Data and migration notes

These browser versions were rebuilt from five supplied Shiny application archives in October 2026. Their controls, component order, climate inputs and intended calculations are retained. The original source archives were not edited.

Corrections made

• Pasvik seasonal bands: summer soil moisture, temperature and precipitation, and spring cloud cover now use the same seasonal bands in maps, status boxes and time series. Some original maps used the first (winter) band.
• Pasvik growing-season classification: an R assignment written as if(val<-2) is corrected to the intended comparison val < -2. In the bird-food-sources priority, changes below −2 are DANGER, above +2 are SAFE, and −2 through +2 are NEUTRAL.
• Pasvik metric definition: the description of summer precipitation refers to summer, matching its selected data band.
• GSI time alignment: all 45 annual values are retained and paired with 1980–2024. The original matrix extractor dropped the first value as if it were an ID column.
• Marine production zones: spatial averaging uses the selected longitude/latitude coordinate pairs. The original array indexing also selected the Cartesian combinations of those coordinates, which could include cells outside the chosen set.
• Marine missing values: missing samples break an exceedance run and entirely missing grid cells remain missing on the map. They are not shown as zero-risk cells.
• Snow missing values: absent data remain absent rather than being interpreted as zero snow depth.
• Map orientation: raster rows are projected from their actual geographic coordinates. No extra, display-only latitude flips are applied.

Important scope details

The marine dataset is a forecast snapshot for April–September 2025. Its original run rule is at least five consecutive six-hourly samples within a calendar month. That rule has been preserved and explicitly labelled; it is not a five-day marine-heatwave-duration test. The monthly probability denominator remains 51 ensemble members. The map is a worst-member category display, separate from the probability table.

The snow files identify six forecast months, December–May, but a verified issue date was not supplied. These forecasts are labelled as archived examples. There is no automatic forecast-refresh service.

The CNRM, ICHEC and IPSL model cases retain the source labels Low, Mid and High. Those labels alone do not establish three distinct emissions pathways. Traffic-light classifications retain the original app’s screening thresholds, except for the specific assignment bug described above. In particular, the original forest-fire rainfall rule marks both large increases and large decreases as SAFE; that scientific choice has not been reinterpreted.

The GSI grid is displayed without an additional external land mask, matching the original application’s fallback when its optional mask package is unavailable. Correlations describe association, not causal attribution. The trend line is shown only when its ordinary least-squares slope has a two-sided Student t test p-value below 0.05, matching the source method.

How the website works

Map layers, time series and data subsets are served as static website files. Gzip-compressed float32 data retain the source grid values to float32 precision. Snow and GSI point queries load small geographic chunks at native resolution. Snow map displays use the source’s factor-four averaging; the GSI display uses factor-two averaging. Point extraction retains full resolution. All tool calculations run locally in the visitor’s browser. Base-map tiles are supplied by OpenStreetMap, with attribution shown on each map.

Browser calculations were checked against independent R/terra interpolation and R linear-model statistics, as well as the supplied park metrics and marine ensemble records.