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

The study addresses whether and how organic agriculture reduces pesticide use compared to conventional farming, a key but contested aspect of sustainability. Despite growth in organic acreage and consumer demand, robust comparisons are hindered by selection bias, lack of field-level data, and differences in crops, soils, and farmer behaviors between systems. Pesticides remain a salient sustainability metric; while modern chemistries aim to reduce human toxicity, many remain ecologically harmful and organic regulations limit types, not amounts, of inputs. The authors aim to quantify differences in total pesticide use and products of potential hazard to ecological and environmental endpoints between organic and conventional fields in a real-world, high-value cropping region. They identify organic fields spatially, use field-level pesticide reporting, and apply double-hurdle models to separate the decision to spray from intensity of use, including crop-specific analyses and adjustments for organic yield gaps. This informs the environmental benefits and trade-offs of organic production at scale.

The paper reviews debates on organic agriculture’s environmental performance, highlighting mixed evidence on yield gaps and variability, and the challenge of valid comparisons due to non-random placement of organic farms and crop/soil differences. Historical pesticide issues (e.g., organochlorines/organophosphates) prompted shifts to newer chemistries with reduced mammal/bird toxicity but persistent ecological risks (e.g., to aquatic life and pollinators). Organic farming is often perceived as chemical-free, yet US regulations primarily constrain which substances can be used; some allowed substances (e.g., copper, pyrethrin, azadirachtin) can be ecotoxic. Prior work indicates residue testing emphasizes human toxicity of synthetics, with little field-level data on organic pesticide use. Policy interest (e.g., EU Farm to Fork) and consumer concerns underscore the need for rigorous, field-level comparisons. The authors also note the paucity of comprehensive toxicity data to construct continuous indices across many active ingredients and endpoints, motivating use of label-based binary hazard indicators.

Study region and data: Kern County, California, a major US producer of high-value fruit and vegetable crops. Sample includes 99,533 permitted fields (2013–2019). Organic field identification combined: (1) CDFA State Organic Program records (APN and PLSS TRS), harmonized and spatially joined to county parcel and field shapefiles; (2) refinement via field-level Pesticide Use Reports (PUR) by iteratively classifying frequently used pesticide products as organic or conventional using EPA labels and OMRI listings, eliminating fields with any conventional products; (3) robustness including 2017–2019 self-reported organic flags in county field attributes. Fields using no pesticides and matching CDFA organic were labeled organic; fields with no pesticides and no CDFA organic remained conventional. Pesticide metrics: From daily field-by-product PUR data, kg of active ingredients (AI) and kg of products (AI + adjuvants) were computed and aggregated to permit-site-year. Products were categorized as insecticides (including IGRs, miticides, repellents; excluding dual-action fungicide/insecticides) and flagged for potential hazard to fish, bees, aquatic species, and drift propensity using CA DPR Product Database label-based binary indicators. Human acute toxicity categories used EPA signal words: higher (categories 1–2) vs lower (3–4 or not required). For environmental hazard endpoints, kg of product per ha was used due to adjuvant impacts; using AI instead did not change results qualitatively. Soil covariates: Soil quality measured using the California Revised Storie Index (SSURGO/NASIS), converted to 60-m raster and area-weighted to fields (1 = highest, 6 = lowest); missing values were few and interpolation robustness checks did not affect results. Statistical analysis: Double (lognormal) hurdle models separated: (1) decision to spray (probit with random intercepts for farm-by-crop family, cluster-robust SEs at same level) and (2) intensity on positive-use fields (log-linear model with farm-by-crop family random intercepts, cluster-robust SEs). Covariates included field size, farm size, and soil quality. Outcomes analyzed: AI kg/ha, product kg/ha, insecticides, drift-prone products, products of potential hazard to fish and bees, and human acute toxicity categories. Crop-specific hurdle models were run for carrot, grape, orange, potato, and onion (widely grown both organically and conventionally). Yield-gap adjustment multiplied organic use rates by crop-group-specific organic-to-conventional yield gaps from Ponisio et al. (2015) across cereals, roots/tubers, oilseeds, legumes/pulses, fruits, vegetables, with an all-crop average where needed; this affects only the second hurdle. Toxicity robustness: where possible, calculated a fish Pesticide Toxicity Index following Nowell et al. (2014), supplementing with Standartox; coverage was ~70% of chemicals, with more missing data on organic fields. Software: Spatial analyses in R (v3.5.3); statistical analyses in Stata 16 MP; α = 0.05; multiple robustness checks (random effects vs within estimators, heteroskedasticity considerations, missing crop family observations dropped).

