PollenWatch

A multi-source European pollen aggregator for Home Assistant. The Home Assistant ecosystem already has ~10 pollen integrations that each wrap a single provider; PollenWatch instead combines independent sources and adds a cross-source analytics layer on top. That combination is the point.

Note

PollenWatch is a personal project shared publicly. Stable means the maintainer relies on it daily. It ships a bundled Lovelace severity-gauge card that is auto-registered on install — one HACS install delivers both. PollenWatch tracks 24 canonical species across trees, grasses, herbs and one spore; you pick which ones to materialize as entities at setup (preselected from your country) and can change the selection any time in Options. See Known limitations below for honest disclosure of the open items.

Minimum Home Assistant 2024.11.0 (see HA_COMPATIBILITY.md for the API audit).

Important

Upgrading from PollenWatch 1.x or 2.x is a breaking change. Version 3.0 reset the config-entry schema and removed the old automatic migration, so an existing pre-3.0 entry will not load on 3.0+. After updating, delete the PollenWatch integration entry and add it again to reconfigure: re-selecting the same species and sources reclaims the same entity_ids (existing dashboards keep working), but you will need to re-enter any source API keys (polleninformation, Google) and per-species sensitivity settings. Fresh installs are unaffected.

Sources

Each source is optional except Open-Meteo, and what you get depends on your location: outside Germany you get no DWD; outside Switzerland no MeteoSwiss; outside Bavaria no ePIN; outside the 13 polleninformation countries you get only Open-Meteo (all of Europe) and Google (global, if enabled).

Source	Coverage	API key	Notes
Open-Meteo (CAMS)	All of Europe	none	Always-on primary. 6 species, hourly, 5-day forecast, 92-day backfill.
polleninformation.at	13 countries¹	free key required	Optional. Daily 0–4 index forecast; more species, country-dependent.
DWD Pollenflug	Germany only	none	Optional. Daily 7-point regional index forecast; you pick your DWD region.
MeteoSwiss	Switzerland only	none	Optional, observation-only. Hourly grains/m³; nearest of 19 automatic stations auto-picked. Covers alder/birch/grass.
ePIN (Bavaria)	Bavaria only	none	Optional, observation-only. 3-hourly grains/m³; nearest of 8 automatic stations auto-picked. No olive.
Google Pollen	Global	billing-gated key²	Optional, consensus-only³. UPI 0–5 index, 5-day forecast, all 6 allergens (only source with olive).

¹ AT, CH, DE, ES, FR, GB, IT, LV, LT, PL, SE, TR, UA. ² Requires a Google Cloud project with the Pollen API enabled and a billing account (free tier ~5,000 req/month, but a payment method must be on file) — more involved than the other sources' free keys. ³ Consensus-only: Google's Maps Platform terms forbid caching/storing forecast results, so Google feeds consensus/divergence and gets a raw sensor + forecast + personal_score, but is never baselined into recent_percentile.

Forecast vs observation: Open-Meteo, polleninformation, DWD and Google provide a forward forecast (today + coming days). MeteoSwiss and ePIN are observation networks — they report measured concentrations up to the latest reading, with no tomorrow value, so their sensors show the current/most-recent reading and today's running peak rather than a multi-day outlook.

PollenWatch tracks 24 canonical species spanning trees, grasses, herbs and one spore — covering everything any of the six sources can publish, plus the EAACI/D'Amato high- and moderate-potency set. Open-Meteo is the largest-coverage source (the CAMS-canonical 6); polleninformation, DWD, ePIN, MeteoSwiss and Google each add their own subset (e.g. ash, oak, hazel, rye, plane_tree, cypress, plantago, alternaria). At setup you choose which species to materialize as entities; the per-country preselection seeds a defensible starting set. A source only produces sensors for species it actually reports at your location. See Entities for the honest data-availability picture.

Analytics

On top of the raw per-source sensors:

• recent_percentile — today's level versus the recent window (per source). Open-Meteo (92-day backfill) and MeteoSwiss (months of recent hourly data) compute it on day one from their own history; polleninformation, DWD and ePIN baseline on Home Assistant recorder history and honestly report "insufficient history" until ~2 weeks accrue (and "off_season" when the whole window is zero). Google is excluded — its licence forbids storing forecasts, so it gets no percentile (it still feeds consensus and gets a raw sensor).
• personal_score — a source's raw value × your per-species sensitivity multiplier (0.0–2.0), for personal-threshold automations.
• consensus + divergence (cross-source) — sources are compared on a common 3-level scale derived from the EAACI / Copernicus CAMS thresholds (each source's native scale is bucketed or collapsed onto it — see ANALYTICS.md, where every threshold is sourced, not invented). consensus is categorical — none / low / high / mixed — and carries source_count + max_possible_sources so the card can render the n/m authority badge. A single-source species (1/m) gets a consensus reading pass-through (no mixed is ever emitted) plus the card's visual humbling. divergence is a binary flag, on whenever ≥2 sources are not unanimous (any disagreement), with a spread attribute grading mild (adjacent, a level is still reported) vs mixed (≥2 levels apart, no single level). This is the honest companion to take-the-higher: a {1,1,2} reading reports consensus high and divergence on, instead of a minority high masquerading as confident consensus (#1, resolved from live multi-source data).

