Recipes

Little Cottonwood Canyon, Alta, UT

Alpine treeline ecotone detection validated against 268 independent field sites. Computes ATEI (Wei et al. 2020) at each site using folia's each: fan-out, then compares predicted treeline elevation against Malanson & Skibbe (2023) ground truth.

Property-level flood risk assessment from raw elevation data, FEMA zones, physics-based flood models, and satellite land cover. Type an address — see terrain drainage, impervious runoff, regulatory flood zones, and a composite risk score. Full provenance. National Harbor, MD — the ClimateTech Connect 2026 venue.

Land cover classification using the RLCMS primitives approach (Poortinga et al. 2019, SIG/SERVIR-Mekong). Biophysical indices (NDVI, NDWI, NDBI, BSI) from Landsat 8 are combined via a threshold decision tree into 5 land cover classes — running entirely in the browser.

Active lava lake monitoring and eruption mapping at Ambrym volcano, Vanuatu, using Sentinel-2 SWIR and NBR. Ambrym is a basaltic caldera that hosted persistent lava lakes at Marum and Benbow craters until a December 2018 submarine dyke intrusion drained them abruptly. The recipe captures the lava-lake era and its sudden end; B12 SWIR tracks the lake thermal signal through near-constant tropical cloud cover. The active-lava-mask layer is the primary signal for the lake period, while dNBR documents the caldera floor change following drainage.

Two-epoch surface-water extent of the Aral Sea from Landsat NDWI — one of the most dramatic environmental collapses visible from space. Soviet irrigation diverted the Amu Darya and Syr Darya; by 2007 the South Aral had lost 90% of its volume, while the North Aral (Kazakhstan) partly recovered behind the 2005 Kokaral dam. Two diverging trajectories in one bounding box.

The annual monsoon flood pulse of the Brahmaputra-Jamuna river system in Bangladesh — one of the most dramatic seasonal water body changes on Earth. Each monsoon, 20-30% of Bangladesh floods; for 160 million people this is not a disaster but the agricultural cycle, depositing silt that renews farmland. SAR (Sentinel-1) is essential: monsoon clouds blanket Bangladesh during peak inundation, making optical useless for the event that matters most.

The Bingham Canyon open-pit copper mine in Utah — 1.2km deep, 4km wide, producing ~25% of US copper, and visible from space since before Landsat existed. ICEYE X-band Spotlight SAR at sub-meter resolution captures the pit wall geometry, tailings pond extent, and the disturbed texture zone left by the 2013 Manefay Slide — the largest non-volcanic landslide in North American recorded history, which removed ~130 million tonnes of material from the NW wall.

Fire scar mapping and peatland degradation in Central Kalimantan, Indonesia. Drainage canals for palm oil plantations expose deep carbon-rich peat to fire; the 2015 Indonesia fires released more CO2 in 6 weeks than Germany emits in a year, and the haze reached Singapore, Malaysia, and Thailand. dNBR maps burn severity per event; annual NDVI tracks recovery — or its absence.

BULC-D (Bayesian Updating of Land Cover, Cardille & Fortin 2016) over any AOI, defaulting to the Coast Range checkerboard. Reference vs target NBR periods, Bayesian evidence accumulation -> per-pixel change_magnitude [-1,1]. Detector for the emapr ensemble (research/change-detection/emapr-ensemble/CONTRACT.md). Shares the annual-NBR front-end with landtrendr-change verbatim (same Landsat NBR, so the detectors are comparable); only the final op differs. Generalizes registry/workspaces/bulc-d.folia.yaml (Camp Fire). Op: packages/compute/src/ops/bulc-d.ts.

Parameterized ALERT LAYER (docs/spec/ALERTS.md §2.2, §10.2 pilot): fire where the current CAMS AOD550 near-real-time value exceeds its own recent history by more than N sigma over a user-chosen baseline window. The threshold is the USER's, not the operator's — this is the gap the incumbent CAMS Aerosol Alerts service leaves open (it bakes fixed high / very-high / extreme tiers against a 2003–2018 climatology; here `sigma` and `baseline_window` are subscription parameters).

Forest-loss ensemble over any AOI (default Coast Range checkerboard): runs the detectors that already exist in Folia (LandTrendr + BULC-D) on a shared annual-NBR front-end, maps each to a 0-100 loss probability, and combines them with the mask-aware naive average (the `ensemble` op), mirroring the eMapR forestChangeEnsemble `getNaiveLossAverage` (research/change-detection/emapr-ensemble/ CALIBRATION.md). Adding CCDC/COLD, OPERA, etc. is just another detector input.

