Skip to content
LuckySap
Method & data

How the county maple atlas is built

Every number on an atlas page is produced by a script from a public dataset; nothing is typed in by hand. This page says which datasets, what the rules are, how well they match what producers actually reported (to USDA in the US, to researchers in Quebec), and where they fall short. Canada has its own part below. Download the data (CSV, US and Canada).

1. Production: the Census of Agriculture

Farms, taps and gallons of syrup per county come from Table 37, “Maple Syrup: 2022 and 2017”, in the 2022 Census of Agriculture county volume for each of 17 states. Figures are copied as published. Where USDA withholds a county's taps or gallons to avoid disclosing an individual operation (shown as “withheld”), we leave it blank rather than estimate it. Counties with at least 5 maple farms in 2022 get an atlas page (385 counties); the rest appear only in their state table. The parser checks that county farm counts add up to each state’s published total for both years before anything is written.

Where USDA withheld a 2022 figure for a county with a page but published it in 2017, the maps, cards and tables show the 2017 figure instead, labelled 2017 (9 counties for taps, 11 for gallons, 11 for gallons per tap, each rate from a single census year). It is never ranked, never moves a map’s colour steps and never enters a chart; the downloads keep the 2022 columns blank and add the 2017 figure in its own columns. A 2017 zero is never used.

2. Weather: 36 seasons of daily temperatures

For each county we take the daily maximum and minimum temperature at its county seat (or largest town; the page says which) from the PRISM 4-km daily grid, July 1990 through June 2026, served by NOAA's Regional Climate Centers (ACIS). PRISM is a model that interpolates station observations to a grid, so it is not a thermometer in anyone's sugarbush, and one cell cannot represent a whole county, least of all a mountainous one. A PRISM “day” ends at 7 a.m. Eastern; we checked against five airport stations (about 1,320 days each) and pair each calendar day's overnight low with that afternoon's high the way the live forecast does.

3. The sap-day rule

A sap day is overnight low under 30.2 °F and afternoon high over 37.4 °F, the freeze-thaw definition from field research in Quebec sugarbushes (de Lima Santos et al., 2025), and the same rule our live forecast uses. A classic day is low 20–29 °F and high 40–49 °F, the textbook pattern extension services describe. Weekly charts count sap days per 7-day bin from December 1; monthly shares use calendar months.

4. Season signals, and how we chose them

USDA NASS published each state's approximate average season opening and closing dates from 2011 to 2022. That gives 117 state-seasons of producer-reported truth across 13 states. We scored dozens of candidate rules against it, weighting each county by its 2022 farm count. Rules built only on freeze-thaw clusters cannot time the opening, because in the southern half of the maple belt freeze-thaw days happen all winter: they land three to six weeks before producers actually tap. Heat accumulation does the work.

Chosen season rules and their fit to USDA-reported dates
SignalRuleMean errorCorrelationWithin-state correlation
ClosingThe last sap day on or before the day thaw degree-days (daily mean above 32 °F, summed from Jan 1) reach 2505.5 days0.930.89
Opening35 days before that degree-day date, never earlier than the first sap day that starts a week with at least 3 sap days6.6 days0.850.76

The fit is the same on odd and even years (closing 5.5 vs 5.5 days; opening 5.8 vs 7.4), so the constants are not an artefact of the years used to choose them. Three quarters of closing estimates land within a week of the reported date; 91–93% within two weeks. The rule is tightest in the northern tier (Maine, New Hampshire, New York, Vermont, Wisconsin, Minnesota: 2–5 days) and runs early at the warm edge (Ohio, Indiana, West Virginia), where producers keep collecting after the heat threshold.

5. Anchoring to USDA

Because the rule carries a state-specific bias, the typical dates shown on a county page are anchored: the state's USDA median date plus the difference between that county's signal and the farm-weighted state signal. That keeps only the within-state differences (south before north, valley before hill) and removes the rule's bias. Illinois, Iowa, Kentucky and Missouri have no USDA dates, so their pages show the raw weather signal with a caveat. Year-to-year spread is each county's own interquartile range of signal dates over 36 seasons.

6. Limits worth knowing

  • One grid cell per county; your woods may run days earlier or later with slope, aspect and elevation.
  • Producer-reported dates reflect decisions (when people chose to tap) as well as weather.
  • USDA records are short for Indiana, Minnesota and West Virginia (three years each).
  • PRISM's authors advise against using it for very long-term trends; trend statements on this site use long-record station observations instead.
  • The Census counts farms; backyard sugarmakers who don't sell are not in it.

7. Downloads and reuse

Both halves of the atlas are free to download as CSV, written by the same scripts from the same data as the pages. A blank cell means the agency withheld or did not publish the value; nothing is estimated, and a missing value is never written as zero. Each file opens with a few comment lines starting with # (read them with comment="#" in pandas or comment.char in R), and the README holds the full column guide and every licence line.

  • county-maple-atlas.csv385 rows · 82 columns · Data 2026-10-06, tree counts 2026-10-07

    US counties

    2022 and 2017 Census of Agriculture farms, taps and gallons (withheld values blank), typical season dates and the spread of the weather signals, sap days for each of 36 seasons, and the Forest Service’s count of sugar and red maples 10 and 18 inches and up, with sampling error and plots, wherever the sampling error is 50% or less.

  • canada-maple-atlas.csv143 rows · 87 columns · Data 2026-10-08

    Canadian census divisions

    Statistics Canada’s 2021 and 2016 farms and taps with their quality grades and suppression symbols (withheld values blank), the PPAQ’s 2026 MRC figures in their own columns, typical dates with the method behind them, and sap days for each of 35 seasons.

