Get 20M+ Full-Text Papers For Less Than $1.50/day. Start a 7-Day Trial for You or Your Team.

Learn More →

Patch spraying of weeds in spring cereals: Simulated influences of threshold level and spraying resolution on spraying errors and potential herbicide reduction

Patch spraying of weeds in spring cereals: Simulated influences of threshold level and spraying... A major obstacle to patch spraying of broad-leaved weeds in cereals is a cost-effective method to assess within-field heterogeneity of the weed population. One method could be a camera mounted in front of the spraying vehicle, online image analysis, and field sprayer shifting between ‘on’ and ‘off ’ as the predefined weed damage threshold level is reached. Because such a camera will capture a very limited area (B /1m ) compared to the sprayer width (several m), success of this method requires that spraying decisions vary little within boom width, thereby causing few spraying errors. This approach was evaluated by simulations for varying boom widths and three levels of a weed damage threshold model. Potential herbicide reductions compared to blanket application were also simulated. The average potential herbicide reductions estimated as proportions of fields below the threshold, were 60%, 64% and 53% for the original, 25% increased and 25% reduced threshold levels, respectively. The simulated herbicide reductions were not influenced by boom width, but varied significantly between fields, and between threshold levels. As evaluated by spraying errors, the suitability of the suggested approach will increase by decreasing boom width, vary between fields, and in some fields vary between the threshold levels. For boom widths of 15 m and 24 m, the spraying errors were about 10% and 15%, respectively, where omission of spraying areas above the threshold constituted 5% and 8%, respectively. Keywords: Herbicide reduction, precision farming, site-specific weed management. Introduction image analysis algorithms based on spectral reflec- tance characteristics, which has proved promising for Weeds generally occur in patches or clusters within row crops (Tian & Slaughter, 1998; Steward & Tian, agricultural fields (e.g. Dessaint & Caussanel, 1994; 1999; Onyango & Marchant, 2003). Therefore, Gerhards et al., 1997; Wallinga et al., 1998; Diele- other than spectral characteristics, e.g. shape and man & Mortensen, 1999; Cousens et al., 2002) contour of plants, have been tested as image analysis which represents a potential for treating only the measures (e.g. Andreasen et al., 1997; Clausen patches. Lack of cost-effective weed detection meth- et al., 2000; Pe ´rez et al., 2000; Søgaard & Olsen, ods on a field scale has so far prevented implementa- 2003). Promising results have been achieved for tion of patch spraying (PS) in cereal crops. single species recognition without overlapping leaves In Europe, cereals are normally grown with a row (Gerhards & Christensen, 2003; Søgaard, 2005). spacing about 1213 cm, and herbicide spraying There are economic, environmental and govern- against seed-propagated broad-leaved weeds is con- mental pressures for reducing herbicide inputs ducted when the crop has 34 leaves. At this stage (Anon., 2004). The average reduction of herbicide crop and weed plants have nearly the same spectral consumption from site-specific weed control in reflectance characteristics. Together with the short spring cereals ranged from 40 to 54%, and from 50 row spacing, this has prevented development of Correspondence: T. Berge, Plant Health and Plant Protection Division, BIOFORSK, Hoegskoleveien 7, NO-1432 Aas, Norway. Tel:  / 47 64 94 93 56/ / 47 64 94 94 00/ / 47 64 94 42 20. Fax:  / 47 64 94 92 26. E-mail: [email protected] (Received 7 September 2005; accepted 6 June 2006) ISSN 0906-4710 print/ISSN 1651-1913 online # 2007 Taylor & Francis DOI: 10.1080/09064710600914202 Patch spraying of weeds in spring cereals 213 to 61% in winter cereals (Heisel et al., 1999; weeds in spring cereals; 2) estimate potential herbi- Nordmeyer et al., 2003; Christensen, 2005). In cide reductions; and 3) test whether results were Scandinavia cereals are the most widespread arable dependent on the actual weed damage threshold crop, and in Norway about 90% of the total pesticide level. use is in cereals, the main proportion being herbi- cides (Bjugstad, 2000). The potential for reducing Material and methods pesticide use through PS in cereals should therefore Weed data were collected in 1993 at three spring be significant. cereal fields in Norway (Table I), originally for Reports on spatial variation of weed density within another type of investigation (Fykse & Wærnhus, cereal fields in Norway are few (Fykse & Wærnhus, 1994). The weeds were counted within quadrats 1994; Fykse, 2004). Fykse (2004) plotted total weed density along 180 m transects from four cereal fields. sized 0.5 /0.5 m when the crop was at Zadoks 1314 (Zadoks et al., 1974), the normal time for Spatial variation in weed density was high, both post-emergence herbicide application. The quadrats within field and between fields. Interestingly, there was clear autocorrelation in weed density: positions were separated by 10 to 18 m (Table I). Spraying decisions ‘spray’/‘not spray’ were based with high density were neighbours with positions with high density, and vice versa. on a threshold model (Fykse, 1991) developed for Despite several proposed economic threshold seed-propagated broad-leaved weeds in spring cer- eals at Zadoks 1314 (Table II). If the total weed models for annual weeds in cereals (e.g. Gerowitt & Heitefuss, 1990; Fykse, 1991; Black & Dyson, density exceeded 175 plants m , or the threshold 1993), the practical use has been limited because density for at least one of five strongly competitive cost-effective methods to assess the weed popula- weed species was passed, the decision was ‘spray’. Furthermore, if the sum of two or more of these five tions have been missing (Wilkerson et al., 2002). Image analysis is probably an appropriate solution, species passed the threshold density for the species and from the above literature study, the necessary with the highest threshold value, the decision was algorithms will probably be available within a few also ‘spray’ (Table II). Presence of Galium aparine often leads to lodging, and the threshold value years. To reduce costs, one digital camera and the ordinary sprayers already in use should be the basis (1 plant m ) expresses the problems connected for PS. Therefore we investigated, by means of with harvest and drying, and not its yield-depressing simulations, the suitability for real-time on/off PS effect. As an attempt to apply the weed densities from the original threshold model with more con- in cereal fields based on the information from one camera. We simulated one camera mounted in front fidence, the original density levels were also in- of the tractor taking images along a narrow path in creased and reduced by 25% (Table II). The the middle of the sprayer boom, and the nozzles original threshold model included also relative switched on and off as the weed infestation was crop/weed ground coverage, but as this parameter above or below the damage threshold. To reduce was not recorded in the current fields, it was omitted spraying errors, this design requires that spraying in this study. decisions are relatively strongly correlated within the Spraying decisions (‘spray’/‘not spray’) were taken actual boom widths. The objectives of this study per quadrat and plotted for each field. These maps were to: 1) assess the suitability of using one camera were the basis for manual simulations of PS with per boom to control ordinary field sprayers on/off for varying sprayer boom widths based on the spraying patch spraying of seed-propagated broad-leaved decision of one quadrat per boom width. Increased Table I. Field and sampling characteristics. Voll Landvik Kvithamar Position 59841?N, 10846?E58820?N, 8831?E63830?N, 10852?E Precipitation, mm 600 1230 892 Growing days ( /58C) 169 202 182 Temperature sum ( /58C), degree days 1091 1442 1064 Sampling date in 1993 June 3 4 June 7 June 28 Number of quadrats 190 84 124 Metres between quadrats, N-S direction: 10, 10.5, 13, 14 11, 12 12, 13, 14 E-W direction: 15, 16, 18 10 10, 13, 14, 15 Normal period 1961 1990. 