Twelve places with a question for a model

News about places arrives faster than anyone can model it. The reports below were saved over one summer because each one made someone want to build a model, and then sat unread. Here each is read for one thing: the question a person in that place would actually put to a model, and what kind of model would answer it. The facts are the reports'. The question and the model are ours, and are marked as such.

The twelve sort themselves by size. Some are a single town, lagoon or hillside. Some are a basin, a sea or a strait. A few are the planet's trade or climate. The size decides what the model has to hold, and who could build it, so the list runs small to large.

One place

A town in Colorado with no one to run the water. On 29 July 2026 the Colorado secretary of state's office announced that Hartman, a town of under sixty people in the state's south-east, had been declared abandoned. Every elected official had resigned, there was no board and no clerk, and a resident had written that the water had gone without chlorine for over a month and smelled; residents had been told to boil it for months. Abandonment, the fifteen-page order said, was the only legal route to clean water, and the treatment plant and tower pass to Prowers County. The Denver Post reported that the quarrelling was intense enough to keep away any operator who might have helped. The question: if a town loses its operator and its board, how long until the water is unsafe, and what does it cost the county to take the system over compared with repairing it now? The model: a small-utility model, one town's tank, treatment, dosing records and repair backlog, with the governance state as an input. There is no engine for this on the shelves; it is a spreadsheet joined to the state's drinking-water compliance records.

A lagoon in Portland that turns toxic. Ross Island, in the Willamette River, is artificial: in 1926 the Army Corps of Engineers walled off a cluster of islands into one, and from 1926 to 2001 a gravel company dug the inside out to about 120 feet. The lagoon that was left grows cyanobacteria that close the river to swimmers, and the Willamette has had significant blooms in four of the last five years. On 25 June 2026 a team from Oregon State University presented two fixes led by Desirée Tullos to about a hundred people: open the lagoon's south-west wall so the river runs through it, or do that and add a second channel on the north-west. The cost is put at about $17 million or $27 million, and the point of either is to flush the lagoon faster than its water warms and the bacteria multiply. The question: does one channel flush faster than the water warms, or does it take two? The model: a hydrodynamic and water-quality model of the lagoon and its reach, which is what the team ran, and which an engine like Delft3D is built for; it needs the pit's bathymetry, river discharge and temperature, nutrients, and past bloom records.

A salmon pen in a Chilean fjord with no cable to shore. Mowi's Huar Norte pen in the Los Lagos region ran a diesel generator day and night for feeders, lights, cameras, oxygen sensors and pumps. Panels on pontoons now ring the cages and charge a battery bank, and a controller runs the generator only for what sun and storage cannot cover. The page puts solar and battery at about 57 percent of the site's annual power, the diesel avoided at roughly 36,750 gallons a year, and says fewer boats now carry fuel out; it also says plainly that no fish gathered under the panels. Its figures come from two industry links, and its lead image is labelled as generated by AI. The question: for this pen in this fjord, how much of the year-round load can sun and batteries carry, and how many hours a year does the generator still run? The model: an hourly energy balance of one site, local irradiance against the pen's metered load, battery size and the generator's fuel curve. Any energy-system tool does this; the hard part is a year of local weather.

A ranch in Montana, thirty-seven years after the cattle left. Ted Turner bought the Flying D, between the Gallatin and Madison rivers south of Bozeman, in 1989, took off the cattle, tore out hundreds of miles of interior fence and brought in bison. Under cattle the ranch had run 7,000 to 8,000 head and some creek bottoms had been grazed to mud. Turner later called the fence removal an early mistake, since the herd grazed some ground too hard and other ground too little, and some fencing went back. The Turner Institute of Ecoagriculture says the ranch now carries about 4,500 bison and has returned over 5,000 acres of cropland to grazing. The column's evidence that the swap matters comes from elsewhere: a 29-year comparison at Konza Prairie in Kansas and a Montana stream study, and its authors say one site's results do not settle how one ranch performs. The question: if a ranch goes from cattle to bison and the interior fences come out, where does the herd overgraze and undergraze, and how do the creek bottoms change over decades? The model: a rangeland model of one property, herd movement relative to water and fence against vegetation and drought, checked against riparian plots over time.

Pine hills in Israel that were planted to be green. A Times of India desk piece, citing a study published by the Society for the Protection of Nature in Israel, says a century of planting turned open shrubland, grassland and the loess plains of the south into dense pine, mostly Aleppo pine, and that the plantations may have replaced habitats with their own specialised plants and animals. Dense pines shade out plants that need sun, the acid needle layer can stop native seed, and pine roots draw down scarce water and dry springs. The piece says that after a run of bad fires, among them the 2021 fire in the Jerusalem hills, the national forestry fund no longer plants pines in natural batha and grassland and now removes dense canopy and seedlings and regrows native oak, carob and terebinth. The question: if the pines on this hillside are thinned, how much of the open ground and its lizards and birds comes back, does the spring below recover, and does the fire risk to the village go down? The model: three coupled ones, land cover and habitat, a water balance, and fire spread, run over scenarios of removal; the data are plantation-age maps, spring records, species surveys and fire history. The site's ecosystems and land question page is where that chain belongs.

