2.1 Overall methodology
Analysing the data obtained from the water management institution at catchment scale (ASA), this study is based on the hypothesis that the farming practices applied in terms of irrigation management are related to some farms’ characteristics, determining their propensity to be heavily or weakly water consumers. This characterization can be the starting point to have an assessment of agricultural water consumption at large scale. Then, to refine the estimation, it is important to consider water applications that vary from one orchard field to another, particularly within farms that are heavily water consumers. The general methodology of this study is described in figure 1. First of all, at farm scale, (1) starting from available data acquired by the ASA and from farmers interviews, information about irrigation water consumption were collected at farm scale for 15 farms; then (2) the water consumptions of each farm was relied to a series of farms structural and biophysical characteristics, in order to understand which factors have an influence on irrigation behaviour (3) using eight statistical methods to classify farms in the territory; (4) the most occurrent statistical classification obtained was applied to the each farm in the area, in order to obtain a classification of all the farms in two main classes: the highest irrigators or the weakest.
This first classification at farm level was one of the main inputs for the orchard fields scale analysis. For the heavily irrigated farms, the following workflow was applied: (1) irrigation water volumes applied were estimated on the field taking into account the accurate information from farmers interviews and orchards characteristics for 66 orchard fields; then (2) the estimated volumes of each field was used to rely them with some biophysical and geographical factors of the parcel particularly belonging to the heavily irrigated farms and finally (3) nine statistical classification methods were applied to each orchard fields which belong to heavily irrigated farms in the area and (4) the most occurrent statistical classification was applied to all the 279 orchard fields belonging to heavily irrigated farms , in order to have an estimation of the overall water consumption for irrigation at the watershed scale.
The estimation obtained was compiled and compared with the available data from ASA at larger scale (municipalities and watershed scales), as a validation of the method. In particular, at municipal scale, the total volume corresponding to the sum of each terminal they monitored in all the municipality was used. These volumes were also compiled at watershed scale, in order to obtain a single amount of water consumed for the watershed.
Fig.1 Overall scheme of the methodology. AWC: Available Water Content; ASA – ‘Association Syndicale Autorisée’: water managers at catchment scale; LPIS: Land Parcel Identification System; RGA - Recensement Général Agricole: French government data on crops grown in municipalities. Grey ovals refer to the sections of the material and methods describing these parts.
2.2 Description of the case study
The Ouvèze-Ventoux region is a part of the Ouvèze river basin, in the South-East of France, a right tributary of the Rhône (Roux et al. 2019). The Ouvèze has a typical torrential flow regime, as it can be subject to heavy flooding as well as periods of very low flow, particularly in summer. The area is around 100km², delimited at North-East by the Ouvèze river, and at South-Est by the Mont Ventoux (map Fig 2). It includes four municipalities with a variety of landscapes: a wide plateau in the Entrechaux area (around 300m asl.) in the North-Ouest and a valley with a colder micro-climate and higher altitude (around 500m asl.) in Beaumont-du-Ventoux. The climate is characterised by rainfall of around 650mm/year, according to the Carpentras weather station, mainly concentrated in autumn-winter period, whereas recent years have been characterized by heavy summer droughts. Heavy rainfall can occur, with extremes events of up to 300mm in a single episode (maximal observation in Entrechaux 22/09/1992) (Piegay and Bravard 1997).
The distribution of water for irrigation is managed by the ASA (Association Syndicale Autorisée, Authorized Syndicate Association) Ouvèze-Ventoux. The managed areas of the ASA are mapped in red on figure 2 and cover 304ha in Beaumont-du-Ventoux, 467ha in Malaucène, 372 ha in Entrechaux and 211 ha in Le Crestet municipality (total 1354ha). Most of these surfaces are occupied by fruit trees (44% of the utilised agricultural areas – UAA) and vineyards (43% of the UAA - 83% of which are wine-growing vines and 17% table vines). Most of the orchards are cherry trees (Prunus avium), which is a very traditional production in the area (IGP “Cerises des coteaux du Ventoux”). However, in the last years they suffered different hazard related to climate change, with increasing water needs (around 540mm/year in the south of France, according to CABRL 2019) and recurrent diseases (Herrera and J 2015; Ali et al. 2017; Medda et al. 2022). In terms of water use, irrigation restrictions have been imposed during dry summers, reducing irrigation authorisations and even banning all irrigations systems, even water-saving drip irrigation (a so-called "crisis" state), as it was the case during the drought of 2022 (Préfecture de Vaucluse 2022).
