A herbicide resistance risk assessment for weeds in wheat and barley crops in New Zealand

We estimated the risk of selecting for herbicide resistance in 101 weed species known to occur in wheat and barley crops on farms in New Zealand. A protocol was used that accounts for both the risk that different herbicides will select for resistance and each weed’s propensity to develop herbicide resistance based on the number of cases worldwide. To provide context we documented current herbicide use patterns. Most weeds (55) were low-risk, 30 were medium-risk and 16 high-risk. The top ten scored weeds were Echinochloa crus-galli, Poa annua, Lolium multiflorum, Erigeron sumatrensis, Raphanus raphanistrum, Lolium perenne, Erigeron bonariensis, Avena fatua, Avena sterilis and Digitaria sanguinalis. Seven out of ten high-risk weeds were grasses. The most used herbicides were synthetic auxins, an enolpyruvylshikimate-phosphate synthase inhibitor, acetolactate synthase (ALS) inhibitors, carotenoid biosynthesis inhibitors, and long-chain fatty acid inhibitors. ALS-inhibitors were assessed as posing the greatest risk for more species than other modes-of-action. Despite pre-emergence herbicides being known to delay resistance, New Zealand farmers only applied flufenacet and terbuthlazine with high frequency. Based on our analysis, surveys for herbicide-resistant species should focus on the high-risk species we identified. Farmer extension efforts in New Zealand should address resistance evolution in cropping weeds.


Abstract:
We estimated the risk of selecting for herbicide resistance in 101 weed species known to occur in wheat and barley crops on farms in New Zealand. A protocol was used that accounts for both the risk that different herbicides will select for resistance and each weed's propensity to develop herbicide resistance based on the number of cases worldwide. To provide context we documented current herbicide use patterns. Most weeds (55) were low-risk, 30 were medium-risk and 16 high-risk. The top ten scored weeds were Echinochloa crus-galli, Poa annua, Lolium multiflorum, Erigeron sumatrensis, Raphanus raphanistrum, Lolium perenne, Erigeron bonariensis, Avena fatua, Avena sterilis and Digitaria sanguinalis. Seven out of ten high-risk weeds were grasses. The most used herbicides were synthetic auxins, an enolpyruvylshikimatephosphate synthase inhibitor, acetolactate synthase (ALS) inhibitors, carotenoid biosynthesis inhibitors, and long-chain fatty acid inhibitors. ALS-inhibitors were assessed as posing the greatest risk for more species than other mode-of-action. Despite pre-emergence herbicides being known to delay resistance, New Zealand farmers only applied flufenacet and terbuthlazine with high frequency. Based on our analysis, surveys for herbicide-resistant species should focus on the high-risk species we identified. Farmer extension efforts in New Zealand should address resistance evolution in cropping weeds.

