Calculation of diversification indicators and other covariates

Author

Romain Frelat

Published

October 23, 2026

Summary

Indicators

Indicator Data Format
perimeter and area of the field RPG+ vectoriel
mean field size within buffer RPG+ vectoriel
crop rotation (N-4:N) RPG+ vectoriel
hedgerows around field RPG+ & BD Haie
Liu 2023		 vectoriel
raster 30m
% land cover within buffer RPG+ & CLCplus raster 10m
edge density Not available NA

Datasets

Field observations

Figure 1: Number of observations per year and per project
Table 1: Number of observations per year and per project
2014 2015 2016 2017 2018 2019 2020 2021 2022 2023 2024 2025 TOTAL
Agrim2019 0 0 0 0 0 40 0 0 0 0 0 0 40
BACCHUS_OPERA 0 0 0 0 40 38 40 40 38 38 38 0 272
BIOMHE 0 0 0 0 0 0 40 0 0 0 0 0 40
BISCO 0 0 0 27 0 0 0 0 0 0 0 0 27
DIVAG 0 0 0 0 0 40 0 0 0 0 0 0 40
DURUM_MIX_GM 0 0 0 0 226 0 0 0 0 0 0 0 226
EXCLU_BVD 0 0 0 0 0 0 0 7 12 16 0 0 35
FRAMEwork_BVD 0 0 0 0 0 0 0 36 0 0 0 0 36
Herrera2026 0 0 0 0 0 0 30 35 40 0 0 0 105
lepibats 0 0 0 0 0 0 0 33 0 0 0 0 33
muesli 0 0 29 0 0 0 0 0 0 0 0 0 29
OSCAR 0 0 0 0 21 44 50 80 100 102 115 0 512
PestiRed 0 0 0 0 0 0 76 60 62 79 62 70 409
Pigot2023 0 0 0 0 0 0 0 156 144 0 0 0 300
SEBIOPAG_BVD 0 0 0 0 0 0 20 21 20 0 0 0 61
SEBIOPAG_Plaine de Dijon 20 20 20 20 20 20 20 20 20 20 20 0 220
SEBIOPAG_VcG 20 19 17 17 17 17 17 17 17 17 0 0 175
SEBIOPAG_ZAAr 0 0 0 0 20 0 0 17 0 16 0 0 53
Seree2022 0 0 0 0 0 14 19 0 0 0 0 0 33
SERIPAGE 0 0 9 0 0 0 0 0 0 0 0 0 9
TOTAL 40 39 75 64 344 213 312 522 453 288 235 70 2655
Figure 2: Map of field observations

More detailled overview of the field observations can be found here

Identification of the crop field

Because of data availability, we will only focus on the observations made in the period 2015-2025 with field coordinates. The project DURUM_MIX_GM has a single coordinate leading to the entrance of the Institut Agro-Montpellier, without registered crop field within 1000m from the coordinates so it was excluded.

We identified the crop field from RPG or Nutzung dataset based on the coordinates and the year of the sampling.

Table 2: Number of observations in fields from RPG
Nobs in_RPG in_VineOrchard Perc
Agrim2019 40 40 0 100.00
BACCHUS_OPERA 272 1 268 98.90
BIOMHE 40 39 0 97.50
BISCO 27 26 0 96.30
DIVAG 40 40 0 100.00
EXCLU_BVD 35 0 35 100.00
FRAMEwork_BVD 36 0 36 100.00
Herrera2026 105 94 0 89.52
lepibats 33 30 0 90.91
muesli 29 29 0 100.00
OSCAR 512 448 11 89.65
PestiRed 339 303 0 89.38
Pigot2023 300 299 0 99.67
SEBIOPAG_BVD 61 0 61 100.00
SEBIOPAG_Plaine de Dijon 200 200 0 100.00
SEBIOPAG_VcG 155 152 0 98.06
SEBIOPAG_ZAAr 53 53 0 100.00
Seree2022 33 30 0 90.91
SERIPAGE 9 9 0 100.00

Remarks:

  • In total, 95 % of the fields observations (N=2204) are covered with information on crop field.
  • It is important to acknowledge that RPG is incomplete but the vineyards and orchards provided by data providers helps characterizing projects with permanent crops.