Figure 3 shows the relation between the water consumption for each year (data from ASA) and the rains: seasonal (April to September included) and annual from the Carpentras weather station (15Km-SSE). A large variability is observed which can appear sometimes correlated with the annual rainfall amount: for drier years (such as 2017 or 2022), higher consumptions are observed, but for most of the time this variability seems not related to rainfalls. The annual rainfall is not the only parameter impacting the ASA water consumption and the variability between annual and seasonal rainfall could affect the water reserved in the soil. The seasonal rainfall (between April and September) is also not completely correlate to the irrigation, even if the low level of seasonal rainfall could explain some years of heavy irrigation (2016, 2017 or 2019 for example).
The taxation of the water consumption by ASA is referring to water terminals which are connected to one or more fields of a same consumer. According to the statements of water managers in the region, the water intake stations of the ASA are not very precise and show up to 30% error comparing to the real consumption, especially when the field is irrigated frequently but in small quantities (as in the case of a drip irrigation orchard or vineyard).
Fig.2 Map of the Ouvèze-Ventoux watershed and ASA-OV managed areas and fields referenced in the LPIS.
Fig.3 Water consumption in the ASA area in the watershed according to the ASA Ouvèze-Ventoux (mm): annual (in red) and seasonal (April-September included, in blue) rains from the Carpentras weather station (20Km SSO) from 2006 to 2022
2.3 Dataset description
2.3.1 Data on farms and fields’ structure and distribution
Table 1 summarizes the available datasets for the Ouvèze-Ventoux watershed. All the data listed and described on the following paragraph have been used to characterize the farming systems of the study area, including all the topographical and farming practices information possibly related to the use of water.
In order to have a comprehensive description of each cultivated parcels, data from the Land Parcel Identification System (LPIS, availed freely by the ‘geoservices’[1] from IGN) was acquired. LPIS is a spatialized dataset on crops declared each year by farmers to received European subsidies. Considering the source, not all farms are declared, depending wherever or not they are asking for support. In particular, perennial crops are often lacking, because they are not submitted to subsidies. Usually, these crops are declared when they are cultivated in mixed farms, where there are also annual and/or herbaceous crops. Information on the location of the agricultural parcels and their crop are given, but no information on irrigation is furnished nether distinction among the different species of orchards, except for olive groves, table grapes, and vineyards. In this study, the last available LPIS data in the region (2020) was considered. This dataset represents 1721 agricultural fields, representing 48% of the total 3581 fields in the area, belonging to 65 farms. In terms of surface, they represent 1001ha, including 504 fields (302ha) of orchards in 2020. The orchard class in the LPIS do not include olive groves and truffle groves which are referenced differently. Considering that the LPIS does not cover all the farms in the area, the database was completed with the agricultural census (RGA - Recensement Général Agricole, partially freely available by the AGRESTE website[2]). This database provides the information reported in table 2 at municipality scale only. In this dataset, irrigated surfaces are referenced but this information is not accessible freely and requires accreditation.
In terms of topographical and soil data, this study is based on the digital elevation model and the pedological map of the study area. The digital elevation model was used to calculate the altitude, slope and exposition of the parcels. Soil data, especially about available water capacity (AWC) which is calculated according to the pedological characteristic of the soil (Cousin et al. 2022), have been derived from the soil map of the society managing the Provence canal (SCP[3] , Societé du canal de Provence) in 2012. The AWC values has been calculated for each farms’ field and then aggregated at farm level. In some cases, where the soil data were lacking, the information was completed with the Infosol unit map of AWC (Román Dobarco et al. 2019b, a; Romàn Dobarco et al. 2022) , available at France scale at a spatial resolution of 90m for the first 2m of soil (Román Dobarco et al. 2019b)[4].
