# Identifying and characterizing pesticide use on 9,000 fields of organic agriculture – Nature.com

We first decide The state of affairs of pure crop fields in Kern County After which estimate whether or not standing as pure versus typical fields decides pesticide use (Fig. 5).

Fig. 5: Methodology overview.

Figure outlines The primary method steps from decideing pure fields to creating the evaluation knowledge to carry outing the statistical analyses. All pictures proven are simplified, seen recurrentations of The informationsets. CDFA Refers again to the California Dehalfment of Meals and Agriculture, wright hereas APN is the Assessor’s Parcel Quantity and TRS is the Township-Differ-Part. Figuring out pure fields combines the created CDFA pure APN, CDFA pure TRS, and pure pesticides knowledge layers collectively to create The final pure versus typical fields layer used Inside the evaluation knowledge section. All evaluation knowledge layers are then inputted into The numerous statistical analyses.

### Figuring out pure fields

We recognized pure fields using A combination of California Dehalfment of Meals and Agriculture (CDFA) data and Kern County Agricultural Commissioner’s Office spatial knowledge (“fields shapefiles”) and pesticide use data. No single supply was full, and as such, we evaluated a number of fullly different strategyes to decideing pure fields.

#### California Dehalfment of Meals and Agriculture (CDFA) data

Data on The state of affairs of pure fields, per the California State Organic Program, for 2013–2019 was obtained by request from the California Dehalfment of Meals and Agriculture (CDFA). The CDFA, by way of the State Organic Program, requires annual registration of licensed pure producers Who’ve an anticipated gross sale of over \$5000. We have been particularally Inside the pesticide facets of pure manufacturing, which is ruled in our research area by the USDA’s Nationwide Itemizing of Allowed and Prohibited Substances. The Nationwide Itemizing of Allowed and Prohibited Substances delineates which synthetic substances Might be make the most ofd and which pure substances Can’t be used for pest administration in US pure manufacturing. Befacets substances particularally (dis)permited on the Nationwide Itemizing, permited substances embrace non-synthetic organic, botanical, and mineral inputs. Area location knowledge have been Inside The Sort of both Assessor’s Parcel Quantity (APN) or PLS System Township-Differ-Part (TRS) worths, although knowledge have been reported with out systematic formatting. We harmonized the CDFA APN worths to merge with the Kern County Assessor’s parcel shapefile (2017), which we then spatially be a part ofed with the Kern fields shapefiles. We adopted An identical course of with PLSS TRS worths, which have been then merged with the Kern County PLS Part shapefile, and be a part ofed to Kern field shapefiles. We Check with our final pure designalation as “CDFA Organic”. Particulars of The information cleansing course of are described Inside the Ancillary Data Processing Strategies section under.

#### Using pesticide use reviews to refine pure field identification

After spot-look ating pesticide use on CDFA Organic fields, it turned clear we had not fullly get rid ofd typical fields. This was probably As a Outcome of of variation in polygon geometries between PLSS Parts, Kern County Assessor parcels, and Kern agricultural fields knowledge. To further refine our classification, we used field-diploma pesticide use, as quickly as extra from the Kern County Agricultural Commissioner’s Office. As hundreds of pesticide merchandise (lively components + adjuvants) are in use in Kern County, we took an iterative strategy to get rid of fields using typical pesticides. We first restricted the universe of pesticides to these utilized to fields that have been CDFA Organic. We then recognized the 50 Principally used pesticide merchandise by Pretty A pair of purposes, and manually categorized every as pure or typical. Having recognized these merchandise as described under, we matched them again in, eliminating fields that used typical merchandise and decideing as “PUR Organic” People who used solely pure merchandise. We repeated this course of, hand decideing In all probability the Principally used merchandise and eliminating fields using typical merchandise till we had remoted fields using solely pure merchandise.