Order of Authors: Zachary Ngow
Richard J. Chynoweth

Chris Evan Buddenhagen
Opposed Reviewers: Response to Reviewers: Reviewer #1: Because frequency and abundance of weed species in the region have not been used in the scoring, the paper provides a picture of risks due to the current use of herbicides in the re-gion, which is informative for the farmer about the intrinsic risk of an abundant species in its own farm. However, although I agree with intrinsic high risk species as stated on line 280-285, it is pity that the overall risk for the region is not presented. The regional risk is assessed! We use a published account of weeds, as well as expert knowledge to identify problem weeds in wheat and barley. The risk assess-ment considers the propensity of those weeds to develop resistance. It uses the most important factors, the crop specific region wide weed flora and region wide data about herbicide use. The selection pressure comes from the herbicides used and so captures the most important dimensions of the regional risk. Spain, France (Columa), Italy, etc., including companies, extension service and farmers, and published in local con-ferences. That is another example of the "focus on one or two problematic species in a given crop" issue we highlight, and therefore does not need to be added.
L142 It can be assumed that weed biology is already included in the inherent species score, but abundance is a key parameter. Abundance includes effects of farming systems: if many individuals still remain in the field, then the risk of developing resistance is higher than if the population is small. A score modifier to account for variation of abundance (e.g. 1 for low density and 3 for more than a million plant/ha, or another scale) could be used. Another factor is the species frequency in the studied region, which doesn't seem to be taken into account (is the Production Wise platform collecting main troublesome weed species?). While that is true, regional abundance and density data do not exist to be able to implement that. We acknowledge that distribution, abundance and phenology are important #(L272). ProductionWise does not gather any information on weeds. Because we identified high-risk species and herbicides the data are useful as a region-al assessment. Farmers and experts can now interpret the observed weed abun-dance with an eye to the likely risk of resistance. We emphasize high use herbicides in our studied regions, because that selection pressure drives the evolution of resistance. We mention in the introduction that in New Zealand, there are only 12 instances of herbicide resistance in arable crops #(L41-43) and in the discussion delve further into New Zealand HR weeds (L238-244). The high and low usage herbicide groups come from ProductionWise data, which can be found on Table 2. The basis of deciding which groups are 'high use in New Zealand' is on L106-108. Cases of resistance have been indicated on Table 3.
Reviewer #2 Dr Stephen Moss: Abstract: Generally fine but as likely to be most-read part I suggest a few minor changes. Do you need to say 'wheat and barley' or would 'cereal' suffice? Ditto in title. We think wheat and barley is a more accurate statement for our case since the herbicide data focuses on that. Wheat and barley is more widely grown than oats, and depending on the reader, corn would also be consid-ered to be a cereal… I suggest: 15-18'We estimated the risk of selecting for herbicide resistance in 100 weed species known to occur in wheat and barley crops on farms in New Zealand. A protocol was used that accounts for both the risk that different herbicides will select for resistance and each weed's propensity to develop herbicide resistance based on the number of cases worldwide.' Done 24'ALS-inhibitors were assessed as posing the greatest risk for more species than any other mode-of-action. Done 25I prefer 'Pre-emergence' but that may depend on journal style. Done 26'……. in this class commonly used by…..' Is this better than 'favoured'? Done Somewhere I think it would be worth stressing in abstract that 7 out of the high-risk 10 are grass-weeds. If space is limiting, I think the last sentence of abstract could go or 'farmer extension efforts' incorporated into previous sentence. Done Introduction: Good -acceptably concise and relevant. I suggest including some brief (one or two sentences) information on arable cropping in NZ and what proportion of that is accounted for by wheat and barley. Or at least to show that wheat and barley comprise a significant proportion of arable crops -you might even wish to mention that NZ currently holds world record for both barley and wheat yield/ha (I think that is correct). I would! Done 32Suggest: '……potential losses….' Losses of 23% don't actually commonly occur. Done 41Clarify 'cases'. To non-specialist reader it is vague and could mean number of fields or farms -although 'specialists' know why that term is appropriate. I would suggest stating number of weed species instead which is 13 according to Heap website country info. That lists 19 'cases' (29 April) although that dupli-cates some species. If it is 25 cases now (according to your reference), perhaps someone in NZ should provide Ian Heap with updated info. This is important if this paper is published. Likewise amend '12' in line 42 if appropriate. Done re-worded to reflect species numbers in New Zealand, and documented instances of resistance, to distinguish from the usage of cases in the Heap database. 45'haphazard' is a bit unfair, although this is a valid point. I would dispute that my sampling and testing over 30 years has been 'haphazard'! Suggest 'unsystemat-ic' as a better word -I would agree with that. End sentence after 'globally'. Then 'It reflects…..' Done 50comma needed after [9][10][11][12][13][14]. These 6 refs are all valid, but do you need them all? They are valid, focus on systematic efforts to detect any and all weed re-sistant cases. Also, they could be hard to find for other researchers working on a similar project. Leaving as is. 53-54Good points -one issue is that if resistance is perceived to be rare then hard to justify cost. If very widespread then why bother to do a survey? Money could be better spent. Also lack of infrastructure and personnel to conduct surveys may be as important as cost. Also, do surveys have much long-term value? Is a 20-year old survey of much value now -perhaps only as a benchmark? Perhaps money better spent on more 'durable' studies? Not saying that you need to con-sider these aspects here -perhaps in discussion as risk assessment relevant to survey priorities. Good points thanks for the insightful commentary. 59Suggest: '…….and their prior record of resistance in cereal fields elsewhere in the world. ' We specifically looked at wheat and barley. It's an important dis-tinction. 62-64This is fine and a good succinct sentence, but I wonder if it is worth emphasising more that risk assessment includes not only herbicide risk but also weed species risk for non-specialist readers, as this is not particularly intuitive? So, could read: 'They present a quantitative risk matrix using both herbicide-risk (some herbicides pose a higher risk than others) and species-risk (some weed species are more resistanceprone than others), with an optional score modifier de-signed to account for agronomic management practices that may reduce the risk.' Done 67Change: 'Individual field and farm scale risk is not assessed as this requires detailed information on past herbicide use, including timing etc…..' (This covers what has been applied -surely most important factor?) We changed the para-graph to emphasize importance of our herbicide application data set: ". This risk assessment is on an industry-wide scale informed by anonymized herbicide application data from wheat and barley fields. Risks were not assessed at the scale of individual farms and fields, this requires detailed information about herbicide timing, mixtures and rotations, and their interactions with weed bi-ology, crop rotations and other cultural practices. All the high-risk weeds iden-tified here should be targeted in surveys designed to detect herbicide-resistant weeds. Materials and methods: Good -makes it clear what was done and why slightly different approaches to published protocol were adopted 76Suggest: 'Most grasses and some …..' The wonders of Google mean that I can see, within 60 seconds, that two grass weed species were identified at species level in the paper cited….. Done 83Suggest: '….legacy herbicide mode of action (MoA) groups……' Done 88-89Suggest: '….legacy HRAC MoA group [24], with risk scores of 1, 2 or 3 given for low, medium or high risk respectively.' Done 96Suggest: '…..scored as '1'.' ('One' is a bit ambiguous). Done 98Suggest: 'Most recent….' Done but, "The most recent…" 108-109This is slightly confusing: 'The most used herbicides for each crop were charac-terized by weighting active ingredient amounts and the number of fields they were applied to.' It almost implies total a.i. weight was used. Not sure this sen-tence is needed. Deleted the sentence. 110Why taxon? Species better? Changed to species. 112Suggest: '…..HRAC MoA group…...' And I suggest elsewhere throughout paper. Done 113Suggest: '….herbicide type…. Done 114Suggest: '……the global number of resistance cases….' Done 115Suggest: 'To obtain the 'high', 'medium' and 'low' risk scores as used in the Moss protocol [17],…..' Done 117-118This sentence is slightly confusing and maybe could be improved. Is this correct? 'We assessed overall species-risk as the sum of the herbicide-risk multiplied by the "inherent" species-risk [17] combined for all relevant HRAC MoA herbicide groups, but only…….' (I think 'once' is confusing). The example you give below is useful. Done 120Suggest: '……weed species….' Done 135Suggest: '……to determine species-risk scores based on the number of cases of resistance worldwide, but we think this……' Done 137Suggest: ' ….45 'high' and 'medium' resistance risk species, many more than Moss et al……….' Done 139Suggest: 'The Moss protocol also used score modifiers that take into account resistance management practices including the use of non-chemical control measures' . Done 139-142Suggest: 'We did not use the score modifiers since these vary from field to field and our objective was slightly different. We acknowledge that actual risk of resistance development is determined mostly by the frequency and type of herbi-cide applied (selection pressure) interacting with characteristics of weed biolo-gy, distribution and abundance [5]. One aspect that seems missing, is some idea of the crop rotations used on NZ farms and the impact this might have. This focus in this paper is on wheat and barley but the frequency of growing these crops and consequent herbicide use will surely impact on the resistance risk? How commonly is arable cropping ro-tated with grass? Are allarable farms common compared with mixed farms? This may explain why herbicide resistance is relatively uncommon in NZ. Also relevant to risk at individual field level which lends itself nicely to comments about where you should set priorities for resistance monitoring, management and farmer KT. No need to go into great detail, but is relevant. We have includ-ed some text about rotations.