214 T.W. Berge et al. Table II. Definition and threshold values (weeds m ) of the model after Fykse (1991). Decision is ‘spray’ if criterion XX is met. Original threshold 9 /25% modified Criteria Definition level threshold levels A Total density of seed-propagated broad-leaved weed species 175 131 219 B Density of Galeopsis ssp. L. 25 19 31 C Density of Chenopodium album L. 45 34 56 D Density of Brassica rapa ssp. oleifera (DC.) Metzger 20 15 25 E Density of Stellaria media (L.) Vill. 45 34 56 F Density of Galium aparine L. 1 1 X Density of at least two of the species B F higher than the threshold for the species with the highest value XX ‘Full model’, one of the criteria A X fulfilled For G. aparine not relevant, explained in text. boom widths were simulated for the N-S direction coefficients. Minitab (Minitab Inc., 2003) was used in all statistical analysis. and the E-W direction by including adjacent quad- rats. For a given boom width, spraying decisions could either be conspecific of type ‘spray’, conspe- Results cific of type ‘not spray’, or heterogeneous including Mean total weed density per field varied more than both ‘spray’ and ‘not spray’ quadrats. Proportions of five-fold, from 53 to 295 plants m (Table III). In the sum of conspecific spraying decisions per given terms of standard deviation and coefficient of varia- boom width, hereafter termed ‘conspecific booms’, tion, the within-field variation of total weed density were plotted against boom width. was high (Table III). No field had random distribu- The following PS approaches, with constant tion of total weed density (pB /0.005, Table III). The detection resolution, but decreasing spraying resolu- frequency distributions of total weed density were tion, were simulated: 1) PS-1: spraying decisions clearly positively skewed at Voll and Kvithamar, but assigned to each quadrat independently, simulating a only slightly skewed at Landvik (Figure 1). Landvik patch sprayer with independently working sections showed flatter than normal distribution; elsewhere and one camera per section, or alternatively a very the distributions were sharper than normal, with narrow sprayer guided by one camera. This would high (Kvithamar) and very high (Voll) positive yield the most flexible spraying, and represented the kurtosis (Table III). best estimate of potential herbicide reductions Conspecific spraying decisions were more or less through PS for the dataset; 2) PS-2: the same aggregated at all fields, and for all threshold levels spraying decision assigned to two adjacent quadrats (Figure 2). Typically, weed patches defined as areas based on weed conditions at the quadrat viewed by consisting of nearest neighbour ‘spray’ quadrats the simulated camera. This simulated a wider boom using the original threshold level increased and/or (with no possibility for independent actions by the merged when the threshold level was reduced, and boom sections) and one camera per boom; 3) PS-3: shrank and/or divided if the threshold level was the same spraying decision assigned to three adjacent increased (Figure 2). quadrats based on the weed status of the middle one, For proportions of ‘conspecific booms’ (Figure 3), simulating an even wider boom (with no possibility field (p5 /0.001), threshold level (p /0.002) and for independent actions by the boom sections) and boom width (p5 /0.001) were main effects, but not one camera per boom. Compared to PS-1, using PS- the direction within field. The average reduction in 2 and PS-3 would inevitably result in ‘spraying the proportion of ‘conspecific booms’ per metre errors’. These were either failures caused by not boom was equal at Voll and Landvik (p /0.918), spraying quadrats with infestations above the thresh- and less than the reduction at Kvithamar old (error type 1), or spraying quadrats with infesta- (p5 /0.001). The intercept for Voll and Landvik tions below the threshold (error type 2). was higher (p /0.008) than the intercept for Kvitha- Differences between directions, threshold levels mar (Figure 3). For simulated boom widths covering and fields were tested with analysis of variance. 2, 3 and 4 quadrats, there is a probability of Tukey tests were applied to detect the actual 50%, 25% and 12.5%, respectively, to obtain con- differences. Linear regression models were fitted to specific decisions if the decisions are randomly the simulation results, and two-sample t -tests distributed. All simulations at Voll and Landvik were used to test differences between regression were well above these limits of randomness, whereas Patch spraying of weeds in spring cereals 215 Table III. Summary statistics of total weed density and threshold model species per field. Voll Landvik Kvithamar Mean total weed density, weeds m 53 295 131 Minimum total weed density, weeds m 00 8 Maximum total weed density, weeds m 432 844 628 Proportion of quadrats without weeds, % 5 1 0 Standard deviation of mean density, weeds m 59 244 115 Coefficient of variation, % 112 83 88 Skewness 2.76 0.66 1.99 Kurtosis 11.46  /0.63 4.72 Total weed density randomly distributed 10.85 2.19 6.55 Mean density of Galeopsis ssp., plants m 30 11 Mean density of C. album , plants m 782 18 Mean density of B. rapa ssp. oleifera , plants m  10 Mean density of S. media , plants m 119 5 Mean density of G. aparine plants m 1 Anderson-Darling normality test statistic. p valueB /0.005 for all fields. a few simulations were below at Kvithamar widths, and the mean proportions of ‘conspecific (Figure 3). Increased threshold levels caused a booms’ for 15 m and 24 m booms were 69% and 61%, respectively. significantly higher proportion of ‘conspecific The proportions of potential reduction in herbi- booms’ than reduced threshold levels, both on cide use (Figure 4) varied significantly between fields an average of all fields (p /0.002), and at Voll (p5 /0.001) and threshold levels (p5 /0.001), but not (p5 /0.001) and Kvithamar (p /0.020), but not at by boom width. Using resolution PS-1, 86%, 36% Landvik (Figure 3). The proportions of ‘conspecific and 59% of the quadrats were below the original booms’ decreased linearly with increasing boom threshold level at Voll, Landvik and Kvithamar, respectively (Figure 4), corresponding to a mean of Voll 60%. Estimates obtained by resolution PS-1 for the increased and decreased threshold levels were 64% and 53%, respectively, but not statistically different from the estimate using the original level. The magnitude of the tested threshold levels was of no influence on reduction at Landvik (Figure 4), whereas at Voll the reduced level gave less herbicide reduction than the increased level (p /0.017), and at Landvik Kvithamar the reduced level gave less than both the increased and the original level (p5 /0.009). By 10 applying wider booms (PS-2 and PS-3) both smaller and greater individual savings than by PS-1 were observed (Figure 4), but neither linear slope fitted to the mean of all fields and all threshold levels, nor fitted linear slopes per threshold level, were signifi- cantly different from zero (p] /0.051). At field level, Landvik was the only one with fitted slope different Kvithamar from zero ( /0.4% m , p /0.040). At every field all spraying error types increased linearly with boom width, but less than 1% m (Figure 5). For 15 m and 24 m booms, predicted average total spraying errors (95% confidence inter- vals in brackets) were about 10% (911%) and 15% (1417%), respectively. Predicted type 1 and type 2 spraying errors for these widths were 5% (45%) 0 120 240 360 480 600 720 840 and 8% (79%), respectively. Generally, Kvithamar had the highest spraying Figure 1. Frequency distributions of total weed density (weeds m ) per field. errors, Voll the second highest and Landvik the Percent 216 T.W. Berge et al. (a) Voll 0 40 80 120 160 200 240 280 320 (b) Landvik 0 50 100 150 200 (c) Kvithamar N S 0 50 100 150 200 250 300 X-coordinate (m) Figure 2. Spraying decision maps for PS-1. Filled (j) and open (I) squares denote ‘spray’ and ‘not spray’ quadrats for original threshold level. Circles (k) mark additionally ‘spray’ quadrats if the threshold level was reduced by 25%. Diamonds (2) mark quadrats that changed to ‘not spray’ if the threshold level was increased by 25%. lowest errors (Figure 5). This ranking was significant (p /0.032). For total error the reduced level also for total and type 1 spraying errors (p5 /0.046). caused higher (p / 0.034) error rates than the Main effect of threshold level on spraying errors was original level (Figure 5, bottom row). At Kvithamar, observed at Voll and Kvithamar, but not Landvik. At both the reduced (p /0.002) and the original (p5 / Voll, the reduced level gave higher errors than the 0.001) level gave higher total error than the in- increased level for total (p /0.017) and type 1 error creased level. Y-coordinate (m) Patch spraying of weeds in spring cereals 217 Voll Landvik Kvithamar 2 2 2 2 3 2 2 4 80 3 2 2 2 2 3 3 2 3 2 4 4 70 3 4 5 2 34 4 3 5 2 3 2 2 3 2 3 5 2 4 3 3 2 2 4 4 4 4 4 2 2 2 3 3 2 quadrats 2 3 Threshold Level 3 3 4 30 4 4 - 25% 3 4 3 quadrats + 25% 3 20 3 original 4 quadrats 15 25 35 15 25 35 15 25 35 Simulated boom width (m) Figure 3. Proportions of booms with conspecific spraying decisions versus simulated boom width. Labels on datapoints are numbers of quadrats per boom width. Horizontal lines mark limits