METHODS
What 10Point measures,
and what it doesn't.
10Point is a ranked scouting signal built from terrain, cover, weather, and access-related landscape structure. This page explains the inputs, the ranking logic, and the readiness gate — so you know what signal you're using and when we hold one back.
THE SYSTEM
Landscape inputs
terrain, cover, access
↓
Species model
per-region training
↓
Percentile ranking
state + national
↓
Readiness gate
shown, limited, or withheld
↓
Hunt decision
your verification
The signal contract
Five things worth being precise about before you read a map: what we measure, what we rank, what the map shows, when forecast claims are withheld, and what you still have to do.
What goes in
Terrain and landscape structure: vegetation type, canopy cover, vegetation height, elevation, slope, aspect, and road proximity. All aligned to a 30m grid across the lower 48 from LANDFIRE, USGS, and NLCD public datasets.
What gets ranked
Relative habitat quality for the selected species — how each 30m cell compares to the rest of the landscape. Not raw scores, and not a yes/no call. A ranking.
What you see on the map
Every habitat cell's rank shown as a state percentile and a national percentile. Top 1%, top 5%, and top 10% tiers are called out so you can scan fast and prioritize where to spend scouting time.
When forecasts are graded or withheld
Sightline — the hour-by-hour movement forecast — grades what it shows. A spot with full live weather and structural context reads as ready; a spot missing a single factor still shows a graded "partial-input" score, ring, and ranking. The Hot Window call needs complete inputs, and the forecast is withheld entirely only when the core inputs are unavailable, stale, or locked — in which case the product says so instead.
What you still verify
Sign. Access. Recent hunting pressure. Crop changes, logging, burns, or other disturbance. Stand-specific microfeatures inside a 30m cell. The map narrows the question — it does not answer it.
How to read the percentile ranking
Habitat quality is shown as percentile ranks within the selected state. If a cell is in the top 5%, it means 95% of habitat in that state scores lower. The tier swatches below are how that ranking shows up on the map.
Elite — top 1%
Rare ground. Worth a real scout.
Prime — top 5%
High-scoring habitat for the species.
Strong — top 10%
Quality habitat. Good secondary option.
Use top-percentile ground to decide where to scout — not to decide whether to scout. A 30m cell can still hide the pinch that actually holds the deer.
Coverage and update cadence
Species coverage
Whitetail covers the lower 48. Turkey, elk, and mule deer cover their established ranges. Coverage expands with each data refresh.
Habitat refreshes
Habitat layers are rebuilt on each LANDFIRE, NLCD, and USGS data cycle — typically annual.
Wind refreshes
Wind forecasts refresh hourly from NOAA's HRRR model. Free gets current conditions. Pro extends the planning horizon where forecast data is available and claim-ready.
Sightline readiness
Forecast claims are graded, not all-or-nothing. With full live weather and structural context a spot reads as ready; with a single missing factor it still shows a graded "partial-input" read that can be scored and ranked. The UI withholds the forecast entirely only when the core inputs are unavailable, and says prediction data is incomplete.
Where coverage is thinner
Public data density varies by region, and source quality varies with it. The map surfaces where things are well-mapped and where they're not.
Honest limits
Not live animal location
A ranked landscape is a probability field, not a tracker. Good ground raises your odds; it does not spawn deer.
Not stand placement
Microfeatures inside a 30m cell — a pinch, a rub line, a blowdown — matter, and they're smaller than we can see.
Not current intel
Yesterday's pressure, this morning's logging crew, last week's crop change — invisible to us. Boots still run that loop.
Not uniform coverage
Some regions have denser public data than others. Source quality varies. The map says so.
Not fallback-as-forecast
If a required forecast or structural path fails, 10Point does not turn backup/default output into a hot ranking.
Under the hood
Species models are ResUNet-34 convolutional neural networks — they read the landscape the way an image model reads a picture — trained per USGS region against GAP species habitat maps and aligned to a 30m grid across the lower 48. Predictions are converted to state and national percentiles so the map reads as a ranking, not a raw score. Movement Forecast grades every read by the data behind it: a spot with complete inputs and a spot missing a factor both get a score, a ring, and a place in the ranking, and the thinner one is labeled a limited read. Naming the Hot Window itself takes complete inputs. When the core inputs are not there, the forecast is withheld.
Data sources
• LANDFIRE Existing Vegetation Type, Cover, and Height (EVT / EVC / EVH)
• LANDFIRE 30m Digital Elevation Model (elevation, with derived slope and aspect)
• NLCD Impervious Surface Descriptor (road proximity)
• USGS GAP Species Habitat Maps (training labels)
• USGS PAD-US (public-land boundaries, map overlay)
• NOAA/NWS NBM (forecast weather)
• NOAA/NWS RTMA (recent temperature)
• NOAA MRMS QPE (recent precipitation)