Conclusions: Generally fine but suggest the following. 343-346Suggest: 'A European protocol [17], designed primarily to assist in herbicide authorization procedures, was adapted to assess the risk of herbicide re-sistance evolving in 100 different weed species known to occur in New Zealand wheat and barley crops. More than half the weeds (55) we assessed were low-risk, 29 were medium-risk and 16 high-risk. The 10 species posing the highest resistance risk were: etc etc' Done 349-351Final two sentences could usefully be a bit 'punchier'. 'Suggest: 'We are planning extensive surveys in New Zealand to detect new cases of herbicide re-sistance. The risk assessment outlined in this paper will enable us to prioritise those weeds identified as posing a high resistance risk and, consequently, make better use of available resources. The risk assessment procedure as described in this paper has the potential to be a very useful tool for evaluating the risk of herbicide resistance in a wide range of different weed species in other coun-tries too'. Done Reviewer #3 (YASEEN KHALIL): The introduction and methodology of the experiments seem to be satisfactory. However, in the methods section, the authors need to explain how they did the statistical analyses and data presentation. We do not carry out any statistical tests. All our data is descriptive, and are visual-ized using graphs . We added a sentence to the last line of the methods section about the graphs.
SPECIFIC COMMENTS: L20-22 Add the classifiers to the scientific names. Alternatively, you may add the classifiers to the scientific names in the Materials and methods section and after that no need for it to be mentioned. We think the reviewer is referring to the specific binomial author authorities. We examined other abstracts in PLOS ONE and do not see the these being used for abstracts unless it is specifically about a taxonomic treatment. L20-22 Add hyphen "herbicide-resistant" Done but line 27 L45 Add the "it reflects the varying" Done L52 Add the classifier to the scientific name "Alopecurus myosuroides ". Done and also for Avena fatua L20-22 Add hyphen "farm-scale" We added this at line 68 though which was the first men-tion L74-76 What about the other 25% of the wheat and barley production regions in New Zea-land. I would prefer to stick to the mentioned region and not extrapolate to the whole coun-try of New Zealand. We intend the weed list to capture our best estimate of the weeds known to occur in wheat and barley fields in New Zealand. We added a new sentence to explain: We expanded the weed species list to include species known to occur in wheat and barley fields in the wider New Zealand context. L78 It will be very useful to mention the regions of the studies conducted by the mentioned researcher's literature [21,22]. We disagree. We checked the regions studied in those articles and knowing them adds lit-tle insight. Also, only a few species were added from these. L79 It is recommended to follow the journal guidelines in this regards instead (Species no-menclature). Taxonomic authorities added at first mention. L91-92 What is the logic and the rationale behind this decision? Worldwide, Group A herbi-cides are the most vulnerable group in terms of weed resistance evolving We do classify Group A herbicides as high risk. We include a bit more detail here and consider the issues in the discussion. It now reads: With group A having 48 cas-es and group G herbicides having 47 cases we chose to place the two groups in the same risk category, with a difference in the numbers of cases of just worldwide we believe they are indistinguishable from the data. The alternative is to use the same threshold as in the Moss protocol, but this would result in group A and G being me-dium risk, which fails to capture the high-risk status of group A herbicides. L122 Add the classifier to the scientific name "Chenopodium album". Done L140 Add the "that the actual" Done L142 Add comma "distribution, and abundance" Done L147 Add of "total of 100 weeds" Done, but it is now 101 weeds. L150 I would recommend adding the S1 Table to the text instead of having it as a  supporting file. Done   L152 I would recommend adding the S2 Table to the text instead of having it as a supporting file. Done We think this one should stay where it is because of its length. L179 Add the "ranked by the wheat" Done L203 Add comma "modes-of-action, and" Done L234 Add comma "Without this list, we" Done L234 replace are with a "As a result" Done L240 Add comma "cases in wheat, and" Done L246 Add s "indicates elevated" Not done. "may indicate" is correct L247 Add comma "surprising when" Not done. Comma is correct L256 delete "Clearly" Done L275, 281, 284 Add hyphens to "spring-sown cereals" Done, "on-field" Not done the acting on field experience" refers to experience gained in the field, "high-risk" Done L289"because of the number" Not done. This makes sense "This was necessary in part be-cause the number of cases for different herbicide groups has increased since the Moss arti-cle was published." L290 have instead of has "groups have increased" Not done. This makes sense "This was necessary in part because the number of cases for different herbicide groups has increased since the Moss article was published." L296"because of worldwide" Not done, it makes sense the way it is.
L308 Surprisingly, there have Done L324 Add commas "may, in fact, implicate" Done Dear PLOS ONE editors (Ahmet Uludag, Ph.D.), Thanks for shepherding our manuscript "A herbicide resistance risk assessment for weeds in wheat and barley crops in New Zealand" through the review process. We are happy to provide you with a much-improved manuscript. Reviewer 1 seemed to focus on specific concerns about aspects of the methods. After a careful re-read of the relevant parts of the article we think those concerns are addressed in the article. We found the comments of Reviewer 2 (Dr. Stephen Moss, whose protocol we adapted) were very helpful, and because of his suggestions we reorganised the discussion substantially and adopted most of his other recommendations. Reviewer 3 also provided helpful suggestions that were mostly adopted. We provide the following line-by-line responses to the reviewers. The most significant of these was to move both the supplemental data files into the main text. We did this for one of them but left the species list as a supplement. It's a big table, so we would defer to your judgement about including it. In the "Response_To_Reviewers.doc" document reviewer comments are followed by our response in bold italics. We also added Solanum americanum, a species we should have included in our assessment bringing the number of species assessed up to 101relevant updates were made throughout the text.
We do not think the text needs to be changed with respect to ZN and CEB affiliations. If you think it needs to be changed it could include this information: AgResearch Ltd is a crown owned research institution doing science research businesses but owned by the Crown (i.e. the Government) in New Zealand. ZN and CEB affiliation to AgResearch Ltd. does not alter our adherence to PLOS ONE policies on sharing data and materials. The funding agency is also the main public science funding organization in New Zealand and provided financial support in the form of authors' salaries and/or research materials. They did not play a role in the study design, data collection and analysis, decision to publish, or preparation of the manuscript.
We have had one prior interaction with PLOS about this manuscript when we first submitted itour previous cover letter was dated 6 April 2020. herbicide resistance cases are rare, and often focus on one or two problematic species in a given crop, 54 for example, Alopecurus myosuroides Huds. in French wheat fields [18] or Avena fatua L. in two 55 Canadian townships [19]. Systematic surveys may be rare because of the cost, a lack of specific 56 pathways for reporting [11], and industry perceptions about the importance of resistance. Surveyors 57 should ideally be open to the discovery of completely novel cases while also being aware of those 58 that are most at risk of developing resistance. 59 Here we adapt a recently published risk assessment protocol [20] to identify those weeds most 60 at risk of developing resistance in New Zealand, given their occurrence in wheat and barley fields, 61 and their prior record of resistance in wheat and barley fields elsewhere in the world. This Moss et al. 62 protocol [20] set out to assess resistance risk as part of a pesticide authorization process in Europe, 63 based on a European Plant Protection Organization (EPPO) protocol, originally developed in 1999 64 [21]. They present a quantitative risk matrix using both herbicide-risk (some herbicides pose a higher 65 risk than others) and species-risk (some weed species are more resistance-prone than others), with an 66 optional score modifier designed to account for agronomic management practices that may reduce 67 the risk. We took advantage of a unique data set about herbicide use in wheat and barley fields in 68 New Zealand to place our risk assessment into context, and construct a framework for herbicide 69 resistance surveys and extension efforts in the New Zealand cropping industry. This risk assessment 70 is on an industry-wide scale informed by anonymized herbicide application data from wheat and 71 barley fields. Risks were not assessed at the scale of individual farms and fields, this requires detailed 72 information about herbicide timing, mixtures and rotations, and their interactions with weed biology, 73 crop rotations and other cultural practices. All the high-risk weeds identified here should be targeted 74 in surveys designed to detect herbicide-resistant weeds. 75

