This release refreshes nearly every indicator with the latest available data (most through 2025, including the 2025 USDA Forest Inventory and Analysis evaluation and 2025 NLCD land cover). It also includes substantive methodology work: all indicators unified onto one scoring framework of three families, removing the non-monotonic “distance-from-a-moving-mean” artifact; a new Stand Complexity metric (a Gini coefficient of tree basal area, scored against published reference bounds); a new robust trend method (Theil-Sen slope with a Mann-Kendall significance test); the earlier Ozone Exposure scoring correction; a Snow Cover station-averaging correction; an area-weighted Mean Forest Patch Size; and new typed uncertainty bands. Several behind-the-scenes data-handling improvements were also made.
Data updates
Indicators were refreshed to the latest year each source has released:
- Forest Damage, Forest Mortality, Invasives (aerial detection survey): through 2025; now computed on a Vermont-only subset.
- Precipitation, Minimum Temperature, Maximum Temperature: through 2025; now computed on a Vermont-only subset (previously aggregated across multiple states).
- Ozone Exposure: through 2025 (also a scoring correction, see below).
- Snow Cover: through 2025; Vermont station coverage expanded from 2 to 12 GHCN-D stations (also a methodology correction, see below).
- Recreation (state-park visits): through 2025.
- Bird Diversity (Vermont Forest Bird Monitoring Program): through 2025.
- Climate Extremes (U.S. Climate Extremes Index, Northeast): through 2025.
- Canopy Density (Condition) and Growing Season Length (Stressor): through 2024 (USGS Remote Sensing Phenology). Note a sensor change at 2021 (eMODIS to eVIIRS, a different platform/resolution) in the underlying series. Growing Season Length values are also corrected this release: a masking fault in the processing had let pixels outside Vermont (inside the state’s bounding box) enter the sample as zero-day seasons, biasing the annual mean low by roughly a third and pinning the lower edge of its variability band at zero. Historical values and scores are recomputed; Canopy Density was unaffected.
- Hunting Harvest: through 2025.
- Acid Deposition (NADP National Trends Network): through 2024.
- Mercury Deposition (NADP Mercury Deposition Network): through 2025.
- Drought (six-month growing-season SPEI): through 2024.
- Maple Syrup Production: through 2024.
- Forest Cover, Forest Connectivity, Mean Forest Patch Size, Impervious Surface (NLCD land cover): through 2025. Forest Cover is computed from NLCD rather than FIA survey estimates, so specific values differ from prior releases though the long-term pattern is unchanged. Mean Forest Patch Size also changes statistic this release (see its section below). NLCD Tree Canopy Cover stays at 2023 (the 2024/2025 canopy product is not yet published).
- FIA indicators (USDA Forest Inventory and Analysis, 2025 Vermont evaluation): Carbon Storage, Tree Growth, Tree Mortality, Damage and Decay, Crown Dieback, Softwood and Hardwood Regeneration, Stand Age Diversity, and Species Diversity, all through 2025. Stand Complexity is refreshed on its new Gini dataset (see its section below). These were on hold for the new Vermont evaluation, which has now published and been ingested.
Methodology correction: Ozone Exposure
What changed: The Ozone Exposure score now uses a data-derived upper bound (the dataset maximum plus 10% of the data range), as described in the published methodology, rather than a hardcoded ceiling of 17 ppm-hour.
Why this matters: The implementation had diverged from the documented methodology and produced overly lenient scores. With the previously hardcoded ceiling, Vermont’s ozone data never approached the top of the scale (Vermont’s observed W126 maximum is ~10.3 ppm-hour).
What this means for users: Historical scores have been recomputed. Years with elevated ozone (notably the mid-1990s and 1999) now score appropriately lower, while modern years remain at 5. The directional finding (Vermont’s substantial long-term ozone decline) is unchanged; its visible magnitude is now more accurately represented. The published methodology is unchanged; this brings the code into agreement with it.
Methodology change: Stand Complexity (new metric)
What changed: Stand Complexity is now measured as a Gini coefficient of tree basal area, computed from the same USDA Forest Inventory and Analysis data, replacing the Shannon-Weiner index over 5-inch diameter classes. It is scored against published reference bounds (0.50 to 0.75; Sterba and Sterba 2018) instead of the observed range of Vermont’s own record, and it now carries a confidence interval (see the uncertainty bands section).
