Field testing of a physical impact mill in the Canadian Prairies

Abstract Herbicide resistance in western Canada has increased interest in alternative weed management strategies. Physical impact mills, a form of harvest weed seed control, have been identified as a strategy that may be well suited for Canadian use. The efficacy of the Harrington Seed Destructor, a physical impact mill, was evaluated in 20 producer fields in Alberta on a wide spectrum of weed species over 3 years. Significant differences in weed densities between the physical impact mill treatment and the regular harvest densities were few; however, some population density reductions were observed. Measurable reductions in weed densities may have been limited by the short timeframe of the experiment, the high initial densities of the weeds, or the targeted weed species having dormancy or longer term seedbanks. Additionally, identified knowledge gaps on how best to optimize physical impact mill efficacy may have reduced the efficacy of the physical impact mill in this study. This study showed no conclusive evidence for the efficacy of a physical impact mill on tested weed species under field conditions. However, it did provide a number of important considerations for future studies.


Introduction
Increasing prevalence of herbicide-resistant weeds has increased the need for alternative weed management strategies (Norsworthy et al. 2012;Beckie et al. 2020).Harvest weed seed control (HWSC) is a unique alternative weed management method that prevents dispersal of weed seeds by the combine during harvest.Instead, utilizing their capture by the combine, weeds are collected, destroyed, or removed from the field (Walsh et al. 2018).Harvest residues are typically separated into the grain, which is kept, or the chaff (finer residue) and the straw, which are expelled (Walsh et al. 2018).Weed seeds can be found in all three fractions but are often concentrated or assumed to be concentrated in the chaff fraction (Walsh et al. 2018).While HWSC was initially developed in Australia (Walsh et al. 2012), research is now ongoing in Europe, the United States, and Canada (Walsh et al. 2018), with low levels of adoption in all three regions.For HWSC to be effective, weeds must retain their seeds above the combine cutting height until crop maturity, and the seeds must be effectively destroyed by the HWSC method of choice (Walsh and Powles 2014;Walsh et al. 2018).Numerous HWSC methods exist, including chaff carts, narrow windrow burning, chaff lining and chaff tramlining, bale-direct systems, and physical impact mills (Walsh et al. 2018).Each of these methods has its own advantages and disadvantages related to costs, nutrient sequestration or residue removal, destruction versus concentration of weed seeds, and time, labor, and additional equip-ment requirements.Best fit tactics will vary by region, farm type, and typical farm practices already in use on the farm.
Canadian research has indicated promise for HWSC technologies to be used in Canadian agriculture.While low levels of seed retention mean reduced likelihood of density reductions in some species via HWSC, other species have been identified with high levels of seed retention, suggesting potentially high efficacy of HWSC methods (Burton et al. 2016(Burton et al. , 2017;;Beckie et al. 2017;Tidemann et al. 2017a).The most applicable method of HWSC for Canadian producers is the use of a physical impact mill, also known as a seed impact mill.This allows the destruction of weed seeds, retention of crop residues, and a logistically practical add-on to the combine (Walsh et al. 2018).Stationary mill unit testing has demonstrated high mortality levels of Canadian weed seeds that pass through the mill system (Tidemann et al. 2017a(Tidemann et al. , 2020)).While this background research has indicated a high likelihood of success in controlling many problem weeds on the Canadian Prairies, these results have not yet been validated in producer fields.Therefore, the objective of this study was to evaluate the efficacy of a physical impact mill in reducing weed densities in producer fields in central Alberta.