Two-epoch ice extent of Columbia Glacier, Alaska, from Landsat NDSI. One of the fastest-retreating tidewater glaciers on record: its terminus has pulled back more than 20 km since the early 1980s and the fjord it vacated is now open water.

San Joaquin Valley cropland block

Per-field crop vigor and phenology from a Sentinel-2 NDVI time series. For an AOI covered by field boundaries, compute NDVI for every clear pass, fit a harmonic model to each pixel's season, and aggregate the phenology to each field: how vigorous the crop is (annual amplitude), when it peaks (phase, day-of-year of maximum greenness), and its multi-year trend. The per-field table is the crop-monitoring product; the NDVI stack and amplitude/phase rasters are the "show your work" layers.

San Joaquin Valley cropland block — Fresno County

San Joaquin Valley cropland block — Fresno County

San Joaquin Valley cropland block — Fresno County

San Joaquin Valley cropland block

San Joaquin Valley cropland — Fresno / Kings Counties

San Joaquin Valley cropland — Tulare / Fresno County

Two-epoch shoreline and water extent at Dubai from Landsat NDWI. Between the early 1990s and today, Palm Jumeirah, Palm Jebel Ali and the World islands turned open Gulf water into land — one of the largest deliberate coastline changes on Earth.

EEFA chapter F4.0 reproduction: filter (bbox, 2000-2010, cloud<50), map (per-scene EVI), reduce (mean + median) over Landsat C2 L2 at Shanghai.

Per-patch embedding change between two quarters at one AOI, from one foundation model release. Embed the AOI at t0 and at t1 with the same model, then take the per-patch cosine distance between the two vector sets and render it as a change surface. High distance means the model sees the place differently than it did before; it does not say why, and the recipe makes no claim about the cause.

San Joaquin Valley cropland block

Pilot alert layer (ALERTS.md §10.1): wrap the NASA FIRMS NRT active-fire feed and deliver detections through Folia's full delivery surface — webhook, MCP push, signed receipt — rather than FIRMS's email-only alerts. The detection is not novel (FIRMS already does it); the value is the delivery fabric + an inspectable firing that carries the FIRMS source row that drove it.

Every nature-based carbon offset project overlapping an area of interest, from Cecil's Forest Carbon Project Boundaries (~575 projects, 55 countries). Clips the project polygons to the AOI and returns each project with its registry (Verra, Gold Standard, ...), methodology, type, developer, and claimed area — the inspectable "who claims to be doing carbon work here, under what standard." Designed as an audit layer: pair it with an independent forest-change observation to check whether the claimed footprint holds up. No magic — the output rows are the raw project records that intersect the AOI.

Seafloor depth profile along a transect line, sampled from GEBCO 2024 global bathymetry (15 arc-second COG). Places evenly-spaced points along the line and reads the elevation pixel under each — negative values are depth below sea level. Answers "how deep is it, and how does the seabed slope, along this path?" for any ocean transect. Windowed COG reads mean a single profile touches only a few kilobytes of the 4.6 GB global grid, not the whole file.

100-year river flood depth over a Rio de la Plata window, read from the JRC GloFAS global flood hazard maps published as a Portolan STATIC catalog on Source Cooperative. The single source layer resolves through the catalog-search static search backend (collection walk + versions.json sha256 pins); a flooded-pixel mask feeds a terminal answer layer reporting the mean modeled inundation depth. The demand fixture for sdk/050 (static-catalog search backend).

Active-fire alerts read from the SOURCE product rather than from a vendor feed: NOAA's GOES-R ABI L2 Fire/Hot-Spot Characterization (ABI-L2-FDC), taken straight off the public NOAA Open Data S3 buckets.

Wetland and forest CHANGE over a Gulf Coast county, delivered as a subscribable alert and ranked by community vulnerability. Two Sentinel-2 composites per AOI — a baseline window and a current window — give NDVI (vegetation vigour) and MNDWI (open water / wetland inundation). The bitemporal deltas fire two independent arms: vegetation LOSS when NDVI drops past a threshold, wetland CHANGE when MNDWI moves either direction past a threshold. A cell fires if either arm crosses, and carries both deltas plus which arm drove it, so a reader sees "NDVI 0.75 -> 0.35 (-0.40) => forest loss" rather than a bare boolean.

