Site-specific weed control has become possible since the introduction of global positioning systems (GPS). GPS technologies allow us to determine position, map fields, including weeds and topographic features then apply treatments to previously identified areas. We can anticipate that real time weed recognition will progress from being able to distinguish weeds in fallow to being able to distinguish between species.
Environmental loading of pesticides could be reduced by site-specific application, along with herbicide costs while maintaining weed control and crop tolerance. GPS and geographic information systems (GIS) technologies may assist with field monitoring and integrate other site-specific operations.
This technological leap has challenged our understanding of weed biology and distribution and our assumptions concerning prescriptions for herbicides. Before potentials can be realized, many questions must be answered. What factors influence weed distribution? How are weeds distributed in fields? Which weeds are suitable for site-specific applications? How can we effectively apply herbicides to different parts of the field and to different weed populations?
A series of experiments were conducted across Alberta to answer some of these questions. Canada thistle, sow thistle and wild oat were selected as target weeds for site specific herbicide applications.
Usefulness of grid sampling to establish the distribution of weeds
Weeds were identified and counted in three fields over three years using different scales of a grid sample. In 1997, weed populations in Field 4 were sampled one day prior to herbicide application. A grid of 124 points was established, approximating one sample per ha. Weeds were identified and counted in four randomly placed, 0.25 m2 quadrates within 3 meters of each grid point for a total of 496 samples.
In 1998 and 1999, weed populations in Field 5 were sampled on a grid pattern with a 50 meter spacing. Using this grid, 240 points were located in the field, approximating one point every 0.27 ha. At these locations, all weeds were identified and counted in 9 regularly spacing, 0.25 m2 quadrates.
In 1999, Field 9 was sampled on a 25 m grid spacing, with each point representing approximately 0.12 ha. Weed counts were conducted using the same 9, 0.25 m2 quadrates.
All grids locations were established using an AIM Navigator and Omnistar to provide real time differential correction. Weed density data were analyzed using kriging techniques on important weed species and weed classes (annual, winter annual, perennial, total weeds). Kriged data was interpolated based on a 5-meter search area.
Fields 1 and 8 were selected for a comparison of site-specific and conventional herbicide applications. Site-specific application maps were derived by weed tagging, along parallel lines at 20-m intervals. Geographic position was established using an AIM Navigator for global positioning and Omnistar for differential correction, mounted on an all terrain vehicle (ATV). Spatial position of Canada thistle and sow-thistle was indicated by the presence of point data in a prescription map.
The fields were divided into twenty, 23 meter wide strips the length of the field, and treatments (site-specific or conventional) randomly assigned to each pair of strips. These treatment pairs constituted 10 replicates. Within each replicate pair, 20 randomly placed quadrates were established prior to spraying. The quadrates were paired in the two treatments (site-specific or conventional), located 10 meters apart, to account for auto-correlation of weed patches. One half of each paired plot was covered with plastic during spraying and served as the untreated control, while the second half of the plot received the treatment. In total, 400 small plots were established, 100 treated in each of the conventional and site-specific areas and 100 untreated in each of the conventional and site-specific areas.
For site-specific treatments, a prescription map was developed. The sprayer was turned on and off in response to thistle presence or absence. For conventional treatments the sprayer remained on at all times. The clopyralid rate applied to site-specific treatments was 201-g ai/ha while blanket treatments received the standard clopyralid rate for the area, 102-g ai/ha. Both site-specific and conventional applications were made using a 23-meter Patriot sprayer equipped with an AIM Navigator and Dickey John controller. Field 1 was sprayed on June 8, 1998 and Field 8 on June 8, 1999.
At BBCH stage 45, 0.25 m2 samples of barley, Canada thistle and sow thistle were removed from Field 1 in 1998 to provide 400 samples for individual dry weights. In 1999, the barley in Field 8 was at BBCH stage 61 to 69 when the same procedure was followed to collect dry weights. Data in 1998 and 1999 was analyzed using analysis of variance in a general linear model (GLM) using SAS. Means were separated using Student-Newman-Keul.