76
Weed list 77 We generated a list of potential target weeds from wheat and barley crops in New Zealand in the same risk category, with a difference in the numbers of cases of just worldwide we believe they 100 are indistinguishable from the data. The alternative is to use the same threshold as in the Moss 101 protocol, but this would result in group A and G being medium risk, which fails to capture the high-102 risk status of group A herbicides. There are different ways we can reasonably set risk thresholds, 103 which are discussed later. Remaining ranking thresholds were not changed: medium risk 5-9%, low 104 risk 1-5% and very low risk <1%. Low-risk and very low risk are both scored as '1'. 105 arable farmers have registered to use the platform, but anonymization was complete, with no unique 110 identifiers for farms or fields. Farmers recorded every spray event (by herbicide product) in their 111 fields. For example, an individual active ingredient used three times in a field is recorded three times. 112 Products with multiple active ingredients were recorded as independent applications. Counts of 113 herbicide use in fields were summarized by active ingredient and legacy HRAC [26] herbicide mode-114 of-action, from product label information. Relative rates of use by mode-of-action were quantified 115 and characterized as very high (>20% of all application instances), high (>10%), moderate (>1%), 116 low (~1%), extremely low (<1%) and nonexistent (0%). 117 118 We assume that the best way to predict resistance in a weed species to any given herbicide (by 119 HRAC group) is proportional to the number of documented cases of herbicide resistance in the 120 same taxon, given the use of the same herbicide type in New Zealand. in the world. We include all cases of resistance for each weed species, rather than restricting our 130 focus to cases from wheat and barley, because we are interested in a species propensity to develop 131 5 of 18 resistance to a herbicide group. For example, the high-risk species Chenopodium album L. has more 132 than 10 documented cases of resistance giving it a species score of 3. Then we consider the 133 herbicide-risk scores for those herbicides where Chenopodium album has evolved resistance 134 somewhere in the world. There were cases in two high-risk herbicide groups B and C1 (each with a 135 herbicide score of 3), and one medium-risk group O (herbicide score of 2). The species and 136 herbicide scores are multiplied and summed (3×3) + (3×3) + (3×2) = 24. We distinguished cases 137 that were in herbicide groups highly-used (or not) by wheat and barley farmers in New Zealand. 138

