Skip to main content
Dryad logo

Landscape drivers of pests and pathogens abundance in arable crops

Citation

Barbu, Corentin et al. (2021), Landscape drivers of pests and pathogens abundance in arable crops, Dryad, Dataset, https://doi.org/10.5061/dryad.w3r2280r5

Abstract

Farmers’ use of fungicides and insecticides constitutes a major threat to biodiversity that is also endangering agriculture itself. Landscapes could be designed to take advantage of the dependencies of pests, pathogens, and their natural enemies on elements of the landscape. Yet the complexity of the interactions makes it difficult to establish general rules. In our study, we sought to characterize the impact of the landscape on pest and pathogen prevalence, taking into account both crop and semi-natural areas. We drew on a nine-year national survey of 30 major pests and pathogens of arable crops, distributed throughout the latitudes of metropolitan France. We performed binomial LASSO generalized linear regressions on the pest and pathogen prevalence as a function of the landscape composition in a total of 39,880 field × year × pest observation series. We observed a strong disequilibrium between the number of pests or pathogens favored (15) and disadvantaged (2) by the area of their host crop in the landscape during the previous growing season. The impact of the host crop area during the ongoing growing season was different on pests than on fungal pathogens: the density of most pathogens increased (11 of 17, and no decreases) while the density of a small majority of pests decreased (7 of 13, and 4 increases). We also found that woodlands, scrublands, hedgerows, and grasslands did not have a consistent effect on the studied spectrum of pests and pathogens. Although overall the estimated effect of the landscape is small compared to the effect of the climate, a territorial coordination that generally favors crop diversity but excludes a crop at risk in a given year might prove useful in reducing pesticide use.

Methods

Relevant extracts from the methods of the paper:

Pests and pathogens data

Since 2008, the French epidemiological services record and centralize observational data of crop pests and pathogens from arable field monitoring. In this study, we made use of two epidemiological information subsystems: Vigicultures ® (Sine et al. 2010) and VIGIBET (ITB – Sugar Beet Research Institute), that covered 17 of the 22 former French administrative regions including approximately two-thirds of its territory over the 2009-2017 period. From these two databases, we extracted information for 30 pests or pathogens on six crops (winter wheat, winter barley, corn, oilseed rape, sugar beet, potato).

We eliminated from the data the observations for which the reported crop didn’t match the crop indicated in the RPG data, considered here as the gold standard as they are tax data and have been successfully used to train automated detection of crops based on satellite imagery (Inglada et al. 2017). Depending on the crop, this could affect 5 to 30% of the field x year combinations in the database. Many of these observations also had little to no observations of pests or pathogens. We understand them as monitoring points entered by mistake and never really monitored.

Data were collected each year during the cropping season from weekly monitoring of georeferenced fields by technicians from various organizations and trained farmers (Table SI.1.). A different set of fields was monitored each year, freely chosen by the technicians performing the surveillance. It was requested that the fields be far enough apart to reflect the diversity of the territory for which the technicians are responsible, but practical access considerations are also taken into account. Possible issues with repeated measurements and auto-correlation in the data are discussed in Supplementary Materials SI.7.

All fields were conventional farming fields. The head of the observation network informed us that some observations were performed in non-treated spots but we could not account for the crop protection practices because the information was often missing in the database.

In each field, several observation types assessing the state of crop epidemics were retrieved through standardized protocols for each monitored pest and pathogen (e.g. damage severity scale on the plant for pathogens, relative or absolute organism abundance observed on the plant or in traps, amount of plants with symptoms, etc.). Not all the observation types were reported in equal numbers.

Here we kept for each organism considered only the observation type with the highest number of field x year observed to maximize the spatiotemporal extent of each pest or pathogen information. We also expected these widely used observation types to be relevant to describe the risk linked to the organisms as they are originally used to motivate pesticide applications. In total, data for 13 pests of winter wheat, corn, and oilseed rape, and 17 pathogens of winter wheat, winter barley, oilseed rape, sugar beet, and potatoes were analyzed. Detailed information on the pests and pathogens studied, observation periods, and observation types we used can be found in tables SI.1 and SI.2.