Field size

We calculated the area and the perimeter of the crop fields corresponding to the samplings.

Figure 3: Field area in ha per project. The dashed line show the median area.
Table 3: Summary statistics per project of the area (in ha) of crop fields
Min. 1st Qu. Median Mean 3rd Qu. Max.
Agrim2019 0.97 2.21 3.02 3.80 4.99 10.21
BACCHUS_OPERA 0.23 0.58 0.78 1.07 1.40 3.15
BIOMHE 0.68 1.59 3.69 4.48 6.29 13.99
BISCO 0.50 1.10 1.71 3.22 3.15 23.69
DIVAG 0.97 2.21 2.98 3.17 4.02 6.01
EXCLU_BVD 0.45 0.58 0.72 0.94 1.29 2.20
FRAMEwork_BVD 0.29 0.58 0.84 1.25 1.77 3.28
Herrera2026 0.87 4.58 7.88 8.76 11.91 37.52
lepibats 1.40 2.77 6.18 11.89 13.96 55.47
muesli 0.41 1.90 3.84 5.46 7.61 34.05
OSCAR 0.22 0.48 1.00 1.71 1.70 21.35
PestiRed 0.10 0.97 1.06 1.32 1.40 5.19
Pigot2023 2.40 7.93 10.46 12.87 17.05 44.15
SEBIOPAG_BVD 0.29 0.60 0.87 1.37 1.77 5.68
SEBIOPAG_Plaine de Dijon 0.53 5.11 6.83 7.46 8.92 17.82
SEBIOPAG_VcG 0.85 2.90 4.27 5.56 7.00 24.06
SEBIOPAG_ZAAr 1.21 3.16 4.73 4.96 6.94 16.22
Seree2022 0.90 2.60 4.81 9.17 12.85 37.52
SERIPAGE 1.55 2.19 3.51 4.04 5.19 7.60
Figure 4: Field perimeter in m per project. The dashed line show the median perimeter.
Table 4: Summary statistics per project of the perimeter (in m) of crop fields
Min. 1st Qu. Median Mean 3rd Qu. Max.
Agrim2019 485.39 691.18 797.06 889.45 1053.13 1674.45
BACCHUS_OPERA 214.18 330.40 398.22 449.45 552.24 1157.85
BIOMHE 354.03 592.12 835.73 907.71 1072.08 1807.17
BISCO 307.54 503.97 593.59 743.08 846.90 2550.45
DIVAG 485.39 654.90 799.78 816.22 937.89 1275.08
EXCLU_BVD 269.15 324.13 459.31 452.72 550.25 669.28
FRAMEwork_BVD 247.58 445.90 509.95 523.05 602.94 938.87
Herrera2026 527.24 1056.08 1423.60 1450.42 1727.66 5991.28
lepibats 488.44 824.10 1260.50 1789.15 2670.28 6828.99
muesli 257.24 608.61 939.47 1060.52 1338.17 4021.15
OSCAR 200.02 358.30 493.46 583.72 749.38 2728.54
PestiRed 278.44 454.92 513.90 551.14 611.12 1126.50
Pigot2023 707.30 1381.42 1616.08 1735.90 2089.44 4574.97
SEBIOPAG_BVD 247.58 366.24 500.45 525.22 643.19 984.37
SEBIOPAG_Plaine de Dijon 393.53 1075.19 1240.09 1297.87 1558.72 2220.91
SEBIOPAG_VcG 377.85 815.28 1171.63 1202.13 1505.30 4249.39
SEBIOPAG_ZAAr 451.97 793.52 949.02 1061.93 1207.41 3466.03
Seree2022 378.99 778.39 1272.30 1482.71 2076.89 5991.28
SERIPAGE 494.30 750.53 776.57 928.22 1227.43 1466.38
Figure 5: Relation between area and perimeter (in log scale)

There is a strong relation between area and perimeter (Figure 5). In median, field size is 2.1 ha and field perimeter is 703m.