for random distributions of spraying decisions for 2, 3 and 4 quadrats per boom. Discussion conspecific spraying decisions to be relatively strongly autocorrelated at the boom-scale too. In Suitability of simulated PS approach due to spraying Norway, sprayer booms have traditionally been less errors than 15 m, but up to 24 m wide booms are of Potential for PS of the three fields was indicated growing interest. The predicted average proportions by the variation in total weed density (Figure 1), of ‘conspecific booms’ (Figure 3) for boom widths visual aggregation of conspecific spraying decisions 15 m and 24 m were relatively high: 69% and 61%, (Figure 2) and proportions of the fields below the respectively. The associated total spraying errors threshold (Figure 4). These were field-scale results. (Figure 5) were about 10% (15 m) and 15% Suitability of the simulated PS approach requires (24 m). During operational PS, it is probably harder Voll Landvik Kvithamar 15 24 15 24 15 24 0 10 20 30 40 0 10 20 30 40 0 10 20 30 40 Simulated boom width (m) Figure 4. Proportions of field simulated unsprayed versus simulated boom width. Diamonds are values for PS-1 and dots are values for PS-2 and PS-3. Legend: see Figure 3. Proportions of field unsprayed (%) Proportions of 'conspecific booms' (%) 218 T.W. Berge et al. Voll Landvik Kvithamar Error, type 1 Error, type 1 Error, type 1 15 24 15 24 15 24 10 20 30 10 20 30 10 20 30 Error, type 2 Error, type 2 Error, type 2 15 24 15 24 15 24 10 20 30 10 20 30 10 20 30 Total Error Total Error Total Error 15 24 15 24 15 24 10 20 30 10 20 30 10 20 30 Simulated boom width (m) Figure 5. Proportions of field with spraying errors versus simulated boom width. Upper row: error type 1, middle row: error type 2, bottom row: sum of error type 1 and error type 2. Legend: see Figure 3. for farmers to accept omission of spraying patches narrower than the boom width and/or are very above the threshold than accepting spraying areas irregular in shape. As the spraying errors decreased with decreasing below the threshold. As the former error type was boom width (Figure 5), suitability of the simulated relatively low (about 5% (15 m) and 8% (24 m)), the simulated PS approach may be acceptable, particu- PS concept will generally increase as boom width decreases. Based on our results, using selective larly among farmers using the 15 m wide booms. The expected short- and long-term herbicide cost control of individual 2 m (Stafford & Miller, 1993) reductions compared to total costs, and the farmers’ and 3 m (Gerhards et al., 2002) boom sections would have reduced the average total spraying personal experiences and interests, will influence acceptable error rates. errors to about 1% and 2%, respectively. These Suitability of the simulated PS approach is likely low error rates require high mapping and spraying to vary between fields due to differences in spraying resolutions which necessarily involve relatively com- error levels. In the present study Landvik had the plex systems that probably are less robust and lowest errors, and Kvithamar the highest (Figure 5). more expensive (Paice et al., 1998) than our This could be explained by the spatial distributions simulated PS approach. In a simulation study, of ‘spray’ and ‘not spray’ quadrats at the boom width Barroso et al. (2004) found that the critical para- scale. Whereas Landvik had relatively elongated and meter that determined the economic viability of well-defined patches of neighbouring ‘spray’ quad- patch mapping and spraying resolution was the rats (Figure 2) and relatively strongly autocorrelated technology costs. Whether very high detection conspecific spraying decisions (Figure 3), Kvithamar and spraying resolution would facilitate adoption of had comparatively frequent shifts between ‘spray’ PS in cereal fields therefore remains an open and ‘not spray’ quadrats (Figure 2) and conspecific question. decisions were less autocorrelated (Figure 3). Gen- Is there a relationship between the level of spray- erally, a field where weed patches fit the width of the ing error of a field and the general weed infestation sprayer boom will have fewer spraying errors than a level? Analysing these variables, it seemed that if the field in which the majority of the patches are average field density of at least one of the threshold Proportions of field with spraying errors (%) Patch spraying of weeds in spring cereals 219 criteria AF (Table II) was close to its threshold (Figure 2) and the limited number of possible value, the field was likely to have relatively high simulations. Effects of boom width on the amount spraying errors. This is because: 1) an average of herbicide reduction would probably happen more density near the threshold value indicates that a often in fields where the patterns of conspecific relatively large proportion of the field has weed spraying decisions are shaped as rectangles and/or densities near that value (cf. Figure 1 for criterion relatively small-sized fields, than in fields with weed A); and 2) weed densities within fields are generally infestations of more irregular pattern and/or larger autocorrelated (e.g. Cardina et al., 1997). Conse- fields. quently, if a quadrat’s density is far below or far The relatively high, but expected field-to-field above the threshold, it is likely that the neighbouring variation in potential herbicide reduction (Figure quadrat is also far below or far above, and 4) is explained by different weed infestation levels of hence these neighbours are likely to have the same the fields (Table III), and is in agreement with, e.g. spraying decision. However, if a quadrat’s density Gerhards et al. (2005) who reported variation of is near the threshold (a little above or a little below), 2080% in herbicide savings by PS against broad- it is likely that the neighbour is also near the leaved weeds in German spring barley. Within a six-year period of PS of dicotyledonous weeds in threshold, but it is more a matter of chance whether these quadrats have the same spraying decisions German winter cereals, mean herbicide reduction compared to neighbours with densities far from could be less than 30% in one year and more the threshold. Hence, areas with densities near the than 60% in another (Nordmeyer & Zwerger, threshold are likely to cause higher quadrat- 2005). Year-to-year variation in savings must also to-quadrat variation in spraying decisions than be expected. areas having densities far from the threshold. As it The proportion of a field being above the damage is this quadrat-to-quadrat variation that generates threshold affects the suitability of PS. Barroso et al. the spraying errors, a field with weed densities (2004) found that the profitability of site-specific close to one or more of the threshold criteria is Avena sterilis control in Spanish winter barley gen- likely to have relatively high spraying errors. For erally increased as the proportion of the field infested example, at Kvithamar the average weed densities of decreased. This is very logical and probably also criteria A, B and C (Table III) were relatively close to valid for our simulated PS approach and data. Hence the threshold densities (Table II), and this field had Voll would be the most, and Landvik the least, the highest spraying errors (Figure 5). profitable field to patch spray (in the year of investigation) (cf. Figure 4). Potential herbicide reduction Influence of the tested threshold levels The average potential herbicide reductions esti- mated by PS-1 were substantial, 5364%, depend- At Landvik neither proportions of ‘conspecific ing on threshold level. These were of the same order booms’ (Figure 3), potential herbicide reduction as reported from practical PS of seed-propagated (Figure 4) nor spraying errors (Figure 5) were broad-leaved weeds in spring barley in Denmark affected by different threshold levels, whereas at (Heisel et al., 1999) and in winter cereals in Voll and Kvithamar these variables were influenced. Germany (Nordmeyer et al., 2003; Gerhards et al., This happened because at Landvik the proportions 2005). of ‘spray’ quadrats were relatively stable for different On average, boom width was of no main influence threshold levels (cf. Figure 2): 68% at reduced level on the potential herbicide reduction, meaning that and 63% at increased level, leading to a relative savings of the same order as estimated by PS-1 can difference of only 8% (100*(68%63%)/63%) be- also be expected for wider booms. However, as both tween the modified levels. At Voll and Kvithamar, on smaller and greater reductions than by PS-1 oc- the other hand, the relative differences in propor- curred from resolutions PS-2 and PS-3 (Figure 4), tions of ‘spray’ quadrats between the modified the actual saving obtained for one field, at least a threshold levels