To categorise a product as pure or typical, we first Looked for every product’s U.S. EPA-registered product label, using The exact product identify and EPA product registration number. If tright here was any indication on the label thOn the product was licensed as pure by the Organic Supplies Consider Institute (OMRI), or said “To be used in pure manufacturing” or “pure”, then the pesticide was recognized as pure (n = 132). If tright here was no pure indication on the product label, we searched the OMRI certification knowledgebase for merchandise with comparable identifys and producers, and recognized merchandise as pure if such certifications existed (n = 39). If all components have been outlined (i.e., no inert or unoutlined components) and have been acknowledged pure lively components, then the pesticide was recognized as pure (n = 1) (Supplementary Data 1). We Did not Search out EPA-registered labels For 3 merchandise and confirmed on the California Dehalfment of Pesticide Regulation internet website That they are both inlively or out of manufacturing (EPA registration numbers: 52467-50008-AA-5905, 36208-50020-AA, 2935-48-AA-120). Each of the three was not often used (n < 4) on CDFA pure fields throughout the 2013 to 2019 knowledge. To be conservative, we outlined them as typical. If a field used no pesticides and was CDFA pure, we labeled it as pure. If it used no pesticides and was not CDFA pure, we primarytained it as typical.

#### Kern County agricultural commissioner’s office spatial knowledge

For 2017–2019, the commodity attrihowevere field Inside the Kern County agricultural fields shapefiles indicated self-reported pure fields with “-ORGANIC” or “-ORG” following the commodity identify. Self-reported pure crop fields Might Even be much less right than these recognized using CDFA knowledge as self-reporting Isn’t validated nor required. After evaluationing a number of huge crop producers in Kern County, we suspect that many crops being grown pureally with CDFA certifications Aren’t being reported as such to the county. Nonethemuch less, we run a strongness look at collectively with these fields Together with the PUR Organic fields for 2017 to 2019 To grab smaller farms That Can be exempt from certifications. Doing so Did not qualitatively change our end outcomes (Supplementary Figs. 2 And three).

### Analysis knowledge

Past decideing pure fields, the objective of this enterprise was To know the influence of pure versus typical manufacturing on fullly different metrics of pesticide use. As such, we calculate numerous pesticide metrics, and farm, field, and crop traits.

#### Kern County fields shapefiles

Annual Kern County fields shapefiles are publicly out tright here and embrace knowledge on rising permit ID (“farm”), website ID, field measurement, crop type, date lively and inlively, amongst fullly different information. A crop field is outlined as a permit-website-yr combination. Permit numbers monitor indivitwin growers by way of time, wright hereas website IDs are a doc Of every permitted website primarytained by a grower Yearly. Site IDs sometimes persist from one yr to The subsequent (notably for perennial crops), They typically Do not Appear to be duplicated Infacet The identical yr such that rotated crops are given distinctive website IDs. In some circumstances, a number of crops are grown on The identical field concurrently. In such circumstances, every crop has A singular website ID. Neverthemuch less, when a number of crops are being grown on The identical field concurrently, The complete space occupied by every crop Isn’t specified, however quite is doced As a Outcome of the complete field measurement. To Scale again bias in pesticide use costs, The sector measurement was divided by the Quantity of crops being grown on a field, beneath The Cas quickly aspt that Every one crops on multi-crop fields occupy an equal space. From crop type, we decided the crop’s taxonomic household For use in later analyses.

#### Pesticide use reviews

Area-by-day-by-product diploma pesticide use knowledge Can be found to The general public on the Kern County Dehalfment of Agriculture and Measurement Requirements internet website. Kilos of pesticide merchandise have been transformed to kg of lively components (AI) and recognized by The Type of product (e.g., pesticides) and “potential hazard” to non-goal organisms (e.g., bees) using the California Dehalfment of Pesticide Regulation (DPR) Product Database (see Ancillary Data Processing Strategies under). We outline pesticides as pesticides, insect progress regulators, miticides, and repellents, excluding merchandise Which have twin movement (insecticide and fungicide), however not excluding insecticide merchandise with adjuvant or fertilizer components. The Product Database embraces binary indicators for whether or not a given product is A potential hazard to fullly different environmental outcomes. Since pesticide adjuvants (“inert” components) are typically used Inside the absence of lively components And might by no implysthemuch less have environmental or wildlife influences per the label, we used kg of the product quite than a kg of lively components as measures of use for merchandise of potential hazard to environment and non-goal organisms. Using lively components Instead Did not qualitatively change our end outcomes (Supplementary Fig. 4). Potential hazards to non-goal organisms are based mostly on the Environmental Hazards assertion on the pesticide label, which is regulated by the EPA, and based mostly on toxicity to birds, fish, invertebcosts, bees, and mammals40. We chosen To make the most of binary metrics quite than an combination index because hundreds of merchandise are Utilized in Kern County, and toxicity metrics for The numerous fullly different chemical compounds And numerous environmental endfactors of curiosity Aren’t Available. Thus, we proceed with measures of chemical product use (kg ha1) of potential hazard to a collection of ecological and environmental endfactors (fish, aquatic species, drift, and so on.). We furtherly embrace merchandise of extreme acute toxicity (EPA signal phrases 1 And a couple of) and decrease acute toxicity (EPA signal phrases 3 and 4 or not required) to people. For fullness, we furtherly combinationd ecotoxicity knowledge for 29 merchandise that recurrent about 50% Of every pure And conventional pesticide use (Supplementary Desk 10) from the Pesticide Products Database61, primarily.