Materials and methods
Twenty producer fields with dense weed patches were identified within a 50 km radius of the Agriculture and Agri-food Canada, Lacombe Research and Development Center (Fig. 1).Weed patch size varied in these fields from a minimum of 7553 m 2 (0.75 ha) up to a maximum of 75 809 m 2 (7.5 ha), with an average patch size of ∼21 991 m 2 (2.1 ha).Numerous weed species were found in these patches, allowing testing of the physical impact mill across a broad spectrum of weeds (Table 1).Each patch was broken into three replicates with two treatments per replicate.Plot sizes therefore ranged from a minimum of 1259 m 2 to a maximum of 12 635 m 2 with an average plot size of ∼3665 m 2 .The two treatments used in this study were harvest with a physical impact mill on or harvest with the physical impact mill by-passed ("regular" harvest).The physical impact mill used in this study was the original tow-behind Harrington Seed Destructor (HSD) (Walsh et al. 2012) (Fig. 2).While physical impact mills that are integrated into combines are now available (e.g., iHSD, Seed Terminator, or Redekop SCU), at the time this study was planned only the HSD was available.Research conducted in Australia has demonstrated similar levels of efficacy between the integrated mills and the original tow-behind mill unit (Walsh et al. 2012(Walsh et al. , 2017)), meaning that these results would be broadly applicable to the more modern and recent integrated innovations.To use the tow-behind HSD, an interface was custom designed by PAMI (Humboldt, Saskatchewan) to connect the unit to a John Deere S690 combine.The interface allowed easy switching between the HSD and HSD by-passed modes for the treatments (Fig. 3).In both cases, the straw moves along the bottom conveyor, through the unit without any processing, and to the straw spreaders located at the back of the HSD.During the HSD treatments, the chaff was directed through the mill itself where it was expelled at the back with enough force to ensure spread.During the regular harvest treatment, the chaff was also directed onto the straw conveyor and would have been spread evenly across the harvest area by the straw spreaders.The mechanism to switch between treatments was adjusted between 2017 and 2018 harvests (Fig. 3) due to logistical complications (machinery and HSD interface impacting one another on field hills); however, this is not expected to have altered the efficacy of the HSD treatments, as the mill was still fully by-passed on non-impact mill treatments.The rigid tube structure shown in Fig. 3C was used in the 2018 and 2019 harvests.
This study was conducted between 2017 and 2020, with treatments applied at harvest in 2017, 2018, and 2019, and weed population density assessments in 2017, 2018, 2019, and 2020.Treatments were applied in the same location in each year to determine the cumulative effect of the physical impact mill system over time in comparison to a regular harvest.Weed density assessments were conducted by species prior to study initiation across a transect of the study area to identify which species were present and assess densities.Treatments were applied at crop harvest each fall.Weed Table 1.Dominant weed species in the 20 producer fields that were statistically analyzed for the impact of the physical impact mill treatment on their densities.Warwick (1979) Note: Weeds are listed by their Latin and common names.Characteristics limiting Harvest weed seed control (HWSC) potential are given with references where possible.N/A is recorded when no seed retention studies have been conducted and there are no obvious defining biological characteristics or data prior to this study.densities were assessed each spring prior to in-crop herbicide applications through seedling counts in 15 0.5 m 2 quadrats per plot in a "W" pattern.Distances between quadrat counts varied by plot to ensure all 15 could fit within the variable plot sizes; these distances were not measured but estimated by each density assessor as they counted.Plot edges were avoided to minimize any crossover effect between plots and treatments.
In the final assessment year, 20 soil cores of 10 cm diameter to a 5 cm depth were taken across each plot to quantify the weed seedbank.Soil cores were washed through a 250 μm screen to remove most of the soil volume and sieved to remove large particles such as crop residue.The remaining material was mixed in a 55 cm × 28 cm × 1.3 cm tray with approximately 3 L of potting soil (JiffyMix, Professional Gardener, Calgary, AB) mixed with fertilizer (Harrell's Pro-Fertilizer (14-14-14: N-P 2 O 5 -K 2 O), Lakeland, FL), covered by an additional 1.5 L, and placed in a growth chamber with a 16 h light/8 h darkness photoperiod and 20 • C/12 • C temperature regime.Trays were watered daily and any emerged seedlings identified and removed weekly.After a 3-week growth period, trays were placed into a −18 • C freezer for 3 weeks for cold stratification.This was then followed by a 3-week growth period under LED grow lights (Monios-L T5 Grow Lights 120 W, Monios-L, online) with a 16 h/8 h light:dark period in a temperature-controlled room at ∼25 • C. Trays were again assessed weekly and emerged seedlings were identified and removed.This was followed by a minimum of 6 weeks in a −18 • C freezer before a final 3-week growth and assessment period, again under the germination lights.If no germination was observed during the second growth cycle, samples were discarded and did not go through this final growth period.Following each cold stratification cycle, potting soil and soil core material in the trays were mixed and disturbed to help stimulate germination.At the end of the third growth cycle, germination of additional weeds was rare and so growth cycles were terminated.The initial soil sieving process also intentionally removed larger weed seeds such as those of Avena fatua (wild oat), which were identified in the screenings and counted by hand.Due to dormancy in wild oat seeds, this method was utilized to more accurately assess the number of wild oat seeds in these samples.Wild oat viability was not assessed; however, only structurally sound seeds were counted in these samples.
While weeds were identified in both seedling assessments and seedbank assessments to species, analysis focused on the dominant species in each field as entire populations (seedlings plus seedbank densities).This included the top five most abundant weeds in each field and/or those present in densities of ≥50 plants m −2 .Densities were analyzed in Proc Glimmix of SAS 9.4 (SAS Institute Inc. 2023) using negative binomial distributions with treatment as a fixed effect and rep(Field) as a random effect.Analyses were conducted across all fields with that weed species identified as dominant, as well as within each field where the weed was dominant.An exploratory principal component analysis (PCA) was also conducted in R version 4.2.1 (R Core Team 2021), using RStudio 2022.12.0, to investigate the relationship between the various weed species and the treatments for both above-ground seedlings and the weed seedbank.In this case, the entire population was broken into its above-and below-ground components to investigate whether either of those components was showing an impact of the treatments that were maybe being masked by the contribution of the other.The above-ground seedlings for all three spring density assessments were included in the PCA, while the seedbank was limited to the only density assessment conducted in 2020.PCA biplots were created using the factoextra package (Kassambara and Mundt 2020).A parallel analysis was conducted using the paran package in R to determine the number of principal components to retain (Dinno 2018); however, only the first two principal components are visualized.In addition, an analysis of multivariate homogeneity of group dispersion was conducted using the betadisper function in the vegan package in R (Oksanen et al. 2022) for the seedling and seedbank PCAs, separately.An Analysis of variance of the homogeneity of dispersions was conducted in addition to a permutation test with 999 permutations.A PERMANOVA was conducted utilizing the adonis2 function in vegan (Oksanen et al. 2022) incorporating a blocking factor by field (strata = field) and 999 permutations.