Cases of resistance by taxon and risk scoring
The summed (cumulative) scoring method described above is not used in the European Moss 139 protocol because its purpose was to regulate herbicide product use [20]. In contrast, we wanted to 140 determine the risk that different herbicides will select for resistance in weed species known to occur 141 in New Zealand's wheat and barley fields. Ultimately we hope to inform sector stakeholders about 142 risk, and to improve herbicide resistance detection. We adapted their protocol to score species-risk in 143 an industry-wide assessment. Unlike Moss et al.
[20] we used an explicit threshold (though arbitrary) 144 to determine species-risk scores based on the number of cases of resistance worldwide, but we think 145 this approach produces credible risk estimates in the light of current knowledge. We examined 101 146 species and ended up with 46 'high' and 'medium' resistance risk species, many more than Moss et 147 al (they scored an example list of only 13 high and medium risk taxa). The Moss protocol also used 148 score modifiers that take into account resistance management practices including the use of non-149 chemical control measures. We did not use the score modifiers since these vary from field to field 150 and our objective was slightly different. We acknowledge that actual risk of resistance development 151 is determined mostly by the frequency and type of herbicide applied (selection pressure) interacting 152 with characteristics of weed biology, distribution and abundance [5]. All graphs were created in R 153 using the ggplot2 package [28,29]. 154   Zealand wheat and barley fields, here ordered by observations in wheat crops. The herbicides that 204

Results
were not ranked in the top ten for each crop respectively were included here for both of the crops for 205 completeness. 206

of 18
Cases of resistance by species and risk scoring 207 The documented cases of herbicide-resistance for medium and high-risk species and all herbicide 208 mode-of-action combinations are shown in have twenty or more recorded cases that occur in five or more unique mode-of-action groups each. 218 Amaranthus powelii Avena sterilis Fallopia  The cumulative risk scores shown in Fig 2 gives a higher score to weeds with resistance to 231 multiple modes-of-action. Because of this species with high overall risk scores may have relatively 232 few cases of resistance detected overall but their score is inflated by cases of resistance to multiple 233 HRAC modes-of-action. For example Lolium perenne (15 cases in 5 modes-of-action) has a slightly 234 lower risk score overall compared to its congener Lolium multiflorum (130 cases in 7 modes-of-235 action). Chenopodium album is an example of a relatively low scoring weed that has had 49 cases of 236 resistance documented but within only HRAC modes-of-action (Fig 2, Table 3 as amongst the least likely to select for resistance [29], and is recognized as medium-risk by Moss. 295 We ranked glyphosate (group G;  (Table 3). Groups F1 and K3 have a relatively low risk of developing 327 resistance based on historical occurrences even though they are highly used in New Zealand. 328 Examining our herbicide use information in the context of published herbicide evolution models 329 can provide important insights. Models based on dryland wheat systems and the weed Lolium rigidum 330 [39] showed that resistance rate evolution was not slowed by simple herbicide rotations (i.e. annual 331 with few herbicides). Importantly the use of soil-applied herbicides, particularly trifluralin (group 332 K1), full-rate mixes of herbicides, and complex 8-year long rotations were shown to delay resistance 333 evolution by years, and in some scenarios by decades [39]. There is a chance that resistance cases in 334 soil-applied pre-emergent herbicides are under-reported compared to post-emergent ones (they are 335 harder to test). For now, resistance evolution in those herbicides appears to be slower. Trifluralin did 336 not feature amongst the most common herbicides in the farmer herbicide use data we obtained for 337 2017 and 2018 (<0.5% of field applications); the only frequently used pre-emergent herbicides were 338 flufenacet and terbuthylazine. Farmers should be informed about the high-risk species and herbicide 339 combinations, so as to avoid high-risk behaviors, or at least keep an eye out for problems they will 340 select for. Farmer decision support platforms and research and extension efforts in New Zealand 341 should emphasize mechanical and cultural control measures, the use of soil-applied pre-emergent 342 herbicides, full-strength label-rate herbicide mixtures, crop rotation to utilize herbicides otherwise 343 unavailable and herbicide rotations of key active ingredients to achieve maximum control and to 344 reduce the rate at which herbicide resistance evolves in weed populations. 345   (some herbi-cides pose a higher risk than others) and species-risk (some weed species are more 73 resistance-prone than others), with an optional score modifier designed to account for agronomic 74 management practices that may reduce the risk.They present a quantitative risk matrix using ranked 75

