Identifying and characterizing pesticide use on 9,000 fields of organic agriculture

Notwithstanding popular perception, the environmental impacts of organic agriculture, particularly with respect to pesticide use, are not well established. Fueling the impasse is the general lack of data on comparable organic and conventional agricultural fields. We identify the location of ~9,000 organic fields from 2013 to 2019 using field-level crop and pesticide use data, along with state certification data, for Kern County, CA, one of the US’ most valuable crop producing counties. We parse apart how being organic relative to conventional affects decisions to spray pesticides and, if spraying, how much to spray using both raw and yield gap-adjusted pesticide application rates, based on a global meta-analysis. We show the expected probability of spraying any pesticides is reduced by about 30 percentage points for organic relative to conventional fields, across different metrics of pesticide use including overall weight applied and coarse ecotoxicity metrics. We report little difference, on average, in pesticide use for organic and conventional fields that do spray, though observe substantial crop-specific heterogeneity.


nature research | reporting summary
April 2020 Field-specific reporting Please select the one below that is the best fit for your research. If you are not sure, read the appropriate sections before making your selection.

Life sciences
Behavioural & social sciences Ecological, evolutionary & environmental sciences For a reference copy of the document with all sections, see nature.com/documents/nr-reporting-summary-flat.pdf Ecological, evolutionary & environmental sciences study design All studies must disclose on these points even when the disclosure is negative. Our goal is to quantify the differences in total pesticide use and pesticides of specific concern to different ecological and environmental end points to further understanding of the environmental benefits and drawbacks of different production systems. We harmonize and aggregate several data sources to identify the spatial location of organic crop fields, and rely on unique, field-level crop and pesticide use data from Kern County, California to understand pesticide use difference. Our investigation primarily relies on double hurdle models to parse apart the decision to spray pesticides from the decision of how much to spray. Using these models, we evaluate (1) overall differences between organic and conventional fields with respect to the decisions to spray and how much to spray for total pesticide use and pesticides of potential hazard to a range of different endpoints, (2) crop-specific differences in pesticide use decisions between organic and conventional fields for five crops commonly grown with both organic and conventional practices, and (3) how adjusting for yield gaps may influence the overall results.
The statistical approach is further described in "Randomization" below.
To assess the potential implications of a yield gap on our results, we modify our per hectare pesticide use rates following Ponisio et al. 2015 (https://doi.org/10.1098/rspb.2014.1396) as a robustness check. We group commodities into cereals, roots and tubers, oilseeds, legumes/pulses, fruits, and vegetables and assign them the Ponisio et al. (2015) yield gap estimates for that group. Crops that did not fall into any of the above groups were (e.g., cannabis) provided the all crop average from Ponisio et al. (2015). Lastly, we analyze how conventional and organic differ with respect to soil quality using a within estimator approach to account for cropspecific differences in soil quality.
Our full sample consisted of 99,533 fields (organic and conventional fields combined). This represents all fields in Kern County, CA based on the Kern County Agricultural Commissioner's spatial data (available on http://www.kernag.com/gis/gis-data.asp) between 2013-2019.
Please see "Research sample" above. The start and end dates (2013-2019) were defined by the availability of CDFA records of organic agriculture.
This study used secondary data from California Department of Food and Agriculture (CDFA) and Kern County Agricultural Commissioner's Office spatial data ("fields shapefiles") and pesticide use records.
All fields in Kern County, CA based on the Kern County Agricultural Commissioner's spatial data (available on http:// www.kernag.com/gis/gis-data.asp) between 2013-2019.
Observations with ambiguous crop family were dropped in any models including family in either the random effects or the cluster robust standard errors. While 7,367 fields were dropped due to missing crop family, 6,684 of those were uncultivated agriculture. A small number of observations (n = 319) were dropped due to missing soil quality data. Including observations with interpolated soil quality has little effect on our results.
Code to reproduce the main results is available in the supplementary data. Data to repeat the main analysis is available on Dryad. This study used exclusively secondary data collected by county and state agencies.
Additional information on the statistical approach and methods considered is described in the randomization section below.
Our statistical analysis proceeded in two steps. First, we evaluated whether conventional versus organic fields differed in pesticide use, modeled as a continuous variable, using pooled ordinary least squares and panel data models to determine the influence of different model specification decisions (see SI text, SI table 2-3). However, pesticide use can be conceived as a two-part decision. First, there is the decision to use pesticides at all, and second is the decision of how much to spray when using pesticides. Tobit models are traditionally used to estimate models with censoring. However, Tobit models force the mechanisms determining whether to spray (i.e., moving from pesticide = 0 to pesticides > 0) to be the same as the mechanisms determining the amount sprayed when some pesticides are used (pesticides when pesticides > 0). Double-hurdle models are an alternative to the Tobit model that allow for the separation of these two decisions.
For the first hurdle, we are interpreting the use of zero pesticides as the true choice of the farmer and are predicting the probability the farmer (of a given field) is "zero type" as a function of being organic or not. We do so using a random effects probit model with covariates for field size, farm size, and soil quality, with random intercepts for farm-by-crop family and with cluster robust standard errors clustered at the farm-by-crop family (SI text). In the second hurdle, we evaluate what drives the amount of pesticide use on