Global grape harvest time series: the 1961-2024 vineyard ledger, delivered as rows

One record owns this query, and Datadory delivers it as plain typed rows: FAOSTAT - Crops and Livestock Products, the UN Food and Agriculture Organization's production account, domain QCL. The full domain packs 4,209,110 observations spanning 312 commodity items - grapes beside wheat, maize, rice, almonds in shell, tobacco leaf and natural rubber - across 244 countries and territories for every year 1961 through 2024. It scores 10 out of 10 on Datadory's field-documentation rubric against a catalog mean of 7.81 across 1,744 records: one of just 145 datasets to reach the ceiling.

Grapes carry the three elements that make a harvest series usable rather than decorative. Area harvested counts the hectares of vines picked in the reference year (element code 5312, measured in ha). Production counts the tonnes harvested (5510, t). Yield is the derived ratio in kilograms per hectare (5412, kg/ha) - computable from the other two, but pre-computed here so nobody rebuilds it by hand. Filter the item column to Grapes and all three series arrive in one pass; that is precisely what normalized long-format rows buy you.

Sizing matters in this slice: Distillers & Vintners holds zero primary datasets among the 1,744 records Datadory catalogs, with six related records pooled in from adjacent industries. The two FAO records are the ones carrying any vineyard-supply signal, which is why they lead.

Coverage runs annually from 1961 through the 2024 reference year in the December 2025 build reviewed - up to 64 observations per country-element cell, on stable codes throughout. A 1961 reading and a 2024 reading land in the same table under the same definitions, which is why this panel supports trend-and-cycle work instead of just latest-year rankings: sixty-four annual yields are enough runway to separate structural drift from weather shocks, enough to test whether supply signals led prices, and enough to feed vintage-cycle features into a demand model.

Fourteen flat columns, no nesting, one row per country-commodity-element-year. Captured observations for the domain's most-cited test case, Italian almonds in shell, exactly as they arrive:

# captured observations -- December 2025 build
Area    : Italy                 Item : Almonds, in shell      Year : 2022
Element : Area harvested        Unit : ha                     Value: 53890.000000
Flag    : A  (official figure)

Area    : Italy                 Item : Almonds, in shell      Year : 2022
Element : Production            Unit : t                      Value: 74590.000000
Flag    : A  (official figure)

Area    : Italy                 Item : Almonds, in shell      Year : 2022
Element : Yield                 Unit : kg/ha                  Value: 1384.100000
Flag    : A  (official figure)

# the grape shelf -- measure slots fill when your sample is cut
Grapes          Area harvested | Production | Yield       ha | t | kg/ha

Swap the item for Grapes and the structure is identical - that is the point of a shared spine. Three columns repay attention before any model touches the panel. Value prints as text to six decimal places, so cast deliberately or a naive integer read corrupts tonnages. Area Code (M49) carries a leading apostrophe that needs stripping before joins to external country tables. And Flag rides along on every row, so an official figure and an agency's estimate separate before they contaminate an average rather than after.

A grapes-only extract collapses cleanly because livestock-specific elements such as Milk Animals, Laying and Producing Animals/Slaughtered simply do not apply to a vine crop: three series per country-year, nothing else. Four lookup dimensions - element, flag, item codes with UN CPC mappings, and area codes with M49 geography - travel beside the main table, so joins learned once run everywhere in the family.

Every value is graded, not just the dataset. A marks an official figure reported by the country, E an estimated value, I a value imputed by a receiving agency, X a figure contributed by an external organization, and M a missing value where the data structurally cannot exist. That column is the difference between a defensible vintage comparison and a smooth lie.

The working rules: filter on A whenever a client-facing claim is country-specific; treat E, I and X stretches as modeled inputs rather than reported history when building trend features; and watch X rows when lining the series up against other international sources, because both sides can be restating the same underlying report. Coverage interacts with grading too - smaller producers often carry official figures only in recent years, so an unfiltered panel quietly mixes reported history with imputed backfill and calls the result continuity.