Conclusions
herbicide-risk and species-risk with an optional score modifier designed to account for agronomic 76 management practices that may reduce the risk. We took advantage of a unique data set about 77 herbicide use in wheat and barley fields in New Zealand to place our risk assessment into context, 78 and construct a framework for herbicide resistance surveys and extension efforts in the New Zealand 79 cropping industry. This risk assessment is on an industry-wide scale informed by anonymized 80 herbicide application data from wheat and barley fields. Field and farm scale risk is not assessedRisks 81 were not assessed at the scale of individual farms and fields, this requires detailed information about 82 herbicide timing, mixtures and rotations, and their interactions with weed biology, crop rotations and 83 other cultural practices. All the high-risk weeds identified here should be targeted in surveys designed 84 to detect herbicide-resistant weeds. 85

86
Weed list 87 We generated a list of potential target weeds from wheat and barley crops in New Zealand Table Table S12 , with risk scores of 1, 2 or 3 given for low, medium or high risk respectivelywith 108 risk scores of 1, 2 or 3 given for low, medium or high respectively. We set the threshold for high-risk 109 at >9% of recorded cases, which captures the original high-risk categories identified in the Moss 110 protocol, but now places group G (glyphosate) in the high category. Moss  Approximately 900 arable farmers have registered to use the platform, but anonymization was 124 complete, with no unique identifiers for farms or fields. Farmers recorded every spray event (by 125 herbicide product) in their fields. For example, an individual active ingredient used three times in a 126 field is recorded three times. Products with multiple active ingredients were recorded as independent 127 applications. Counts of herbicide use in fields were summarized by active ingredient and legacy 128 HRAC [26][25][24] herbicide mode-of-action, from product label information. Relative rates of use 129 by mode-of-action were quantified and characterized as very high (>20% of all application instances), 130 high (>10%), moderate (>1%), low (~1%), extremely low (<1%) and nonexistent (0%). The most 131

of 22
used herbicides for each crop were characterized by weighting active ingredient amounts and the 132 number of fields they were applied to. 133 134 We assume that the best way to predict resistance in a weed taxon species to any given 135 herbicide (by HRAC group) is proportional to the number of documented cases of herbicide 136 resistance in the same taxon, given the use of the same herbicide type in New Zealand. species had cases of resistance documented somewhere in the world (Table S1Table 1). We include 148 all cases of resistance for each weed species, rather than restricting our focus to cases from wheat 149 and barley, because we are interested in a species propensity to develop resistance to a herbicide 150 group. For example, the high-risk species Chenopodium album L. has more than 10 documented 151 cases of resistance giving it a species score of 3. Then we consider the herbicide-risk scores for 152 those herbicides where Chenopodium album has evolved resistance somewhere in the world. There 153

Cases of resistance by taxon and risk scoring
were cases in two high-risk herbicide groups B and C1 (each with a herbicide score of 3), and one 154 medium-risk group O (herbicide score of 2). The species and herbicide scores are multiplied and 155 summed (3×3) + (3×3) + (3×2) = 24. We distinguished cases that were in herbicide groups highly-156 used (or not) by wheat and barley farmers in New Zealand. 157 (though arbitrary) to to determine species-risk scores based on the number of cases of resistance 171 worldwide, but we think this set species-risk scores based on the number of cases (see methods) of 172 6 of 22 resistance but we think this approach produces credible risk estimates in the light of current 173 knowledge. We examined 1010 species and ended up with 465 'high' and 'medium' resistance risk 174 species, many more than Moss et al 45 high and medium risk species, much more than Moss et al. 175 (they scored an example list of only 13 high and medium risk taxa). The Moss protocol also used 176 score modifiers that take into account resistance management practices including the use of non-177 chemical control measuresThe Moss protocol also used score modifiers that take into account 178 resistance management practices. We did not use the score modifiers since these vary from field to 179 field and our objective was slightly different. We acknowledge that actual risk of resistance 180 development is determined mostly by the frequency and type of herbicide applied (selection pressure) 181 interacting with characteristics of weed biology, distribution and abundance We did not use the score 182 modifiers since these vary from field to field. We acknowledge that actual risk of resistance 183 development is determined mostly by the frequency and type of herbicide applied (selection pressure) 184 interacting with features of weed biology, distribution and abundance [5]. All graphs were created in 185 R using the ggplot2 package [28,29]. 186 galli, Erigeron spp. and Raphanus raphanistrum are notable emerging weeds, so they were included. 193

Weed list
Some taxa noted by Bourdôt were resolved to species, such as Trifolium spp., which became Trifolium 194 repens and Trifolium pratense. Sixteen high-risk and 29 moderate-risk species are shown in Table  195 S1Table 1, and the other low-risk species are discussed below. A full list of weeds considered for 196 New Zealand wheat and barley crops is displayed in Table S2Table S1. 197

