---
title: how to find the keystone parcels in a watershed
canonical_url: https://ensurance.app/guide/how-to-find-the-keystone-parcels-in-a-watershed
markdown_url: https://ensurance.app/guide/how-to-find-the-keystone-parcels-in-a-watershed.md
subtitle: "position, not acreage, with public data and a map"
category: how-to
---

# how to find the keystone parcels in a watershed

*position, not acreage, with public data and a map*

To prioritize land for conservation by what it holds up, start from what must keep working downstream, draw the area that serves it, and score each parcel by position: watershed position, wetland, ecosystem edges, corridor pinch, people within reach, and whether anything could replace it. Rank, then list who depends on each.

That is the whole method. The rest of this page is the steps, the public layers that make each one possible, and the places where the method is honest about what it cannot see.

A [**keystone parcel**](/guide/what-is-a-keystone-parcel?from=guide) is a piece of land whose condition holds up more than its own acres. The parcels you will find are living things: a headwater forest, a wetland at a low point, a strip of cottonwoods along a dry-country creek. They exist whether or not anyone pays for them. [**Ensurance**](/?from=guide) is how the parcels that stand on them can pay them. It is not the forest.

:::johnson
**position, not acreage.** — A 40-acre wetland at a catchment's one low point can outrank a 4,000-acre block in the middle of a bigger one.

[see which places hold others up →](/explore?from=guide)
:::

## step 1: start from what must keep working

Most conservation planning asks what is rarest or what prevents extinction, and it has good tools for that. This method asks a different question: what do the town, the intake, and the orchard stand on? Start from the thing downstream that cannot fail. A water intake. A hospital's cooling plant. A rail embankment. A town on a floodplain. A valley of orchards. Write it down with a number attached to it: customers served, beds, trains a day, acres in fruit.

This is the inward trace described in [the dependency your map can't see](/guide/the-dependency-your-map-cant-see?from=guide). Everything after this step is finding the parcels that trace lands on.

## step 2: draw the shed

Draw the area that delivers the service to the thing in step 1. For water, that is the catchment above the intake. Public layers do most of the work:

- **Parcels.** The county's parcel layer; most US counties publish one. Everything below gets scored against these boundaries.
- **Basins and flow.** HydroBASINS gives nested catchments worldwide, each with a pointer to the next basin downstream, and places the shed. In the United States, NHDPlus adds every stream reach and the land that drains to it; use its catchments or a flow-accumulation grid for position at parcel grain.
- **Wetlands.** The National Wetlands Inventory in the United States. Any mapped wetland in the shed is a candidate before you score anything else. Outside the United States, wetland maps are thinner; expect to draw some by hand.
- **Land cover.** The National Land Cover Database, or a global equivalent, for forest, grassland, cropland, and built ground.
- **Existing protection.** PAD-US in the United States, the World Database on Protected Areas elsewhere. A parcel next to a protected area tends to receive spillover from it. Use GAP status 1 and 2 as protected for biodiversity; GAP 3 and 4 land is managed for other uses and can still be a keystone. A protected keystone is still a keystone; protection does not pay for its condition.

For pollination, microclimate, and fire, the shed is a ring rather than a catchment. [What is a serviceshed](/guide/what-is-a-serviceshed?from=guide) gives the typical reach for each: 600 meters to 3 kilometers for pollinators, 50 to 200 meters for shade, 100 meters to a kilometer for fire, by the fuel.

## step 3: score position

Within the shed, every parcel gets six marks. None of them is size.

| proxy | what to look for | the layer |
|---|---|---|
| Watershed position | Headwater parcels score high; the water for everyone below starts there | HydroBASINS, NHDPlus |
| Wetland presence | Any mapped wetland is a candidate; drop excavated and impounded classes (farm ponds, ditches) and check the map's date | National Wetlands Inventory |
| Ecotone density | More ecosystem transitions per unit area: forest to meadow to water | Land cover, in a moving window |
| Corridor bottleneck | Is this the narrowest point in a habitat corridor between two larger blocks? | Land cover, roads, protected-area edges |
| Beneficiary density | Count the built assets and people inside each mechanism's reach of this parcel | Buildings, roads, intakes, census blocks |
| No substitute | Does another parcel in the shed do the same job for the same people? If no, this one holds a position; if yes, it is matrix | Your own reading of the first five |

There is no calibrated cutoff. Mark each parcel high where it sits in the top quarter of the shed on that proxy, rank by how many marks it carries, and write down why each parcel near the top is there. Expect overlap: headwaters, wetlands, and edges often come together, so a parcel can collect three marks for one fact.