  • README.txt · column guide and licence lines for both files
Column guide: county-maple-atlas.csv (82 columns)
Columns of county-maple-atlas.csv
ColumnWhat it holds
fipsFive-digit county FIPS code.
countyCounty name as the atlas page shows it.
stateState name.
uspsTwo-letter state code.
reference_townThe town whose weather stands for the county: the county seat, or the largest town where no seat is listed (GeoNames).
reference_kindHow the reference town was chosen (county seat, largest town, historic county seat in Connecticut).
latReference town latitude, decimal degrees.
lonReference town longitude, decimal degrees.
prism_cell_elev_ftElevation of the PRISM 4-km grid cell holding the reference town, feet.
farms_2022Farms with maple taps, 2022 Census of Agriculture (Table 37). Farm counts are never withheld.
taps_2022Maple taps, 2022 Census of Agriculture. Blank when USDA withheld it.
gallons_2022Gallons of maple syrup, 2022 Census of Agriculture. Blank when USDA withheld it.
withheld_2022true when USDA withheld the county’s 2022 taps or gallons to avoid disclosing individual operations “(D)”.
farms_2017Farms with maple taps, 2017 Census of Agriculture.
taps_2017Maple taps, 2017. Blank when withheld.
gallons_2017Gallons of maple syrup, 2017. Blank when withheld.
withheld_2017true when USDA withheld the county’s 2017 taps or gallons.
typical_open_doyTypical season opening, day of year (Jan 1 = 1). Anchored to the state’s USDA-reported median where one exists.
typical_close_doyTypical season close, day of year.
typical_methodanchored = moved onto the state’s USDA-reported dates; weather = the rule’s own estimate, where USDA reported no dates (Illinois, Iowa, Kentucky).
signal_open_median_doyMedian opening signal of the freeze-thaw rule over the seasons, day of year.
signal_open_q1_doyOpening signal, first quartile (an early year).
signal_open_q3_doyOpening signal, third quartile (a late year).
signal_close_median_doyMedian closing signal, day of year.
signal_close_q1_doyClosing signal, first quartile.
signal_close_q3_doyClosing signal, third quartile.
mean_sap_daysAverage sap days a season, December 1 to May 31. A sap day has an overnight low under 30.2 °F and afternoon high over 37.4 °F.
mean_classic_daysAverage classic sap days a season (low 20–29 °F and high 40–49 °F).
sap_days_1991 … sap_days_2026Sap days in the season ending in that year (December of the year before through May), PRISM daily temperatures.
fia_plot_yearsYears of FIA plot visits behind the county’s estimates (FIA measures a different share of its plots each year).
fia_sugar_maple_10in_treesSugar maples 10 in DBH and up, live, on forest land: FIA’s estimate, in trees (unrounded; the site rounds to two significant figures). Blank unless the sampling error is 50% or less.
fia_sugar_maple_10in_se_pctSampling error of that estimate: one standard error, % of the estimate. Blank with the estimate.
fia_sugar_maple_10in_plotsInventory plots where at least one such tree was tallied. Blank with the estimate.
fia_sugar_maple_18in_treesSugar maples 18 in DBH and up, live, on forest land: FIA’s estimate, in trees (unrounded; the site rounds to two significant figures). Blank unless the sampling error is 50% or less.
fia_sugar_maple_18in_se_pctSampling error of that estimate: one standard error, % of the estimate. Blank with the estimate.
fia_sugar_maple_18in_plotsInventory plots where at least one such tree was tallied. Blank with the estimate.
fia_red_maple_10in_treesRed maples 10 in DBH and up, live, on forest land: FIA’s estimate, in trees (unrounded; the site rounds to two significant figures). Blank unless the sampling error is 50% or less.
fia_red_maple_10in_se_pctSampling error of that estimate: one standard error, % of the estimate. Blank with the estimate.
fia_red_maple_10in_plotsInventory plots where at least one such tree was tallied. Blank with the estimate.
fia_red_maple_18in_treesRed maples 18 in DBH and up, live, on forest land: FIA’s estimate, in trees (unrounded; the site rounds to two significant figures). Blank unless the sampling error is 50% or less.
fia_red_maple_18in_se_pctSampling error of that estimate: one standard error, % of the estimate. Blank with the estimate.
fia_red_maple_18in_plotsInventory plots where at least one such tree was tallied. Blank with the estimate.
fia_flagall_shown, or each blank FIA estimate and why, separated by semicolons: none_tallied (no such tree on any plot; never printed as zero), se_over_50 (too few plots for a reliable county estimate), no_sampling_error. no_fia_record = the county has no FIA estimate at all.
taps_fallback_valueWhere USDA withheld taps_2022 and the county has a page: the 2017 tap count the site shows instead (published and above zero). Blank otherwise. Not comparable with 2022 values; the site never ranks it.
taps_fallback_yearThe census year of taps_fallback_value (2017). Blank when there is none.
gallons_fallback_valueWhere USDA withheld gallons_2022: the 2017 gallons the site shows instead (published and above zero). Blank otherwise.
gallons_fallback_yearThe census year of gallons_fallback_value (2017). Blank when there is none.
Column guide: canada-maple-atlas.csv (87 columns)
Columns of canada-maple-atlas.csv
ColumnWhat it holds
cd_codeStatistics Canada census-division code (SGC 2021).
dguidStatistics Canada DGUID of the division.
divisionOfficial 2021 census-division name, accents kept.
name_2016The 2016 name, where it differs.
cd_typeStatistics Canada division type: MRC (regional county municipality), TÉ (equivalent territory), CTY (county), DIS (district), RM (regional municipality) and others.
provinceProvince.
province_codeQC, ON, NB or NS.
reference_townThe town whose weather stands for the division: its most populous municipality, or its largest population centre (2021 Census).
reference_kindHow the reference town was chosen.
latReference town latitude, decimal degrees (GeoNames).
lonReference town longitude, decimal degrees.
daymet_cell_elev_mElevation of the Daymet 1-km cell holding the reference town, metres.
daymet_cell_elev_ftThe same elevation in feet.
farms_2021Farms reporting maple taps, 2021 Census of Agriculture (Table 32-10-0362-01).
farms_2021_qStatistics Canada’s data-quality grade for farms_2021: A excellent, B very good, C good, D acceptable, E use with caution.
taps_2021Maple taps, 2021. Blank when Statistics Canada did not publish it.
taps_2021_qGrade or symbol for taps_2021: A–E as above, F = too unreliable to be published (value blank).
withheld_2021true when the 2021 tap count was withheld.
farms_2016Farms reporting maple taps, 2016 Census of Agriculture (Table 32-10-0423-01).
farms_2016_qSymbol for farms_2016, when the table printed one.
taps_2016Maple taps, 2016. Blank when withheld.
taps_2016_qSymbol for taps_2016: x = suppressed to meet the confidentiality requirements of the Statistics Act (value blank).
withheld_2016true when the 2016 tap count was withheld.
ppaq_yearEdition of the PPAQ Fiches région the next columns come from (Quebec only). The PPAQ counts enterprises under the joint marketing plan, not census farms: never add these to the census columns.
ppaq_regionPPAQ sugaring region.
ppaq_mrcThe MRC block’s name as the PPAQ prints it.
ppaq_partialtrue when the PPAQ block covers only part of the division.
ppaq_enterprisesMaple enterprises in the block.
ppaq_tapsTaps in the block.
ppaq_taps_roundedtrue when the sheet prints the tap count rounded (to a tenth of a million).
ppaq_production_lbAnnual syrup production, pounds: the PPAQ’s estimate from Groupe AGÉCO’s yields, not a measured harvest.
ppaq_production_roundedtrue when the sheet prints production rounded.
ppaq_value_cadAnnual production value, Canadian dollars.
ppaq_value_roundedtrue when the sheet prints the value rounded.
ppaq_private_forest_pctShare of the block’s taps in private forest, %.
ppaq_public_forest_pctShare of the block’s taps in public forest, %.
ppaq_source_urlThe PPAQ sheet the block comes from.
typical_open_doyTypical season opening, day of year (Jan 1 = 1; 0 and below fall in December).
typical_close_doyTypical season close, day of year.
typical_methodanchored = Quebec: the division’s weather signals moved by the province-wide gap to producers’ reported seasons (Houle et al. 2015); weather = the rule’s own estimate (Ontario, New Brunswick, Nova Scotia).
typical_anchorThe record an anchored date is tied to (houle2015).
signal_open_median_doyMedian opening signal of the freeze-thaw rule, day of year (unadjusted weather signal in all four provinces).
signal_open_q1_doyOpening signal, first quartile.
signal_open_q3_doyOpening signal, third quartile.
signal_close_median_doyMedian closing signal, day of year.
signal_close_q1_doyClosing signal, first quartile.
signal_close_q3_doyClosing signal, third quartile.
mean_sap_daysAverage sap days a season, December 1 to May 31: the same rule as the US (overnight low under 30.2 °F and afternoon high over 37.4 °F; −1 °C and +3 °C).
mean_classic_daysAverage classic sap days a season (low 20–29 °F and high 40–49 °F).
sap_days_1991 … sap_days_2025Sap days in the season ending in that year, Daymet daily temperatures (December 31 is missing in leap years).
taps_fallback_valueWhere Statistics Canada withheld taps_2021: the most recent earlier published tap count the site shows instead (2016, or 2011 where 2016 was withheld too; Table 32-10-0423-01), above zero. Blank otherwise. Statistics Canada cautions that the censuses are not strictly comparable; the site never ranks it.
taps_fallback_yearThe census year of taps_fallback_value (2016 or 2011). Blank when there is none.
ccs_floor_2021Where Statistics Canada withheld taps_2021 (Ontario, New Brunswick, Nova Scotia) but published 2021 tap counts for some of the division’s census consolidated subdivisions (Table 32-10-0362-01): the sum of those published parts, when above 0. A floor (at least this many taps), not the division’s total. Blank otherwise.