Why this matters: The Shannon index treats diameter classes as unordered categories and is maximized by a flat distribution: an equal number of trees in every size class. The ecological hallmark of a complex, resilient, uneven-aged stand is different: a “reverse-J” distribution with many small stems alongside fewer large, dominant trees. The Gini coefficient respects the order and magnitude of tree diameters and rewards exactly that structure; comparative evaluations rank it the best-performing diameter-diversity index (Lexerod and Eid 2006). Fixed, science-set bounds also mean the score no longer rescales the past whenever a new extreme year arrives.
What this means for users: The displayed value changes scale (a Gini near 0.71-0.73 for Vermont, rather than a Shannon index value), and historical scores are recomputed. Vermont scores high (about 4.4-4.6 of 5) and slowly rising over 2005-2025, a real structural-maturation signal, now measured against an external reference with headroom in both directions.
Methodology correction: Snow Cover station averaging
What changed: The annual snow-cover value is now the average of each station’s departure from its own long-term normal (plus a fixed constant so the series still reads in snow-days), instead of a plain average of raw station values. Stations need at least 15 qualifying snow-years to be included, and the confidence interval is the standard error of that mean departure.
Why this matters: The station list is fixed, but which stations report a complete snow year varies, and their long-term normals span roughly 82 to 197 snow-days in Vermont alone. Four of the twelve Vermont stations have no record in the 1961-1990 baseline window, so a plain average compared a snowier historical station mix against a less-snowy modern one: about 60% of the apparent decline against the baseline came from the changing mix, not from snow. Measuring every station against itself removes that artifact; a station joining or leaving the record can no longer tilt the series.
What this means for users: Historical values and scores are recomputed. Vermont’s baseline-to-recent decline shrinks from about 21 snow-days to about 9 (still a real decline, now the part attributable to weather), and the Snow Cover score rises accordingly.
Methodology change: Mean Forest Patch Size (area-weighted)
What changed: The indicator now scores the area-weighted mean patch size, each forest patch weighted by its share of total forest area, instead of the plain mean of patch areas. It answers “how large is the patch containing a randomly chosen forest hectare.” The displayed value changes scale substantially (Vermont 2025: about 35,160 ha, versus 75 ha under the plain mean).
Why this matters: The patch population is numerically dominated by tiny classification slivers a few pixels across, while a handful of very large contiguous patches hold nearly all the forest area. The plain mean is therefore pulled between those extremes and largely tracks the proliferation of edge fragments; the median fails outright here (half of all mapped “patches” are 3-7 pixels, so the median is quantized to two or three pixel-multiples per region and cannot carry a trend). Area weighting is the standard remedy in the landscape-ecology literature (the effective-mesh-size family) and makes slivers contribute almost nothing.
What this means for users: Historical values and scores are recomputed, and Vermont’s trend reverses: the plain mean showed a significant decline, driven by growing numbers of small edge fragments, while the area-weighted mean shows the large core patches, where nearly all the forest area actually is, consolidating: a significant increase. Both patterns are real; the indicator now reports the one that describes the forest rather than the fragments. Read alongside Forest Cover and Forest Connectivity, which capture amount and linkage.
Methodology corrections: unified scoring across all indicators
What changed: The dashboard’s 33 indicators now share one coherent scoring framework of three families. About half had been scored by distance from a recomputed long-term mean (“tent” scoring), whose reference drifts as new data arrives, so several reported their trend backwards or washed it out. The three families:
- Directional (Category 1): more of a good thing scores higher, more of a bad thing scores lower, across the observed range. Six indicators that used a compressed “value over the maximum” formula (hardwood and softwood regeneration, growth, timber production, hunting, stream indicators) now use the full 1-5 range; Forest Cover, Forest Connectivity, Mean Forest Patch Size, and Stand Age Diversity were confirmed directional (their tent scoring had inverted or flattened real trends).