Results
Unsurprisingly, in a group of 20 fields and 13 farmers, cropping rotations, harvest management practices, and herbicide strategies varied dramatically (Table 2).Weed population data were obtained on all 20 fields throughout the study period, with one field having harvest treatments implemented only twice due to the growth of hemp in 2018 (Table 2).Thus, we were able to track weed populations in all fields over the intended target period.
There were limited significant differences (p < 0.1) between populations after 3 years of harvest treatments with and without a physical impact mill.When tested across all fields containing weed species of interest, only two instances of significance were observed (Table 3).Across fields with volunteer canola (Brassica napus) and sowthistle species (Sonchus spp.), populations of those weeds were higher in the physical impact mill treatments (Table 3).There were more instances of significance when investigating weed populations in individual fields, which is logical when analyzing across fields combines many cropping rotations and harvest management practices.There were 5 fields out of the 20 where significant differences between physical impact mill treatments were observed for specific species.In field 2, wild oat populations were increased in the HSD treatment, while hempnettle (Galeopsis tetrahit) populations were reduced in the HSD treatment (Table 3).In field 8, false cleavers (Galium spurium) and hempnettle populations were reduced in the HSD treatment, while sow thistle species populations were higher in the HSD treatment (Table 3).In field 15, wild buckwheat (Fallopia convolvulus) and chickweed (Stellaria media) populations were lower in HSD treatments (Table 3).Finally, in field 16 and field 17 wild oat and false cleavers, respectively, were lower in the HSD treatments.
In the PCA, the first two principal components described 11.5% and 10.5% variation, respectively, for the above-ground seedling densities and 13.5% and 10.8% variation of the seedbank populations (Figs.4A and 4B).There was no differentiation between the treatment ellipses that were overlapping; however, they did demonstrate slightly different shapes.The parallel analysis suggested retention of six principal   components for the seedling data and two principal components for the seedbank data.The homogeneity of group dispersion tests was not significant for the seedling (p = 0.97) or the seedbank (p = 0.87) densities.Additionally, the PER-MANOVA test indicated no significant difference between treatments (p = 0.954).Overall, the PCA showed no difference between the treatments in both the above-ground and seedbank analyses.