• Sample and organic identification: 99,533 fields (2013–2019). Approximately 9,100 organic fields identified. Annual organic field share ~7–11% of permitted fields.
• Structural differences: Organic fields are ~44% smaller than conventional on average, but farms with both organic and conventional fields are ~4× larger than purely conventional farms (1240 ha vs 319 ha). Organic fields are on higher-quality soils (lower Storie Index values) on average, including after crop-specific controls.
• Overall probability of spraying (first hurdle): Being organic reduces the probability of using any pesticides by about 31 percentage points for AI (0.31 ± 0.03), and by 18–31 percentage points (0.18 ± 0.02 to 0.31 ± 0.03) across other outcomes (products, insecticides, drift-prone, potential hazard to fish and bees, and human acute toxicity categories). Larger field and farm sizes and higher soil quality generally increase the probability of spraying.
• Intensity on treated fields (second hurdle): For most outcomes, organic status is associated with an estimated ~1–17% decrease in use, but not statistically significant, except: • High acute human toxicity products (EPA categories 1–2): significant 27% ± 11% reduction for organic. • Low acute human toxicity products (categories 3–4/not required): significant 28% ± 14% increase for organic.
• Crop-specific results (AI kg/ha): Organic reduces probability of any use by 21–51 percentage points across carrot, grape, orange, potato, onion. Among fields that spray: • Carrot: −87% ± 3% (significant). • Grape: +132% ± 33% (significant increase). • Orange: −64% ± 11% (significant). • Potato: −81% ± 5% (significant). • Onion: +54% ± 54% (not significant). For carrot and grape across multiple pesticide outcomes, organic reduced carrot use by 80–98% (except low-toxicity +72% ± 12%), while for grape organic increased use by 126–286%.
• Yield-gap adjustment: Adjusting organic use rates by Ponisio et al. (2015) crop-group yield gaps leaves the first-hurdle effects unchanged; second-hurdle effects shift toward zero and are generally near zero and not significant (except low-toxicity products), indicating per-output differences in intensity largely diminish.
• Toxicity-weighted robustness (fish PTI): In a subsample with ~70% chemical coverage (biased by missing toxicity data, especially for organic), organic fields show roughly half the toxicity-weighted use for fish compared to conventional.
• Descriptive statistics: Average field ~31 ha; average AI use ~25 kg/ha and product use ~45 kg/ha annually across all fields; 1,293 farms/year, average farm size 451 ha.

The findings indicate that, in a large, intensive production region, organic fields are more likely to be pesticide-free in a given year, evidenced by an 18–31 percentage point reduction in the probability of spraying across outcomes. Among fields that do spray, organic and conventional fields generally apply similar amounts per area across most metrics, with notable exceptions: organic applies significantly fewer high-acute-human-toxicity products and more low-acute-toxicity products. Crop-specific analyses reveal substantial heterogeneity: some crops (carrot, orange, potato) show large reductions in pesticide use under organic management, whereas others (grape) show substantial increases. Adjusting for estimated organic yield gaps brings intensity differences closer to zero per unit output, implying that per-area reductions do not necessarily translate into per-output reductions in all contexts. Structural differences—organic fields being smaller, on better soils, and within much larger mixed farms—highlight selection and scale dynamics influencing pesticide decisions. Field size increases the propensity to spray, and larger farms tend to spray more frequently but use less per treated field, suggesting differing economic thresholds and potential ecological mechanisms. The results refine the sustainability narrative: organic management tends to reduce the frequency of pesticide use, limiting exposure and potential ecological impact on many fields, but when pesticides are used, intensities are often comparable, with outcomes depending on crop and toxicity category. These insights suggest targeted, crop-specific policies and the importance of toxicity profiles alongside total use when evaluating environmental performance.

This study provides the first spatially explicit identification of thousands of organic fields in a major US agricultural county and rigorously compares pesticide use between organic and conventional systems. Key contributions include: (1) a reproducible framework integrating certification records, parcel/PLSS geographies, and pesticide use reports to locate organic fields; (2) evidence that organic fields are smaller, part of larger farms, and on better soils; (3) robust reductions in the probability of pesticide use under organic management across multiple metrics, with similar per-area intensities among fields that do spray except for shifts toward lower acute human toxicity; (4) pronounced crop-level heterogeneity, including cases where organic use is higher. Adjusting for estimated yield gaps moves intensity differences toward zero per unit output. Future work should develop comprehensive toxicity-weighted indices covering the breadth of products and endpoints, acquire field-level yield data to enable per-output comparisons, expand analyses to other regions and farm structures, and investigate spatial arrangements and farm-level decision-making that influence pesticide dynamics.

• Potential misclassification of organic fields due to spatial joins among PLSS sections, parcels, and field polygons; some non-certified fields could be misidentified, and some certified fields may not be captured.
• Not all organically managed fields are certified; uncertified organic fields in the conventional group could bias estimates toward zero. Including self-reported organics in robustness checks did not qualitatively change results.
• Lack of field-level yield data necessitated per-area comparisons; yield gap adjustments rely on meta-analysis estimates that may not reflect local realities.
• Toxicity measures primarily binary label-based hazard indicators; comprehensive continuous toxicity indices lacked sufficient coverage across chemicals/products and endpoints, with non-random missingness greater for organic fields.
• Unobserved confounding cannot be fully ruled out despite random effects and robustness checks (e.g., farmer characteristics, enterprise-level decisions).
• Generalizability may be limited: results are from one intensive, high-value cropping region with many mixed farms; patterns may differ elsewhere.
• Some observations dropped due to missing crop family or soil data; however, robustness checks suggest limited impact on conclusions.