Outliers

Figure 6: Field with small area and large perimeter
Figure 7: Field with large area

Remarks:

The overall pattern of field size per projects matches the expectations. Project focussing on wineyards (e.g. Bacchus, OSCAR), orchards (in BVD - Basse Vallée de la Durance) have smaller fields than average; while project focusing on large cereals (e.g. muesli, Plaine de Dijon) have larger fields.

Yet, when looking for outliers, we can see that:

  • Some fields are defined as a strip (large perimeter, small area), e.g. Figure 6
  • Some fields include many sub-plots, e.g. Figure 7.

Field size within buffer

Using the coordinates of the sampling sites, we calculated the average area of all crop fields within a buffer (500m and 1000m).

Table 5: Summary statistics of the field area (in ha) within different buffer size
B_500m B_1000m
Min. 0.02 0.16
1st Qu. 0.73 0.72
Median 1.29 1.19
Mean 2.39 1.97
3rd Qu. 2.94 2.37
Max. 21.98 16.35
NA’s 2.00 1.00
Figure 8: Average field size with buffer of 500m
Figure 9: Average field size with buffer of 1000m

We see that some observations don’t have crop field within 500m (N=2 in Table 5). These observations (listed in Table 6) would need to be checked and ensure that they are close to agricultural lands.

Table 6: Observations with no fields within a 500m buffer.
Study_ID Plot_ID Year
181 OSCAR 33_2011_00002 2018
481 OSCAR 11_2023_00004 2023
Figure 10: Correlation among field areas per buffer size

Outliers

Figure 11: Highest average field size within 1000m buffer
Figure 12: Lowest average field size within 1000m buffer
Figure 13: Highest difference between 500 and 1000m buffer
Figure 14: Lowest difference between 500 and 1000m buffer

Remarks:

  • The field size within buffer is a good proxy for agricultural practices at landscape level.
  • The values with a buffer of 500m and 1km are highly correlated (r = 0.86). Yet values with a buffer of 1km are slightly less extreme (higher minimum and lower maximum).

Crop rotation (N-4:N)

Figure 15: Length of the land cover time series per project.
Table 7: Number of observations per length of crop rotation data (in years) available
0 1 2 3 4 5
Agrim2019 0 0 0 0 1 39
BACCHUS_OPERA 1 0 0 1 2 268
BIOMHE 1 0 1 0 0 38
BISCO 1 0 0 26 0 0
DIVAG 0 0 0 0 0 40
EXCLU_BVD 0 0 0 0 0 35
FRAMEwork_BVD 0 0 0 0 0 36
Herrera2026 9 0 0 0 2 94
lepibats 2 1 4 0 3 23
muesli 0 0 29 0 0 0
OSCAR 34 1 26 36 79 336
PestiRed 1 2 7 38 144 147
Pigot2023 1 0 0 0 0 299
SEBIOPAG_BVD 0 0 0 0 0 61
SEBIOPAG_Plaine de Dijon 0 20 20 20 20 120
SEBIOPAG_VcG 1 18 17 17 19 83
SEBIOPAG_ZAAr 0 0 0 0 20 33
Seree2022 3 0 0 0 0 30
SERIPAGE 0 0 9 0 0 0

The number of observations in the period N-4:N in Table 7 is a results of the number of observations per year (Table 1). Most observations have 5 years of data on crop rotation (73%, N=1682).

Let’s zoom on the crop rotations over the 5-year period (i.e. excluding observations with incomplete crop rotation data).

Figure 16: Number of different crop cultivated in the period N:N-4. Only observations with complete information on land cover for the 5 years are included.
Figure 17: Number of different crop groups cultivated in the period N:N-4. Only observations with complete information on land cover for the 5 years are included.
Figure 18: Number of different crops cultivated in the past 5 years per project
Figure 19: Number of different crop groups cultivated in the past 5 years per project

From the 1682 observations with complete crop rotation information in [N-4:N], 535 have the same crop for the whole time period, while 112 fields have five different crop groups in the past 5 years (Figure 16).

