Introduction
Mechanical cultivation is the primary means of in-crop weed control on many organic farms (Baker and Mohler Reference Baker and Mohler2014; Gallandt Reference Gallandt, Chauhan and Mahajan2014). It is also an important aspect of Integrated Weed Management because it can help reduce the selection pressure of herbicides, thereby contributing to the delayed development of herbicide-resistant weeds (Norsworthy et al. Reference Norsworthy, Ward, Shaw, Llewellyn, Nichols, Webster, Bradley, Frisvold, Powles, Burgos, Witt and Barrett2012). In many specialty crops for which herbicide options are limited, row cultivation is an integral tactic for controlling early season “escapes” (Fennimore and Doohan Reference Fennimore and Doohan2008). But while between-row (or interrow) cultivation can be effective, cultivation in the 10-cm wide in-row (or intrarow) zone, is typically only marginally effective (Gallandt et al. Reference Gallandt, Brainard, Brown and Zimdahl2017; Vanhala et al. Reference Vanhala, Kurstjens and Ascard2004). Uncontrolled intrarow weeds often require additional control efforts and may contribute to crop yield losses (Ascard and Fogelberg Reference Ascard and Fogelberg2008) and credits to the weed seedbank (Melander et al. Reference Melander, Rasmussen and Bàrberi2005). Precision-based mechanical weed control advances continue through machine learning and vision guidance systems, but many technological and logistical challenges remain before they can be used on a practical basis (Fennimore et al. Reference Fennimore, Slaughter, Siemens, Leon and Saber2016; Lu et al. Reference Lu, Zhang, Wang and Hu2025), thereby extending the relevance of basic cultivation.
While spring tine harrows, disk hillers, or even sweeps can sometimes be adjusted to disturb weeds in the intrarow zone, implements such as torsion weeders or finger weeders exclusively target the intrarow zone (Mohler et al. Reference Mohler, Teasdale and DiTommaso2021). Each tool may be used individually, but previous studies have shown that using more than one in sequence in the same pass, or “stacking” them, can create synergies among implements (Brown and Gallandt Reference Brown and Gallandt2018; Gallandt et al. Reference Gallandt, Brainard, Brown and Zimdahl2017). The most consistently synergistic combination—torsion weeders followed by finger weeders followed by a harrow—appeared to combine different mechanisms of mechanical weed control such as severing (Mohler Reference Mohler, Liebman, Mohler and Staver2001), uprooting (Kurstjens et al. Reference Kurstjens, Perdok and Gonese2000), and burial (Baerveldt and Ascard Reference Baerveldt and Ascard1999; Kurstjens and Perdok Reference Kurstjens and Perdok2000), respectively. But this combination caused unacceptable crop damage and was less selective (i.e., the ratio of efficacy to crop damage was not favorable) than most of the implements used alone (Brown and Gallandt Reference Brown and Gallandt2018).
Plant height is a proxy for tolerance to the forces involved with cultivation, since larger plants become more robust and well-rooted (Connors Reference Connors2022; Mohler Reference Mohler, Liebman, Mohler and Staver2001). Therefore, to improve the selectivity of intrarow cultivation, it may be necessary to employ multiple weed control tactics to maximize the crop’s size advantage relative to the weeds and to maintain crop yield potential through reduced competition (McDonald et al. Reference McDonald, Riha and DiTommaso2010). But crop cultivar selection as well as cultivator adjustment are also critical factors because implements should disturb the soil as closely as possible to crops and employ the right level of soil disturbance for the given conditions (Gallandt et al. Reference Gallandt, Brainard, Brown and Zimdahl2017; Zawada et al. Reference Zawada, Legutko, Gościańska-Łowińska, Szymczyk, Nijak, Wojciechowski and Zwierzyński2023).