Active lava detection at Big Ben (Mawson Peak), Heard Island, a remote subantarctic glaciated stratovolcano in the southern Indian Ocean, using Sentinel-2 SWIR and NBR. Big Ben shows intermittent summit lava and flank flows that interact with the island's glaciers to produce lahars. It sits under near-permanent cloud, making the B12 SWIR thermal anomaly the only reliable orbital confirmation of eruptive activity; the cloud ceiling is raised to 95 because visible bands are almost never clear enough to permit optical mapping here.

SLC 90-minute drive envelope

Mount Timpanogos Wilderness

Parameterized ALERT LAYER (docs/spec/ALERTS.md §2.2, §10.3 pilot): fuse four independent near-real-time forest-disturbance systems and fire where at least `min_agreement` of them flag the same H3 cell. This is Folia's reproduction of GFW's Integrated Disturbance Alerts (Reiche et al. 2024, ERL ad2d82) — but the fusion RULE is the user's, not the operator's. GFW bakes one confirmation rule; here `min_agreement` and each system's confidence gate are subscription parameters (ALERTS §2.2), so a 2-of-4 subscriber and a stricter 3-of-4 subscriber share this one layer definition without materializing variants.

Active lava flow mapping on the Big Island of Hawai'i using Sentinel-2 SWIR and NBR. The 2018 Lower East Rift Zone eruption destroyed 716 structures and added 3.5 km² of new land; ongoing Halemaʻumaʻu summit activity from 2020 continues. SWIR bands (B12) saturate at active lava temperatures and penetrate volcanic SO2 plumes where visible bands cannot — essential for mapping through the eruption haze.

Two-epoch surface-water extent of Lake Powell from Landsat NDWI. The second largest reservoir in the United States fell to roughly a quarter of capacity during the 2000-2022 Colorado River drought, exposing canyon walls and the "bathtub ring" that make the decline legible from orbit.

Per-pixel LandTrendr trajectory segmentation over any AOI, defaulting to the Oregon Coast Range clearcut checkerboard (Landsat C2 L2 annual NBR). First instance of the detector output contract (research/change-detection/emapr-ensemble/CONTRACT.md): the LandTrendr layer's `magnitude` and `year` bands map to the contract's `magnitude` / `onset_year`.

Single-chain LandTrendr demo: annual NBR composites (cloud tier) feeding the LandTrendr temporal segmenter (client tier). Derived from change-ensemble with the second detector and ensemble removed so the composite->landtrendr chain is single-consumer and fully quotable.

Where did the forest change, what year, and how is recovery going — the Kennedy recovery-questions deliverable (sdk/011, signed I5+I6) on one LandTrendr-segmented NBR trajectory. Runs OFFLINE end to end on a vendored real annual-NBR stack (Landsat C2 L2, 2021 Bootleg Fire bbox, 8 years 2017-2024) via local://, so it reproduces on a laptop with no network. Three layers answer the three questions from the SAME fitted trajectory:

Two-epoch ice extent of the Larsen B Ice Shelf, Antarctic Peninsula, from Landsat NDSI. Roughly 3,250 km² of shelf that had stood for millennia disintegrated over about five weeks in early 2002; the baseline epoch here is the intact shelf and the recent epoch is the season after the collapse.

Active thermal anomaly and flow mapping at Láscar volcano, Chile, using Sentinel-2 SWIR and NBR. Láscar is the most active volcano of the central Andes, with a history of vulcanian explosions and a persistent high-temperature fumarole field at the summit crater. Its high-altitude arid setting in the Atacama gives some of the clearest viewing conditions of any active volcano on Earth, making the B12 SWIR thermal anomaly over the active crater unusually clean and detectable even at low eruption intensity.

Active lava lake monitoring at Masaya volcano, Nicaragua, using Sentinel-2 SWIR and NBR. Masaya hosts one of the planet's few persistent lava lakes, active in the Santiago crater since December 2015. The primary signal here is the continuous B12 SWIR thermal anomaly from the active-lava-mask layer, which tracks the lake's persistent thermal emission; dNBR is retained for consistency with the family but is secondary because no large discrete flow field has developed outside the crater rim.

Two-epoch surface-water extent of the Mesopotamian Marshes of southern Iraq from Landsat NDWI. Drained by canal and embankment after 1991 to roughly a tenth of their former area, then partially re-flooded from 2003 onward — a loss and a deliberate restoration in the same bounding box.

Pellworm, Schleswig-Holstein, DE

The 2022 Pakistan megaflood — 33 million people displaced, 1,700 killed, one-third of the country submerged at peak. Pakistan emits less than 1% of global CO2 but bore enormous climate damage; a defining moment in Loss and Damage negotiations at COP27. SAR (Sentinel-1) is the only sensor that cut through the monsoon cloud cover to map actual flood extent in near real-time. This recipe implements change-detection flood mapping against a 2021 reference.