Using The sectors shapefiles, pesticide use costs have been summarized for every permit-website-yr grouping for numerous pesticide outcomes, collectively with kg ha−1 of a pesticide product, lively components, and merchandise of potential hazard to fish, bees, and/or aquatic species, merchandise Susceptible To float, these categorized as pesticides, and extreme and low acute toxicity merchandise, as described above. Not all environmental outcomes have been equally probably, and Particularly a number of pesticide use outcomes of potential cas quickly asrn lacked a enough Quantity of observations on pure fields for strong comparability to their typical counterhalfs (Supplementary Desk 11). We statistically analyze the subset of environmental outcomes that have been fairly widespread in each pure And conventional fields, and thus extra More probably To current reliable estimates.

#### Soil spatial knowledge

To deal with the potential that pure And conventional fields have systematically fullly different soil extreme quality, we used the California Revised Storie Index. The Storie Index land classification system is extensively used throughout California To evaluate soil extreme quality and agricultural productiveness39 and is currentd in Soil Survey Geographic Database (SSURGO) tabular knowledge for A lot of the state. Scores are systematically decided by a mannequin Inside the Natural Resupplys Conservation Service (NRCS) Nationwide Soil Information System (NASIS) Computer software, based mostly on tabular knowledge Inside the SSURGO knowledgebase. Soil traits used Inside the mannequin embrace soil profile, floor textual content materialure (e.g., loamy to clay-rich, excluding pure horizons), topographical options, rising season size, and dynamic properties (e.g., drainage, alkalinity, acidity, erosion)39. Fertility and fullly different readily modified traits are excluded. The system makes use of six rating levels—” Grade One” being The very Highest extreme quality soil relevant For many crops and “Grade Six” being unproductive land. We embrace the Storie Index as a measure of soil extreme quality in all analyses. Further, we evaluate whether or not pure And conventional fields differ systematically in soil extreme quality, using each The general Storie Index As properly as to extensively measured elements of dynamic properties, which theoretically could be affectd by on-farm administration (Supplementary Desk 1).

Storie Index worths have been transformed from a shapefile to a 60-m raster using R’s quickerize package deal. The soil raster was used to extract Storie Index worths for every Kern field polygon, using an space-weighted imply carry out. 319 fields had no Storie Index ratings. To assess the signalificance Of these lacking fields, we interpolated the Storie Index worths using an inverse distance weighting carry out in R’s gstat package deal62,63. The accuracy of interpolated worths was look ated by making use of a Depart-One-Out Cross-Validation carry out to 500 randomly chosen factors. Including these fields Did not qualitatively change our end outcomes.

### Statistical evaluation

Our statistical evaluation proceeded in two steps. First, we evaluated whether or not typical versus pure fields differed in pesticide use, mannequined as a regular variable, using pooled odd least squares and panel knowledge fashions To discover out the affect Of numerous mannequin particularation selections (see Ancillary Statistical Strategies under, Supplementary Notes, Supplementary Desks 2 And three). Neverthemuch less, pesticide use can be cas quickly asived as a two-half choice. First, Tright here’s The selection To make the most of pesticides In any respect, and second is The selection of how a lot to spray when using pesticides. Tobit fashions are conventionally used to estimate fashions with censoring. Neverthemuch less, Tobit fashions strain the mechanisms figuring out whether or not to spray (i.e., shifting from pesticide = 0 to pesticides > 0) to be The identical as the mechanisms figuring out The quantity sprayed when some pesticides are used (pesticides when pesticides > 0). Double-hurdle fashions64 are An alternate selection to the Tobit mannequin That permits for the separation Of these two selections.