Discussion
Weed species present in the study made up a good crosssection of problematic and common weeds to the local area (Leeson et al. 2019).Fields were selected that had relatively high densities of weeds that had varying potential success for HWSC.For example, fields with volunteer canola and cleavers were selected to ensure inclusion of weeds that were expected to be good targets for impact mills (Tidemann et al. 2017a(Tidemann et al. , 2017b) ) and would show responsiveness to the HSD treatment.The study included weeds such as wild oat (high seed dormancy and low seed retention) (Tidemann et al. 2017b) and chickweed (grows below cutting height) that were expected to be less responsive to the HSD treatment due to their biology not matching well with HWSC tactics.Limited impact of the physical impact mill across all species was unexpected.The lack of efficacy on volunteer canola and sow thistle populations was particularly unexpected for volunteer canola (Tidemann et al. 2017b).It is less surprising for sow thistle that has been shown to have low seed retention in producer fields (Beckie et al. 2017) as many seeds are wind dispersed before harvest.Annual and perennial sowthistle species were not differentiated due to challenges in identifying them with confidence to species at 1-2 leaf stage across a number of people and a number of samples.Perennial sowthistle populations have a low likelihood of control with impact mills due to their perennial nature and lack of reliance on seed to maintain density (Mohler et al. 2021).Many of the weeds where significant population reductions occurred in the impact mill treatment are weeds identified as good targets for HWSC methodologies.This includes false cleavers (Tidemann et al. 2017b) and wild buckwheat (Burton et al. 2016(Burton et al. , 2017)).Also, from previous anecdotal observations of high hempnettle seed retention at maturity, it was speculated to be a good HWSC target prior to this study.Mixed results with wild oat are not surprising; while low levels of seed retention have been measured (Burton et al. 2016(Burton et al. , 2017;;Tidemann et al. 2017b), high levels of variability in wild oat seed retention have been observed in the field, and high levels of retention have also been measured (Shergill et al. 2023).Reduction in chickweed populations after impact mill treatment is surprising as chickweed typically grows below the conventional harvest cutting height.In the first year of the study in that particular field (field 15), the crop was peas.Peas are typically cut just above the soil, often due to crop lodging.It is possible that the impact on the chickweed relates to that first, low-cut crop year.
Use of pre-seeding herbicide applications was surprisingly limited by the producers in this study or significantly underreported by those producers.Most of the fields where a positive impact of the HSD was noted were fields where a preseeding herbicide application was utilized (Tables 2 and 3).The products used were primarily contact or short-lived residuals (Table 2).The small plot research on seed retention that has been conducted in Canada included a pre-seeding herbicide application strategy (Burton et al. 2016;Tidemann et al. 2017b).If, based on our sample of producers, many producers are not utilizing a pre-seeding herbicide application, seed retention values from plot studies may be optimistically high, particularly for early emerging weed species.A theoretical example of what may be happening is that in retention studies, a pre-seeding herbicide application is conducted and weed seed retention is measured on the weeds that emerge after that initial herbicide application and activity.If producers are not utilizing a pre-seeding herbicide application, it is possible that weeds are emerging earlier, maturing earlier, and retaining fewer seeds than have been measured in the seed retention studies conducted to date.It would be worthwhile to determine how frequent a pre-seeding herbicide application occurs, or what factors farmers take into consideration when making that decision.Use of a pre-seeding herbicide application may be an important strategy to optimize control of weeds using a physical impact mill.
The scarcity of weed density reductions in HSD treatments for identified good target species in western Canada was initially surprising.However, after some thought, the lack of impact should not, perhaps, be unexpected.The trial sites were in fields where weed populations were dense, and in the densest patches of weeds in those fields, as that is where the weed populations were also most consistent.However, it is also highly likely that densities in these locations were not seed limited (Selig et al. 2022), meaning that reducing seedbank inputs is unlikely to have an impact on the seedling population.We also evaluated weed densities in the seedbank; however, these patches are also where the seedbank would be the highest and yearly contributions would only be a small proportion of the total density.This study was also conducted over a 3-year time period, limited by logistical and financial constraints.However, the vast majority of the weeds included in the study form longer lasting seedbanks than the timeline of this study due to dormancy; for example, wild oat seeds last on average for 4-5 years, with a proportion of the seeds surviving for up to 9 years, with higher persistence in the soil than on the soil surface (Beckie et al. 2012).This adds to the limited impact that yearly seedbank contributions would make to the overall population size, limiting ability to measure differences in population in a 3-year timeframe.This is in contrast to the clear results observed in Walsh et al. (2013); however, in that study even those fields without HWSC incorporated showed a distinct decline in weed pressures.A shorter seedbank weed like kochia (∼90% loss of viable seed in the seedbank in 1 year) (Schwinghamer and Van Acker 2008;Beckie et al. 2018) may better demonstrate physical impact mill efficacy in a shorter term study such as this one.Investigating the crop rotations of the fields where weed population impacts occurred does not highlight an obvious reason for differences being found in those fields and not others.All fields included canola and wheat in their rotations, two of the fields (15 and 17) included peas in rotation, which does tend to be an earlier maturing crop relative to wheat and canola, and two of the fields (16 and 17) utilized swathing in 1 year as preharvest management.While earlier maturing crops and swathing do increase the efficacy of HWSC (Burton et al. 2016;Beckie et al. 2017;B. Tidemann, unpublished data), these factors were not unique to the fields where differences in weed populations were observed.It is not clear why only these field by weed combinations showed responses to the treatments.