In order to complete the description of the area with information on farming practices related to irrigation, a field surveys were carried out between 2019 and 2023, enquiring 21 farmers in the watershed area, 13 of which depending on the ASA for their water supplies, in farms which have a significant proportion of perennial crops (e.g. orchards or vineyards). During the interviews the following information were collected: location of all the farms’ fields (749 plots), land cover and crop type (with cultivar detail); for orchard crops, the spacing between trees and between rows, the type of drippers used, their flow and dispositions in the crop, and the irrigation schedule.
Combining all the information listed, a geo-database including 3581 parcels was obtained which represent the whole agricultural areas in the watershed. Among this total, 710 parcels, corresponding to 393 ha (22% of the total agricultural surfaces), belong to the farms which were interviewed and for which more detailed information about farming practices were available.
Tables 1: List of the available dataset
|
Source
|
Resolution
|
Information type
|
LPIS 2020 – Land parcel information system
|
Data.gouv (French gouvernemental data) [5]
|
Plot scale
|
Crop typology (38 classes)
-
Farm to which the parcel belongs
|
Surveys
|
Farmers
|
Plot scale (for 21 farms)
|
Crops, cultural practices, trees implantation pattern, irrigation time
|
RGA 2020
|
INSEE (partially confidential database)
|
Municipal scale
|
Crops typology and areas, including irrigated areas
|
ASA water consumption
|
ASA – Ouvèze-Ventoux
|
Farm scale
Municipal scale
Watershed scale
|
Water consumption taxed of 13 farms
-
Consumption at each water distribution terminal
-
Total water consumption provided by ASA (for each year)
|
DEM
|
Geoportail
|
1m
|
altitude
|
Soil map
|
SCP
|
1/25000 aggregated at Pedologic unit scale
|
Pedological information (Dominant texture and Pedological names of first layers)
|
AWC map 1
|
Link to the SCP soil map
|
Pedologic unit scale
|
AWC
|
AWC map 2
|
Infosol INRAE unit
|
Pixel size: 90m x 90m
|
AWC estimated from modelling for the theorical 2 first meters of soil
|
Table 2 Areas (ha) referred in the RGA data for main orchards and vineyards typologies in the four municipalities in the Ouvèze-Ventoux area.
|
Total (ha)
|
Irrigated (ha)
|
|
Beaumont
|
Crestet
|
Entrechaux
|
Malaucène
|
Beaumont
|
Crestet
|
Entrechaux
|
Malaucène
|
Cherries
|
72
|
3
|
7
|
66
|
50
|
2
|
2
|
54
|
Apricots
|
34
|
3
|
18
|
55
|
20
|
0
|
11
|
38
|
Plums
|
44
|
6
|
2
|
33
|
30
|
6
|
1,5
|
25
|
Olive groves
|
5
|
10
|
11
|
24
|
1
|
1
|
0,6
|
11
|
Truffle tree groves
|
3
|
1
|
3
|
17
|
0,05
|
0
|
1
|
6
|
Others orchards
|
4
|
1
|
0,2
|
17
|
1
|
0
|
0,2
|
11
|
Table vineyards
|
17
|
19
|
30
|
26
|
13
|
13
|
15
|
9
|
Wine vineyards
|
82
|
86
|
212
|
200
|
8
|
0
|
6
|
18
|
2.3.2 Data on water consumption at farm and field scale
In this study, the available information (provided by ASA) about water distribution and consumption on the study area, was used as dependent variable on the statistical modelling analysis, and as a validation for the overall consumption. Water resources for irrigation are managed by the ASA, which delivers water to each farmer via a pressure network and irrigation terminals. Each water access terminal can normally be linked to one or more fields of a same farm. Each farmer declares an irrigated surface and volumes are recorded on terminals. As mentioned before, the ASA estimates the sensor uncertainty at around 30%. This data is then summarised by the ASA to define a water consumption for each farm for the water taxation. The ASA provided the total consumption volumes for 13 of the 21 interviewed farms.