The mechanisms beneathlying The two selections (to spray, how a lot to spray if spraying) can differ such that fullly different covariates can describe every course of, and The identical covariates are permited To impact The two course ofes In a number of methods (i.e., signal and magnitude can differ). The primary, binary choice Is usually mannequined with a probit mannequin.

$${{{{{rm{P}}}}}}left(y=0|{{{{{bf{x}}}}}}proper)=1-Phi left({{{{{bf{x}}}}}}gammaproper)$$

(1)

Then, the second choice is mannequined as a linear mannequin with pesticide use following a lognormal distrihoweverion, conditional on constructive pesticide use64

$$log (y)|{{{{{bf{x}}}}}},y , > , 0 sim {{{{{rm{Regular}}}}}}({{{{{bf{x}}}}}}{{{{{mathbf{upbeta }}}}}},{sigma }^{2})$$

(2)

wright here Φ is The normal normal cdf, x is a vector of explanatory variables collectively with pure standing, y is pesticide use, and ({{{{{mathbf{upbeta }}}}}}) is a vector of coefficients. We use a lognormal hurdle mannequin quite than a truncated normal hurdle mannequin since pesticide use Is very non-normal, and Q-Q plots suggested substantial mannequin enchancment using a lognormal quite than normal distrihoweverion. In distinction to the panel knowledge fashions described Inside the Ancillary Statistical Strategies under, our estimation equation used pure log-reworked variables for pesticides (and field and farm measurement) quite than inverse hyperbolic sine (IHS) transformation since solely constructive observations are embraced Inside the second hurdle mannequin. Following insights derived from our panel knowledge fashions (Supplementary Notes), we construct on The important hurdle mannequin cas quickly aspt using the farm-by-crop household intermovement as a random intercept in each the first and second hurdle. We chosen the farm-by-crop household intermovement quite than a crossed random influence As a Outcome of of computational feasibility with hundreds of permits and lots of of crops, As a Outcome of of comparableity of end outcomes to the within estimator mannequin (i.e., fixed influences in causal inference time periodinology; Supplementary Notes, Supplementary Desk 2), And since of AIC/BIC (Supplementary Desk 3). Further, We uncover proof of heteroskedasticity from each seen inspection and Levine’s look at, which provides further problems to computing crossed random influences. Thus, we proceed with the farm-by-crop household intermovement in a random intercept mannequin with cluster strong normal errors clustered On The identical grouping. In doing so, observations, wright here the taxonomic household of the crop was unclear, have been dropped. Of the 7367 fields that have been dropped As a Outcome of of lacking crop households, 6684 have been uncultivated agriculture.

Our knowledge are influenceively repeated cross-sections quite than A exact panel since fields are outlined by the farm-website-yr combination and thus usually change yr-to-yr or when crops rotate. We mannequin it as such. This suggests We do not require observations to Have not any spray in all time durations, as Can be the case in a double hurdle panel mannequin. Linking field IDs over time by way of spatial course ofing is difficult by crop rotations Of numerous measurement spaces. Since farmers might farm a number of fields beneath fullly different administration methods, as we illustrate right here, and have fullly different contrexact obligations at a sub-farm diploma, requiring farms to by no implys use pesticides on all fields Isn’t reflective of on-the-floor selections.

We repeated the lognormal hurdle fashions indivitwinly for carrots, grapes, oranges, potatoes, and onions, which have been The one extensively-grown crops with Greater than 100 pure fields. This permited for A particular slope and intercept by crop type.