Based on slow-motion video recordings of the cultivation implements used in studies presented by Brown and Gallandt (Reference Brown and Gallandt2018), which included torsion weeders, finger weeders, and harrows used in a wide range of conditions, we identified several opportunities to potentially reduce crop damage while maintaining weed control efficacy. First, despite being made of spring steel, the torsion weeders often appeared to cause severe crop damage if they contacted the crop. If their aggressive role in loosening the soil could be replaced by sweeps located farther from the crop, this component of crop damage could be eliminated. Second, the finger weeder and harrow operated directly in the crop row, but if these implements could be spaced farther from the crop while still disturbing intrarow weeds (as suggested by Parks et al. Reference Parks, McCollough and Gallandt2025), selectivity would likely improve. From preliminary trials, a mini disk hiller also showed potential to excel in this role (D.C. Brainard, personal communication). Finally, previous studies were conducted using manually steered planters and cultivators, but precision could likely be improved through use of GPS guidance (Zawada et al. Reference Zawada, Legutko, Gościańska-Łowińska, Szymczyk, Nijak, Wojciechowski and Zwierzyński2023). We hypothesized that by making the changes described above, we could achieve our overall objective of improving the selectivity of intrarow cultivation.
Materials and Methods
Trials were conducted in 2018 and 2019 at the Cornell AgriTech in Geneva, New York (42.88°N, 77.03°W). All fields had a history of vegetable production and were prepared with a fall-planted winter-killed cover crop, one or two passes of primary tillage with chisels or disks, and then a secondary tillage with a Perfecta harrow (Unverferth Manufacturing Co., Kalida, OH) to smooth the soil surface and control weeds immediately prior to planting. We conducted two trials with snap bean and four trials with beet. Dates and conditions associated with each trial are shown in Table 1. Snap bean (‘Provider’) were planted on 76.2-cm-wide rows with 4.1-cm in-row spacing at 3.0-cm depth, and beet (‘Ruby Queen’) were planted on 76.2-cm rows, with 1.9-cm in-row spacing (11.2 kg ha−1) at 1.5-cm seeding depth with a precision vacuum planter (Monosem, Edwardsville, KS). All plots were 76.2 cm wide, centered on individual crop rows with one row per plot. Plot length was 27.3 m for snap bean and 12.1 m for beet. No fertilizer, herbicides, or irrigation were used. All planting and cultivation operations were guided using GPS (Ag25 Antenna, TMX 2050 Display, NavController III, and AG GPS 542 base station; Trimble Inc., Westminster, CO).
Location, crop, soil, and weed conditions associated with each cultivation trial.
Table 1. Long description
The table presents data from trials conducted in 2018 and 2019, focusing on various fields with loam soil type and organic matter content of 2.7. The table has 7 columns and 6 rows. Column headers are Year, Field, Soil type, Soil organic matter, Crop, Planting date, Cultivation date, Crop stage, Weed stage, Date of evaluation, and Major weed species. Row 1: 2018, C, Loam, 2.7, Snap bean, July 5, June 14, 1st, trifoliate. Row 2: 2019, D, Loam, 2.7, Snap bean, July 18, June 28, 1st, Chenopodium album, Amaranthus species, Portulaca oleracea. Row 3: 2018, E, Loam, 2.7, Beet, July 30, August 16, 2-leaf, 2-4 leaves, Chenopodium album, Tripleurospermum inodorum, Capsella bursa-pastoris. Row 4: 2018, E, Loam, 2.7, Beet, July 30, August 24, 4-leaf, 6-8 leaves, Chenopodium album, Tripleurospermum inodorum, Capsella bursa-pastoris. Row 5: 2019, C, Loam, 2.7, Beet, June 28, July 15, 2-leaf, 2-4 leaves, Amaranthus species, Portulaca oleracea. Row 6: 2019, C, Loam, 2.7, Beet, June 28, July 25, 4-leaf, 6-8 leaves, Amaranthus species, Portulaca oleracea.
Treatments entailed single cultivations using combinations of one, two, or three different cultivation implements used on the same toolbar in the same pass. Treatments were implemented using a two-row toolbar with rear manual steering correction (Argus; KULT Kress, Vaihingen an der Enz, Germany) to carry the following implements at a ground speed of 4.0 km h−1.
In snap bean plots, treatments included 1) sweeps, 2) sweeps set closer to the crop row, 3) sweeps followed by a small disk hiller, 4) sweeps followed by finger weeders, 5) sweeps followed by a row harrow (a rake-like implement with an adjustable drop weight), 6) sweeps followed by finger weeders followed by the disk hiller, 7) sweeps followed by finger weeders followed by the row harrow, and 8) sweeps followed by finger weeders that were set up farther from the crop row followed by the row harrow with the drop weight providing the least possible downward pressure. More details on the treatments are provided in Figure 1 and Table 2. A slow-motion video of the sweep+finger weeder+disk combination is included in Supplementary Figure S1. All treatments were used only once per plot at the first trifoliate stage of the snap bean. Experiments were arranged in a randomized complete block design with four replications per trial. The initial trial was conducted in 2018, and it was repeated on a different field in 2019 (Table 1).