The cheapest high-value habit is publishing two views from one extract: a strict A-only view for anything a client will read aloud, and an all-flags view for modeling. When the two disagree materially for a large producer, that gap is not noise to smooth over - it is a finding worth checking against the next annual release before it reaches a deck.

Three structural choices decide whether a grape panel survives contact with production use.

First, the aggregate trap. World totals and FAO regional groupings sit in the same Area column as sovereign states. Convenient for quick world sums; hazardous for everything else, because summing labels naively double-counts China several times over before the first chart renders. Tag aggregate rows at ingest and drop them deliberately - especially before computing any country's share of a total.

Second, key extracts on the M49 code rather than display names whenever external joins are planned, because M49 is the convention other statistical agencies publish against. Identity fixed by code survives commodity renames and area re-basing intact; identity fixed by name string does not.

Third, keep the long format until the last step. Pivoting the three elements into area_ha, production_t and yield_kg_ha columns yourself keeps the flag aligned with every value; reaching for a pre-pivoted variant forfeits that guarantee. And mind the ceiling: the grain here is country x commodity x element x year, with nothing below the nation. Anyone promising province-level tonnage from these rows is selling a different dataset.

Volumes alone do not answer a sourcing, pricing or investment question. Four companions complete the arc, all on the same shelf.

OIV Statistics Database - World & Country Statistics is the book of record for vitiviniculture itself: vineyard area under vines, wine production and consumption in hectolitres, table-grape versus dried-grape volumes, export volumes and values, and re-exports, harmonized with the FAO's Statistics Division so a hectolitre means the same thing everywhere. Depth comes in layers - online yearbooks 1999 through 2014, thematic briefs running 2014 to 2025, and outlook reports referencing the 2024/2025 campaigns, with preliminary current-season estimates arriving each October and the full prior-year picture each April.

FAOSTAT - Detailed Trade Matrix answers where the fruit goes: 52,410,630 bilateral flows across roughly 572 items including fresh grape, dried grape and must/wine lines, traded between about 232 reporter and 255 partner areas annually from 1986 through 2024, with the same per-row provenance flag. A shortfall in one reporter's harvest can be traced to offsetting corridors rather than argued about.

FAOSTAT Food and Agriculture Data supplies the fermentation-input basket - sugar cane (item 156), sugar beet (157), oranges (490) - beside producer price indexes monthly since January 2010 and balance-sheet series that split into legacy and current-methodology editions around 2010, a seam any long panel must respect. And Open Food Facts - Open Database & API lands the story on the shelf: roughly 4.7 million barcoded products with brand, ingredients and category facets, the closest proxy to wine-and-spirits assortment tracking this slice carries. The table below lines the five records up.

Market researchers & consultants size vineyard supply markets from hectares and tonnes measured identically in every reporting economy - one comparable panel beats stitching national statistics bureaus together by hand. Investors & quants take the sixty-four-year span as a regressable supply factor, usually beside the OIV consumption columns and the bilateral trade flows on matching grains. Data scientists get the rare cross-country panel that joins clean on M49 codes, item codes and a plain year column, with a quality grade attached to every observation instead of buried in a methodology PDF. Sales & growth teams in the beverage equipment and input trades read harvested area as derived demand - hectares are what put planters, harvesters and trellis systems in fields. Journalists & academics quote it because each figure traces to a documented compilation method and a stated provenance grade, which is more than most harvest numbers circulating in decks can say.

What makes the panel unusually cooperative downstream is that the hard parts are handled before delivery: elements pivoted wide where a model wants width and kept long where a pipeline wants append rows; units held beside amounts so hectares, tonnes and kilograms per hectare never mix; item and area codes resolved to names with the numeric originals retained as join keys; provenance flags preserved as a first-class field; aggregates tagged separately from sovereign states. When the next annual build lands, it passes through the same normalization, so anything prototyped on the sample survives into production unchanged.

Start with a sample: tell us the markets and the horizon and the extract comes back shaped like your question.

This page is one thread of a broader guide. Start with the distillers-vintners data guide for how the whole six-record pool fits together, then best distillers-vintners datasets lines the quality scores up side by side and the distillers-vintners data hub browses the same catalog as records.