Ranking herbicide groups by resistance cases
198 Herbicide mode-of-action risk rankings are displayed (Table 1Table  and wheat have shared patterns of herbicide usage with respect to mode-of-action (Table 2Table 23)  214 and active-ingredients. Synthetic auxins (HRAC group O) were represented in higher proportions 215 than any other class of herbicide (a total usage rate of 26%). ALS-inhibitors (B), PDS-inhibitors (F1) 216 and EPSPS-inhibitors (G) were highly used herbicide groups with >10% total usage, and acetyl 217 coenzyme-A carboxylase inhibitors (A) and photosystem-II disrupters (C) were moderately used 218 (total <10%). EPSPS-inhibitors (G) were used in larger proportions in barley (18%) versus compared 219 to wheat (12%); alternatelyconversely, farmers used K3 herbicides significantly more in wheat at 220 12% compared to barley at 4%. Glyphosate (MoA group G) is mostly used (>95%) used to control 221 weeds pre-sowing of the cereal crops, for termination of the previous crop or pre-establishment weed 222 control. It is very rarely used as crop pre-harvest desiccant. 223  The ten most used herbicides as a percentage of total application instances documented by farmers in 239 New Zealand wheat and barley fields, here ordered by observations in wheat crops. The herbicides 240 that were not ranked in the top ten for each crop respectively were included here for both of the crops 241 for completeness. 242

243
The documented cases of herbicide-resistance for medium and high-risk species and all herbicide 244 mode-of-action combinations are shown in Table S1Table  and Avena fatua have many twenty or more recorded cases in that occur in five or more unique mode-254 of-action groups eachfive or more unique mode-of-action each (>20). 255   Cumulative The cumulative risk scores depicted shown in (Fig 2) weight species that have high 269 "intrinsic" species-risk scores, with cases in high-risk modes-of-action and showgives a higher score 270 to weeds with resistance to multiple modes-of-actionmodes of action. A large proportion of the 271 cumulative species risk is explained by multiple cases within high-use modes-of-action. We see that 272 someBecause of this species with high overall risk scores may have have fewerrelatively few cases 273 of resistance detected overall but their score is inflated by cases of resistance to multiple HRAC 274 modes-of-action. For , for example Lolium perenne (15 cases in 5 modes-of-action) has a slightly 275 lower risk score overall compared to its congener Lolium multiflorum (130 cases in 7 modes-of-276 action) but significantly fewer cases. In New Zealand at least, these species freely hybridize, making 277 identification difficult. Species-risk scores emphasize cases where resistance to multiple modes-of-278 action have been found. Chenopodium album is an example of a relatively low scoring weed that has 279 had many 49 cases of documented resistance documented but in within only few HRAC modes-of-280 action (Fig 2,   axis is the sum of the herbicide-risk × species-risk scores (from Table S1Table  but when sowings get delayed because of wet spring weather then spraying will impact these 320 seedlings. Farmer implementation of integrated weed management will vary at a fine spatial scale 321 too, impacting selection pressure. Nevertheless, the information we gathered provides a useful 322 indicator wheat and barley weeds with a known propensity to develop resistance. By being aware of 323 all the weed species that are high-risk we shoulds improve detection of resistance cases in future. By 324 acting on field experience herbicide-resistance researchers may be biased toward the most frequent 325 survivors of weed management, and may not sample less abundant but high-risk species. Farmers 326 should be informed about the high-risk species and herbicide combinations, so as to avoid high-risk 327 behaviors, or at least keep an eye out for problems they will select for. The unique data we obtained 328 about actual herbicide use (Table 2) will also help us to prioritize herbicides for testing of weeds that 329 we will collect in future surveys designed to detect herbicide resistance. 330 Estimating the overall prevalence of herbicide resistance in all the major farming sectors in New 331 Zealand could cost $1-3 million NZD depending on sampling rates [8]. An obvious concern is that 332 detection rates for herbicide-resistant weeds will necessarily underestimate the true rate, given that 333 surveyors Reviewer #1: Because frequency and abundance of weed species in the region have not been used in the scoring, the paper provides a picture of risks due to the current use of herbicides in the region, which is informative for the farmer about the intrinsic risk of an abundant species in its own farm. However, although I agree with intrinsic high risk species as stated on line 280-285, it is pity that the overall risk for the region is not presented.

L50-54
There is also an important survey on blackgrass in UK (see Hicks et al Nature Ecology & Evolution 2(3) 2018, DOI: 10.1038/s41559-018-0470-1). Probably other reports are available on surveys developed by national weed research associations in Spain, France (Columa), Italy, etc., including companies, extension service and farmers, and published in local conferences.

That is another example of the "focus on one or two problematic species in a given crop" issue we highlight, and therefore does not need to be added.
L142 It can be assumed that weed biology is already included in the inherent species score, but abundance is a key parameter. Abundance includes effects of farming systems: if many individuals still remain in the field, then the risk of developing resistance is higher than if the population is small. A score modifier to account for variation of abundance (e.g. 1 for low density and 3 for more than a million plant/ha, or another scale) could be used. Another factor is the species frequency in the studied region, which doesn't seem to be taken into account (is the Production Wise platform collecting main troublesome weed species?).  Table 3.