were much higher: 54% and 74%, relatively small one, might depend on the relative respectively, and high enough to yield significant camera position and starting point of spraying. In the effects of the threshold levels on the tested variables. long term, i.e. several years and/or larger fields, there In addition to the expected increase in herbicide will be no boom width influence on herbicide savings due to reduced threshold level seen at Voll, savings. The significant, but weakly decreasing and particularly at Kvithamar (Figure 4), higher reduction by the increasing boom width seen at proportions of ‘conspecific booms’ (Figure 3) Landvik (Figure 4) was probably due to the rectan- and reduced total spraying errors (Figure 5) were gular-like pattern of conspecific spraying decisions observed for the increased compared to the reduced 220 T.W. Berge et al. Cardina, J., Johnson, G. A., & Sparrow, D. H. (1997). The nature level. The two latter variables, which both depend and consequence of weed spatial distribution. Weed Science , on the local (boom-scale) distribution of conspecific 45 , 364 373. spraying decisions, were affected because the local Christensen, S. (2005). Table I. Potential savings of herbicides in patterns between the modified levels were signifi- percentage of common practice in 59 trials with site-specific cantly different due to the substantial relative differ- weed management in Europe. http://www.ewrs-sswm.org/ homepage.htm, March 2005. ences in ‘spray’ quadrats. At Landvik, however, the Clausen, S., Nicolas, S., Kirkhus, T., & Fykse, H. (2000). pattern differed only marginally between the differ- Automatic weed mapping in cereal fields using image ent threshold levels (cf. Figure 2). processing techniques. In: Proceedings of NOBIM-2000, the Because the threshold level affected the spraying Norwegian Image Processing and Pattern Recognition Confer- errors significantly in some fields, the suitability of ence , June 6 7, 2000. Trondheim, Norway. 5 pp. the simulated PS approach may vary with the Cousens, R. D., Brown, R. W., McBratney, A. B., Whelan, B., & Moerkerk, M. (2002). Sampling strategy is important for applied threshold level. Even if the 25% increased producing weed maps: a case study using kriging. Weed threshold level generally posed least spraying errors Science , 50 , 542 546. in the present study, there are no logical reasons to Dessaint, F., & Caussanel, J.-P. (1994). Trend surface analysis: a believe this is a general rule. Whether the increased simple tool for modelling spatial patterns of weeds. Crop or reduced threshold level will give the fewest Protection , 13 , 433 438. spraying errors depends on the field-specific pattern Dieleman, J. A., & Mortensen, D. A. (1999). Characterizing the spatial pattern of Abutilon theophrasti seedling patches. Weed and level of weed infestations. Research , 39 , 455 467. Fykse, H. (1991). Damage thresholds for weeds in spring cereals. Statens fagtjeneste for landbruket. Faginfo 2 (Informasjonsmøte Conclusion i plantevern), 165 173. (In Norwegian.) The simulated PS approach caused relatively small Fykse, H. (2004). Herbicide resistance. Plantemøtet Østlandet 2004, Grønn kunnskap , 8 , 347 357. (In Norwegian.) spraying errors for the 15 m (10%) and 24 m (15%) Fykse, H. & Wærnhus, K. (1994). Experimental fields for organic booms. Therefore, the simulated approach appears cropping systems. Characterization of the weed flora at the to be a worthwhile method for spring cereal fields. start of the experiment. Norwegian Agricultural Research , 8 , To support this statement, future studies should 217 233. (In Norwegian.) however include more fields, finer sampling grids, Gerhards, R., Wyse-Pester, D. Y., Mortensen, D., & Johnson, G. and image datasets should be the basis for the A. (1997). Characterizing spatial stability of weed popula- tions using interpolated maps. Weed Science , 45 , 108 119. simulations. Gerhards, R., So ¨ kefeld, M., Timmermann, C., & Ku ¨ hbauch, W. (2002). Site-specific weed control in maize, sugarbeet, winter wheat, and winter barley. Precision Agriculture , 3,25 35. Acknowledgements Gerhards, R., & Christensen, S. (2003). Real-time weed detec- tion, decision making and patch spraying in maize, sugar- The Norwegian Institute for Agricultural and En- beet, winter wheat and winter barley. Weed Research , 43 , vironmental Research (BIOFORSK) is gratefully 385 392. acknowledged for financial support. The authors Gerhards, R., Oebel, H., & Dicke, D. (2005). Practical experi- ences with a system for site-specific weed control using real also thank Kjell Wærnhus and Kristin Eide, BIO- time image analysis and GPS-controlled patch spraying FORSK, for performing the fieldwork. (TURBO). In: Proceedings of the 13 th EWRS Symposium . June 19 23, 2005. Bari, Italy. Gerowitt, B., & Heitefuss, R. (1990). Weed economic thresholds in cereals in the Federal Republic of Germany. Crop References Protection , 9 , 323 331. Heisel, T., Christensen, S., & Walter, A. M. (1999). Whole-field Andreasen, C., Rudemo, M., & Sevestre, S. (1997). Assessment experiments with site-specific weed management. In J V of weed density at an early stage by use of image processing. Stafford (Ed.), Proceedings of the Second European Conference Weed Research , 37,5 18. on Precision Agriculture Part 2 (pp. 759 768). Sheffield Anonymous. (2004). Action plan on reducing risk connected to Academic Press. the use of pesticides (2004 2008). Adopted by the Ministry Minitab Inc. (2003). MINITAB Statistical Software, Release 14 of Agriculture (of Norway) February 16, 2004. 14 pp. for Windows, State College . Pennsylvania. Barroso, J., Fernandez-Quintanilla, C., Maxwell, B. D., & Rew, L. Nordmeyer, H., Zuk, A., & Ha ¨ usler, A. (2003). Experiences of J. (2004). Simulating the effects of weed spatial pattern and site-specific weed control in winter cereals. In J Stafford, & A resolution of mapping and spraying on economics of site- Werner (Eds.), Proceedings of Precision Agriculture 2003, peer- specific management. Weed Research , 44 , 460 468. reviewed papers . Wageningen Academic Publishers. Bjugstad, N. (2000). Precise and secure application of pesticides. Nordmeyer, H. & Zwerger, P. (2005). Long-term investigations in Kompendium i kursene TLT100 og TLT200. Institutt for th tekniske fag, Norges Landbrukshøyskole. 175 pp. (In Nor- precision weed control. In: Proceedings of the 13 EWRS wegian.) Symposium . June 19 23, 2005. Bari, Italy. Black, I. D., & Dyson, C. B. (1993). An economic threshold Onyango, C. M., & Marchant, J. A. (2003). Segmentation of row model for spraying herbicides in cereals. Weed Research , 33 , crop plants from weeds using colour and morphology. 279 290. Computers and Electronics in Agriculture , 39 , 141 155. Patch spraying of weeds in spring cereals 221 Paice, M. E. R., Day, W., Rew, L. J., & Howard, A. (1998). A Søgaard, H. T. (2005). Weed classification by active shape models. stochastic simulation model for evaluating the concept of Biosystems Engineering , 91 , 271 281. patch spraying. Weed Research , 38 , 373 388. Tian, L. F., & Slaughter, D. C. (1998). Environmentally adaptive Pe ´rez, A. J., Lo ´ pez, F., Benlloch, J. V., & Christensen, S. (2000). segmentation algorithm for outdoor image segmentation. Colour and shape analysis techniques for weed detection in Computers and Electronics in Agriculture , 21 , 153 168. cereal fields. Computers and Electronics in Agriculture , 25 , Wallinga, J., Groeneveld, R. M. W., & Lotz, L. A. P. (1998). 197 212. Measures that describe weed spatial patterns at different Stafford, J. V., & Miller, P. C. H. (1993). Spatially selective levels of resolution and their applications for patch spraying application of herbicide to cereal crops. Computers and of weeds. Weed Research , 38 , 351 359. Electronics in Agriculture , 9 , 217 229. Wilkerson, G. G., Wiles, L. J., & Bennett, A. C. (2002). Weed Steward, B. L., & Tian, L. F. (1999). Machine-vision weed management decision models: pitfalls, perceptions, and density estimation for real-time, outdoor lighting conditions. possibilities of the economic threshold approach. Weed Transactions of the ASAE (Am. Soc. Agric. Eng.) , 42 , 1897 Science , 50 , 411 424. Zadoks, J. C., Chang, T. T., & Konzak, G. F. (1974). A decimal Søgaard, H. T., & Olsen, H. J. (2003). Determination of crop rows code for the growth stages of cereals. Weed Research , 14 , by image analysis without segmentation. Computers and 415 421. Electronics in Agriculture , 38 , 141 158. http://www.deepdyve.com/assets/images/DeepDyve-Logo-lg.png Acta Agriculturae Scandinavica Section B Taylor & Francis

Patch spraying of weeds in spring cereals: Simulated influences of threshold level and spraying resolution on spraying errors and potential herbicide reduction

Download PDF
10 pages

Loading next page...