A basin, a sea, a strait

Mussels dying in the rivers of the American South, and no one knows why. North America has about 300 documented freshwater mussel species. Roughly one in ten is gone and a third of the rest are listed. The losses kept going after the obvious causes eased: dam building had largely ended by the mid-1980s and the rivers were cleaner under the Clean Water Act, yet most populations kept falling. Wendell Haag of the U.S. Forest Service watched Horse Lick Creek in Kentucky lose its mussels from the 1990s, a loss mostly complete by 2005. In 2018 he caged captive-reared juveniles at seventeen sites in the Rockcastle River; growth was poor at some, cold water slowed it, mine contamination showed little sign, and the strongest correlation was with the density of the invasive Asian clam. His $1.3-million follow-up put juveniles in about ninety streams in thirteen states over three summers, with algae, sediment, chemistry, fish and insect data beside them; he is still working through it, warns it may not be conclusive, and says the clam is so far the one factor present in most declining streams. Hatcheries in Alabama and Kentucky breed and release rare species meanwhile, and a 2024 federal grant of $3.4 million is to recover six of them. The question: in this creek, is it the clams, the sediment, the runoff or something else that stops young mussels growing, and if hatchery mussels are planted here, will they live long enough to breed? The model: a stream-reach model linking food, clam density, temperature, sediment and host fish to juvenile survival, so a basin can rank reaches for reintroduction. Haag's ninety streams are the calibration set such a model needs and does not yet have.

A town on the Rio Grande that must give water back to Texas. Texas sued New Mexico at the Supreme Court in 2013 over groundwater wells that it said took river water owed under the 1938 compact. A settlement among the three states and the federal government was approved this May. New Mexico has until May 2036 to pump 18,200 acre-feet a year less from wells between Caballo Reservoir and the state line, about 6 to 7 percent of the groundwater used there, and the river will be gauged at the line for the first time. The state's first move is $150 million to buy groundwater rights from farmers, retire them and let the land go fallow; officials are weighing a cap on sales around Hatch, whose vegetable growers are more exposed than the pecan orchards of the Mesilla Valley. If the buy-back falls short, cuts come by priority, newest users first, or by a plan now in confidential mediation. Snowmelt supply in the basin is down 17 percent since 2000, Elephant Butte Reservoir fell to 3 percent after the spring release, and a 2.5-gigawatt data centre is being built at Santa Teresa on rights now used by a sod farm. The question: which wells, in which valley, so that the river at the Texas line meets the decree every year to 2036 without hollowing out Hatch? The model: a coupled surface and groundwater model of the reach, pumping by well, demand by crop, return flows, under warming. The sibling basin to the west has a playable version on this site, Who Drinks the River?.

The Great Lakes, refilled once by glaciers. An Earth.com explainer, reporting a study in JAWRA that it does not name, makes a case that the lakes' size matters less than their refill rate. Meltwater filled the five basins about 10,000 years ago; precipitation, runoff and groundwater now add under one percent of the volume in a year, and the St. Lawrence takes away about as little, so hydrologists treat the lakes as non-renewable on a human timescale, and pollutants stay for decades. Industry, power plants, farms and cities take roughly 44 billion gallons a day, most of it returned. Ice is the second lever: peak winter cover has been falling around five percent a decade in NOAA records since 1973, and winter 2024 set a record low near 4 percent. Open water in midwinter evaporates, lowers levels and feeds heavier lake-effect snow, and whitefish lose the ice over their eggs. Farm phosphorus feeds the Lake Erie blooms, one of which shut off drinking water for about 500,000 people around Toledo in August 2014. Some 40 million people drink from the lakes. The question: at 44 billion gallons a day with shrinking ice and rising evaporation, what happens to lake level, intake quality and the whitefish over fifty years, and which withdrawals or phosphorus loads would have to change to hold the line? The model: a basin water balance coupled to lake hydrodynamics, ice and water quality, which is a research-institute model rather than a laptop one; the NOAA ice record and the tributary nutrient loads are the data.

A fly that eats living flesh, and the cattle it would stop moving. The New World screwworm is a fly whose female lays eggs in any open wound or natural opening, and whose maggots eat only living tissue. The Meat Institute's briefing to packers says a detection in the United States must be reported at once to state animal health officials and the federal area veterinarian, that infested animals are quarantined where they stand and fly traps go out to find the spread, and that neighbouring states may stop animal movements from an affected state until surveillance shows the fly would not travel. The fly was eradicated from the United States in 1966 and a small Florida Keys outbreak was ended in 2017, both by releasing sterile males, since a female mates only once. The question: if the fly turns up on a ranch in this county, how far and how fast does it spread through cattle and wildlife, where do the traps and sterile releases go, and which movement routes get shut? The model: a spread model of the fly, driven by temperature and host density, coupled to the livestock movement network. Sterile-insect programmes plan with exactly this kind of model; what a county needs is the premises and movement records to run it locally.

