<p>For most of its history, GIS answered questions in the past tense. Where did the flood reach? Which parcels changed hands? How far did the fire spread? The map was a record — an authoritative account of something that had already happened.</p><p>That's changing. The most interesting geospatial work in 2026 increasingly answers questions in the future tense: Where <em>will</em> the ground fail? Which neighborhoods <em>will</em> urbanize next? Where is the next maintenance failure most likely to happen? This shift — from describing the past to forecasting the future — is quietly one of the biggest changes in what GIS is for.</p><p>It's worth understanding what's actually behind it, because the hype tends to outrun the substance.</p><p><br></p><h2>What "predictive GIS" actually means</h2><p>Strip away the marketing, and predictive GIS is the marriage of two things that were, until recently, kept in separate rooms: spatial data and machine learning.</p><p>The idea is straightforward. Predictive analytics uses statistical modeling and machine learning on historical data to estimate future outcomes. Predictive <em>GIS</em> does the same thing, but with location as a first-class variable — so the model isn't just asking "what happens next," it's asking "what happens next, <em>here</em>, given everything we know about this place and the places around it."</p><p>That last part is the whole point. Spatial data has a property that most datasets don't: things near each other tend to be related. A landslide is more likely on a steep, saturated slope adjacent to a slope that already failed. Urban growth clusters around existing roads and amenities. Ignore that spatial structure and you throw away the most useful signal you have. Predictive GIS is, at its core, the discipline of keeping it.</p><p><br></p><h2>Where it's already working</h2><p>This isn't speculative. Predictive spatial models are running in production across several domains right now:</p><p><strong>Geohazards.</strong> Machine learning has become the predominant approach in landslide and geohazard modeling. Researchers combine slope, soil saturation, rainfall, and historical failure data to produce susceptibility maps that forecast <em>where</em> the ground is most likely to give way — and increasingly <em>when</em>. One published system in Chiang Rai, Thailand builds its entire landslide dataset in QGIS, then feeds it into a machine learning model to predict risk dynamically.</p><p><strong>Land use and urban growth.</strong> Models like the Land Transformation Model pair neural networks with GIS to forecast how urbanization will spread, weighing factors like road networks, recreational amenities, and agricultural density. Planners use these to test scenarios before committing to infrastructure.</p><p><strong>Public health.</strong> Predictive GIS forecasts disease outbreaks by modeling how cases spread across population and mobility data — turning a map of where people are sick into a map of where they're likely to get sick next.</p><p><strong>Infrastructure and utilities.</strong> This is where the money is moving fastest. Utilities are using predictive models to anticipate failures in buried infrastructure and flag emerging maintenance hotspots — shifting decisions earlier, before the pipe bursts rather than after.</p><p><strong>Disaster response.</strong> Flood, wildfire, and earthquake impact models increasingly forecast likely progression rather than just recording current extent — feeding proactive evacuation and resource allocation.</p><p><br></p><h2>The part that gets skipped</h2><p>Here's where we'll be less breathless than most coverage of this topic.</p><p>A predictive model is only as good as the spatial data underneath it — and spatial data is unusually easy to get subtly wrong. A model trained on features in mismatched coordinate reference systems will produce confident, precise, completely wrong forecasts. Bad geometry doesn't announce itself; it just quietly corrupts the output. The most sophisticated Bayesian network in the world can't compensate for a training set where "adjacent" parcels aren't actually adjacent because someone's projection was off.</p><p>This is the unglamorous truth of predictive GIS: the machine learning is often the easy part. The hard part is the data discipline that comes before it — provenance, coordinate systems, topology, validation. The same fundamentals that have always separated good GIS work from bad, now with higher stakes because a model will amplify whatever you feed it.</p><p>There's also a subtler trap. A forecast carries an air of authority — it looks like knowledge about the future. But every predictive model encodes uncertainty, and a map that shows a single crisp "predicted flood boundary" without communicating how uncertain that line is can be actively dangerous. The 2026 Esri User Conference had entire technical sessions on exactly this: helping GIS professionals embrace and communicate uncertainty rather than hide it behind a clean-looking output. That's a sign of a field maturing past the "look what we can predict" phase.</p><p><br></p><h2>What this means for the people doing the work</h2><p>If you work in GIS, this shift changes the job description more than it might first appear.</p><p>The valuable skill is no longer just producing an accurate map of what exists. It's understanding enough about both the spatial data <em>and</em> the modeling to know when a forecast can be trusted — and when it's a confident-looking artifact of a flaw in the pipeline. That's a hybrid skill set: spatial fundamentals plus enough statistics and machine learning literacy to interrogate a model's output rather than just accept it.</p><p>It's the same pattern we keep coming back to in this newsletter. The tooling gets more powerful. The need for a human who understands what the data actually means gets <em>more</em> acute, not less. A predictive model doesn't replace the GIS professional's judgment — it relocates it, from "is this map accurate?" to "should we believe this forecast?"</p><p>That second question is harder. It's also a lot more valuable.</p><p><br></p><h2>The takeaway</h2><p>GIS is shifting tense — from past to future, from record to forecast. It's a genuine and useful evolution, already delivering real value in geohazards, planning, public health, and infrastructure.</p><p>But a forecast is a claim about the future built on the quality of your data about the past. The organizations and professionals who get real value from predictive GIS won't be the ones with the fanciest models. They'll be the ones whose spatial data was clean enough to trust the model in the first place.</p><p>The future tense is only as reliable as the past tense it's built on.</p>