- Departure from a fixed reference (Category 2): the score is highest at a frozen baseline and declines smoothly for departures either way. Climate indicators (precipitation, minimum and maximum temperature, snow, drought) use the 1961-1990 baseline; shorter records (growing season, tree growth, damage and decay, tree mortality, tree health) use the average of their first several years, held constant. Tree health, tree growth, damage and decay, and tree mortality are treated as stability measures: dieback, death, decay, and growth are natural forest processes, so an unusually high or unusually low value both score below the healthy baseline. Climate Extremes is scored one-sided (fewer extremes than the historical normal is simply good, not a stressor), with the top score anchored one standard deviation below the 1961-1990 mean: genuinely calm years score highest, a year at the baseline average reads about 3.7 of 5, and only extremeness above that anchor pulls the score down. (Anchoring below the mean keeps the long-term rise in extremes visible as a score trend, instead of pinning half the record at exactly 5.)
- External thresholds (Category 3): acid deposition, mercury deposition, ozone, and tree species diversity keep their published, science-set targets. Tree species diversity also had a bug fixed that had been pushing its score too high. Stand Complexity joins this externally-referenced family this release: its new Gini metric is scored against published reference bounds rather than the observed range (see its section above).
- Scores that were stuck near the top of the range: review of the recomputed scores found that Stand Age Diversity and Maple Syrup Production were not using the full 1-5 scale; a fault in how their scoring bounds were calculated meant the lowest years were never allowed to score low. Stand Age Diversity was the clearest case: its measured value has declined significantly over the past two decades, but its score had stayed in a narrow band near 4.4-4.7, hiding that decline. Both now use the full range, so earlier years score noticeably lower than before and the long-term trend is visible on the dashboard for the first time. Mercury deposition also had a rounding fault that could push a score just past 5.0.
What this means for users: Affected historical scores are recomputed and some trend arrows change direction to correctly reflect better-vs-worse. Stand Age Diversity and Maple Syrup Production show the largest shifts, because their scores now span the full scale. The underlying data is unchanged, only how it is scored.
New trend method: robust Theil-Sen slope + Mann-Kendall test
What changed: The trend arrow and the dotted trend line now use a Theil-Sen slope (a robust median-of-pairwise-slopes, resistant to a single unusual year) with a Mann-Kendall rank-based significance test, replacing the previous ordinary-least-squares fit and parametric t-test. Direction and significance are also separated: the arrow direction follows the trend of the health score (so it always points toward better or worse), while whether an arrow is shown at all (significance and magnitude) is judged on the raw data. An arrow appears only when the data trend is statistically significant and large enough to matter. The Overall and category summary rows pass the same significance and magnitude gates as individual indicators (their magnitude is judged against the score’s average level, since a summary has no separate raw data series); with the current record the Overall arrow reads flat.
What this means for users: Trend arrows are more robust to outlier years and more faithful to the long-term direction. Some arrows may change relative to prior releases; for example, an indicator with a weak but real long-term data trend can now show an arrow where a noisy least-squares fit previously read flat.
New capability: typed uncertainty bands
Where it is statistically appropriate, the dashboard now carries and can display an uncertainty band around each annual value, and it labels the kind of uncertainty, so a confidence interval is never confused with other kinds of spread.
- Confidence interval, design-based (from the source): the FIA-derived indicators (Carbon Storage, Tree Growth, Tree Mortality, Damage and Decay, Crown Dieback, Softwood/Hardwood Regeneration) carry the standard error FIA publishes with each estimate. Part of this release, with the 2025 FIA data.
- Confidence interval, computed from field samples: Acid Deposition (annual mean pH), Stream Indicators (EPT richness), and Snow Cover (departures averaged across reporting weather stations) compute the standard error of the annual value directly from the underlying samples.
- Confidence interval, bootstrap: Bird Diversity (Living Planet Index) carries the bootstrap interval the index method already produces. Stand Complexity (new metric this release) builds its interval by resampling FIA field plots and recomputing the Gini; the middle 95% of those recomputations is the band.
- Confidence interval (diversity measures): Species and Age Diversity now show a confidence interval, so you can see how precisely each diversity value is estimated. New this release. (Stand Complexity’s interval is the bootstrap band above.)
- Spatial variability (not a confidence interval): Canopy Density and Growing Season Length can show the interquartile range of pixel values across the state, the typical spread of the measurement across Vermont. It is drawn and labeled distinctly, because it describes how variable the landscape is, not how precisely the average is known.