Active lava flow mapping at Piton de la Fournaise, Réunion Island, using Sentinel-2 SWIR and NBR. One of the most active basaltic shield volcanoes on Earth, Piton de la Fournaise erupts frequently with effusive flows confined largely to the Enclos Fouqué caldera; EONET tracked an eruptive episode closing April 2026. B12 SWIR saturates over incandescent basaltic channels and reveals flow extent even under the eruption's gas plume where visible bands are obscured.

Hourly clear-sky solar irradiance on a horizontal plane at a single point, for one date. NOAA closed-form ephemeris for solar az/el, Kasten-Young air mass + Meinel transmission for DNI, flat 10% diffuse for DHI. Horizontal global irradiance GHI = DNI * sin(elevation) + DHI. Optionally scales to expected PV output: nameplate_w * (GHI / 1000) * derate. Intended for quick "does my panel reading make sense?" checks — a point companion to the raster horizon-shade recipe (same transform module, no horizon walk; assumes open sky, SVF = 1).

Scrub through annual composites of a normalized-difference index over any AOI and date window. Sources Sentinel-2 L2A from Microsoft Planetary Computer, gates per-scene pixels with the SCL classification mask, computes a per-scene index, then collapses each calendar year to a single median raster. Output is a year-keyed band stack that folia-view exposes as a scene scrubber via `folia-raster-bands`; the runner mounts a minimal year slider on top.

Two-epoch NDVI loss over the Rondônia fishbone clearing pattern in the Brazilian Amazon, along the BR-364 corridor. Landsat surface reflectance is composited over an early-2000s baseline epoch and a recent epoch; the drop in NDVI between them is the clearing proxy.

Two-epoch extent of centre-pivot irrigation in the Wadi As-Sirhan basin, Saudi Arabia, from Landsat NDVI. Circular fields watered from a fossil aquifer expanded across bare desert through the 1990s and 2000s; groundwater policy and depletion have since retired many of them.

Five canonical spectral indices over a named AOI from Sentinel-2 L2A: NDVI, NDSI, NDWI, NBR and BSI. Each is computed per scene, cloud-masked from the scene classification layer, and reduced to a median over the date window. One snapshot per run, not a time series; rerunning on a schedule updates the snapshot.

Storm damage totals by event type within an area of interest, 1950-present. Reads NOAA's Storm Events Database (~2.0M events) as cloud-native GeoParquet, clips to the AOI by bounding-box overlap, and aggregates event counts, property + crop damage, deaths, and injuries grouped by event type. Answers "what has hit this place, how often, and how bad?" for any US location. The filtered event rows are the show-your-work — every dollar in the total traces back to a specific NCEI event record.

National sourcing sheds for all 319 Brazil sugarcane mills: partition Brazil's land cells to each mill by allocation method (straight-line nearest = voronoi, or capacity-weighted gravity = huff), for a methods-comparison map. Capacity weighting is the axis where this exceeds the open incumbents (PalmWatch is unweighted).

Facility -> plausible sourcing shed (capacity-weighted Huff allocation of H3 cells within a haul bound) -> from-class -> to-class land-use-change transition matrix per shed, plus the native-conversion summary. Commodity and region specifics are parameters, not code.

National (all-Brazil) profile of supply-shed-change: 319 sugar mills -> capacity-weighted sourcing sheds -> cane land-use-change per shed, cane-masked to real MapBiomas cane. Straight- line; a drive-time profile adds the OSRM distances table.

Default — SLC 90-minute drive isochrone bbox

Two-epoch land extent of the Wax Lake Delta, Louisiana, from Landsat NDWI — the inverse of the usual coastal story. An artificial 1941 outlet channel let the Atchafalaya build a new delta from open bay, and it has been gaining land ever since while the rest of the Louisiana coast erodes.

Wildfire burn history for an area of interest from the USGS Wildland Fire Combined Dataset (national perimeters, 1878-2021). Clips the fire perimeters to the AOI, then returns the burn footprint: how many distinct fires, total burned area, the most recent burn year (hence years-since-last-fire), and the largest single fire. Answers "has this place burned, how often, and how recently?" for any US location. The clipped perimeters are the show-your-work.

Two-epoch NBR over the 1988 Yellowstone fires, which burned about 3,200 km² of the park in one season. The baseline epoch is the pre-fire forest and the recent epoch is today's canopy, so the difference reads as RECOVERY rather than severity: the stand-replacing lodgepole burn has been regrowing for more than three decades.