Landscape composition data

The delimitation of all French agricultural fields subsidized within the framework of the European Common Agricultural Policy is provided through the "Registre Parcellaire Graphique" (RPG). For annual crops, it is reputed to be nearly exhaustive. The geometry of the fields is described by farmers based on the aerial photographs of the BD Ortho®, a departmental orthophotography of 50 cm resolution provided by the French National Institute for Geographic and Forestry Information (IGN) (ASP et IGN 2019; Font 2018). From 2006 to 2014, fields were described by islets, a group of contiguous fields, but 80% of them had only one type of crop. In each islet, the detailed areas were given by crop types (28 crop types for 329 crops registered). Here we used six of them: winter wheat, oilseed rape, winter barley, corn (including both silage and grain corn), other industrial crops (mainly and considered here to be beet), and flowering vegetables (mainly and considered here to be potatoes). From 2015 to 2017, the description of crops in the RPG was available by species (not crop type) and by field (not islet) and we used this more precise information.

The semi-natural components considered were woods, grasslands, scrublands, and hedgerows. The RPG provided us with grassland delineations for the year of the observation (temporary and perennial grasslands are not distinguished here). The BD TOPO® (vegetation layer version 2.2 2017), a vector map with a resolution of 1 m (IGN 2016) drawn from the BD ORTHO® by the French National Institute for Geographic and Forestry Information, provided us with the geometry of the other components: woods, hedgerows, and scrublands, considered to be stable over the studied years. From this database, we grouped as “woodlands” the broadleaved, coniferous, and mixed woodlands, with closed or open canopy.

Variables preparation and control variables

Pest and pathogen abundance measurements were not normally distributed, often rounded informally, and sometimes distributed into categories. Also, the number of observations of a given pest or pathogen varied by field and year. As a result, we simplified the data into two counts per field and year: the count of observations above and under the median of the observations for all fields and years (SI.2.). For half of the organisms, only presence-absence data were available (SI.2) we then used the counts of observations with or without the pest or pathogen among the observations of the year in a given field. In both cases (with/without or above/under median), the two counts have by construction, a binomial distribution and describe the risk of being above a threshold (presence or median), hereafter referred to as the risk.

We quantified the landscape composition by measuring the area (m2) of semi-natural components and of the pest or pathogen host crop around each observation in buffers with radii of 200 m, 1 km, 5 km, and 10 km. As the abundance of a crop in the landscape could be correlated with its recurrence in the rotation at the field level, the field level rotation effect could be attributed by the regression models to landscape variables. To avoid such confusions we explicitly considered two crop rotation variables: the time elapsed in the observed field since 1) the host crop or 2) grassland, were cultivated. As only 2 years of RPG data were available before the first observations of pests and pathogens, we simplified these variables to three values: 1, 2, and 3 years or more. We discarded the points when the host crop or the grassland was not alone in the islet the last time it appeared.

To account for the potential effect of annual weather and the heterogeneity of crop management in different sub regions, we added two variables to the pool of variables: first, a categorical variable by year and region based on a supra-regional zonation of agroclimatic conditions (SI.3, Figure SI.1b) aggregating French départements (Lorgeou et al. 2012) and second, a sub-regional zonation of homogeneous farming systems (SI.3, Figure SI.1a), as defined by the French technical institute for cereals Arvalis, Institut du Végétal, (Arvalis 2011).

Usage Notes

This dataset has one line per observation site - year - pest.

In columns are

1) the observations of the pests (number observations above the threshold in the year, total number of observations (positive or negative) of the pest in the year.

2) the landscape variables: for each landscape component and buffer (200m, 1km, 5km, 10km), the square meters covered by the landscape component

3) the control variables: region, year, homogeneous practice small region.

Funding

Agence Nationale de la Recherche, Award: ANR-11-LABX-0034

Agence Nationale de la Recherche, Award: ANR-001368-P00004321

GIS GCHP2E, Award: 2015-2016

GIS GCHP2E, Award: 2015-2016