Reviewer #2 Dr Stephen Moss:
Abstract: Generally fine but as likely to be most-read part I suggest a few minor changes. Do you need to say 'wheat and barley' or would 'cereal' suffice? Ditto in title.

We think wheat and barley is a more accurate statement for our case since the herbicide data focuses on that. Wheat and barley is more widely grown than oats, and depending on the reader, corn would also be considered to be a cereal…
I suggest:

15-18
'We estimated the risk of selecting for herbicide resistance in 100 weed species known to occur in wheat and barley crops on farms in New Zealand. A protocol was used that accounts for both the risk that different herbicides will select for resistance and each weed's propensity to develop herbicide resistance based on the number of cases worldwide.' Done 24 'ALS-inhibitors were assessed as posing the greatest risk for more species than any other mode-of-action. Done 25 I prefer 'Pre-emergence' but that may depend on journal style. Done

26
'……. in this class commonly used by…..' Is this better than 'favoured'? Done Somewhere I think it would be worth stressing in abstract that 7 out of the highrisk 10 are grass-weeds. If space is limiting, I think the last sentence of abstract could go or 'farmer extension efforts' incorporated into previous sentence. Done Introduction: Good -acceptably concise and relevant. I suggest including some brief (one or two sentences) information on arable cropping in NZ and what proportion of that is accounted for by wheat and barley. Or at least to show that wheat and barley comprise a significant proportion of arable crops -you might even wish to mention that NZ currently holds world record for both barley and wheat yield/ha (I think that is correct). I would! Done Suggest: ' ….45 'high' and 'medium' resistance risk species, many more than Moss et al……….' Done 139 Suggest: 'The Moss protocol also used score modifiers that take into account resistance management practices including the use of non-chemical control measures' . Done 139-142 Suggest: 'We did not use the score modifiers since these vary from field to field and our objective was slightly different. We acknowledge that actual risk of resistance development is determined mostly by the frequency and type of herbicide applied (selection pressure) interacting with characteristics of weed biology, distribution and abundance [5].' Done Results: Generally good and concise. One key paragraph needs improvement to improve clarity  Table 2 Suggest last 10 categories are simply summarised to save space and some statement added within table. e.g. 'The following 10 HRAC MoA groups each accounted for less than 1% of herbicide applications, N,

202-215
This paragraph covers the key output of this paper but is confusing and really needs a thorough re-write. Expand if necessary, to clarify results. Fig 2 contains a lot of information. I did wonder how much the 'cases' adds to this but I can see that it is relevant. Some definition of 'cases' is needed, as mentioned earlier. This is tricky as not easy to explain in a few words, but if this is explained in the M&M then no need to explain again, except perhaps to say refer to M&M. Should be made clear in Figure that cases is worldwide, as might be interpreted by casual reader as in NZ. Done Figure 2

Resistant Weeds as unique combinations of weed species and HRAC herbicide mode-ofaction (species x site of action)."
212-215 I would make the point that 7 out of 10 are grass-weeds. Grass-weeds are overrepresented in global cases of resistance. Some comment on this in discussion maybe. Done Discussion: The paper would benefit from a more concise discussion with improved, more logical structure. Content is OK but lacking in focus and a bit rambling. I suggest reducing it to about 60% max of current length by heavily reducing some paragraphs or omitting altogether. The focus should be on the lessons and implications from what you have presented in the paper. Some of Discussion, while valid, reduces the impact of the paper rather than enhances it. 288-306 Again lacks focus. Glyphosate comments relevant. But to make some suggestion but then say 'Estimations of these other factors can be made, but are likely to be inaccurate.' Is a case of 'shooting yourself in the foot.' Ditto 'Using Beckie's as our herbicide-risk ranking would produce different risk-scores.' Omit. Several sentences deleted and reworded to avoid "shooting ourselves in the foot" but keeping the main ideas as a reasonable caveat around interpretation of our results.

226-235
Other paragraphs -consider what content really adds to paper and what can be omitted.
One aspect that seems missing, is some idea of the crop rotations used on NZ farms and the impact this might have. This focus in this paper is on wheat and barley but the frequency of growing these crops and consequent herbicide use will surely impact on the resistance risk? How commonly is arable cropping rotated with grass? Are all-arable farms common compared with mixed farms? This may explain why herbicide resistance is relatively uncommon in NZ. Also relevant to risk at individual field level which lends itself nicely to comments about where you should set priorities for resistance monitoring, management and farmer KT. No need to go into great detail, but is relevant. We have included some text about rotations.
Conclusions: Generally fine but suggest the following.

343-346
Suggest: 'A European protocol [17], designed primarily to assist in herbicide authorization procedures, was adapted to assess the risk of herbicide resistance evolving in 100 different weed species known to occur in New Zealand wheat and barley crops. More than half the weeds (55) we assessed were low-risk, 29 were medium-risk and 16 high-risk. The 10 species posing the highest resistance risk were: etc etc' Done

349-351
Final two sentences could usefully be a bit 'punchier'. 'Suggest: 'We are planning extensive surveys in New Zealand to detect new cases of herbicide resistance. The risk assessment outlined in this paper will enable us to prioritise those weeds identified as posing a high resistance risk and, consequently, make better use of available resources. The risk assessment procedure as described in this paper has the potential to be a very useful tool for evaluating the risk of herbicide resistance in a wide range of different weed species in other countries too'. Done