Table 8: Most populat crop group rotation cultivated in the 5 years before observations
Var1 Freq
vignes 488
vergers 132
ble_tendre_hiver,colza_hiver,orge_hiver 75
divers,vignes 26
gel_sans_prod,vignes 24
ble_tendre_hiver,prairies_temporaires 17
ble_tendre_hiver,colza_hiver,proteagineux 16
ble_tendre_hiver,mais_ensilage 14
ble_tendre_hiver,colza_hiver,orge_hiver,proteagineux 13
ble_tendre_hiver,colza_hiver,lin_non_textile,proteagineux 12
ble_tendre_hiver,colza_hiver,orge_printemps,proteagineux 12
autres_cult_indus,ble_tendre_hiver,orge_hiver 11
ble_tendre_hiver,mais_grain_semence 10
mais_grain_semence,vignes 10
ble_dur_hiver,colza_hiver 9
ble_dur_hiver,ble_tendre_hiver,colza_hiver,orge_printemps 8
ble_tendre_hiver,colza_hiver,mais_ensilage,orge_hiver 8
autres_cult_indus,ble_tendre_hiver,mais_grain_semence,orge_hiver 7
ble_dur_hiver,ble_tendre_hiver,tournesol 7
ble_tendre_hiver,colza_hiver,fourrage_fabacees 7

Remarks:

  • The level of crop rotation depends strongly on the project targeting either annual vs perrenial cropping systems.
  • In France, the spatial dataset about orchard and wineyards was set as constant during the study period, which artificially ignores any changes in agricultural practices in these perrenial systems. So crop rotation information must be ignored for the perrenial crops in France.

Hedgerows length

We overlaid the field as defined in RPG+ (+ a 10m buffer) with the tree cover density from Liu et al. 2023 and with other national information on hedgerows (BD Haie and Nutzung).

Figure 20: Correlation among hedgerow indicators
Figure 21: Hedgerow cover (%) per project from Liu 2023. The dashed line show the median biomass.
Figure 22: Hedgerow cover (%) per project from national database. The dashed line show the median biomass.
Table 9: Percentage of fields with some hedgerow detected per data source
Liu National
Agrim2019 100.00 100.00
BACCHUS_OPERA 40.07 26.10
BIOMHE 92.50 97.50
BISCO 96.30 96.30
DIVAG 92.50 95.00
EXCLU_BVD 100.00 100.00
FRAMEwork_BVD 100.00 100.00
Herrera2026 51.43 49.52
lepibats 54.55 48.48
muesli 100.00 96.55
OSCAR 58.01 47.07
PestiRed 45.72 7.67
Pigot2023 29.00 38.00
SEBIOPAG_BVD 100.00 100.00
SEBIOPAG_Plaine de Dijon 54.50 39.50
SEBIOPAG_VcG 89.03 98.06
SEBIOPAG_ZAAr 94.34 100.00
Seree2022 24.24 45.45
SERIPAGE 100.00 100.00

Outliers

Figure 23: Field with high hedgerows cover in BD Haie but not in Liu 2023
Figure 24: Field with low hedgerows cover in BD Haie but high in Liu 2023

Remarks:

  • There is a relatively good correlation among the percentage of hedgerows cover in Liu_2023 dataset and the national dataset (r= 0.66 for France, and 0.46 for Switzerland).
  • Liu et al. 2023 might consider orchards as hedgerow as seen for the BVD projects in Figure 21.
  • In Switzerland, the percentage of fields with hedgerows is extremely low (7.67 %, Table 9) while the values are not so different when considering Liu_2023. This observation seems to indicate a difference in how hedgerows were defined in France and in Switzerland and would suggest the use of Liu_2023 when comparing both datasets.