The data are yours to reuse with attribution (“LuckySap County Maple Atlas, getluckysap.com”) alongside the sources’ own credits. For the US file:

  • USDA NASS Census of Agriculture: U.S. Government work, public domain.
  • PRISM: PRISM Group, Oregon State University, https://prism.oregonstate.edu, data accessed 2026-10-05.
  • Town coordinates: GeoNames (geonames.org), CC BY 4.0.
  • Tree counts: USDA Forest Service, Forest Inventory and Analysis Program, Wed, 07 Oct 2026 14:17:12 GMT. Forest Inventory FIADB-API web-application Version 2.0 St. Paul, MN: U.S. Department of Agriculture, Forest Service, Northern Research Station. [Available only on internet: http://apps.fs.usda.gov/fiadb-api/evalidator]

The Canadian file carries Statistics Canada’s, the Open Government Licence’s, Daymet’s, GeoNames’ and the PPAQ’s lines, word for word, in its header and in section 16.

8. Canada: production from Statistics Canada

The Canadian pages work at the level of the census division: Quebec's regional county municipalities (MRCs) and equivalent territories, Ontario's counties, districts and regional municipalities, and the counties of New Brunswick and Nova Scotia. Farms reporting maple taps and the number of taps come from Statistics Canada's Census of Agriculture, Table 32-10-0362-01 (2021) and Table 32-10-0423-01 (2016), copied as published. Divisions with at least 5 farms reporting taps in 2021 get a page: 83 in Quebec, 43 in Ontario, 10 in New Brunswick and 7 in Nova Scotia, 143 in all. The parser checks that division farm counts add up to each province's total for both years.

Suppressed values stay blank and read "withheld by Statistics Canada": in 2021 a value too unreliable to publish is marked F, in 2016 a confidential one x. 49 of the 143 page divisions have their 2021 tap count withheld (24 in Quebec, 14 in Ontario, 6 in New Brunswick, 5 in Nova Scotia); farm counts are never suppressed. Where the 2021 count was withheld, the maps, cards and tables show the most recent earlier census that published one above zero, labelled with its year: 2016 for 41 divisions and 2011 (same table, carried only where 2016 was withheld too) for 4; 4 stay withheld. Those figures are never ranked or charted, and never move a map’s colour steps. 2021 estimates graded E carry Statistics Canada's "use with caution" flag. The division tables have no syrup volume, so Canadian pages show no gallons. Statistics Canada's own notes on comparing the two censuses:

  • 2021 Census of Agriculture: changes to the census-farm definition; compare with 2016 with caution (table note 1).
  • 2021: the main-farm-location rule was applied more stringently, so fewer geographic refinements were made; compare CDs with earlier censuses with caution (table note 4).

9. Quebec's second source: the PPAQ fact sheets

The Producteurs et productrices acéricoles du Québec (PPAQ) publish a fact sheet for each sugaring region (12 sheets, 2026 edition) with each MRC's enterprises, taps, annual production in pounds and dollars, and the share of taps in private and public forest. 67 of the 83 Quebec page divisions carry these figures. They count something different from the census (enterprises registered under the joint marketing plan, by where the sugarbush is, five years later), so they sit in their own block, labelled with their year, and are never merged with or substituted for census values, not even where the census tap count is withheld. Figures the sheets print rounded to a tenth of a million are printed the same way. Annual production is PPAQ's estimate from Groupe AGÉCO's yield figures, not a measured harvest. Blocks that combine several MRCs are left unattached, and Francheville carries a block that covers only part of the division, marked as partial. Across the province the sheets add up to 8,663 enterprises and about 57.9 million taps.