In Fields 21 and 22, wild oat distributions were tagged, a map prescribing variable rates of clodinafop was developed, the fields were sprayed and control assessed. In the one field, high and low weed populations were identified and 44.4 ac received the low rate (60 ml/ac) while 108.6 ac received the higher rate (85 ml/ac). In the second field, low, medium and high weed populations were identified and 48 ac received the low rate, 6.5 acre received the medium rate (73 ml/ac) and 99.5 ac received the high rate. Herbicides were applied with a Flexicoil sprayer equipped with dual booms. At least 40 paired plots were established in areas identified by weed tagging as having low, medium and high wild oat populations with one of the paired plots covered during spraying. Wild oat were counted in treated and untreated paired plots and four weeks following application, wheat and wild oat biomass was determined. Data was analyzed using GLM in SAS. Means were separated using Student-Newman-Keul.
Usefulness of grid sampling to establish the distribution of weeds
Grid sampling of weeds in fields is laborious but maps of weed distribution can easily be produced by kriging. However, these maps smooth the variability of density in the field by over-estimating low densities and under-estimating high densities. Grid size was decreased from 0.96 ha per point in 1997 (Field 4) to 0.27 ha (Field 5, 1998, 1999) and down to 0.12 ha per point (Field 9, 1999). Reducing the scale of grid sampling did not reduce the difficulties associated with smoothing. The smoothing function of kriging creates inaccurate maps that are of limited use in site-specific spraying. An example of a map created by kriging is shown in Figure 1.
In Fields 1 and 8, where site-specific experiments were conducted, there was equal control of Canada thistle or sow thistle in blanket and site specific application. Depending on the percentage of the field containing Canada and sow thistle, herbicide use was 20 to 60% lower than conventional blanket applications.
The clumped distribution of Canada and sow thistle made sampling of the field difficult. Because paired plots were placed randomly, many plots contained no Canada thistle. For example, in Field 1 only 12% of the samples included Canada thistle while 47% contained sow thistle. In Field 8 only 28% of the plots contained Canada thistle.
In the two fields (Fields 21 and 22) where wild oat was treated with site-specific variable rate herbicide applications, wild oat populations varied considerably. In Field 21, wild oat averaged 125 plants/m2 and ranged from 0 to 456 plants/m2 while in Field 22, wild oat averaged 84 plants/m2 and ranged from 0 to 212 plants/m2.
In Field 21, field scouting correctly identified significant differences in both wild oat density and biomass (Figure 2). There were also significant differences in wild oat biomass between the treated and untreated plots. The low herbicide rate reduced wild oat biomass by 91.8 % while the high rate reduced biomass by 94.1 %. Wheat biomass was significantly higher in areas that received the herbicide and in areas of weed tagging, as have low wild oat populations.
In Field 22, there were no significant differences in wild oat density or biomass in areas that were identified as having low, medium or high wild oat populations (Figure 3). While herbicide treatment significantly decreased biomass, there were no differences in the percentage reduction in biomass in the areas which received different herbicides rates. The low rate of clodinofop provided a 94.5 % reduction in biomass, while medium and high rates provided 96.4 and 96.3%, respectively. In this field, where wild oat populations were more uniform, scouting failed to correctly differentiate between levels of wild oat populations and increasing the herbicide did not alter the control of wild oat. There were no measurable differences between wheat biomass in areas that were identified as having differential wild oat populations, but herbicide treatment did significantly increase wheat biomass.
Site specific clodinofop application reduced herbicide use by 8.5 and 9.6% in Field 21 and Field 22, respectively.
Gathering information and creation of accurate maps of weed populations are a major barrier to the use of site-specific applications. Grid sampling and mapping using kriging techniques provide limited information or even misinformation about weed populations, Weed-scouting procedures for wild oat have not proved consistently accurate but scouting for Canada thistle was more successful.
Variable herbicide rates can be applied using currently available application equipment. While control with Lontrel is relatively consistent on Canada thistle and sow thistle, the response of wild oat populations to variable herbicide rates is more difficult to predict.
Figure 1. Distribution and density of total weeds in Field 4, 1997, overlaying topography. The map was derived by kriging weed densities obtained by grid sampling.
Figure 2. Wild oat biomass after herbicide application in treated and untreated plots established in Field 21 in areas that had been identified by weed tagging as having high or low wild oat populations.
Figure 3. Wild oat biomass after herbicide application in treated and untreated plots established in Field 22 in areas that had been identified by weed tagging as having high, medium or low wild oat populations.