The data about the farm consumption were calculated for 5 years (2016-2020) and adjusted for the total surface of the farm to estimate the average quantity of water applied in millimetres per year, assuming that the total farm area does not change from one year to another during the 5 years.
Figure 4, shows the average water consumption per farm for the 13 farms from ASA information and 2 others (R3 and T1) for which the sum of irrigation was calculate (describe in 2nd part of part 2.3.2). Two groups can be clearly identified: some farms consume more than 100mm per year while other consume less than 40mm. Dashed lines show the mean values of heavily irrigated (163.8mm) and weakly irrigated farms (12.9mm). According to this information, farms were separated in two groups: heavily irrigated, if they applied more than 100mm/year and weakly irrigated, if they applied less than 40mm/year.
Fig.4 Water consumption according to the ASA data for the surveyed farms. Dashed red: average consumption of higher consumers (in red), Dashed blue: average consumption of lower consumers (in blue)
Moreover, 2 more farms in the watershed were added not belonging to the territory managed by the ASA. These two farms have cherry orchards as only relevant irrigated crop. Volumes of water were estimated by the sum of the water use in each orchard’s plot, obtained according to some relevant characteristics assessed during interviews, namely: (1) the spacing between rows of trees (Srows in m) and between trees on a same row (Strees in m); (2) the type of irrigation equipment, namely r flow rate in litres per hour (Flow in L/h) and repartition in the field by the spacing between 2 drippers (Sdrip in m) or the number of drippers per trees (case of micro-sprinklers) (Ns in drippers/trees); (3) the irrigation schedule, which depends on the type of orchard cultivar, compiled as a number of hours of irrigation per year (t in hour/year). The assessment of the water quantity is thus made by following equation 1 or 2 according to the type of irrigation equipment. In the equation 1, relative to micro-sprinklers, the number of trees is computed according to the distances between rows and trees, considering a squared field of 1ha where 1m is lost on each side of the plot. In the Equation 2, referred to drip irrigation, the number of rows is computed for a squared field of 1ha where 1m is lost for the borders, this value divided by the spacing between rows is equivalent to the number of rows in this plot and, this value multiplied by 98m of row longer in the square plot (the 100m side of the square - 2m border) provide the length of piper’s tube in the field.
These 2 equations give an estimation of the annual water consumption for an irrigated field of cherry orchard. However, this estimation does not reflect potential interannual variability. These equations were applied to 66 cherry orchard fields belonging to heavily irrigated farms and to the 68 cherry orchard field belonging to weakly irrigated farms. Figure 5, shows the high diversity in terms of water consumption for the orchards of heavily irrigated farms: we observed two main behaviours, corresponding to the two pics. Dashed lines show the mean value for each class.
Fig.5 Density of plots by their water consumption estimated for the 66 orchards of heavily consumers farms, in red the threshold between the 2 groups of plots, in dashed green the mean values of each groups (110mm and 538mm).
Considering the general distribution of the water consumption at field scale, the two most occurrent means values (110mm and 538mm) were applied at each orchard field scale to sum the overall water use at municipal and watershed scale.
For the weakly irrigated farms, less variability was observed and the mean value of water consumption at field scale for cherry orchards was 26mm, used as reference value.
For the remaining irrigated fields, not covered by orchards, we used as value of water amount yearly applied the quantity declared by farmers during the interviews and also by the local experts, in particular the ASA technicians. The yearly volumes declared are showed on table 3.
Table 3 Irrigation apply according to the typology of perennial crops
Crop
|
Water volume per field (mm)
|
Orchards heavily irrigated of heavily irrigated farm
|
538
|
Orchard weakly irrigated of heavily irrigated farm
|
110
|
Orchard of weakly irrigated farm
|
26
|
Olive groves
|
150
|
Truffle tree groves
|
40
|
Table vineyards
|
300
|
Wine vineyards
|
100
|
The two additional farms for which were estimated the water consumption (R3 and T1 in figure 4) considering the sum of each orchard consumption have respectively 173mm/year for the R3 farm and 224mm/year for the T1. These two farms are high consumers where the main crop is orchards which represent the biggest part of them area (86% for R3, 69% for T1).
