores advanced topics in DGGS development, including datacubes, big data integration, sensor networks, and point cloud management.

As Discrete Global Grid Systems (DGGS) continue to evolve, new research opportunities are emerging across analytics, platform development, higher-dimensional modeling, and integration with cutting-edge technologies.

The concept of Digital Earth plays multiple roles across research, education, and public engagement. These roles have been broadly categorized into five types: mapping and visualization tool, analytical and modeling tool, underlying basis for derivative applications, data storage structure, and platform for Volunteered Geographic Information (VGI) [1].

Good to begin well, better to end well. – June K. Robinson

Good to begin well, better to end well. – June K. Robinson

Good to begin well, better to end well. – June K. Robinson

Good to begin well, better to end well. – June K. Robinson

Good to begin well, better to end well. – June K. Robinson

Good to begin well, better to end well. – June K. Robinson

This is a blog post on Esri ECCE website: Modeling Line Features on DGGS Grids in the R-ArcGIS Environment

EXECUTIVE SUMMARY This study explored the possibility of using Landsat-8 OLI Imagery to estimate spruce budworm (Choristoneura fumiferana Clem.; SBW) defoliation levels in Bas-Saint-Laurent-Gaspésie regions of Québec, where an outbreak has been expending since 2014. Seven vegetation indices were tested for their relative influence in predicting SBW defoliation levels. A combination of Green Chlorophyll Index (Chlgreen) and Normalized Difference Moisture Index (NDMI), as the top two predictors...

I shared a blog post on the Esri Canada ECCE site detailing a Python script tool I developed for ArcGIS. The tool summarizes tree characteristics within a moving search window, making it easier to analyze spatial patterns in forest inventory data.