We conduct a number of strongness look ats. First, We Do not have knowledge on crop yields. Neverthemuch less, To evaluate the potential implications of a yield hole on our end outcomes, we modify our per hectare costs following Ponisio et al.15 as a strongness look at. We group commodities into ceexacts, roots and tubers, oilseeds, legumes/pulses, fruits, and greens and assignal them the Ponisio et al.15 yield hole estimates for that group. Crops That did not fall Right into any of the above teams (e.g., hashish) have been currentd the all-crop common from Ponisio et al.15. Second, we analyze how typical and pure differ with respect to soil extreme quality using a within estimator strategy to account for crop-particular variations in soil extreme quality. Third, binary toxicity metrics, wright hereas useful given the Quantity of chemical compounds and endfactors of curiosity right here, by no implysthemuch less fail To fullly differentiate gradations of toxicity for chemical compounds above (or under) the regulatory threshold. As converseed about above, The information needed to calculate many combination indices (e.g., Pesticide Load57 and Environmental Impact Quotient58) Aren’t Available for All of the chemical compounds in our research. For fullness, we tried to calculate the Pesticide Toxicity Index for one properly-studied endpoint, fish. We supplemented knowledge currentd in Noproperly et al.41 with knowledge from Standartox42. Neverthemuch less, solely about 70% of the chemical compounds Utilized in our research matched, and pesticide merchandise used on pure fields have been extra More probably to lack toxicity information for A Quantity of chemical compounds. We briefly converse about the extremely preliminary investigation, given the non-random lacking toxicity knowledge.

All spatial analyses have been carried out in R Statistical Software v 3.5.3 and all statistical analyses have been carried out Stata 16 MP. For all look ats, statistical signalificance was based mostly on two-tailed look ats with (alpha =0.05.)

### Ancillary knowledge course ofing methods

#### Cleaning parcel knowledge

To spatially discover pure fields, We would have appreciated to match the Assessor’s parcel numbers (APNs) currentd Inside the CDFA tabular knowledge to APNs Inside the Kern County Parcel shapefile (from 2017). Over 90% of the APN entries Inside the CDFA knowledge have been Inside the format [xxx-xxx-xx], although a number of APNs have been typically currentd in The identical cell separated by line breaks, semi-colons, commas, and/or spaces. We made preliminary edits separating worths into indivitwin cells in Microsoft Excel since formatting was extremely inconsistent. Observations whose APNs Weren’t Inside the [xxx-xxx-xx] have been modified so thOn their format matched. In the R environment, dashes have been inserted after the third, sixth, and eighth characters (1234567895 turned 123-456-78-95) for APNs That did not already contaInside them. Occasionally, APN numbers have been Provided with dashes, however with segments of inright size (e.g., 12-34-567). In these circumstances, APN segments have been both trimmed from The biggest or padded with a zero on the left So as that they matched the [xxx-xxx-xx] format. This strategy yielded The biggest Quantity of matches and was look ated for accuracy as described under. Additional segments (from APNs with Greater than two dashes and eight numeric characters) have been dropped. A handful of APNs with fewer than eight numeric characters and no dashes have been dropped fullly.

The edited CDFA APNs have been then be a part ofed with the Kern County Assessor’s parcel shapefile, creating the “CDFA pure shapefile”. In complete, 1637 of 1829 indivitwin CDFA data be a part ofed effectively. To guage the accuracy of be a part ofs between CDFA tabular knowledge, Kern County parcel, and Kern County agricultural spatial knowledge, we spot-look ated possession information using “Agency” (CDFA) and “PERMITTEE” (Kern County agricultural knowledge) worths.

To then decide the crop fields Infacet the pure parcels, we carried out a spatial be a part of between the CDFA pure shapefile and the Kern County fields shapefiles. Earlier to carry outing the be a part of, the CDFA parcels’ dimensions have been lowered with a 50-m buffer to get rid of spatial be a part ofs between CDFA parcels and crop fields that have been solely touching the parcel margins. Of 5 fullly different buffer widths evaluated, 50 m lowered the Quantity of false constructives and negatives, as decided by evaluating the “Agency” and “PERMITTEE” worths. We Check with The sectors that match as “APN Organic”.

#### Cleaning PLSS Township-Differ-Part worths

Each yr a number of producers reported Township, Part, and Differ (TRS) worths, According to the PLS System (PLSS), quite than APN worths. We used these TRS worths to decide PLSS Parts that includeed pure fields.

We separated any cell includeing a number of TRS worths and eliminated any prefixes Similar to “S”, “Part”, “Sec.”, “T”, and “R” Which might forestall be a part ofing to Kern County PLSS spatial knowledge in Excel. In the R environment, we padded the left facet of the “S” worth with a 0 if it was a single digit, then concatenated the three columns Right into a “TRS” column. We be a part ofed TRS from the CDFA tabular knowledge to PLSS spatial knowledge, which recognized 563 Parts as includeing pure fields, from 2013 to 2019, out of An complete of 664 distinctive TRS codes Inside the CDFA knowledgeset. We then carried out a spatial be a part of between PLSS Parts that include pure fields and Kern County fields shapefiles, to decide all agriculture fields that overlap with these Parts. Additional course ofing using the Pesticide Use Reports is described above.