Cultivation implements and their adjustment for each trial. Implement sequences are listed from front to back.
Table 2. Long description
A table comparing cultivation implements and adjustments for different trials. The table has three columns: Trial, Implement, and Adjustment. It contains data for various trials including Snap, KULT, Finger weeders, Row harrow, Sweep, Sweep (aggressive), Sweep+finger weeder+row harrow (light setting), and Beet. Each trial lists specific implements and their adjustments. For example, the Snap trial uses beans with a disk adjustment, while the KULT trial uses Kress Duo disks with specific dimensions and spacing. The Finger weeders trial specifies medium hardness, diameter, depth, and spacing. The Row harrow trial details the arrangement and spacing of tines and drop weight. The Sweep trial describes blade width and working depth. The Sweep (aggressive) trial has a similar description but with a different intrarow space. The Sweep+finger weeder+row harrow (light setting) trial combines elements from previous trials with specific adjustments. The Beet trial uses a row harrow and spring tine harrow with specific settings. All other tools are described as in snap bean trials.
In beet plots, treatments used the same toolbar, ground speed, and implements as outlined for snap bean treatments. But we also included a spring tine harrow (Rabe Werk, Bad Essen, Germany) treatment, which was dragged at a faster speed (11.2 km h−1 following the treatment described by Brown and Gallandt [Reference Brown and Gallandt2018]). More details are included in Figure 1 and Table 2. Treatments were implemented once per plot and were arranged in a randomized complete block design with four replications per trial. Trials were conducted in 2018 and 2019 on separate fields each year, and separate trials were conducted at the 2-leaf and 4-leaf beet stage each year (Table 1).
Photographs of each cultivation implement used in these trials. Treatments included A) sweeps; B) sweeps (left) followed by finger weeders (center in yellow) followed by a row harrow (right in blue with the drop weight providing the lowest possible downward pressure); C) sweeps (left) followed by a small disk hiller with gauge wheels (center); and D) a spring tine harrow.
Data Collection
We sampled total weed density and crop stand 3 to 6 d after each cultivation by counting the number of living plants present in 0.1-m-wide quadrats centered on the intrarow zone (see Table 1 for dates and site details). The length of crop row included in the quadrats varied by trial to ensure statistically sufficient quantities (Vanhala et al. Reference Vanhala, Kurstjens and Ascard2004). In snap bean, weed density in untreated plots was an average of 52.6 and 84.7 weeds m−2 in 2018 and 2019, respectively. In 2018, weed density in untreated plots was an average of 55.7 and 55.5 weeds m−2 in 2-leaf and 4-leaf beet trials, respectively. In 2019, weed density in untreated plots was an average of 46.1 and 60.7 weeds m−2 in 2-leaf and 4-leaf beet trials, respectively. In snap bean plots, weed and crop quadrats were 27.3 and 5.5 m long, respectively. In beet plots, weed and crop quadrats were 12.2 and 3.0 m long, respectively. Weed control efficacy was calculated as the percentage difference in the mean number of weeds in the treated plots compared with the untreated plots. Crop survival was determined as the percentage of precultivation crop plants (counted within 24 h prior to cultivation) that remained intact in the postcultivation count.
Statistical Analyses
We completed all statistical analyses in JMP 17 software (SAS Institute Inc., Cary, NC). We statistically analyzed weed control through log-transformed postcultivation counts (Vanhala et al. Reference Vanhala, Kurstjens and Ascard2004), but in our results and discussion, we present untransformed weed control efficacy. For each trial, we used multiple linear regression models of each response variable with fixed effect terms for block and treatment. Each trial year was analyzed separately due to preceding multiple regression models with additional terms for year and year-by-treatment interactions, which showed interactions between year and treatment on efficacy in snap bean (P = 0.002) and on crop survival in 2-leaf beet (P = 0.001). Assumptions of constant variance and normality of residuals were assessed visually in residual by predicted plots and Shapiro-Wilk tests, respectively. If the multiple linear regression indicated a significant effect of treatment, we used contrasts to compare categorizations of one, two, or three tools, and then between setups within each category.
