How LakeRadar scores a lake

Where every value comes from

Every number on a lake page carries one of four badges, and the badge is always visible next to the value it describes. Nothing is shown without one, and nothing is shown at all when there is no source behind it.

Each badge renders one of these sentences beside its value. The station, source and timestamp below are illustrative; the wording is the wording the site uses, produced by the same code that produces it on a lake page.

Measured

Measured by USGS 09380000; as of Mar 14, 1:45 PM UTC; provisional data subject to revision.

Forecast

Forecast from the National Weather Service; updated Mar 14, 1:45 PM UTC.

Calculated

Calculated by LakeRadar from the conditions shown; see methodology.

Curated

Curated from a published state agency source; verified Mar 14, 1:45 PM UTC.

Two further sentences are fixed and appear verbatim wherever they apply:

• Not available from a verified source for this lake.
• Latest source update is older than expected. Use caution and check the official source.

Starting confidence by kind

A badge also sets the ceiling on how much that value can contribute to a score's confidence. A measurement from an instrument outranks a model run, and a model run outranks nothing at all.

How long a value stays current

Freshness is judged per data type, not site-wide. An alert that is twenty minutes old is unusable; a curated fact about a boat ramp is still good after a year. Inside the first window a value counts as current and its confidence factor is 1.00. Past it, the factor decays linearly toward 0.50 at twice the window. Past the third column the value is excluded from scoring entirely rather than quietly trusted.

A gauge whose documented upstream cadence is slower than the generic window is judged against its own cadence instead, so a daily-updated station is never marked stale for behaving normally. What the visitor sees is one of four labels:

• Up to date
• Older than expected
• Stale
• Timestamp unavailable

How a score is calculated

Scores are computed from stored readings at render time under versioned constants, currently lakeradar-score/1.0.0. Every persisted score records that version, so a published number can always be reproduced from the weights that produced it. Bumping a weight, a curve, a threshold or a cap means a new version — a shipped version is never edited.

1. Each factor for the activity is evaluated against the stored weather or water values and produces a 0–100 sub-score.
2. Missing optional factors are dropped and the remaining weights are re-normalized, so an absent input does not silently count as zero.
3. A missing required factor suppresses the score entirely. No number is produced from an unrepresentative subset of the inputs.
4. The weighted sub-scores are summed, then the safety caps below are applied. Caps run last, so no amount of good weather can out-weigh an active warning.
5. A confidence value is computed from the source confidence and freshness of the inputs actually used. Below 0.60 no numeric score is displayed at all — only the reason.

An input that is missing contributes 0.50 to the confidence mean while being excluded from the score's weights. Unknown data halves the confidence contribution rather than zeroing it, so one absent optional input does not suppress an otherwise well-sourced score.

Score labels

The same bands apply to every activity and to the overall score. A label is a summary of a number, not a verdict: it never appears without the score and the confidence beside it.

What each activity weighs

Each activity has its own weight table summing to 100 points. The tables below are the tables the engine reads, so what is written here is what is computed.

The overall score

The overall score is a weighted mean of the five activity scores. Fishing and boating carry the most weight because they are the primary audiences; sailing, kayaking and swimming are scored identically in their own right and simply count for less of the single headline number. An activity whose score is suppressed contributes nothing, and the remaining weights are re-normalized.

Safety caps and official alerts

An active National Weather Service alert overrides the arithmetic. The alert itself is always rendered above any score, with its event, severity, expiry and a link to the official NWS page, so a warning is never something a visitor has to scroll past a number to find.

Event names are matched case-insensitively against the NWS event field, and the thresholds are configuration rather than hardcoded prose, so operations can tighten them without a code change. A cap lowers a number. It is not a statement that conditions are acceptable, and LakeRadar does not state or imply that any lake is without lightning hazard, that any water is at a safe level, or that any trip is worth taking.

Best windows

A best window is a span of hours in the forecast during which an activity scores well. It is derived from the same hourly rows as the hourly table, and it is only claimed when there is enough forecast data to support it.

When the forecast is too short, or no span reaches the minimum average, LakeRadar shows no window. An empty window slot is a true statement; an optimistic one is not.

Which lakes get published

A URL is worth indexing because it satisfies a distinct task with enough unique value, not because it exists. LakeRadar scores every lake against the signals below. The maximum is 110 points and a lake enters the index at 70 or above — but three signals are required regardless of the total, because no quantity of other points can substitute for them:

• Verified identity, coordinates and state
• Working NWS mapping
• Official or authoritative source link

A verified USGS gauge is explicitly optional. A lake with no gauge can still be published; its water level and water temperature are simply omitted rather than estimated from a neighbouring lake. A curated summary must reach 80 words to count as useful unique content and is guided to stay under 180. At least 3 lake facts must carry a source, and at least 4 related lakes must be computed so a page is never a dead end.

A state guide is listed in the state directory once at least 5 of its lakes are published. Below that, the state page still exists and still links to the lakes it does have, but it is marked noindex,follow and left out of the sitemap. LakeRadar would rather publish fewer pages: a directory of thin pages teaches a crawler to distrust the directory.

What LakeRadar does not do

• It does not estimate a value from a different lake. If a reading does not exist for the lake you are looking at, it is not shown.
• It does not report a boat ramp as open without an authoritative real-time source. Access points carry their source and the date they were verified, and nothing more is claimed.
• It does not list a fish species without a source, and it does not publish catch reports, hotspots or stocking claims.
• It does not treat its scores as opinions about a business, so they are never marked up as reviews or aggregate ratings.
• It does not ask for your location, load third-party scripts, or set advertising cookies. See the privacy policy.
• It does not replace official warnings. Where an NWS alert exists, the alert is the authoritative statement and the score is secondary.