French wine dataset to mapping the expected harvest value by county

This database is built from open data as described in the paper entitled  ‘French wine: Combination of multiple open data sources to mapping the expected harvest value’ (2024).

Related Research Paper

This dataset supports the research article: French wine: Combination of multiple open data sources to mapping the expected harvest value. The full paper is available via HAL: https://univ-lemans.hal.science/hal-04627672.

Methodological Overview

This study addresses the challenge of reconstructing macro-scale economic values for French vineyards from incomplete public information. Due to privacy regulations (GDPR), granular production data is often protected, leaving researchers with marginal totals (sums by region and sums by appellation) but lacking the cross-tabulated matrix required for precise risk assessment.

We treat this gap as an inverse problem solvable through mathematical programming. The core methodology employs a constrained optimization algorithm (implemented in SAS PROC OPTMODEL) to estimate the latent surface area s_ac (hectares) for each appellation a in each county c. The model respects the following constraints:

• Marginal Consistency: Sum of estimated areas must not exceed known totals by appellation (s_a) and by county (s_c).
• Geographical Authorization: Areas are forced to zero where an appellation is not authorized in a specific county (data provided by INAO).
• Objective Function: Maximization of AOP priority to reflect economic weight.

Once the surface map is reconstructed, expected harvest values are computed using Olympic average yields (to mitigate extreme weather years) and official insurance scale prices. This results in a high-resolution map of economic exposure, distinct from physical crop load mapping used in Precision Viticulture.

Data Sources

All data utilized are open-source or publicly accessible government records. Below is the list of primary sources consulted (last viewed June 26, 2024):

Official Statistics & Registries

• Vineyard Registry (CVI): Customs service statistics on area and production by appellation and county.
  https://www.douane.gouv.fr/la-douane/opendata?f[0]=categorie_opendata_facet:467
• National Institute of Origin and Quality (INAO): Catalog of counties authorized for AOP/PGI production and quality specifications.
  https://www.inao.gouv.fr
  https://www.data.gouv.fr/fr/datasets/?q=inao
• Agreste / Ministry of Agriculture: National agricultural accounts and agricultural regions definitions.
  https://agreste.agriculture.gouv.fr/agreste-web/methodon/Z.1/!searchurl/listeTypeMethodon/
  https://agreste.agriculture.gouv.fr/agreste-web/download/publication/publie/Dos2203/2Pages%20de%20Dossier2022-3_CCAN_ChapitreII.pdf
• Insurance Specifications: Official scale values for crop insurance premium calculation (Ministry of Agriculture).
  https://info.agriculture.gouv.fr/boagri/document_administratif-4b9ef75e-29a7-449d-9e40-7e5253bfd642/telechargement

Supplementary Data

• Champagne Area Distribution: Detailed breakdown of Champagne vineyard surfaces by county.
  https://maisons-champagne.com/fr/appellation/aire-geographique/
• RICA (FADN): Farm Accountancy Data Network (referenced for context on farm structures).
  https://www.casd.eu/source/reseau-dinformation-comptable-agricole/?tab=16

Simplified SAS code

DATA LINAO;
INPUT INSEE_COM CVI AUTHORIZED SINIT;
DATALINES;
01001 3B011 0.33 0
01001 3B011M 0.33 0
01001 3B012 0.33 0
01001 3B012M 0.33 0
01001 3B013 0.33 0
....
;;;;
RUN;

DATA LA;
INPUT  CVI S;
DATALINES;
1B001D 78.714523339
1B001M 3064.7940186
1B001S 9474.2135637
1B002D 3
1B002S 12.987045088
....
;;;;
RUN;

DATA LC;
INPUT INSEE_COM SDC;
DATALINES;
01001 0.2
01002 5.0859
01003 0.2
01004 1.0499999999
01005 0.2
....
;;;;
RUN;

ODS OUTPUT SolutionSummary=SolutionSummary;
PROC OPTMODEL PRESOLVER=AUTOMATIC ;
SET <STR> Icvi;
SET <STR> Icom;
SET <STR,STR> Iinao;
NUM Authorized {Iinao};
NUM SInit {Iinao};
NUM S {Icvi};
NUM SDC {Icom};

READ DATA LA INTO Icvi=[cvi] S;
READ DATA LC INTO Icom=[Insee_com] SDC;
READ DATA LINAO INTO Iinao=[cvi Insee_com ] Authorized SInit;

SET NODES = union {<cvi,Insee_com> IN Iinao} {cvi,Insee_com};
VAR SurfModel  {<cvi,Insee_com> IN Iinao} INIT SInit[cvi,Insee_com] >= 0 <= MAX(0, MIN(S[cvi],SDC[Insee_com])) ;

MAX obj= SUM {<cvi,Insee_com> IN Iinao} SurfModel[cvi,Insee_com]*Authorized[cvi,Insee_com];

CON SurCVI {cvi IN Icvi}:
    SUM {<(cvi),Insee_com> IN Iinao} SurfModel[cvi,Insee_com] <= S[cvi]; 

CON SurCom {Insee_com IN Icom}:
    SUM {<cvi,(Insee_com)> IN Iinao} SurfModel[cvi,Insee_com] <= SDC[Insee_com]; 

SOLVE WITH NLP;
CREATE DATA optmodel(RENAME=(SurfModel=SurfModel&i)) FROM [cvi Insee_com] SurfModel;
QUIT;

File Description

• optimization_results.csv: The output table of estimated surfaces (ha) per appellation and county, along with calculated harvest values.