### Ancillary statistical methods

We started with a pooled odd least squares (OLS) mannequin that, as the identify suggests, swimming pools observations over farms, yrs, and crop varieties. Neverthemuch less, tright here Might Even be attrihoweveres of crops or farms That Can be systematically fullly different between pure And conventional, and this systematic distinction could bias our pooled OLS end outcomes. To deal with this, we first confacetred propensity rating strategyes however have been unable To discover a enough stability of our covariate distrihoweverion between pure And conventional fields. As an various, we restricted our pattern to fields with overlapping farmers and crop varieties. In fullly different phrases, we focused on the subset of fields That are grown by farmers producing each pure And conventional fields and to crops That are produced each typically and pureally. Neverthemuch less, this shrunk our knowledgeset by two-thirds.

To leverage extra of our knowledge, we subsequent confacetred panel knowledge fashions as A method to tackle unobserved variables. We confacetr each within-estimator fashions (Also referred to as “fixed influences” in causal inference time periodinology, however fullly different from the biostatistical use of the time period) and random influences fashions (with random intercepts), looking for To grab traits of the crop, grower, and yr. The benefit of a within-estimator strategy is thOn the omitted variables are eliminated (by way of differencing) and thus, They Are typically correlated with covariates with out biasing the estimation. In fullly different phrases, pesticide use and all covariates are distinctiond from their crop-particular imply (or crop household, farmer, and so on. particular imply, Counting on the mannequin). In doing so, the propensity Needless to say crops (crop household, farmer) to be grown pure or to be quick or sluggish adopters Of lalook at utilized sciences is eliminated. The disbenefit is that traits shared by all fields of a crop (e.g., worth) are misplaced Inside the differencing, and extra primarily, thOn the differencing Isn’t simply translated to nonlinear fashions that we make use of later Inside the evaluation. Random influences are extra simply translated to nonlinear fashions. The disbenefit of random influences is the strong assumption thOn the unobserved variables are uncorrelated with the covariates18,65, which is required for random influences coefficient estimates to be unbiased. Here, we see the distinction in coefficient estimates between the within-estimator and random influences fashions are quite small (Supplementary Desk 2).

Random influences notably crossed random influences with hundreds of permits and lots of of crops, introduce computational challenges As a Outcome of Of huge, sparse matrices. Further, We uncover proof of heteroskedasticity from each seen inspection and Levine’s look at, which provides further problems to computing crossed random influences. We proceed using the farm-by-crop household intermovement in a random intercept mannequin with cluster strong normal errors clustered On The identical grouping based mostly on AIC/BIC (Supplementary Desk 3), computational feasibility, and comparableity to the within-estimator end outcomes (Supplementary Desk 2). Observations, wright here the taxonomic household of the crop was unclear, have been dropped in any fashions collectively with household in both the random influences or the cluster strong normal errors. Of the 7367 fields that have been dropped As a Outcome of of lacking crop households, 6684 have been uncultivated agriculture.

In the panel knowledge fashions, we used IHS transformations to accommodate extremely non-normal pesticide (and field and farm measurement) knowledge. IHS is Simply like pure log transformation66 however is outlined at zero, which Is important given A huge frmovement of our observations have zero pesticide use. As with log–log transformations, IHS–IHS transformation can be interpreted as efinalicities. We pre-multiply pesticide use by 100 To reinformationrce estimation66, although This Does not have an effect on interpretation. As described above, we leverage insights on mannequin particularation from the panel knowledge fashions, however Rely upon the double hurdle fashions to parse ahalf The selection to spray from The selection of how a lot to spray.

### Reporting abstract

Further information on evaluation designal Is out tright here Inside the Nature Research Reporting Summary linked to This textual content material.

#### agriculture

agriculture

Hamilton Packing Agency proprietor, Jason Schlange, speaks with Appearing Director of Mt. Dept. of A…….

agriculture

#### Payette County is small in size but big in agriculture – Post Register

PAYETTE — Payette County is Idaho’s smallest county in measurement however ranks Inside The ver…….