## step 4: count the mechanisms

Go back to the seven ways a parcel reaches past its fence, listed in [does protecting one parcel help the land around it](/guide/does-protecting-one-parcel-help-the-land-around-it?from=guide): pollination, predators, water, shade, soil, fire, genes. Mark how many of them this parcel contributes. Count the strong ones first: water, shade, and pollination where crops sit within reach. Mark predators, soil, and genes as possible; the evidence for them is moderate or weak. A riparian forest often carries several at once. A parcel that carries several is structurally more important than one that carries one, because its loss cuts several services at the same time.

## step 5: subtract the bad neighbor

Ask what the parcel exports that nobody wants. A patch of habitat can harbor the pest that eats the orchard beside it. A dense stand can carry a fire run into a subdivision that a mowed field would have stopped. The evidence on pest control in particular is inconsistent: a 2018 analysis across 6,759 sites found predator responses to the surrounding landscape were inconsistent across systems. A candidate that exports harm is not disqualified, but the harm goes on the same sheet as the help, and a plan to manage it comes before any claim.

## step 6: rank, then ask who is inside

You now have a short list: the handful of parcels that hold the shed up. For each one, list who sits inside its reach. The intake. The hospital. The farms. The road. The insurer of the road. That list is the set of parties with a reason to keep the parcel in condition, and the start of [the free rider is standing on a keystone parcel](/guide/the-free-rider-is-standing-on-a-keystone-parcel?from=guide).

## going further: the modeled version

The six proxies are a screen you can run with a laptop and public data. A modeled version does three more things, and the ecologist Jay Gutierrez has built one.

His Spillover Predictor ingests public layers (StreamCat, NHDPlus, land cover, soils, LANDFIRE fuels, forest inventory, USGS water quality), builds a feature matrix for the area, and maps spillover pathways on a hydrologic graph: parcel to catchment to stream to beneficiary. It then runs hierarchical Bayesian inference over seven mechanism states (hydrology, sediment, microclimate, fire admissibility, biodiversity support, deployability, timber resilience; a different split from the seven families in step 4) and scores each parcel by its posterior probability of being what he calls an "investable keystone", with the uncertainty stated. The last step is a value-of-information calculation: where would one more measurement change the ranking most? That is where to put a sensor or a survey first.

On 42 synthetic parcels in a Central Virginia test it ranked seven as pilot-ready, four where monitoring should come first, and four where mitigation should come first or run in parallel. The sets overlap. Its inputs are public. The model is his, and it is unpublished. One modeled run. It shows the model works end to end. It does not test the six-proxy screen above, and it does not show that any real parcel is keystone.

## the limits, stated

**Proxies are proxies.** Watershed position and a wetland flag are strong signals. Ecotone density and beneficiary counts are weaker. Replaceability is a judgment.

**Below a data threshold, say so.** Where the layers are thin, report that the parcel cannot yet be scored rather than a confident number. The modeled version does the same: it reports deployability, not a verdict, when the evidence is too sparse.

**The established tools solve a different problem.** Marxan, Zonation, and prioritizr pick a set of sites that meets targets for the features you give them at least cost, and you can give them service layers. InVEST maps where a service is produced and, in some models, where it is delivered. None of them, by default, scores one parcel by how many people and assets stand on it. You can feed the six marks into prioritizr as a feature. Marxan's *irreplaceability* means something else: how often a site is needed to meet a target. Which parcels spill most, for whom, is still an open research question, and this screen does not settle it. The method here is a complement, and it is still a method, not a validated product.

**It is not a title search.** Knowing a parcel is keystone says nothing about who owns it, whether they would talk, or what a deal would look like. That is a separate conversation. No rights, no model: talk to whoever holds and cares for the parcel before the map leaves the room.