Installation

PollenWatch is not in the default HACS store yet. Add it as a custom repository:

1. In Home Assistant, open HACS → Integrations → ⋮ → Custom repositories.
2. Add https://github.com/TheDave94/pollenwatch with category Integration.
3. Install PollenWatch, then restart Home Assistant (minimum HA 2024.11.0).
4. Go to Settings → Devices & Services → Add Integration and search for PollenWatch.

Setup & configuration

Setup is a two-step wizard with Open-Meteo on by default.

Step 1 — Location. Pick the point to monitor on the map (defaults to your Home Assistant location). Open-Meteo snaps it to the nearest ~10 km CAMS grid cell; setup is refused if the location is outside CAMS European coverage. Location is fixed once set (to keep recorder history and the percentile baseline coherent). To move it, remove and re-add the integration. You can add PollenWatch more than once for several locations.

Step 2 — Species. A checkbox list of all 24 species, preselected from your Home Assistant country via the region-defaults table (e.g. AT/DE/CH get the Central-European set + ash/oak; IT/ES add olive/cypress/nettle_family; SE skips Mediterranean species; UK gets plane_tree). The maintainer's recommendation is the starting point — uncheck what you don't want, check extras that matter to you. alternaria (a fungal spore, not pollen) is the single deliberate opt-in: never preselected, available if you want it.

In the integration's Options (after setup) you can change the selection and update interval, set personal sensitivity multipliers per species, and enable the optional sources:

• polleninformation.at — toggle on, pick your country, and paste a free API key (requested from polleninformation.at). Stored encrypted in HA.
• DWD — toggle on and pick your DWD region (Germany only; enabling it for a non-German location is rejected as out-of-coverage).
• MeteoSwiss (Switzerland) and ePIN (Bavaria) — just toggle on; no key, no region. The nearest measuring station to your location is picked automatically (shown in the option description and as a station attribute on the sensors). Enabling either outside its country/region is rejected as out-of-coverage.
• Google Pollen (global) — toggle on and paste an API key. Read the option text first: this needs a Google Cloud project with the Pollen API enabled and a billing account attached (there is a free tier, but Google requires a payment method on file) — it is more involved than the other keys. Stored encrypted in HA. Google is consensus-only (no recent_percentile).

Changing the selection later prunes any deselected species cleanly — the entities and their device are removed from the registry, not left as permanently-unavailable orphans.

Entities

Raw sensors live under a per-source device (e.g. "PollenWatch Open-Meteo", "PollenWatch MeteoSwiss", "PollenWatch ePIN"): sensor.pollenwatch_<source>_<species> (state = that source's current value; attributes include a daily-peak forecast and provenance — and, for the station-based sources, the picked station). Source-specific derived sensors sit alongside them: ..._<species>_recent_percentile and ..._<species>_personal_score.

The cross-source metrics live under a separate "PollenWatch Analytics" device: sensor.pollenwatch_analytics_<species>_consensus (categorical none/low/high/mixed; carries source_count + max_possible_sources so the card can render the n/m badge) and binary_sensor.pollenwatch_analytics_<species>_divergence (binary flag created for any ≥2-source species; on when the sources are not unanimous, with a spread attribute grading the disagreement).

How many entities will I see?

The entity-count table below is a ceiling, not a promise — it's the count you'd get if every source you enable also covered every species you select. Real installations almost always fall under it because of the data-availability matrix.

The formula has two parts:

• Per-source entities = species × (3 if the source has recent_percentile, 2 if not). The five non-Google sources have percentile; Google omits it (its licence forbids storing forecasts).
• Analytics entities = species × (2 if covered by ≥2 enabled sources: consensus + divergence; 1 if covered by 1: consensus only).

Selected species × sources enabled	Per-source ceiling	Analytics ceiling	Total ceiling
6 species, 1 source (Open-Meteo only)	6 × 3 = 18	6 × 1 = 6	≤ 24
6 species, 2 sources (Open-Meteo + DWD)	6 × 3 × 2 = 36	6 × 2 = 12	≤ 48
8 species, 4 sources (AT default + DWD + ePIN, no Google)	8 × 3 × 4 = 96	8 × 2 = 16	≤ 112
+ adding Google to any of the above	+ species × 2	unchanged	+ species × 2

A real install almost always lands well under the ceiling: actual coverage is sparse (e.g. ePIN measures plantago + urtica but not olive; DWD has no olive; MeteoSwiss only covers alder/birch/grass + beech).