Other factors that may have contributed to limited impact on populations would include measurements that have not been taken yet in Canada to have a full picture of weed seed movement.As an example, seed retention work in Canada has focused on retention at the time of crop maturity (Burton et al. 2016(Burton et al. , 2017;;Beckie et al. 2017;Tidemann et al. 2017b).However, it is almost inevitable that there will be additional seed losses at the combine header, due to the disturbance of the combine and header moving through the crop.With some species in the United States that loss has been measured to be up to 89% of the seeds retained at harvest (Winans et al. 2023).While the vast majority of the weed seeds (of other species) in that study were identified to be moving through the combine, the header loss component could be a big missing piece to understand which weed species can be targeted with HWSC methods in western Canada.Additionally, tracking weed seeds through the combine to determine where they are expelled (chaff as assumed, grain or straw) and how that changes with different weed species and crop combinations or various combine settings is an area to be explored.Information from Australia suggests that it is possible to optimize combine settings to collect target weed seeds in the chaff (Broster et al. 2016;Weedsmart 2018).While some of this knowledge was available prior to the current study, it was not considered when setting the combine for this study.Harvested grain was being returned to the producers; therefore, the combine was set to ensure high-quality harvested grain, without consideration of whether weed seed concentration in the chaff was being optimized.Our goal was to ensure the satisfaction and engagement of the producer so that the study could continue to be conducted on the same land over the 3 years of the project.
The PCA described an extremely limited amount of variance in the first two principal components (Figs.2A and 2B).This suggests a large amount of variation in weed populations, not associated with the implemented treatments, or that there was very little variation at all between the populations.It is likely that even though we were in the densest weed patches we could find in the fields, the populations were variable across the study locations.There is no obvious solution to this; however, longer term studies where differences between treatments in theory should get larger over time would allow more detectable differences between the treatments.One interesting aspect of the PCA is that while the centers of the treatment ellipses are directly overlapping, there are differences in shape and alignment with the principal components, however any difference here is extremely limited based on the statistical analyses.There was no difference in centroid location or in the homogeneity of dispersions between treatments.The principal component axes seem to be most strongly associated with weeds only found in one or two fields and in lower densities than some of the major weeds of concern.Unfortunately, this does not help draw any firm conclusions about the efficacy of the impact mill treatment.
Overall, this study provides no conclusive evidence of the efficacy of physical impact mills in a short period of time in western Canada.There are some populations where weeds were reduced, although there were also a few instances of increased populations in the HSD treatment.However, although only two comparisons of cleavers populations showed significant reductions in the HSD treatment, 9 out of 15 fields that had cleavers as a top weed showed numerically lower populations in the HSD treatment, suggesting that had the study continued longer term, it is possible that more significant differences would have been found.While the research is not conclusively showing efficacy of physical impact mills, the above-described issues with timeline, missing measurements of additional seed loss or where weed seeds are expelled, and trial locations suggest some reasons why this may be the case.Interestingly, early adoption of physical impact mills on the Canadian Prairies has started.At the time of submission, between 20 and 30 mill units are actively being used by Canadian farmers (T.Thiessen, J. Lade, personal communications).Anecdotal feedback from some of these early adopters suggests they are initially seeing changes in spatial weed distribution, prior to seeing changes in weed densities.For example, one producer described being able to pinpoint where they utilized their combine with the mill going through a kochia patch versus where they took a combine without a mill through a kochia patch.The former remained the same-sized patch, while the latter resulted in kochia spread across the field.This suggests that short-term physical impact mill or HWSC projects may be better focussed on spatial weed distribution instead of overall weed density, utilizing longer term projects to investigate impacts on densities.
Future studies investigating the impact of HWSC would benefit from consideration of the findings of this study.Initiating a study where populations are not seed limited will reduce the ability to determine the efficacy of the mills in a short time period, particularly for weeds with relatively longterm seedbanks.This, however, increases the challenge of conducting this type of study on producer fields, under realistic conditions, as those areas where populations are not seed limited are likely also areas where the population is highly variable.Targeting weed species that are short-lived in the seedbank will aid in measuring impacts on weed densities in a shorter term study.Prior to moving to in-field testing, filling gaps of measurements such as header loss could be a key component to understanding field-based results.Additionally, having prior knowledge on how to set the combine to successfully direct the vast majority of weed seeds into the chaff fraction to optimize the proportion of weed seeds targeted is important.This will likely vary with crop type, weed species, and potentially even environmental conditions when harvesting, so while Broster et al. (2016) have defined some settings for Australia, additional work is needed in Canada.Finally, in terms of managing logistics, it is more practical to find and partner with one early-adopting farmer who is willing to accommodate research trials, than for research groups to invest in their own machine, invest in a commercial-sized combine to operate the machine, and coordinate with multiple farmers for fields on which to test the unit.Working with fewer producers and having farmers run their own combines reduce biosecurity concerns as well.While this study does not provide conclusive evidence of efficacy of the mill, it has provided many scientific learning opportunities, discussion opportunities with farmers, new research questions and opportunities, and a "front-row seat" to early adoption of this technology.It is important, now, to help optimize use of the technology, and understand why and how producers are using this technology on the farm and what benefits are being observed.