10. Canada: 35 seasons of Daymet temperatures

PRISM stops at the border, so the Canadian divisions use Daymet V4 R1 daily tmax/tmin, 1-km grid (ORNL DAAC single-pixel API). Each division's series is the daily maximum and minimum at its reference town, 1990-11-01 to 2025-12-31, which gives 35 seasons (1991–2025). The US atlas runs one season longer because Daymet's 2026 is not published yet. Daymet drops Dec 31 in leap years; 1287 CD-days (143 CDs × 9 seasons whose December falls in a leap year) carry no value and never count as sap days.

No list of division seats is published, so the reference town is chosen by population: the division's most populous municipality in the 2021 Census (128 divisions), or, where that municipality is a rural subdivision or holds no town of its name, the division's largest 2021 population centre (15 divisions). Coordinates come from GeoNames, and every point is checked to lie inside its division's boundary. One town cannot stand for a whole division, least of all a big northern one; the 1-km cells sit between 4 and 439 m (13 to 1,440 ft). Every page names its town.

11. Lining up Daymet's days

As with PRISM, we checked which calendar day a Daymet "day" is: Daymet's cell at 5 Environment Canada airport stations, Jan–Apr 2015–2025, dated a day early, the same day and a day late against each station's own day. The same day wins clearly for both the high and the low, so each sap day pairs that day's minimum (normally reached around dawn) with that day's maximum, the way the live forecast does.

Daymet day alignment against five ECCC airport stations: mean absolute error in °F
VariableDaymet a day earlySame dayDaymet a day late
Daily maximum6.991.27.51
Daily minimum8.232.146.88

12. Daymet and PRISM side by side

To make sure Canadian and US pages can be read together, we ran both grids through the same rules at 6 US reference towns near the border and beyond, 1991–2025. On average they agree: 38.6 sap days a season from PRISM against 38.97 from Daymet (+0.37, or +0.95%), with a correlation of 0.88 across town-seasons. Opening and closing signals differ by a median of 0 days for both (mean absolute difference 1.3 and 1.6 days). Daymet runs about 0.47 °F warmer on highs and 0.39 °F on lows. At a single town the two grids can still differ by several sap days a season, mostly through small offsets around the thresholds and different cell elevations, so a gap of a few days between a Canadian and a US place should not be read as real. This comparison is the only place the atlas mixes the two grids; Canadian pages name Daymet and nothing else.

13. Checking the rules against producers: Quebec and Ontario

Quebec has a published record of season dates; Ontario has producers' reports good for a rough check (below); New Brunswick and Nova Scotia have neither. In Quebec, Houle and co-authors collected weekly syrup production from 121 producers in 11 regions, 1999–2011, and published each region's mean first and last week of production (their Table 1; the per-year data are confidential). We mapped the regions, the producer groups of the joint marketing plan at the time, onto 2021 census divisions and averaged the rule's signals over the same seasons, weighting divisions by their 2021 farms.

Season start and end by region: Houle et al. (2015) means against the rule, 1999–2011
RegionProducersDivisionsStart (Houle)Opening (rule)End (Houle)Closing (rule)
Bas-Saint-Laurent1510Mar 24Apr 1Apr 26Apr 28
Côte-du-Sud123Mar 18Mar 26Apr 25Apr 23
Beauce154Mar 14Mar 23Apr 17Apr 23
Québec1511Mar 12Mar 23Apr 20Apr 22
Mauricie84Mar 18Mar 22Apr 18Apr 20
Lanaudière86Mar 14Mar 19Apr 16Apr 17
Estrie127Mar 14Mar 19Apr 18Apr 20
Laurentides, Outaouais, Abitibi-Témiscamingue1016Mar 14Mar 19Apr 20Apr 18
Centre-du-Québec104Mar 12Mar 19Apr 15Apr 18
Saint-Hyacinthe88Mar 7Mar 14Apr 12Apr 13
Valleyfield87Mar 6Mar 13Apr 11Apr 12
How the rule compares with Houle et al. across the 11 regions (days; bias = rule minus Houle)
ComparisonMean errorBiasCorrelationError once the bias is removed
Opening, farm-weighted divisions6.81+6.810.921.61
Closing, farm-weighted divisions2.09+1.380.891.31
Opening, earliest division in the region3.74+3.550.852.46
Closing, latest division in the region6.7+6.610.83.61

The rule puts the regions in the right order (correlation 0.92 for the opening, 0.89 for the closing) but opens the season about 6.81 days later than Houle's first week of production, and closes it 1.38 days later. Part of the opening gap is what Houle measured: the first week in which any surveyed producer in the region made syrup, not an average producer's date. Taking the earliest division in each region instead cuts the opening bias to 3.55 days.

Ontario: what producers reported

Ontario has no season record like Houle's. What it has is OMAFRA's weekly Maple Production Agriphone, based in part on producers' weekly reports, which said through the 2006–2011 seasons when sap ran and when production ended across the province (we read the Internet Archive's copies), and which in 2007 posted the summary of an OMSPA producer survey. We set the rule's dates beside both. It is a qualitative check: the reports name broad areas, not divisions, and the survey covers one season. Ontario stays a weather estimate, with no anchor and no offset.

Ontario publishes no year-by-year season dates. From 2006 to 2011, though, OMAFRA’s weekly Maple Production Agriphone, based in part on reports from producers across the province, gives our weather estimate a rough check.

Producer survey, 2007

119 producers with 21% of Ontario’s taps (Stats Canada 2006) answered the OMSPA survey.

Producers’ averages and our weather rule’s dates, Ontario, 2007
PointProducers (avg)Our rule
First boil / openingMar 15Mar 14
Last boil / closingApr 13Apr 10

First boils ran Feb 27 to Apr 5, last boils Mar 20 to Apr 27. Our rule: the average of Ontario’s 43 atlas divisions, weighted by farms reporting taps in the 2016 census.

Within 3 days of the producers’ averages. Encouraging, but it is one season: a check, not a calibration.