The honest data-availability picture (one paragraph): not every species is reported by every source — the matrix is asymmetric by design (DWD covers Germany-relevant species, ePIN covers Bavaria-specific species like plantago and urtica, MeteoSwiss measures only alder/birch/grass and beech, Google adds olive globally). A species you've selected that no enabled source covers at your location simply does not materialize an entity — selection bounds the blowup; you don't get permanently-unavailable orphans. A species one source covers still gets a consensus entity (badge reads 1/n, with the gauge in single-source mode — desaturated, with an explicit "single source" label, so the honesty gradient is visible at a glance). The 3-level scale itself (none/low/high) and its grains/m³ boundaries are sourced from EAACI/Pfaar position papers and used by CAMS/Climate-ADAPT (see ANALYTICS.md). Every raw and consensus sensor carries a threshold_status attribute classifying the species into one of 5 evidence tiers — species_specific (peer-reviewed per-species cutoff exists; 13 species), family_eaaci (EAACI's actual family group, no species refinement; 5 species), established_no_threshold (characterised allergen but no published numeric cutoff, working bracket carried; 3 species), family_analogy (analogy-only, weakest; 2 species), and fungal (Alternaria, separate evidence base). Six numeric brackets use refined per-species values (ragweed 5/20, olive 10/200, birch 20/100, alder 45/80, hazel 35/80, mugwort 3/50) — cited basis per species in ANALYTICS.md.

Dashboard card

A bundled Lovelace card ships with the integration and is auto-registered as a resource on install — no manual resource step. It offers four layouts — pick one in Options → Default card layout, and every card on every dashboard follows it. Per-card YAML overrides are supported for power users.

Using the Oriel dashboard strategy? It auto-detects PollenWatch and renders a first-party pollen card + badges automatically — no card configuration needed.

Layout	What it shows	Best for
gauge	One species, full categorical gauge + per-source breakdown on click	Single-allergy "is it bad right now?" view
bars	One row per species, severity-tinted fill bar + level word	Flagship multi-species overview — your whole allergy picture in one card
compact	Dense dot-grid, multi-column	Many species (8+); maximum density
tiles	Severity-tinted icon + name + level, tile grid	Visual scan; mirrors a tile-style dashboard aesthetic

The same data in each layout:

Bars — flagship multi-species overview	Compact — dense dot-grid
Tiles — severity-tinted icon grid	Gauge — single species + per-source breakdown

Light and dark themes are both supported:

Minimal YAML

If you've set the default layout in Options, the card needs no config at all — it picks up your layout choice and your selected species automatically:

type: custom:pollenwatch-card

Per-card YAML

To pick a different layout for one card, or to customise:

# Single-species gauge (original view)
type: custom:pollenwatch-card
species: grass            # required for gauge — any canonical species key

# Multi-species overview (bars / compact / tiles)
type: custom:pollenwatch-card
layout: bars              # one of: gauge | bars | compact | tiles
# species: [grass, birch, oak]   # optional — omit to auto-discover from your selection
title: "Today's pollen"   # optional header
show_inactive: false      # optional — true also shows species currently at 'none'

Canonical species keys (any of the 24): alder, birch, grass, hazel, mugwort, olive, ragweed, rye, ash, oak, beech, carpinus, juglans, elm, plane_tree, cypress_family, holm_oak, plantago, urtica, nettle_family, rumex, chenopodium, asteraceae, alternaria. Gauge-only extras: show_mixed_span: true names the conflicting span on a mixed reading; expanded_default: true opens the per-source breakdown by default.

Honest states (every layout)

Every layout uses the same six categorical states — none, low, high, mixed, unknown, nodata — with deliberately distinct treatments for missing data (gray, no severity colour, no needle in the gauge). An empty reading never visually resembles a safe-low one ("gray, never green"). The mixed state uses a 45° hatch in the overview layouts, drawn consistently across bars/compact/tiles so it reads the same wherever it appears. In the bars gallery above, Birch reads Sources disagree (the 45° hatch) and Hazel reads Unknown (gray, no fill) — a degraded single-source reading that never poses as a safe low. The same fail-safe in the gauge:

Provenance marker

Every row and every gauge carries a small marker next to its level word indicating where the underlying threshold comes from: peer-reviewed species-specific evidence, the EAACI family bracket, a working bracket without a published number, or analogy-only. Hover for the full explanation. The honesty is layered: the level word tells you how bad, the marker tells you how solid the bracket behind that word is. See ANALYTICS.md for the sourcing.