2.4 Statistical approach
In terms of statistical methods, for both the classification at farm and field level, a benchmark of classification models was applied, namely: Random Forest (RF), Support vector machine (SVM), Principal component analysis (PCA), Neural Network (NN), Naive Bayesian classification (NB), logistical regression (LR), K nearest neighbours (KNN), linear discriminant analysis (LDA) and Factor Analysis of Mixed Data (FAMD) (only for plot scale considering some non-quantitative variables). The list and description of the statistical methods applied are showed on Tab.4. The different statistical methods are applying in order to give more robustness to the modelling approach. We then considered as the correct class the one resulting on more of the models applied. The Pearson correlation coefficient (Pearson 1920) was estimate between each variables of our datasets and variables not significantly independent were not kept for the NB, kNN and NN classifications.
Table 4. Statistical methods descriptions (default parameters are the default parameters in the R function used)
Statistical approach
|
Description
|
Hyper-parameters applied
|
Principal Component Analysis (PCA)
|
Dimensionality reduction technique identifying the principal components, which are orthogonal linear combinations of the original variables (Gifi 1990; Husson et al. 2005)
|
ncp: 5
Size of confidence ellipses: 95%
|
Support Vector Machine
|
Machine learning algorithm used for classification tasks which find the best hyperplane that separates different classes in the data space (Chen et al. 2004)
|
Kernel: radial
Regularization parameter: default
|
Naive Bayesian
|
Application of the Bayes theorem at a dataset of objects with independent characteristics to define classes (Rish 2001)
|
None
|
Random Forest
|
Machine learning method used for classification and regression tasks. It constructs multiple decision trees during training and combines their predictions through voting (for classification) or averaging (for regression) to improve accuracy and robustness (Breiman 1996, 2001)
|
Tree number,Predictor number per split, Leaf size :
default
|
k-Nearest Neighbor
|
It predicts the class or value of a new data point by considering the majority class or average value of its k-nearest neighbor in the training data (Keller et al. 1985)
|
k: 3
|
Logistical regression
|
Binary classification, estimating the probability that a given input belongs to one of two classes based on predictor variables (Lee et al. 2006)
|
Family: "binomial"
Regulatization parameters: default
|
Factor Analysis of Mixed Data
|
Multivariate statistical technique used for analyzing datasets containing both quantitative and qualitative variables. It explores underlying structures and relationships between variables through dimensionality reduction and factor analysis (Escofier 1979; Kiers 1991)
|
Number of dimensions kept (ncp): 5
|
Neural Network
|
Training a model composed of interconnected nodes (neurons) to classify data based on patterns learned from input-output pairs (LeCun et al. 2015; Chollet 2017)
|
2 layers (5,5 neurons in farm's classification 5,6 in the plot's one)
|
Linear Discriminant Analysis
|
Finds directions that maximize class separation, projects data onto these directions, and predicts class membership based on the projected values (Friedman 1989)
|
Regulatization parameters: default
|
For each statistical method, a cross-validation was applied for the two training datasets (15 farms – 66 orchards) and the methods not providing a sufficient score were not considered.
After applying each classification method independently for each holding, the allocation to the most common class was retained. Similarly, each classification was applied to each of the orchard fields of the heavily irrigated farms and the most common allocation was retained.
2.4.1 Farm scale classification
For the farm classification, the characteristics selected had to meet two main criteria: they had to be accessible via the data available at watershed level, and they had to correspond to characteristics that have a strong influence on water consumption, since the differences observed between the consumption of heavily and weakly irrigated farms are very large (figure 4).