OMAFRA Maple Production Agriphone, July 12, 2007 (archived)

Where they differ

In the south they agree: the rule closes Essex (Windsor) on Mar 27, 2010, the week southern producers reported ending.

In the north the rule’s closings ran late in 2009 and 2011: it closes Algoma (Sault Ste. Marie) on Apr 29, 2009 and May 5, 2011, after the reports here say production had ended.

Ontario’s dates stay a weather estimate; read the northern closing dates as leaning late.

From the agriphone (archived)

  1. 2009

    “Sap collection and maple syrup production ended over the April 10 to 12 weekend in most northern and eastern late-season areas of Ontario.”
    OMAFRA Maple Production Agriphone, April 15, 2009 (archived)
  2. 2010

    “… a small sap flow did occur last weekend February 20 and 21 in early southwestern regions of the province …”
    OMAFRA Maple Production Agriphone, March 1, 2010 (archived)
  3. 2010

    “Many maple syrup producers in southern areas of the province ended sap collection for 2010 by the end of the week, March 27.”
    OMAFRA Maple Production Agriphone, March 29, 2010 (archived)
  4. 2011

    “Sap collection ended in Ontario by the Easter weekend to bring the 2011 maple syrup processing season to a close.”
    OMAFRA Maple Production Agriphone, April 26, 2011 (archived)

14. Anchoring Quebec, and only Quebec

US county dates are anchored to USDA's reported averages. Quebec's are anchored the same way to the producers' record: each division's median weather signals, moved by the province-wide gap to Houle's regions (6.8 days earlier for the opening, 1.4 for the closing). That keeps the differences between divisions, puts the province on the producers' calendar and keeps the Quebec when-to-tap page and the atlas in agreement. No published per-year season record was found for Ontario, New Brunswick or Nova Scotia; their dates are the weather signal only. In Ontario, producers' reports in OMAFRA's Maple Production Agriphone and a 2007 producer survey give a rough check, not an anchor; New Brunswick and Nova Scotia have nothing local to check against. Their pages say "weather estimate". One consequence is a step at the Quebec border: a Quebec division's dates can sit about a week earlier than its Ontario neighbor's for reasons of method, not climate. The year-by-year bars and spreads on every page are the unadjusted weather signal in all four provinces.

15. Limits worth knowing, Canada

  • One 1-km Daymet cell per division, at its reference town. Large or hilly divisions (the northern Ontario districts, the big Quebec MRCs) have sugarbushes far from, and often higher than, that point.
  • Daymet is a model interpolated from weather stations; where it matches a station it partly does so by construction.
  • 35 seasons, one fewer than the US atlas, and December counts in leap years can be one day short.
  • The rules were calibrated on US data. Quebec checks them against 11 regional means from 121 producers; Ontario only loosely, against producers' reports from 2006–2011; New Brunswick and Nova Scotia have nothing local to check against.
  • In 13 division-seasons the rule finds no season at all (1997 and 2012): warmth piled up before any freeze-thaw week. Producers still made syrup those years; the rule can't place it.
  • 49 of 143 page divisions have 2021 taps withheld, and 2016 to 2021 comparisons carry Statistics Canada's caution.
  • PPAQ figures are five years newer than the census, count a different thing, are often rounded, and their production is an estimate.
  • Reference towns are chosen by population, not by seat; some are amalgamated municipalities rather than towns.
  • Map boundaries are simplified twice and clipped at the top edge, so the far north of the largest divisions is cut off.

16. Licences and attribution, Canada

  • Adapted from Statistics Canada, Census of Agriculture 2021, 2016 and 2011 (Tables 32-10-0362-01 and 32-10-0423-01), 2021 Census population and dwelling counts (Tables 98-10-0002-01 and 98-10-0011-01) and 2021 Census cartographic boundary files. This does not constitute an endorsement by Statistics Canada of this product.
  • Contains information licensed under the Open Government Licence – Canada.
  • Thornton, M.M., R. Shrestha, Y. Wei, P.E. Thornton, S-C. Kao, and B.E. Wilson. 2022. Daymet: Daily Surface Weather Data on a 1-km Grid for North America, Version 4 R1. ORNL DAAC, Oak Ridge, Tennessee, USA. https://doi.org/10.3334/ORNLDAAC/2129
  • Place coordinates: GeoNames (geonames.org), CC BY 4.0.
  • Québec MRC figures: Producteurs et productrices acéricoles du Québec, Fiches région 2026 (production estimates from Groupe AGÉCO).

The atlas's gridded temperatures (PRISM in the US, Daymet in Canada) are short records, and PRISM's authors advise against using it for very long-term trends, so the trend study uses weather-station records instead, scored with the same sap-day rule and season signals as the atlas (one shared implementation for both countries). A season runs December 1 to May 31 and counts only if at least 95% of its days have both a maximum and a minimum; a station needs at least 60 complete seasons since 1951. Each station's trend is the Theil–Sen slope in days per decade, tested with Mann–Kendall (p < 0.05). "Then" is 1951–1980; "now" is each country's latest 30 seasons.

United States: 108 NOAA ThreadEx long-term records, 1951–2026, via ACIS. ThreadEx stitches each city's successive sites (downtown office, then airport) into one daily record and leaves the observations as they were. Canada: 85 stations from Environment and Climate Change Canada's Adjusted and Homogenized Canadian Climate Data, the third-generation homogenized daily maximum and minimum temperatures (update to 2023), 1951–2023: 36 in Quebec, 35 in Ontario, 7 in New Brunswick, 6 in Nova Scotia, 1 in Prince Edward Island. The open.canada.ca record still describes the second generation; the files it links to are the third (Vincent, Hartwell and Wang 2020), and those are what we use. Contains information licensed under the Open Government Licence – Canada.

Which Canadian stations: of 325 homogenized stations in Quebec, Ontario, New Brunswick, Nova Scotia and Prince Edward Island, 229 lie in a 2021 census division with at least 5 farms reporting maple taps in the 2021 Census of Agriculture (the atlas page rule) (111 QC, 86 ON, 19 NB, 11 NS, 2 PE); 4 coastal stations whose rounded coordinates fall just offshore take the nearest division within 5 km. 85 of them have 60 complete seasons. The rule keeps the far north (Nord-du-Québec, Kenora, Thunder Bay, the Côte-Nord) out, and Montréal in.