- Map-accuracy band (Forest Cover): Forest Cover now shows a band based on how accurately the underlying satellite land-cover map classifies forest, a published map-accuracy figure rather than a survey sample. It is drawn and labeled distinctly from a confidence interval. New this release.
- No band shown: full counts (harvest, park visits), complete-coverage or derived indices (drought, the climate extremes and ozone indices), and statewide averages (temperature, precipitation) are not sample estimates. Among the land-cover landscape measures, Forest Connectivity and Mean Patch Size do not yet show a band: a sound way to express their uncertainty is still in development, deferred to a later release, not dismissed.
How to read it: a wider band means less certainty (or, for spatial variability, more spread across the landscape); the legend states which kind each indicator shows. For the confidence-interval indicators the size varies a lot: a large, well-sampled total like statewide carbon has a tight interval (about 1%), while a rare subpopulation like heavy crown dieback has a wide one (about 20%); both are correct. Published values and scores are unchanged; the band only adds context.
New capability: Annual Variability badge
The dashboard now offers an optional third lens on each indicator: how stable it is year to year, separate from its score (where it sits) and its trend (which way it is moving). Annual Variability compares the typical year-to-year change in the indicator’s raw values against the record’s overall variation (a mean successive squared difference ratio), binned into low / medium / high against fixed thresholds (the ratio is unit-free, so the badge is comparable across indicators despite their differing units). A slowly trending indicator correctly reads as stable; only a series that genuinely see-saws — reversing direction year to year faster than chance scatter would — reads high.
- A color-coded rating next to each indicator’s score marks its Annual Variability bin (green Low, amber Medium, red High). An indicator with too short a record to bin reliably shows no rating.
- The badge can be turned off from a new Settings tab in the About box (on by default); the preference is remembered in your browser.
Why it matters: two indicators can share a score but tell different stories. Snow Cover sitting at 3.5 with low variability is a steady warming signal; the same 3.5 with high variability is winters bouncing between deep and thin, a destabilization signal the score and trend alone do not capture. Scores, trends, and data are unchanged; this only adds context.
Indicators not updated this cycle
These rely on source data that has not yet been released for the most recent year; they will be refreshed in a future release as data becomes available:
- Stream Indicators: latest manual IWIS export is through 2022.
- Timber Production (VT FPR harvest summary): latest is through 2022.
- Tree Canopy Cover (NLCD TCC): the 2024/2025 canopy product is not yet published, so it stays at 2023 (the NLCD land-cover metrics, by contrast, reached 2025).
- Acid Deposition, Maple Syrup Production, and Drought: sources not yet released for 2025 (at 2024).
Behind-the-scenes improvements
Beyond the scoring, trend, and uncertainty methodology changes described above, a number of internal data-handling improvements were made to the indicator processing pipelines so they keep running correctly as upstream data sources evolve. These do not affect any indicator’s score.
One user-visible display fix is included: score-trend lines had been drawn on a 0-5 scale instead of the dashboard’s 1-5 scale (so they sat too low and could clip at the bottom of the chart), and the downloadable CSV had a mislabeled trend column and a doubled .csv file extension. All three are corrected; no stored score or value changed.
About reproducibility and methodology updates
The Vermont Forest Indicators Dashboard is maintained by the Forest Ecosystem Monitoring Cooperative at the University of Vermont. We are committed to transparent methodology. When implementation discrepancies are identified (as with the ozone scoring correction in this release), we correct them, document the change publicly, and recompute affected historical scores. We do not silently revise published values.
If you have referenced specific scores from a previous release of the dashboard in your work, you may notice changes. The full methodology for each indicator is available within the dashboard under “How is this calculated?” The version of the dashboard used for any given citation is recorded in the dashboard’s URL parameter and persists for reference.
For questions about specific indicators or to report a discrepancy, contact the Forest Ecosystem Monitoring Cooperative at femc@uvm.edu.
Citation
When citing this version of the dashboard, please use:
Forest Ecosystem Monitoring Cooperative. 2026. Vermont Forest Indicators Dashboard, Version 5.0. Available at: https://www.uvm.edu/femc/indicators/vt