Per-species detail

In gauge layout, clicking the body opens the per-source breakdown — each source's native reading (grains/m³, DWD's 7-point string, polleninformation's 0–4 index, Google's UPI 0–5). In the overview layouts, clicking any row / tile opens HA's more-info dialog for that species' consensus sensor.

The card adapts to HA's light + dark themes; the brand severity ramp stays constant per brand/GAUGE_SPEC.md.

Using a third-party card

The raw per-source sensors are standard HA entities, so Home Assistant's built-in cards (Entities, Gauge, History, Statistics, custom-button/template cards, …) work with them directly. On the raw per-source sensors the normalized severity is in the level (0-based integer) and level_label (none/low/high/…) attributes — point a gauge or template card at those rather than at the state, which carries the source's native value (grains/m³ for Open-Meteo, a native index for the others). On the consensus sensor the state is the level word (none/low/high/mixed), with the integer in the level attribute.

Note on pollenprognos-card: it is not compatible with PollenWatch, including via its manual entity-prefix mode. That card derives each allergen's level from the sensor's state on a fixed per-integration scale (e.g. a 0–6 index), whereas PollenWatch keeps the source's raw concentration/index in the state and exposes the normalized level only in the level / level_label attributes. Pointed at pollenwatch_open_meteo_* it reads a raw concentration where it expects a level and reports no data. There is no PollenWatch adapter in that card; use PollenWatch's own bundled card, or a built-in card reading the level_label attribute, instead.

Attribution

PollenWatch's data carries these required attributions:

Generated using Copernicus Atmosphere Monitoring Service information. Pollen data via Open-Meteo.com.

© Polleninformation Austria

© Deutscher Wetterdienst (DWD)

Source: MeteoSwiss

Source: ePIN, Bayerisches Landesamt für Gesundheit und Lebensmittelsicherheit (LGL)

Source: Includes pollen data from Google

Known limitations

Honest disclosures, not blockers — these describe the state of a project the maintainer uses daily.

• Consensus rounds toward the more cautious reading. With take-the-higher, a single higher source pulls the consensus level up ({1,1,2} → high) — a deliberate health-conservative choice. The {1,1,2} "lone-higher" edge (#1) where this read as confident consensus is resolved: divergence now flags any non-unanimity (not just the spread->1 mixed case), so such a reading reports high and divergence on. See ANALYTICS.md for the data behind the call.
• Per-source maturity is uneven. Open-Meteo, polleninformation and Google run on the maintainer's live HA. DWD, MeteoSwiss and ePIN are validated against the live feeds and exercised on a maintainer-side throwaway HA in Munich; they have not yet run in normal end-user installations, so enabling one of those is its first real-world run. Please open an issue if anything looks off.
• Per-species threshold-evidence tiers are now classified but the underlying number is evidence-graded, not certainty-graded (#3). The 3-level scale (none/low/high) and its grains/m³ boundaries are EAACI/Pfaar-sourced and CAMS/Climate-ADAPT-used. Per-species evidence sits in 5 tiers exposed as the threshold_status attribute on every raw and consensus sensor — see the Entities section above. Important honest caveats from the review: Tier-2 "numbers" are ranges, not points (birch 20–155 across studies, olive 162–400, grass explicitly "no consensus"); the threshold concept itself is contested (a recent EJACI study argues there's no threshold below which sensitive people feel nothing — symptoms rise from the first grains); per-spore allergen content varies up to 15× day-to-day for alternaria. The five-tier label tracks evidence provenance, not clinical certainty.
• Six species use refined per-species brackets rather than the family default — ragweed, olive, birch, alder, hazel, mugwort. Cited basis in ANALYTICS.md § Per-species refinements; full provenance in docs/THRESHOLD_PROVENANCE_REVIEW.md.
• alternaria is a fungal spore, not pollen — kept opt-in (never preselected). Useful for people who track it alongside pollen; safe to ignore otherwise.
• Per-region default selection is a starting recommendation, not a prescription. The country-default table is maintained by the project, not a clinical authority; it's a defensible v1 of "what most people in this country are likely to want." Adjust freely in Options at any time.

Brand & design

Brand identity, design tokens, the gauge spec and reference state SVGs live in brand/ — the design source-of-truth.

Related projects

PollenWatch works on its own, and is also deliberately built to work alongside:

• AirWatch — a companion outdoor air-quality integration that shares PollenWatch's architecture (multi-source aggregation, cross-source consensus/divergence, a bundled severity card). AirWatch's card was derived from PollenWatch's design, so the two read consistently side by side.
• Oriel Dashboard — a Lovelace dashboard strategy that auto-detects PollenWatch and renders it as a first-party pollen card + badges, with no manual card configuration.

License

MIT