Estimating Selectivity
Selectivity is typically defined as the ratio between weed control and crop damage (Gallandt et al. Reference Gallandt, Brainard, Brown and Zimdahl2017). But selectivity ratios defined in this way can cause cultivation tools with low efficacy and zero crop damage to appear more favorable than tools with much greater efficacy that incur a trivial amount of crop damage (Rasmussen Reference Rasmussen1992). We chose instead to define selectivity based on an estimation of crop mortality associated with a fixed level of weed control (80%), following the methodology described by Brown and Gallandt (Reference Brown and Gallandt2018). In brief, this process involved first estimating the number of whole or fractional cultivation passes required for each tool combination to achieve 80% weed control, assuming a fixed percentage of weed mortality with each pass. We then used this estimated number of passes to calculate the associated crop mortality (M80) assuming a fixed level of crop mortality with each successive pass.
Visualizing the Relationship Between Tool Stacking and Selectivity Under Varying Crop-Weed Conditions
To visualize the effect of tool stacking on selectivity under various levels of crop size advantage, we graphed M80 against the ratio of crop height to weed height at the time of cultivation. To measure these heights, four randomly selected crop plants and weed plants were measured and the averages were used for each treatment in that trial. For each level of tool stacking, data were fit with the following exponential decay function:
$$Y = {A^{ - B*X}}$$
where Y is M80 and X is the ratio of the height of crop and weeds. Parameters A and B were estimated in JMP 17 software using the Exponential Two Parameter model from the Fit Curve library. We evaluated the curves visually, rather than statistically, since the data were from disparate trials that were not fully crossed.
Results and Discussion
Potential Benefit of Stacked Cultivation
In snap bean, efficacy of one-, two-, and three-implement combinations differed, with more tools (i.e. stacking) resulting in greater efficacy, but not affecting crop survival (Table 3). In beet, stacked tool combinations resulted in greater weed control, and did not affect survival of 4-leaf beet (Table 4). But stacked combinations reduced survival of 2-leaf beet. These results are consistent with those of previous research showing low tolerance of young beet to disturbance (McCollough et al. Reference McCollough, Poulsen and Melander2024) and suggests that weed control should be achieved with other means if needed prior to the 4-leaf stage.
Overall, compared to implements used individually, the stacked cultivation treatments showed potential to improve weed control efficacy and they did not reduce survival of well-established crops. This highlights the success of the changes we made in this study compared to previous treatments that resulted in unacceptably high crop damage (Brown and Gallandt Reference Brown and Gallandt2018). Of our changes (using GPS guidance for planting and cultivation, using sweeps rather than torsion weeders for initial loosening of the soil, and spacing tools so that contact with the crop was minimized) we think that each likely contributed to the improved selectivity.
Table 3. Long description
The table presents data on weed control and crop survival in snap bean for the years 2018 and 2019. It includes various treatments with one, two, and three implements, as well as a significance row. The table has 12 rows and 5 columns. Column headers are Treatment, Weed control 2018 (percent), Weed control 2019 (percent), Crop survival 2018 (percent), and Crop survival 2019 (percent). Row labels include different treatment methods such as Sweep, Sweep (aggressive), Sweep+disk, Sweep+finger weeder, Sweep+row harrow, Sweep+finger weeder+disk, Sweep+finger weeder+row harrow, Sweep+finger weeder+row harrow (light), Significance, and Treatment. Each row lists the corresponding weed control percentages for 2018 and 2019, followed by crop survival percentages for the same years. Notable trends include varying levels of weed control and crop survival across different treatments and years.
a Implement sequences are listed from front to back. Analyses are based on multiple linear regression with fixed effect terms for block and treatment. The blocking term is not displayed since it was not significant. If there was an effect of treatment, contrasts were used to compare between the number of implements and also between the treatments within a given number of implements.
b Within each column, uppercase letters refer to comparisons between one, two, three implements, whereas lowercase letters involve comparisons within each group of one, two, or three implements. Treatments (or treatment categories based on number of implements) not connected by the same letter are significantly different. Untransformed weed control efficacy is presented, but statistical analyses were performed on log-transformed, postcultivation weed counts
c Survival higher than 100% is indicative of new crop emergence or counting error.
d Treatment significance: *** P < 0.001; NS, not significant.