 
/lp/taylor-francis/patch-spraying-of-weeds-in-spring-cereals-simulated-influences-of-aCbgeGK6dG

References (38)

Publisher
Taylor & Francis
Copyright
Copyright Taylor & Francis Group, LLC
ISSN
1651-1913
eISSN
0906-4710
DOI
10.1080/09064710600914202
Publisher site
See Article on Publisher Site

Abstract

A major obstacle to patch spraying of broad-leaved weeds in cereals is a cost-effective method to assess within-field heterogeneity of the weed population. One method could be a camera mounted in front of the spraying vehicle, online image analysis, and field sprayer shifting between ‘on’ and ‘off ’ as the predefined weed damage threshold level is reached. Because such a camera will capture a very limited area (B /1m ) compared to the sprayer width (several m), success of this method requires that spraying decisions vary little within boom width, thereby causing few spraying errors. This approach was evaluated by simulations for varying boom widths and three levels of a weed damage threshold model. Potential herbicide reductions compared to blanket application were also simulated. The average potential herbicide reductions estimated as proportions of fields below the threshold, were 60%, 64% and 53% for the original, 25% increased and 25% reduced threshold levels, respectively. The simulated herbicide reductions were not influenced by boom width, but varied significantly between fields, and between threshold levels. As evaluated by spraying errors, the suitability of the suggested approach will increase by decreasing boom width, vary between fields, and in some fields vary between the threshold levels. For boom widths of 15 m and 24 m, the spraying errors were about 10% and 15%, respectively, where omission of spraying areas above the threshold constituted 5% and 8%, respectively. Keywords: Herbicide reduction, precision farming, site-specific weed management. Introduction image analysis algorithms based on spectral reflec- tance characteristics, which has proved promising for Weeds generally occur in patches or clusters within row crops (Tian & Slaughter, 1998; Steward & Tian, agricultural fields (e.g. Dessaint & Caussanel, 1994; 1999; Onyango & Marchant, 2003). Therefore, Gerhards et al., 1997; Wallinga et al., 1998; Diele- other than spectral characteristics, e.g. shape and man & Mortensen, 1999; Cousens et al., 2002) contour of plants, have been tested as image analysis which represents a potential for treating only the measures (e.g. Andreasen et al., 1997; Clausen patches. Lack of cost-effective weed detection meth- et al., 2000; Pe ´rez et al., 2000; Søgaard & Olsen, ods on a field scale has so far prevented implementa- 2003). Promising results have been achieved for tion of patch spraying (PS) in cereal crops. single species recognition without overlapping leaves In Europe, cereals are normally grown with a row (Gerhards & Christensen, 2003; Søgaard, 2005). spacing about 1213 cm, and herbicide spraying There are economic, environmental and govern- against seed-propagated broad-leaved weeds is con- mental pressures for reducing herbicide inputs ducted when the crop has 34 leaves. At this stage (Anon., 2004). The average reduction of herbicide crop and weed plants have nearly the same spectral consumption from site-specific weed control in reflectance characteristics. Together with the short spring cereals ranged from 40 to 54%, and from 50 row spacing, this has prevented development of Correspondence: T. Berge, Plant Health and Plant Protection Division, BIOFORSK, Hoegskoleveien 7, NO-1432 Aas, Norway. Tel:  / 47 64 94 93 56/ / 47 64 94 94 00/ / 47 64 94 42 20. Fax:  / 47 64 94 92 26. E-mail: [email protected] (Received 7 September 2005; accepted 6 June 2006) ISSN 0906-4710 print/ISSN 1651-1913 online # 2007 Taylor & Francis DOI: 10.1080/09064710600914202 Patch spraying of weeds in spring cereals 213 to 61% in winter cereals (Heisel et al., 1999; weeds in spring cereals; 2) estimate potential herbi- Nordmeyer et al., 2003; Christensen, 2005). In cide reductions; and 3) test whether results were Scandinavia cereals are the most widespread arable dependent on the actual weed damage threshold crop, and in Norway about 90% of the total pesticide level. use is in cereals, the main proportion being herbi- cides (Bjugstad, 2000). The potential for reducing Material and methods pesticide use through PS in cereals should therefore Weed data were collected in 1993 at three spring be significant. cereal fields in Norway (Table I), originally for Reports on spatial variation of weed density within another type of investigation (Fykse & Wærnhus, cereal fields in Norway are few (Fykse & Wærnhus, 1994). The weeds were counted within quadrats 1994; Fykse, 2004). Fykse (2004) plotted total weed density along 180 m transects from four cereal fields. sized 0.5 /0.5 m when the crop was at Zadoks 1314 (Zadoks et al., 1974), the normal time for Spatial variation in weed density was high, both post-emergence herbicide application. The quadrats within field and between fields. Interestingly, there was clear autocorrelation in weed density: positions were separated by 10 to 18 m (Table I). Spraying decisions ‘spray’/‘not spray’ were based with high density were neighbours with positions with high density, and vice versa. on a threshold model (Fykse, 1991) developed for Despite several proposed economic threshold seed-propagated broad-leaved weeds in spring cer- eals at Zadoks 1314 (Table II). If the total weed models for annual weeds in cereals (e.g. Gerowitt & Heitefuss, 1990; Fykse, 1991; Black & Dyson, density exceeded 175 plants m , or the threshold 1993), the practical use has been limited because density for at least one of five strongly competitive cost-effective methods to assess the weed popula- weed species was passed, the decision was ‘spray’. Furthermore, if the sum of two or more of these five tions have been missing (Wilkerson et al., 2002). Image analysis is probably an appropriate solution, species passed the threshold density for the species and from the above literature study, the necessary with the highest threshold value, the decision was algorithms will probably be available within a few also ‘spray’ (Table II). Presence of Galium aparine often leads to lodging, and the threshold value years. To reduce costs, one digital camera and the ordinary sprayers already in use should be the basis (1 plant m ) expresses the problems connected for PS. Therefore we investigated, by means of with harvest and drying, and not its yield-depressing simulations, the suitability for real-time on/off PS effect. As an attempt to apply the weed densities from the original threshold model with more con- in cereal fields based on the information from one camera. We simulated one camera mounted in front fidence, the original density levels were also in- of the tractor taking images along a narrow path in creased and reduced by 25% (Table II). The the middle of the sprayer boom, and the nozzles original threshold model included also relative switched on and off as the weed infestation was crop/weed ground coverage, but as this parameter above or below the damage threshold. To reduce was not recorded in the current fields, it was omitted spraying errors, this design requires that spraying in this study. decisions are relatively strongly correlated within the Spraying decisions (‘spray’/‘not spray’) were taken actual boom widths. The objectives of this study per quadrat and plotted for each field. These maps were to: 1) assess the suitability of using one camera were the basis for manual simulations of PS with per boom to control ordinary field sprayers on/off for varying sprayer boom widths based on the spraying patch spraying of seed-propagated broad-leaved decision of one quadrat per boom width. Increased Table I. Field and sampling characteristics. Voll Landvik Kvithamar Position 59841?N, 10846?E58820?N, 8831?E63830?N, 10852?E Precipitation, mm 600 1230 892 Growing days ( /58C) 169 202 182 Temperature sum ( /58C), degree days 1091 1442 1064 Sampling date in 1993 June 3 4 June 7 June 28 Number of quadrats 190 84 124 Metres between quadrats, N-S direction: 10, 10.5, 13, 14 11, 12 12, 13, 14 E-W direction: 15, 16, 18 10 10, 13, 14, 15 Normal period 1961 1990. 