Pairing the night with the day

Environment Canada's daily maximum and minimum are not always calendar-day values. At stations reporting around the clock, since July 1, 1961 the climatological day ends at 06 UTC; before that, the maximum's day ended at 12 UTC and the minimum's at 00 UTC of the following day. At stations read morning and evening, the maximum's day ends at the next morning's reading and the minimum's at the current day's afternoon reading. In every case the minimum dated D is the night that ends on the morning of D and the maximum dated D is D's afternoon, so each sap day pairs that day's minimum with that day's maximum, exactly as the US station records are paired. The homogenized records also move the post-1961 minimum at the main stations back onto the earlier window, where the 06 UTC day had caught the following night.

We checked it the way we checked PRISM: five Canadian stations against the nearest US ThreadEx record across the border, December–May, Canadian value dated a day early, the same day and a day late. The same day wins for both variables in both periods at all five pairs.

Canadian homogenized values against nearby US ThreadEx records: mean absolute difference in °F, pooled over five cross-border pairs
Variable, periodCanada a day earlySame dayCanada a day late
Maximum, before July 19616.513.327.42
Minimum, before July 19617.563.436.39
Maximum, July 1961 on6.762.547.27
Minimum, July 1961 on7.743.386.79

Pairs: Sault Ste. Marie, ON ↔ Sault Ste. Marie, MI; Windsor, ON ↔ Detroit, MI; Welland-Pelham, ON ↔ Buffalo, NY; Montréal, QC ↔ Plattsburgh, NY; Aroostook, NB ↔ Caribou, ME. The stations sit 12 to 95 km apart, so the same-day differences are mostly real weather differences.

Homogenized or raw? Measured

The US records are raw and threaded; the Canadian ones are homogenized. To measure what that does, the Canadian stations were rebuilt from raw ECCC daily data (the api.weather.gc.ca climate-daily archive), threaded by AHCCD's own list of joined stations and dates, so the only difference is the adjustments. 67 of 85 stations have 60 complete raw seasons over the same 1951–2023 window (the homogenized files hold more days than the archive under the listed station IDs, so the rest fall short).

The same 67 Canadian stations, 1951–2023: homogenized vs raw threaded (slopes in days per decade)
SignalMedian slope, homogenizedMedian slope, rawMedian change per stationSignificant only homogenizedSignificant only raw
Closing signal−1.1−1.00.1101
Opening signal−1.0−0.80.1112
Sap days0.00.00.31010
Five stations, closing / opening / sap-day slopes (days per decade; * = p < 0.05)
StationMinima changedClosing: homog. / raw / raw to 2026Opening: homog. / rawSap days: homog. / raw
Montréal (airport), QC19.6%−0.9* / −0.9* / −0.9*−1.3* / −1.3*0.0 / +0.5
Québec (Jean Lesage airport), QC19.4%−0.7 / −0.5 / −0.70.0 / 0.00.0 / +0.6
Ottawa (research farm), ON77.3%−1.2* / −1.1* / −1.2*−1.6* / −1.5*+0.5 / +0.4
Fredericton (research farm), NB24.8%−0.8 / −0.5 / −0.7−0.7 / −0.6−0.9* / −0.3
Lennoxville, QC0%−0.9* / −0.9* / −1.0*−0.8 / −0.8*−0.7 / −0.8

Across the network the adjustments change the December–May minimum at 59 of 85 stations (22.1% of days at the median station; the station-average shift runs from −4.92 to +2.32 °F) and the maximum at 32. The main stations carry the 1961 observing-time fix: Montréal (airport) 19.6%, Québec (Jean Lesage airport) 19.4%, Fredericton (research farm) 24.8% of December–May minima differ from the raw values, mostly raised, which is what removes a spurious rise in sap days from their raw records. The median slopes hardly move; what moves is significance: 38 of the 67 stations close significantly earlier with homogenized data, 29 with raw. Raw data also run on to 2026: over 1951–2026, 38 of 71 stations close significantly earlier (median −1.0).

The published Canadian rows are the homogenized ones. ECCC built them for trend analysis; they remove steps that are documented and measured (station joins checked against parallel observations, the 1961 change in the minimum, relocations), and those steps act directly on the freeze-and-thaw thresholds the sap-day rule counts. Using raw data to match the US treatment would knowingly keep errors that ECCC has already corrected. The price is a record that ends with the 2023 season and a cross-border comparison between two kinds of record, which the trend page says plainly.

Limits of the trend study

  • Two treatments: US rows raw and threaded (NOAA ThreadEx), Canadian rows homogenized (ECCC AHCCD). The US records are not corrected for site moves; their trends could be muted or exaggerated where a city's thermometer moved.
  • Different periods: 1951–2026 in the US, 1951–2023 in Canada, so "now" means 1997–2026 on one side and 1994–2023 on the other.
  • Different networks: ThreadEx follows major cities and their airports; the Canadian set is mostly climate stations in small towns and farm country, with research farms and airports among them.
  • Resolution: US values are whole °F, Canadian values tenths of a °C (converted at 0.01 °F, so −1 °C is exactly 30.2 °F). A Canadian high of 37.6 °F counts as a thaw; a US high needs 38 °F. Counts shift slightly; trends do not depend on it.
  • The Mann–Kendall test here ignores serial correlation, and the season signals are a weather rule, not producers’ dates.
  • A season with gaps (up to 5% of days) is filled by carrying the previous day forward before scoring, as in the US.

Sources: NOAA ThreadEx; ECCC AHCCD daily temperature (open.canada.ca) and its files; Vincent, Hartwell and Wang (2020), Atmosphere-Ocean; ECCC technical documentation for daily data (climatological day); raw daily data from api.weather.gc.ca.