Land cover within buffer

Table 10: Number of land cover classes and percentage of cover from RPG+ in 1000m buffer per project
n_classes av_perc_rpg
Agrim2019 36 76.36
BACCHUS_OPERA 32 70.18
BIOMHE 40 71.01
BISCO 30 70.56
DIVAG 38 74.85
EXCLU_BVD 39 48.25
FRAMEwork_BVD 38 44.08
Herrera2026 51 77.05
lepibats 36 52.21
muesli 36 73.81
OSCAR 51 61.48
PestiRed 40 56.05
Pigot2023 47 78.15
SEBIOPAG_BVD 42 47.28
SEBIOPAG_Plaine de Dijon 44 68.43
SEBIOPAG_VcG 44 75.48
SEBIOPAG_ZAAr 45 73.34
Seree2022 42 76.67
SERIPAGE 28 76.63

After grouping, there are 54 different categories covered by the 1000m buffers .

Table 11: Top 10 most frequent land cover classes in cover 
                  forest                   vignes         ble_tendre_hiver 
                   15.51                    14.96                    10.43 
    permanent.herbaceous     prairies_permanentes                   sealed 
                    9.38                     5.63                     5.45 
             colza_hiver     prairies_temporaires low.growing.woody.plants 
                    3.48                     3.46                     3.27 
              orge_hiver 
                    3.10

In average, 67 % of the buffer areas are filled with land cover classes from RPG+ (Table 10). The remaining 33% are filled with CLCplus classes, especially forest (15.5%), permanent harbaceous (9.4%) and sealed (5.4%) (Table 11).

Figure 25: Average land cover per project with buffer of 1000m
Figure 26: Average forest cover per project with buffer of 1000m
Figure 27: Average sealed cover per project with buffer of 1000m
Figure 28: Average vineyard cover per project with buffer of 1000m
Figure 29: Average wheat cover per project with buffer of 1000m
Figure 30: Shannon diversity in land cover buffer of 1000m. The dashed line show the median area.
Table 12: Summary statistics per project of the Shannon diversity in land cover buffer of 1000m
Min. 1st Qu. Median Mean 3rd Qu. Max.
Agrim2019 2.02 2.27 2.33 2.34 2.41 2.63
BACCHUS_OPERA 0.56 1.07 1.16 1.15 1.32 1.73
BIOMHE 2.07 2.28 2.35 2.35 2.44 2.61
BISCO 1.99 2.12 2.21 2.19 2.27 2.35
DIVAG 2.22 2.30 2.36 2.39 2.48 2.63
EXCLU_BVD 1.14 1.50 1.66 1.69 1.84 2.24
FRAMEwork_BVD 1.27 1.70 1.87 1.84 1.99 2.16
Herrera2026 1.25 2.09 2.32 2.28 2.49 2.83
lepibats 1.10 1.54 1.88 1.78 2.05 2.35
muesli 1.76 2.22 2.30 2.27 2.39 2.62
OSCAR 0.96 1.42 1.74 1.75 2.02 2.84
PestiRed 1.14 2.04 2.31 2.22 2.51 2.84
Pigot2023 1.29 1.98 2.09 2.14 2.30 2.85
SEBIOPAG_BVD 1.23 1.74 1.87 1.87 2.00 2.25
SEBIOPAG_Plaine de Dijon 1.02 1.99 2.21 2.13 2.34 2.78
SEBIOPAG_VcG 1.86 2.25 2.37 2.35 2.45 2.66
SEBIOPAG_ZAAr 1.97 2.24 2.34 2.34 2.45 2.70
Seree2022 1.73 1.96 2.14 2.17 2.33 2.71
SERIPAGE 1.88 2.16 2.18 2.13 2.19 2.28

Remarks:

  • All buffers include some amount of forest (5-20%) and sealed area (5-10%), while the crop composition is highly project specific with project with perrenial crops (Bacchus, BVD, Oscar) being less diverse than project focussing on annual crops.