214 T.W. Berge et al. Table II. Definition and threshold values (weeds m ) of the model after Fykse (1991). Decision is ‘spray’ if criterion XX is met. Original threshold 9 /25% modified Criteria Definition level threshold levels A Total density of seed-propagated broad-leaved weed species 175 131 219 B Density of Galeopsis ssp. L. 25 19 31 C Density of Chenopodium album L. 45 34 56 D Density of Brassica rapa ssp. oleifera (DC.) Metzger 20 15 25 E Density of Stellaria media (L.) Vill. 45 34 56 F Density of Galium aparine L. 1 1 X Density of at least two of the species B F higher than the threshold for the species with the highest value XX ‘Full model’, one of the criteria A X fulfilled For G. aparine not relevant, explained in text. boom widths were simulated for the N-S direction coefficients. Minitab (Minitab Inc., 2003) was used in all statistical analysis. and the E-W direction by including adjacent quad- rats. For a given boom width, spraying decisions could either be conspecific of type ‘spray’, conspe- Results cific of type ‘not spray’, or heterogeneous including Mean total weed density per field varied more than both ‘spray’ and ‘not spray’ quadrats. Proportions of five-fold, from 53 to 295 plants m (Table III). In the sum of conspecific spraying decisions per given terms of standard deviation and coefficient of varia- boom width, hereafter termed ‘conspecific booms’, tion, the within-field variation of total weed density were plotted against boom width. was high (Table III). No field had random distribu- The following PS approaches, with constant tion of total weed density (pB /0.005, Table III). The detection resolution, but decreasing spraying resolu- frequency distributions of total weed density were tion, were simulated: 1) PS-1: spraying decisions clearly positively skewed at Voll and Kvithamar, but assigned to each quadrat independently, simulating a only slightly skewed at Landvik (Figure 1). Landvik patch sprayer with independently working sections showed flatter than normal distribution; elsewhere and one camera per section, or alternatively a very the distributions were sharper than normal, with narrow sprayer guided by one camera. This would high (Kvithamar) and very high (Voll) positive yield the most flexible spraying, and represented the kurtosis (Table III). best estimate of potential herbicide reductions Conspecific spraying decisions were more or less through PS for the dataset; 2) PS-2: the same aggregated at all fields, and for all threshold levels spraying decision assigned to two adjacent quadrats (Figure 2). Typically, weed patches defined as areas based on weed conditions at the quadrat viewed by consisting of nearest neighbour ‘spray’ quadrats the simulated camera. This simulated a wider boom using the original threshold level increased and/or (with no possibility for independent actions by the merged when the threshold level was reduced, and boom sections) and one camera per boom; 3) PS-3: shrank and/or divided if the threshold level was the same spraying decision assigned to three adjacent increased (Figure 2). quadrats based on the weed status of the middle one, For proportions of ‘conspecific booms’ (Figure 3), simulating an even wider boom (with no possibility field (p5 /0.001), threshold level (p /0.002) and for independent actions by the boom sections) and boom width (p5 /0.001) were main effects, but not one camera per boom. Compared to PS-1, using PS- the direction within field. The average reduction in 2 and PS-3 would inevitably result in ‘spraying the proportion of ‘conspecific booms’ per metre errors’. These were either failures caused by not boom was equal at Voll and Landvik (p /0.918), spraying quadrats with infestations above the thresh- and less than the reduction at Kvithamar old (error type 1), or spraying quadrats with infesta- (p5 /0.001). The intercept for Voll and Landvik tions below the threshold (error type 2). was higher (p /0.008) than the intercept for Kvitha- Differences between directions, threshold levels mar (Figure 3). For simulated boom widths covering and fields were tested with analysis of variance. 2, 3 and 4 quadrats, there is a probability of Tukey tests were applied to detect the actual 50%, 25% and 12.5%, respectively, to obtain con- differences. Linear regression models were fitted to specific decisions if the decisions are randomly the simulation results, and two-sample t -tests distributed. All simulations at Voll and Landvik were used to test differences between regression were well above these limits of randomness, whereas Patch spraying of weeds in spring cereals 215 Table III. Summary statistics of total weed density and threshold model species per field. Voll Landvik Kvithamar Mean total weed density, weeds m 53 295 131 Minimum total weed density, weeds m 00 8 Maximum total weed density, weeds m 432 844 628 Proportion of quadrats without weeds, % 5 1 0 Standard deviation of mean density, weeds m 59 244 115 Coefficient of variation, % 112 83 88 Skewness 2.76 0.66 1.99 Kurtosis 11.46  /0.63 4.72 Total weed density randomly distributed 10.85 2.19 6.55 Mean density of Galeopsis ssp., plants m 30 11 Mean density of C. album , plants m 782 18 Mean density of B. rapa ssp. oleifera , plants m  10 Mean density of S. media , plants m 119 5 Mean density of G. aparine plants m 1 Anderson-Darling normality test statistic. p valueB /0.005 for all fields. a few simulations were below at Kvithamar widths, and the mean proportions of ‘conspecific (Figure 3). Increased threshold levels caused a booms’ for 15 m and 24 m booms were 69% and 61%, respectively. significantly higher proportion of ‘conspecific The proportions of potential reduction in herbi- booms’ than reduced threshold levels, both on cide use (Figure 4) varied significantly between fields an average of all fields (p /0.002), and at Voll (p5 /0.001) and threshold levels (p5 /0.001), but not (p5 /0.001) and Kvithamar (p /0.020), but not at by boom width. Using resolution PS-1, 86%, 36% Landvik (Figure 3). The proportions of ‘conspecific and 59% of the quadrats were below the original booms’ decreased linearly with increasing boom threshold level at Voll, Landvik and Kvithamar, respectively (Figure 4), corresponding to a mean of Voll 60%. Estimates obtained by resolution PS-1 for the increased and decreased threshold levels were 64% and 53%, respectively, but not statistically different from the estimate using the original level. The magnitude of the tested threshold levels was of no influence on reduction at Landvik (Figure 4), whereas at Voll the reduced level gave less herbicide reduction than the increased level (p /0.017), and at Landvik Kvithamar the reduced level gave less than both the increased and the original level (p5 /0.009). By 10 applying wider booms (PS-2 and PS-3) both smaller and greater individual savings than by PS-1 were observed (Figure 4), but neither linear slope fitted to the mean of all fields and all threshold levels, nor fitted linear slopes per threshold level, were signifi- cantly different from zero (p] /0.051). At field level, Landvik was the only one with fitted slope different Kvithamar from zero ( /0.4% m , p /0.040). At every field all spraying error types increased linearly with boom width, but less than 1% m (Figure 5). For 15 m and 24 m