Sources · 15

  1. 1.PRISM daily temperature grids (AN81d, 4 km), accessed 2026-10-05 · PRISM Group, Oregon State University, https://prism.oregonstate.edu
  2. 2.Applied Climate Information System (ACIS) web services · NOAA Regional Climate Centers
  3. 3.USDA NASS maple syrup reports — approximate average season opened and closed dates, 2011–2022 · USDA National Agricultural Statistics Service (mirrored by the University of Vermont)
  4. 4.Sap flow in sugar maple: timing and weather drivers (Frontiers in Agronomy, 2025) · de Lima Santos et al.
  5. 5.2024 U.S. Gazetteer Files — counties · U.S. Census Bureau
  6. 6.GeoNames gazetteer (cities500), CC BY 4.0 · GeoNames
  7. 7.Maple taps, Census of Agriculture, 2021 (Table 32-10-0362-01) · Statistics Canada
  8. 8.Maple taps, Census of Agriculture, 2011 and 2016 (Table 32-10-0423-01) · Statistics Canada
Show 7 more
  1. 9.Population and dwelling counts, 2021 Census (Tables 98-10-0002-01 and 98-10-0011-01) · Statistics Canada
  2. 10.2021 Census cartographic boundary files (census divisions, population centres) · Statistics Canada
  3. 11.Fiches région 2026 (12 regional sheets) · Producteurs et productrices acéricoles du Québec
  4. 12.Daymet: Daily Surface Weather Data on a 1-km Grid for North America, Version 4 R1 · ORNL DAAC
  5. 13.Impacts of climate change on the timing of the production season of maple syrup in eastern Canada (PLOS ONE 10(12): e0144844, 2015) · Houle et al.
  6. 14.Maple Production Agriphone, weekly reports 2006–2011, with the 2007 OMSPA producer survey summary (archived) · Ontario Ministry of Agriculture, Food and Rural Affairs (OMAFRA)
  7. 15.Forest Inventory and Analysis: live trees on forest land by species and diameter (FIADB 1.9.5, EVALIDator / FIADB-API), accessed 2026-10-07 · USDA Forest Service, Forest Inventory and Analysis Program

Tree counts

How many tappable maples: the Forest Service inventory

County, state and atlas-wide maple counts come from the USDA Forest Service’s Forest Inventory and Analysis program (FIA). FIA keeps a permanent grid of field plots across every state and re-measures a share of them each year. From those plots it publishes estimates for any area, each with a sampling error. We do not touch the plot data: we ask FIA’s own estimation engine (the FIADB-API fullreport endpoint that powers EVALIDator) for each number and copy the estimate, its sampling error and its plot count.

The query

Attribute "Number of live trees (at least 1 inch d.b.h./d.r.c.), in trees, on forest land (attribute 4)", grouped by county and species, filtered to live trees of the species below with a trunk diameter at breast height (4.5 ft) of at least 10.0 inches, and separately 18.0 inches: SQL filter TREE.SPCD IN (318,316,314,317,313) AND TREE.DIA >= 10.0. Ten inches is the most common extension minimum for one tap (UMaine, Cornell; the 2006 Producers Manual’s traditional table) and 18 inches is where most guidelines allow a second tap on a healthy tree; both match our taps-per-tree guide and calculator. The 18-inch trees are a subset of the 10-inch ones; never add the two.

FIA species codes used
SpeciesScientific nameFIA codeWhere it is used
Sugar mapleAcer saccharum318Every page
Red mapleAcer rubrum316Every page
Black mapleAcer nigrum314Every page
Silver mapleAcer saccharinum317Every page
BoxelderAcer negundo313Every page
Florida mapleAcer barbatum311Outside-the-belt state pages
Bigleaf mapleAcer macrophyllum312Outside-the-belt state pages
Norway mapleAcer platanoides320Outside-the-belt state pages
Chalk mapleAcer leucoderme323Outside-the-belt state pages

Bigtooth and Rocky Mountain maples are left out: FIA treats them as western “woodland” species and measures them at the root collar, so a 10-inch threshold would not mean the same thing. County and state figures are separate API requests; for the 17 states the county request’s state subtotals and a separate statewide request agree exactly in all 170 species-and-size pairs. The 17-state and national figures are single estimates the API computes from the combined evaluations, with their own sampling errors, not sums we made.

The evaluations

Each state uses its most recent statewide “current volume” evaluation. Because FIA visits a different share of its plots every year, an evaluation pools the latest full cycle of visits: 5 to 9 years in the atlas states. A county figure is not a snapshot of one year; it describes the forest over that span.

FIA evaluations used for the 17 atlas states
StateEvaluationPlots measuredPlotsForested plotsCounties with a reliable sugar maple figure
Connecticut092025 (EVALID 092501)2019–20254883116 of 8
Illinois172025 (EVALID 172501)2019–20255,9051,0396 of 100
Indiana182025 (EVALID 182501)2019–20253,75193919 of 90
Iowa192025 (EVALID 192501)2019–20255,9626152 of 92
Kentucky212024 (EVALID 212401)2016–20244,2172,37455 of 120
Maine232025 (EVALID 232501)2021–20253,4913,10910 of 16
Massachusetts252025 (EVALID 252501)2019–20257945335 of 13
Michigan262025 (EVALID 262501)2019–20256,4594,13434 of 83
Minnesota272025 (EVALID 272501)2021–202517,3946,14621 of 83
Missouri292025 (EVALID 292501)2019–20257,2533,12215 of 114
New Hampshire332025 (EVALID 332501)2019–20251,1531,0409 of 10
New York362025 (EVALID 362501)2019–20254,5032,95549 of 60
Ohio392025 (EVALID 392501)2019–20254,0581,41429 of 87
Pennsylvania422025 (EVALID 422501)2019–20254,4352,90232 of 66
Vermont502025 (EVALID 502501)2019–20251,11791713 of 14
West Virginia542025 (EVALID 542501)2019–20252,3582,01639 of 55
Wisconsin552025 (EVALID 552501)2019–202511,5526,21739 of 72

The 25 US states with an "outside the maple belt" page use their own latest statewide evaluations (plots measured between 2013 and 2025). Two need a note: Texas uses the statewide 2023 evaluation rather than the newer East Texas one, so the Edwards Plateau is included, and Alaska's current evaluation covers coastal Alaska only.

When a number is shown

  • A figure appears only when its sampling error is 50% or less. FIA's sampling error is one standard error as a percent of the estimate: the true count falls within that range about two times in three, and within twice the range about 19 times in 20. At 50% the likely range already runs from half the estimate to one and a half times it; past that a county figure says little, so the page says "too few inventory plots for a reliable county estimate" and shows the state figure instead.
  • A county where no plot tallied a tree of that species and size has no estimate at all. That is not the same as having none, and the page never prints it as zero.
  • Figures are rounded to two significant figures, or three for counts of a hundred million or more. The sampling error, the one-standard-error range and the number of plots that tallied the species sit in a quieter line beside each figure.
  • “Plots” means plots where at least one tree of that species and size was tallied; “forested plots” are plots with any forest land. Of the 1,083 counties FIA reports in the 17 states, 383 have a sugar maple figure (10 inches and up) that passes the rule, including 228 of the 385 counties with an atlas page.