Fig. 1 .
Fig. 1.The location of each of the 20 producer fields where the physical impact mill study was conducted.The Lacombe Research and Development Centre of Agriculture and Agri-Food Canada is indicated with a star.Field locations are within an approximately 50 km radius of the Research and Development Centre.Map created using Google Earth.

Fig. 2 .
Fig. 2. The combine and tow-behind Harrington Seed Destructor set up utilized in the research study.

Fig. 3 .
Fig. 3. Visual depiction of how treatments were changed between Harrington Seed Destructor and regular harvest.(A) The chaff output is being fed into the bypass tube for the regular harvest treatment.The chaff is placed onto the same conveyor as the straw typically travels and not processed by the mill.(B) The chaff output is being fed directly into the mill intake tube where it will be processed by the mill for the Harrington Seed Destructor treatment.In 2018 and 2019, the interface shown in (C) was used where the switch between regular harvest and Harrington Seed Destructor treatments was made at the white circle and the upper tube remained rigid.Photo credit (A) and (B): K.N.Harker.Photo credit (C): J. Kirsch.

Field
Number of significant differences in weed densities between the physical impact mill and regular harvest treatments, indicating which weed species had significantly different densities and whether their populations were increased (+) or decreased (−) in the physical impact mill treatment.Comparisons for weeds were made across all fields containing that species as a dominant species and in each individual field containing those dominant weed species.

Fig. 4 .
Fig. 4. Principal component analysis (PCA) biplots for (A) the seedling weed densities by treatment and (B) the seedbank weed densities by treatment.Ellipses indicate the treatments; the arrows and the length of the arrows indicate the species contribution to that principal component.Axes indicate the principal components and the percentage of variance explained by those components.Treatments are described by Unt = untreated and HSD = physical impact mill/Harrington Seed Destructor treatment.

Table 2 .
Cropping rotation, harvest management practice, and herbicide regime in each producer field in each year of the study, provided by the producers.