booms, predicted average total spraying errors (95% confidence inter- vals in brackets) were about 10% (911%) and 15% (1417%), respectively. Predicted type 1 and type 2 spraying errors for these widths were 5% (45%) 0 120 240 360 480 600 720 840 and 8% (79%), respectively. Generally, Kvithamar had the highest spraying Figure 1. Frequency distributions of total weed density (weeds m ) per field. errors, Voll the second highest and Landvik the Percent 216 T.W. Berge et al. (a) Voll 0 40 80 120 160 200 240 280 320 (b) Landvik 0 50 100 150 200 (c) Kvithamar N S 0 50 100 150 200 250 300 X-coordinate (m) Figure 2. Spraying decision maps for PS-1. Filled (j) and open (I) squares denote ‘spray’ and ‘not spray’ quadrats for original threshold level. Circles (k) mark additionally ‘spray’ quadrats if the threshold level was reduced by 25%. Diamonds (2) mark quadrats that changed to ‘not spray’ if the threshold level was increased by 25%. lowest errors (Figure 5). This ranking was significant (p /0.032). For total error the reduced level also for total and type 1 spraying errors (p5 /0.046). caused higher (p / 0.034) error rates than the Main effect of threshold level on spraying errors was original level (Figure 5, bottom row). At Kvithamar, observed at Voll and Kvithamar, but not Landvik. At both the reduced (p /0.002) and the original (p5 / Voll, the reduced level gave higher errors than the 0.001) level gave higher total error than the in- increased level for total (p /0.017) and type 1 error creased level. Y-coordinate (m) Patch spraying of weeds in spring cereals 217 Voll Landvik Kvithamar 2 2 2 2 3 2 2 4 80 3 2 2 2 2 3 3 2 3 2 4 4 70 3 4 5 2 34 4 3 5 2 3 2 2 3 2 3 5 2 4 3 3 2 2 4 4 4 4 4 2 2 2 3 3 2 quadrats 2 3 Threshold Level 3 3 4 30 4 4 - 25% 3 4 3 quadrats + 25% 3 20 3 original 4 quadrats 15 25 35 15 25 35 15 25 35 Simulated boom width (m) Figure 3. Proportions of booms with conspecific spraying decisions versus simulated boom width. Labels on datapoints are numbers of quadrats per boom width. Horizontal lines mark limits for random distributions of spraying decisions for 2, 3 and 4 quadrats per boom. Discussion conspecific spraying decisions to be relatively strongly autocorrelated at the boom-scale too. In Suitability of simulated PS approach due to spraying Norway, sprayer booms have traditionally been less errors than 15 m, but up to 24 m wide booms are of Potential for PS of the three fields was indicated growing interest. The predicted average proportions by the variation in total weed density (Figure 1), of ‘conspecific booms’ (Figure 3) for boom widths visual aggregation of conspecific spraying decisions 15 m and 24 m were relatively high: 69% and 61%, (Figure 2) and proportions of the fields below the respectively. The associated total spraying errors threshold (Figure 4). These were field-scale results. (Figure 5) were about 10% (15 m) and 15% Suitability of the simulated PS approach requires (24 m). During operational PS, it is probably harder Voll Landvik Kvithamar 15 24 15 24 15 24 0 10 20 30 40 0 10 20 30 40 0 10 20 30 40 Simulated boom width (m) Figure 4. Proportions of field simulated unsprayed versus simulated boom width. Diamonds are values for PS-1 and dots are values for PS-2 and PS-3. Legend: see Figure 3. Proportions of field unsprayed (%) Proportions of 'conspecific booms' (%) 218 T.W. Berge et al. Voll Landvik Kvithamar Error, type 1 Error, type 1 Error, type 1 15 24 15 24 15 24 10 20 30 10 20 30 10 20 30 Error, type 2 Error, type 2 Error, type 2 15 24 15 24 15 24 10 20 30 10 20 30 10 20 30 Total Error Total Error Total Error 15 24 15 24 15 24 10 20 30 10 20 30 10 20 30 Simulated boom width (m) Figure 5. Proportions of field with spraying errors versus simulated boom width. Upper row: error type 1, middle row: error type 2, bottom row: sum of error type 1 and error type 2. Legend: see Figure 3. for farmers to accept omission of spraying patches narrower than the boom width and/or are very above the threshold than accepting spraying areas irregular in shape. As the spraying errors decreased with decreasing below the threshold. As the former error type was boom width (Figure 5), suitability of the simulated relatively low (about 5% (15 m) and 8% (24 m)), the simulated PS approach may be acceptable, particu- PS concept will generally increase as boom width decreases. Based on our results, using selective larly among farmers using the 15 m wide booms. The expected short- and long-term herbicide cost control of individual 2 m (Stafford & Miller, 1993) reductions compared to total costs, and the farmers’ and 3 m (Gerhards et al., 2002) boom sections would have reduced the average total spraying personal experiences and interests, will influence acceptable error rates. errors to about 1% and 2%, respectively. These Suitability of the simulated PS approach is likely low error rates require high mapping and spraying to vary between fields due to differences in spraying resolutions which necessarily involve relatively com- error levels. In the present study Landvik had the plex systems that probably are less robust and lowest errors, and Kvithamar the highest (Figure 5). more expensive (Paice et al., 1998) than our This could be explained by the spatial distributions simulated PS approach. In a simulation study, of ‘spray’ and ‘not spray’ quadrats at the boom width Barroso et al. (2004) found that the critical para- scale. Whereas Landvik had relatively elongated and meter that determined the economic viability of well-defined patches of neighbouring ‘spray’ quad- patch mapping and spraying resolution was the rats (Figure 2) and relatively strongly autocorrelated technology costs. Whether very high detection conspecific spraying decisions (Figure 3), Kvithamar and spraying resolution would facilitate adoption of had comparatively frequent shifts between ‘spray’ PS in cereal fields therefore remains an open and ‘not spray’ quadrats (Figure 2) and conspecific question. decisions were less autocorrelated (Figure 3). Gen- Is there a relationship between the level of spray- erally, a field where weed patches fit the width of the ing error of a field and the general weed infestation sprayer boom will have fewer spraying errors than a level? Analysing these variables, it seemed that if the field in which the majority of the patches are average field density of at least one of the threshold Proportions of field with spraying errors (%) Patch spraying of weeds in spring cereals 219 criteria AF (Table II) was close to its threshold (Figure 2) and the limited number of possible value, the field was likely to have relatively high simulations. Effects of boom width on the amount spraying errors. This is because: 1) an average of herbicide reduction would probably happen more density near the threshold value indicates that a often in fields where the patterns of conspecific relatively large proportion of the field has weed spraying decisions are shaped as rectangles and/or densities near that value (cf. Figure 1 for criterion relatively small-sized fields, than in fields with weed A); and 2) weed densities within fields are generally infestations of more irregular pattern and/or larger autocorrelated (e.g. Cardina et al., 1997). Conse- fields. quently, if a quadrat’s density is far below or far The relatively high, but expected field-to-field above the threshold, it is likely that the neighbouring variation in potential herbicide reduction (Figure quadrat is also far below or far above, and 4) is explained by different weed infestation levels of hence these neighbours are likely to have the same the fields (Table III), and is in agreement with, e.g. spraying decision. However, if a quadrat’s density Gerhards et al. (2005) who reported variation of is near the threshold (a