For scale: about 908 million sugar maples 10 inches and up on forest land in the 17 atlas states (±1.3%, 84,890 plots) and about 959 million in all 50 states (±1.3%).

Checked against FIA’s own reports

To check the pipeline end to end, we asked the same API for the evaluations behind 10 published FIA state reports and compared its answers with the numbers printed in them. 22 of 22 match: whole-state tree totals to the exact tree, species shares to the printed tenth of a percent, and counts to the printed million.

Pipeline check against published FIA figures
Published figureAs printedSame evaluation via the APIDifference
Vermont, all live trees 1 in and upForests of Vermont, 2023 (FIA Annual Snapshot, FS-582), p. 1: “Total number of live trees (≥1.0 in. diameter) on forest land” and “Top species by count”3,232,385,314 (±2.77%)3,232,385,314exact
Vermont, sugar maple share of all live trees13.4%13.36%-0.04 points
Vermont, all live trees 1 in and upForests of Vermont, 2024 (FIA Annual Snapshot, FS-635), p. 1: “Total number of live trees (≥1.0 in. diameter) on forest land” and “Top species by count”3,237,616,927 (±2.91%)3,237,616,927exact
Vermont, sugar maple share13.1%13.05%-0.05 points
New York, all live trees 1 in and upForests of New York, 2023 (FIA Annual Snapshot, FS-595), p. 1: “Total number of live trees (≥1.0 in. diameter) on forest land” and “Top species by count”11,504,833,605 (±1.41%)11,504,833,605exact
New York, red maple share11.6%11.64%+0.05 points
New York, sugar maple share9.2%9.18%-0.02 points
Michigan, all live trees 1 in and upForests of Michigan, 2023 (FIA Annual Snapshot, FS-593), p. 1: “Total number of live trees (≥1.0 in. diameter) on forest land” and “Top species by count”13,962,507,492 (±1.41%)13,962,507,492exact
Michigan, red maple share13.7%13.74%+0.04 points
Michigan, sugar maple share9.0%9.00%0.00 points
Ohio, all live trees 1 in and upForests of Ohio, 2023 (FIA Annual Snapshot, FS-630), p. 1: “Total number of live trees (≥1.0 in. diameter) on forest land” and “Top species by count”3,677,608,751 (±2.50%)3,677,608,751exact
Ohio, sugar maple share14.4%14.40%0.00 points
Ohio, red maple share11.5%11.45%-0.05 points
Pennsylvania, all live trees 1 in and upForests of Pennsylvania, 2023 (FIA Annual Snapshot, FS-626), p. 1: “Total number of live trees (≥1.0 in. diameter) on forest land” and “Top species by count”7,751,239,121 (±1.77%)7,751,239,121exact
Pennsylvania, red maple share17.2%17.22%+0.02 points
Pennsylvania, red maples 5 in and upPennsylvania Forests 2019: Summary Report (NRS-RB-131), p. 10587 million586,951,931-0.01%
Pennsylvania, sugar maples 5 in and up182 million181,520,560-0.26%
Vermont, sugar maples 5 in and upForests of Vermont, 2016 (FS-119), Table 2, p. 3, “Trees (millions)”, at least 5-inch trees156 million155,852,579-0.09%
Vermont, red maples 5 in and up108 million107,769,687-0.21%
New Hampshire, red maples 5 in and upForests of New Hampshire, 2016 (FS-124), Table 2, p. 3, at least 5-inch trees171 million170,804,659-0.11%
New Hampshire, sugar maples 5 in and up65 million64,637,246-0.56%
New York, red maples 5 in and upNew York Forests 2017 (NRS-RB-121), p. 37573 million573,181,221+0.03%

Limits worth knowing

  • FIA counts trees on forest land only. Yard, street and farm trees are not counted, so true numbers are higher. Sugarbushes in pastures, fencerows and yards are missing; so are city street trees.
  • FIA designs its sample for states and large regions. A county estimate rests on the plots that happen to fall in it: a few dozen in a small or lightly forested county, which is why so many fail the 50% rule and why county ranks whose ranges overlap could swap.
  • Trunk size is only the first test of a tappable tree. These counts include trees on steep, roadless, reserved or unwilling owners’ land, unhealthy trees and trees in stands no one will ever tube. They are an upper bound on what is practical to tap, not a count of available taps.
  • Black maple and sugar maple are hard to tell apart and hybridize, so the split between them is less certain than their total.
  • Norway maple is counted only where it has escaped into forest land; planted Norway maples along streets are not in the figure.
  • Estimates change as FIA adds each year’s plots and retires the oldest; these were read on 2026-10-07 from database version FIADB_1.9.5.00. Re-running the pipeline will move them slightly.

Canada

We looked for official estimates of how many sugar or red maples grow in Quebec, Ontario, New Brunswick and Nova Scotia. No government publishes a count of trees by trunk size for these species. What exists is standing wood volume: Canada’s National Forest Inventory reports maples only as one genus and only by ecozone; Ontario (Forest Resources of Ontario 2021), New Brunswick (2015 inventory report) and Nova Scotia (State of the Forest 2016) publish provincial sugar and red maple volumes without sampling errors; Quebec publishes them per public-forest management unit, plus a preliminary map-based count of potential taps for two regions. None of these measures what the US figures measure, so the Canadian pages carry no tree counts rather than a number that would not compare.

Citation

USDA Forest Service, Forest Inventory and Analysis Program, Wed, 07 Oct 2026 14:17:12 GMT. Forest Inventory FIADB-API web-application Version 2.0 St. Paul, MN: U.S. Department of Agriculture, Forest Service, Northern Research Station. [Available only on internet: http://apps.fs.usda.gov/fiadb-api/evalidator]

LuckySap · free on iPhone & Android

Log every run. Know your season.

LuckySap is the maple syrup tracker for iPhone and Android: a sap-flow outlook for your trees, a quick log for every collection and boil, and season stats you can compare year to year.

Download on the App StoreGet it on Google Play