Firstly, we assumed that a farm that will have a predominance of irrigated perennial crops (table vines/ orchards) will consume more water. This hypothesis is also supported by other previous studies which show that farms that do not have enough water will turn to less water-consuming crops (meadows, cereals, etc.) rather than apply deficit irrigation on heavily irrigated crops (Schuck and Green 2001; Gómez‐Limón and Riesgo 2004). Then, the characteristics of the farms tend to define their water consumption. In this study, the choice was to characterise farms by their soil via the average AWC (average of the average AWC of each field weighted by their area), considering that AWC is one of the major characteristics of a plot's water requirements (Pereira et al. 2015). On the other hand, the average altitude of the field is one of the main factors in the characteristics of the field and enables fields to be distinguished by their microclimate (temperatures/winds/rains) (Jacobsen et al. 1997; Sevruk 1997; Archer and Caldeira 2009). So, in this study was kept the average altitude of the farm (mean altitude of each plot weighted by their area) as an important factor of the decision of agricultural practices (Schoenly et al. 1996)
Table 5 Used parameters for the farms’ classification
Data
|
Source
|
Area of heavily irrigated crops (orchards + Table vineyards)
|
LPIS
|
Area of weakly irrigated crops (Cuve vineyards + other crops)
|
LPIS
|
Average available water capacity (AWC) of the farm
|
SCP / Infosol
|
Average altitude of the farm
|
DEM
|
2.4.2 Field scale classification
Based on the previous assessment, the classification was carried out at field scale for the orchard fields belonging to heavily irrigated farms. The variables used for the classification are listed on table 6.
As previously, for this classification each variable was selected to describe the water needs of a crop in this field as a representative proxy of the irrigation need. Altitude was kept as a descriptor of the microclimatic cultural conditions. The percentage of pixels exposed to south was add as a description of the solar direct radiation which affect highly the climatic condition (Dufour 1887). For soil, the main drivers of water needs are relative to the AWC. The available water is affected by the quality of the soil from the texture (sandy soil or not) and by water losses, which are partly linked to runoff from the slope (Gaetano et al. 2017). The proximity of watercourse could also affect the available water and the characteristic of soil (Struyf et al. 2009). Finally, from the point of view of human labour, according to the farmers surveyed, the proximity of a field will have an impact on the frequency of visits by the farmer and therefore on whether or not water is supplied.
Table 6 Variables for the orchard fields’ classification
Data
|
Source
|
Altitude of the centroid of the plot
|
DEM
|
Average slope of the plot
|
DEM
|
Percentage of pixels exposed to the south
|
DEM
|
Dominant soil texture
|
Donesol / pedological map
|
AWC
|
SCP - Infosol
|
Distance to nearest watercourse
|
IGN[6]
|
Distance from the centre of the farm
|
LPIS
|
2.5 Overall water consumption for irrigation at watershed scale
The values of irrigated fields referenced in the table 2 are weighted considering the municipality and the crop type in the field according to the agricultural census data. In fact, the census indicates the percentage of irrigated surface area per municipality, so it was considered a ratio applied to the overall estimation. For the orchard species characterization, the values of water consumption applied at field scale (table 3) were calculated on cherry orchard for a large majority (76%), whereas apricot and other orchards can be considered to consume around 70% of the cherries, based on farmers surveys and irrigation manual (CABRL 2019). For this reason, a correction factor was applied based on the census data about the percentage of each crop and its rate of irrigation in the municipality.
For the LPIS possible bias (not all plots are declared), we had to consider that these data are declarative and as previously indicated, in the study area they cover only 48% of the lands. To compare this value to the total irrigated area in the ASA perimeter, a factor was applied which assumes that the ratio of surface area allocated to each crop in the LPIS is representative to the ratio between surfaces areas allocated to each crop in the all watershed (referred in the table 7). To know the total agricultural area, in the case of this study, a general shapefile was created by the observations in the territory and the cadastral data. This map referred to all the agricultural areas in the territory. The ratio between this area and the area referenced in the LPIS (percentage of crops in the LPIS according to the totality of crops) is used to correct the missing part of the LPIS surfaces. The table 7 shows the percentage of represented area in the LPIS.
Table 7 Representativity of the LPIS surface in comparison of all the parcels in the territory of ASA in each municipality
|
Beaumont
|
Crestet
|
Entrechaux
|
Malaucène
|
LPIS area (ha)
|
119
|
44
|
98
|
185
|
Total agricultural area (ha)
|
164
|
110
|
210
|
254
|
Ratio of LPIS representativity (%)
|
73
|
40
|
47
|
73
|