Parkland in Argentina and Chile, handed back with a 99-year clause. Kris Tompkins, who ran the clothing company Patagonia and left it in 1993, and her late husband Doug Tompkins spent roughly thirty years and, the page says, more than $345 million buying worn-out ranch, forest and wetland in the two countries, repairing it and giving nearly all of it to the governments as parkland. The Argentine gift came with conditions: national parks open to the public, protected for at least 99 years. Repair meant pulling about 700 kilometres of barbed wire on the Chilean properties, where fences had cut the grassland into paddocks for a century, and returning jaguars to the Iberá wetlands after seventy years without them. The page's acreage figures do not agree with each other and it gives no date for the Argentine transfer. The question: with the fences out and the jaguars back, how fast is the grassland recovering across the donated parks and the provincial land next door, and where should the next reintroduction or fence removal go to join the pieces into one landscape? The model: land cover from satellite over the old paddocks, with a connectivity model of the kind Circuitscape runs, and collar and camera-trap data for the cats.

Ships idle in the Gulf, growing a coat. Open hostilities between the United States and Iran began on 28 February 2026 and the Strait of Hormuz has been held shut since. The Times of India, reporting a study in Biological Invasions by a team of 24 marine scientists led by the University of Maryland Center for Environmental Science, gives estimates of about 1,500 vessels stuck in the Persian Gulf and several hundred more in the Gulf of Oman. Hulls at anchor for months grow slime, then weed, barnacles and shellfish, in the season when Gulf organisms grow fastest, and the authors warn that when the ships move again that growth could seed ports worldwide. The International Maritime Organization says a ship idle more than thirty days needs its fouling dealt with before sailing; divers are being sent to Gulf ports to scrape hulls, and the fee per vessel could rise by as much as 60 percent, to around $8,000. The question: when the strait opens, which of these ships carry the heaviest fouling, which ports do they reach first, and which hulls should the divers clean before they leave? The model: a vessel-level one, anchorage dwell time from ship-tracking data against water temperature to estimate fouling per hull, each ship routed to its likely next port, and the port scored for climate match with the Gulf's species.

Trade and the planet

The same strait, seen from a rice field in Bangladesh and an oil port in Saudi Arabia. NASA Harvest's Harvest2Market, an open web tool presented in May 2026, puts in one view what food-security analysts otherwise find scattered: satellite crop condition, the monthly GEOGLAM Crop Monitor, trade figures, shipping and food-security assessments. Its worked example is Bangladesh, which grows about 7 percent of the world's rice and gets much of its fertiliser by sea through Hormuz; the tool flags the closure as a risk to the coming rice crop. From the other side, DW reported on 14 September 2026 that the Houthis had taken the Hanish islands after seizing Mayun Island and the port of Mokha the week before, closing the Bab el-Mandab route that Saudi Arabia had turned to when Hormuz shut, so that Saudi crude reaches Asia only by Suez or around the Cape. The question: a strait closes; how far does the shock travel, through fertiliser into next year's harvest, through insurance and routing into the price of fuel, and through a hull into a harbour on the other side of the world? The model: no single one. Harvest2Market is a trade-network and crop-condition view; the routing question wants a chokepoint model over ship-tracking and port capacity; the invasion question is the one above. They hand off to each other, which is the general case for a question at this size, and the reason this site has a data shelf and an ontologies shelf and not only engines. A tool like DSSAT answers the crop end; GCAM sits at the trade and energy end.

Tropical forests nearing the temperature where leaves stop working. A study in PNAS by Nina van Tiel and colleagues, read here from its abstract, used satellite surface temperature and species range maps to follow the margin between the critical photosynthetic temperature of 208 tropical tree species and the hottest canopy temperature they experienced, over 2001 to 2020, in South America, Southeast Asia and Central Africa. The mean critical temperature was 46.1 degrees. The margin fell consistently, by 0.4 degrees a decade on average and 0.5 in South America, and the area whose canopies exceeded the mean threshold grew from 43 to 57 million hectares over the twenty years. Earth system model projections under medium-to-high emissions put that area at possibly 83 million hectares by 2050 and 160 million by 2100. The question: which stands in this forest, and which species in them, cross the threshold in the next heat wave, and in the 2040s? The model: the study's own, a gridded overlay of canopy temperature on range maps with per-species tolerances, which anyone with the satellite record and a tolerance table can rerun for one forest. It is the only item on this list whose model exists and is published.

What the list says

Count them. The lagoon, the forest and the trade view already have their model; somebody built it and published. The ranch, the parks, the fly and the river have a known kind of model and no local data to feed it, which is the ordinary condition of a place that wants to know something about itself. The mussels, the lakes, the salmon pen and the abandoned town are not modelling problems yet. They are measurement problems, and a model built before the measuring would only be a tidier way of guessing. That leaves the strait, which is a chain. Its questions sit on the seams between models, and the work is in the hand-offs: a crop model does not know what an underwriter charges, and a routing model does not know what a barnacle does in a northern harbour. The guides are about those hand-offs. The use cases are where each of these belongs once someone writes it down.