**The past is not a forecast.** Effect sizes were measured in the climate of recent decades and may not hold under a changing one.

## what we run

Inside our own shop, an admin tool asks three questions of any place we hold a record for, answered from the records we already have: what the place needs and which stored parcels provide it; which stored parcels it depends on and which depend on it; and who already depends on it, so who to call. It does not score the six marks above. It reads relations already on file, and the water answer uses the stored basin chain, one basin up and one down. A missing answer is a hole with a next step. We do not sell it. The public door is [explore](/explore?from=guide), and when a record allows, a property's card shows the answers.

## frequently asked questions

### how do you prioritize land for conservation?

Start from what must keep working downstream, draw the area that serves it, and score each parcel by position rather than size, using the six marks in step 3. Rank, then list who is inside the reach of each top parcel.

### what data do you need to find keystone parcels?

Basin and stream layers (HydroBASINS, NHDPlus), a wetland inventory, a land-cover map, and a protected-area layer. All are public. Building footprints, roads, and census blocks fill in the beneficiary count.

### is a bigger parcel more likely to be keystone?

Not by size alone. A large headwater forest can be keystone, but most interior acres in a large block are redundant for any one beneficiary. Keystone parcels often sit at seams: a headwater, a riparian strip, a corridor's pinch, a wetland at a low point. [The missing object is the graph](/guide/the-missing-object-is-the-graph?from=guide) makes the point that the biggest patch is rarely the load-bearing one.

### can this method tell me whether to buy a parcel?

It can tell you whether a parcel holds other things up and who those things are. Whether to buy, lease, or pay for its condition is a separate decision that depends on the owner, the price, and the people on the list.

## next

The next post in this series turns to the people on the list: [the free rider is standing on a keystone parcel](/guide/the-free-rider-is-standing-on-a-keystone-parcel?from=guide).

If you run a watershed council or a landscape partnership and have already drawn the shed, [regional collaboratives](/solutions/regional-collaboratives?from=guide&topic=keystone-parcels) shows how we work beside one. If you have drawn a short list, [tell us what you found](/contact?from=guide&topic=keystone-parcels). We will say what our records hold for those places, and say so when they hold nothing.

## sources

[HydroBASINS](https://www.hydrosheds.org/products/hydrobasins) — nested catchments with downstream pointers, worldwide.

[NHDPlus High Resolution, USGS](https://www.usgs.gov/national-hydrography/nhdplus-high-resolution) — stream reaches and their catchments, United States.

[National Wetlands Inventory, U.S. Fish and Wildlife Service](https://www.fws.gov/program/national-wetlands-inventory) — mapped wetlands.

[National Land Cover Database, USGS](https://www.usgs.gov/centers/eros/science/national-land-cover-database) — land cover.

[PAD-US, USGS](https://www.usgs.gov/programs/gap-analysis-project/science/pad-us-data-overview) — protected areas, United States.

[Karp, D.S. et al. 2018. Crop pests and predators exhibit inconsistent responses to surrounding landscape composition. *PNAS*.](https://doi.org/10.1073/pnas.1800042115) — 132 studies, 6,759 sites; pest and predator responses inconsistent across systems.

Jay Gutierrez, the Spillover Predictor, 2026 synthesis and Central Virginia demonstration — public inputs, hierarchical Bayesian inference, value-of-information for monitoring; 42 synthetic parcels. Unpublished synthesis; named in earlier posts on this guide.

## the series

1. [what is a keystone parcel](/guide/what-is-a-keystone-parcel?from=guide)
2. [does protecting one parcel help the land around it](/guide/does-protecting-one-parcel-help-the-land-around-it?from=guide)
3. [what is a serviceshed](/guide/what-is-a-serviceshed?from=guide)
4. [how to find the keystone parcels in a watershed](/guide/how-to-find-the-keystone-parcels-in-a-watershed?from=guide)
5. [the free rider is standing on a keystone parcel](/guide/the-free-rider-is-standing-on-a-keystone-parcel?from=guide)
6. [how a keystone parcel gets paid](/guide/how-a-keystone-parcel-gets-paid?from=guide)