Correlation

Figure 31: Correlation among field diversity metrics

Remarks:

  • There is a high positive correlation between field size and the median field size within buffer : the size of the field is related to the average field size in the landscape.
  • There is a high positive correlation between spatial and temporal crop diversity (land cover within a buffer and crop rotation).
  • Landscape with high percentage of vineyards have low diversity of landscape and low crop rotation. On the contrary, landscape with high percentage of wheat have larger field size, and higher diversity of crop spatially and temporally. This observation might be bias by the high number of crop groups compare to other classes of land cover considered.

Summary

Material and method

To characterize the cropping practices, we used the Registre Parcellaire Graphique (RPG) for France and the Geodienste - nutzungsflaechen (Nutzung) for Switzerland. The two datasets are vectorial spatial data that provide the crop field boundaries and categorize the crops into refined categories: 366 crops are considered in France, and 146 in Swizerland. RPG is available annually since 2015 while Nutzung is available since 2019. Because RPG is known to lack data for small wineyards and orchards (ref needed?), we complemented the crop data with local dataset for the samples made in regions dominated by permanent crops.

We intersected the sampling coordinates with the national crop dataset corresponding to the sampling year and extracted the perimeter and the area of the fields. Additionnally, we calculated the mean field size within a buffer of 1km around the sampling position. When available, we exctracted information about the cropping patterns for the 4 years before sampling.

We complemented the crop dataset with the land use land cover data from Corine Land Cover plus (CLCplus). It caracterize the land cover into 11 classes. The dataset is available for Europe with a spatial resolution of 10 m. There has been three release: 2018, 2021, and 2023. We filled the holes of the national crop dataset with the CLCplus information closest to the year of sampling. Then, we extracted the percentage of cover per buffer of 1km around the sampling location. We grouped the land cover and crop classes into 57 classes and calculated the Shannon index of the simplified land cover classes.

Hedgerow density was quantify from national datasets: BD Haies v2 for France in 2021 and Geodienste - nutzungsflaechen for Switzerland in 2023. or Hedgerow density was quantify using European dataset (Liu et al. 2023) providing estimates of the cover and the biomass of trees at 30m resolution for 2019. We extracted the length/area of hedgerows within the fieldw and a 10m buffer around them.

All analysis were performed in R 4.4.1() using terra() and exactextract() packages.

Results

In total, 95 % of the fields samplings are covered with information on crop field, and 73% of the samples (N=1682) have complete 5 years crop rotation data.

  • Most observations have a corresponding crop field in RPG dataset or in complementary dataset (Table 2).
  • All buffers include some amount of forest (5-20%) and sealed area (5-10%), while the crop composition is highly project specific.

References

Baston, Daniel. 2023. Exactextractr: Fast Extraction from Raster Datasets Using Polygons. https://CRAN.R-project.org/package=exactextractr.
Hijmans, Robert J., Andrew Brown, and Márcia Barbosa. 2026. Terra: Spatial Data Analysis. https://CRAN.R-project.org/package=terra.
Institut national de l’information géographique et forestière (IGN). 2024. BD Haie v2: Couche nationale de référence des haies linéaires en France métropolitaine. https://geoservices.ign.fr/bdhaie.
Institut national de l’information géographique et forestière (IGN). 2025. Registre Parcellaire Graphique (RPG) Version 2.0 - Millésime 2015-2025. https://cartes.gouv.fr/rechercher-une-donnee/dataset/IGNF_RPG.
Konferenz der kantonalen Geoinformations- und Katasterstellen (KGK). 2025. Nutzungsflächen (Landwirtschaftliche Bewirtschaftung) - MGDM ID 153.1, Datenstand 2020-2025. https://geodienste.ch/services/lwb_nutzungsflaechen.
Liu, Siyu, Martin Brandt, Thomas Nord-Larsen, et al. 2023. “The Overlooked Contribution of Trees Outside Forests to Tree Cover and Woody Biomass Across Europe.” Science Advances 9 (37): eadh4097. https://doi.org/10.1126/sciadv.adh4097.
R Core Team. 2024. R: A Language and Environment for Statistical Computing. R Foundation for Statistical Computing. https://www.R-project.org/.