Table 4. Long description
The table presents data on weed control and crop survival in beet resulting from various cultivation treatments. It has 12 rows and 8 columns. The columns are labeled as follows: Treatment, Weed control (Two-leaf beet 2018, Two-leaf beet 2019, Four-leaf beet 2018, Four-leaf beet 2019), and Crop survival (Two-leaf beet 2018, Two-leaf beet 2019, Four-leaf beet 2018, Four-leaf beet 2019). The treatments include one implement (Spring tine harrow, Sweep), two implements (Sweep+disk, Sweep+finger weeder, Sweep+row harrow), and three implements (Sweep+finger weeder+disk, Sweep+finger weeder+row harrow). The table shows percentages for weed control and crop survival across different treatments and years. Notable trends include varying effectiveness of different treatments on weed control and crop survival, with some treatments showing higher percentages in specific years and conditions.
a Implement sequences are listed from front to back. Analyses are based on multiple linear regression with fixed effect terms for block and treatment. The blocking term is not displayed since it was not significant. If there was an effect of treatment, contrasts were used to compare between the number of implements and also between the treatments within a given number of implements
b Within each column, uppercase letters refer to comparisons between one, two, three implements, whereas lowercase letters involve comparisons within each group of one, two, or three implements. Treatments (or treatment categories based on the number of implements) not connected by the same letter are significantly different. * Untransformed weed control efficacy is presented, but statistical analyses were performed on log-transformed, post-cultivation weed counts. The blocking term was not displayed since it was not significant.
c Survival higher than 100% is indicative of new crop emergence or counting error.
d Treatment significance: * P < 0.05, **P < 0.01, ***P < 0.001; NS, not significant.
Implement Combination Comparisons
There were few differences between implement combinations within categorizations of one, two, or three tools. All of the differences were based on the minimal effect of the row harrow. For example, in snap bean, 2018 trials with sweep+finger weeder demonstrated greater weed control than sweep+row harrow, and in 2019, the sweep+finger weeder+disk had greater efficacy than the sweep+finger weeder+row harrow (light) treatment (Table 3). Likewise, in 2-leaf-stage beet, combinations with the row harrow resulted in lower efficacy, but also less crop damage (Table 4). Overall, these results indicate that the row harrow trended toward less aggressive action than finger weeders or disks. Likewise, the spring tine harrow provided lower weed control in comparison to the stacked, guided implements (Table 4). Spring tine harrows are some of the most common cultivation tools used on farms (Mohler et al. Reference Mohler, Teasdale and DiTommaso2021), but both this study and previous studies (Melander et al. Reference Melander, Rasmussen and Bàrberi2005; Van der Weide et al. Reference Van der Weide, Bleeker, Achten, Lotz, Fogelberg and Melander2008) show that with low risk to crops, guided cultivation offers potential for increased efficacy compared to so-called blind cultivation.
Regarding ease of use, we found finger weeders to be the most user-friendly of the intrarow tools. Their working ends are made of rubber, which seemed to provide some built-in crop safety compared to the metal tools. Contrastingly, the disks required precise adjustment to throw enough soil to bury small weeds without burying the crop. But burial can be a very effective means of selective intrarow weed control (Merfield et al. Reference Merfield, Bennett, Berry, Bluon, O’Connell and Hodge2020). The disks appeared to work especially well when employed third in the tool sequence so that weeds partially dislodged by previous implements could be more easily buried.
Anecdotally, in the field used for the 2019 snap bean trials, an adjacent part of the field had 8-fold greater weed emergence due to seed rain the previous year. But when we cultivated the snap bean in that part of the field using the same sweep+finger weeder and sweep+finger weeder+disk used in the main trials, the resulting efficacy (74% and 94%, respectively) was similar to our results at lower weed densities (Table 3). These results support the notion that cultivation efficacy is density-independent (Gallandt et al. Reference Gallandt, Brainard, Brown and Zimdahl2017), at least at levels not dense enough to create a mat (Mohler et al. Reference Mohler, Teasdale and DiTommaso2021), and demonstrate the importance of depleting weed seedbanks to reduce the number of weeds surviving a given level of efficacy.
Selectivity
Farmers in central New York indicated that achieving 80% intrarow weed control in one pass of cultivation would be very satisfactory (B. Brown, personal communication). Over all our trials, this was attained in 17%, 20%, and 50% of the one-, two-, and three-tool treatments, respectively. Although crop survival was less than 95% in 33%, 52%, and 69% of the one-, two-, and three-tool treatments, respectively, the farmers indicated that stacked tools may still be preferable because crop damage could be overcome with higher seeding rates.
Calculating the number of cultivation passes required for each treatment to achieve 80% intrarow weed control allowed us to standardize efficacy so that crop damage could be viewed over the range of crop to weed height ratios present in all our trials (Figure 2). Although we could not make statistical comparisons across experiments, visual examination reveals the importance of timing cultivations while weeds are very small relative to the crop. Indeed, of all our trials, the 2018 snap bean trial represented the most ideal conditions, a well-established crop and primarily cotyledon-stage weeds, resulting in our most selective cultivations (Table 3). But loose soil conditions also appeared to benefit the action of the cultivators in that trial.
Cultivation selectivity, as estimated through crop mortality at the calculated number of passes required to achieve 80% efficacy, graphed over the average crop to weed height ratio of each trial. Points were moved slightly left and right to enhance visibility using the automatic jitter function in JMP 17 software. Points represent cultivation setups that used one (dot), two (triangle), or three (square) implements and were fit with exponential decay equations y = 79.9−0.36x, y = 80.8−0.47x, and y = 178.8−0.98x, in the black, gray, and light gray lines, respectively.
The exponential decay curves in Figure 2 demonstrate that in favorable crop to weed height ratios, the combinations of three tools appear more selective than one- and two-tool setups. At 80% intrarow weed control, a level of 5% crop mortality was viewed as acceptable by farmers in central New York (B. Brown, personal communication). Using Equation 1, we calculated that the minimum crop to weed height ratio necessary to achieve less than or equal to 5% crop mortality was 6.2, 5.6, or 3.6 for one-, two-, or three-tool combinations, respectively. However, this trend reversed when crops and weeds were similar in size, since in that situation, stacked tools were overly aggressive on the crop. Overall, Figure 2 highlights the importance of giving the crop an early size advantage; perhaps achieved through prior weed seedbank management, false or stale seedbed periods (Mohler et al. Reference Mohler, Teasdale and DiTommaso2021), or the use of preemergence herbicides (Norsworthy et al. Reference Norsworthy, Ward, Shaw, Llewellyn, Nichols, Webster, Bradley, Frisvold, Powles, Burgos, Witt and Barrett2012).
Practical Implications
Our results indicate several possible benefits of stacked three-tool cultivation. Shallow cultivation requires minimal tractor horsepower; therefore, more implements could easily be added to maximize the effectiveness of each pass. Increased efficacy may allow farmers to save time or money by achieving the desired level of weed control with fewer passes. But regardless of stacking, two keys to our success that could improve farmers’ existing cultivators are tool spacings that minimize crop contact and use of GPS-guided tractors.
This study also revealed the benefit of using intrarow cultivation on very small weeds (Gallandt et al. Reference Gallandt, Brainard, Brown and Zimdahl2017). Additionally, the stacked combinations were able to achieve satisfactory results with larger than usual intrarow weeds, which could allow for improved weed control, reduced yield losses, and decreased weed seed production. However, the drawbacks include potentially greater risk of crop damage, having more tools to adjust, and creating more opportunities for parts to break. These tradeoffs will likely be influenced by the unique management goals and constraints of each farm.
Supplementary material
To view supplementary material for this article, please visit https://doi.org/10.1017/wet.2026.10127
Acknowledgments
We are thankful for the support and agricultural expertise of Abby Seaman, John Wallace, Matthew Christiansen, and Elizabeth Maloney. Thank you to Daniel Brainard for extensive preliminary review of the manuscript. We are grateful to have met Haudenosaunee farmers through this work, and we hope to help make up for the painful history of these lands.
Funding
This project was supported by Grant 2017-70006-27142/project accession 1014000, awarded through the Crop Protection and Pest Management Program by the U.S. Department of Agriculture–National Institute for Food and Agriculture.
Competing interests
The authors declare they have no competing interests.
Open access