little above or a little below), 2080% in herbicide savings by PS against broad- it is likely that the neighbour is also near the leaved weeds in German spring barley. Within a six-year period of PS of dicotyledonous weeds in threshold, but it is more a matter of chance whether these quadrats have the same spraying decisions German winter cereals, mean herbicide reduction compared to neighbours with densities far from could be less than 30% in one year and more the threshold. Hence, areas with densities near the than 60% in another (Nordmeyer & Zwerger, threshold are likely to cause higher quadrat- 2005). Year-to-year variation in savings must also to-quadrat variation in spraying decisions than be expected. areas having densities far from the threshold. As it The proportion of a field being above the damage is this quadrat-to-quadrat variation that generates threshold affects the suitability of PS. Barroso et al. the spraying errors, a field with weed densities (2004) found that the profitability of site-specific close to one or more of the threshold criteria is Avena sterilis control in Spanish winter barley gen- likely to have relatively high spraying errors. For erally increased as the proportion of the field infested example, at Kvithamar the average weed densities of decreased. This is very logical and probably also criteria A, B and C (Table III) were relatively close to valid for our simulated PS approach and data. Hence the threshold densities (Table II), and this field had Voll would be the most, and Landvik the least, the highest spraying errors (Figure 5). profitable field to patch spray (in the year of investigation) (cf. Figure 4). Potential herbicide reduction Influence of the tested threshold levels The average potential herbicide reductions esti- mated by PS-1 were substantial, 5364%, depend- At Landvik neither proportions of ‘conspecific ing on threshold level. These were of the same order booms’ (Figure 3), potential herbicide reduction as reported from practical PS of seed-propagated (Figure 4) nor spraying errors (Figure 5) were broad-leaved weeds in spring barley in Denmark affected by different threshold levels, whereas at (Heisel et al., 1999) and in winter cereals in Voll and Kvithamar these variables were influenced. Germany (Nordmeyer et al., 2003; Gerhards et al., This happened because at Landvik the proportions 2005). of ‘spray’ quadrats were relatively stable for different On average, boom width was of no main influence threshold levels (cf. Figure 2): 68% at reduced level on the potential herbicide reduction, meaning that and 63% at increased level, leading to a relative savings of the same order as estimated by PS-1 can difference of only 8% (100*(68%63%)/63%) be- also be expected for wider booms. However, as both tween the modified levels. At Voll and Kvithamar, on smaller and greater reductions than by PS-1 oc- the other hand, the relative differences in propor- curred from resolutions PS-2 and PS-3 (Figure 4), tions of ‘spray’ quadrats between the modified the actual saving obtained for one field, at least a threshold levels were much higher: 54% and 74%, relatively small one, might depend on the relative respectively, and high enough to yield significant camera position and starting point of spraying. In the effects of the threshold levels on the tested variables. long term, i.e. several years and/or larger fields, there In addition to the expected increase in herbicide will be no boom width influence on herbicide savings due to reduced threshold level seen at Voll, savings. The significant, but weakly decreasing and particularly at Kvithamar (Figure 4), higher reduction by the increasing boom width seen at proportions of ‘conspecific booms’ (Figure 3) Landvik (Figure 4) was probably due to the rectan- and reduced total spraying errors (Figure 5) were gular-like pattern of conspecific spraying decisions observed for the increased compared to the reduced 220 T.W. Berge et al. Cardina, J., Johnson, G. A., & Sparrow, D. H. (1997). The nature level. The two latter variables, which both depend and consequence of weed spatial distribution. Weed Science , on the local (boom-scale) distribution of conspecific 45 , 364 373. spraying decisions, were affected because the local Christensen, S. (2005). Table I. Potential savings of herbicides in patterns between the modified levels were signifi- percentage of common practice in 59 trials with site-specific cantly different due to the substantial relative differ- weed management in Europe. http://www.ewrs-sswm.org/ homepage.htm, March 2005. ences in ‘spray’ quadrats. At Landvik, however, the Clausen, S., Nicolas, S., Kirkhus, T., & Fykse, H. (2000). pattern differed only marginally between the differ- Automatic weed mapping in cereal fields using image ent threshold levels (cf. Figure 2). processing techniques. In: Proceedings of NOBIM-2000, the Because the threshold level affected the spraying Norwegian Image Processing and Pattern Recognition Confer- errors significantly in some fields, the suitability of ence , June 6 7, 2000. Trondheim, Norway. 5 pp. the simulated PS approach may vary with the Cousens, R. D., Brown, R. W., McBratney, A. B., Whelan, B., & Moerkerk, M. (2002). Sampling strategy is important for applied threshold level. Even if the 25% increased producing weed maps: a case study using kriging. Weed threshold level generally posed least spraying errors Science , 50 , 542 546. in the present study, there are no logical reasons to Dessaint, F., & Caussanel, J.-P. (1994). Trend surface analysis: a believe this is a general rule. Whether the increased simple tool for modelling spatial patterns of weeds. Crop or reduced threshold level will give the fewest Protection , 13 , 433 438. spraying errors depends on the field-specific pattern Dieleman, J. A., & Mortensen, D. A. (1999). Characterizing the spatial pattern of Abutilon theophrasti seedling patches. Weed and level of weed infestations. Research , 39 , 455 467. Fykse, H. (1991). Damage thresholds for weeds in spring cereals. Statens fagtjeneste for landbruket. Faginfo 2 (Informasjonsmøte Conclusion i plantevern), 165 173. (In Norwegian.) The simulated PS approach caused relatively small Fykse, H. (2004). Herbicide resistance. Plantemøtet Østlandet 2004, Grønn kunnskap , 8 , 347 357. (In Norwegian.) spraying errors for the 15 m (10%) and 24 m (15%) Fykse, H. & Wærnhus, K. (1994). Experimental fields for organic booms. Therefore, the simulated approach appears cropping systems. Characterization of the weed flora at the to be a worthwhile method for spring cereal fields. start of the experiment. Norwegian Agricultural Research , 8 , To support this statement, future studies should 217 233. (In Norwegian.) however include more fields, finer sampling grids, Gerhards, R., Wyse-Pester, D. Y., Mortensen, D., & Johnson, G. and image datasets should be the basis for the A. (1997). Characterizing spatial stability of weed popula- tions using interpolated maps. Weed Science , 45 , 108 119. simulations. Gerhards, R., So ¨ kefeld, M., Timmermann, C., & Ku ¨ hbauch, W. (2002). Site-specific weed control in maize, sugarbeet, winter wheat, and winter barley. Precision Agriculture , 3,25 35. Acknowledgements Gerhards, R., & Christensen, S. (2003). Real-time weed detec- tion, decision making and patch spraying in maize, sugar- The Norwegian Institute for Agricultural and En- beet, winter wheat and winter barley. Weed Research , 43 , vironmental Research (BIOFORSK) is gratefully 385 392. acknowledged for financial support. The authors Gerhards, R., Oebel, H., & Dicke, D. (2005). Practical experi- ences with a system for site-specific weed control using real also thank Kjell Wærnhus and Kristin Eide, BIO- time image analysis and GPS-controlled patch spraying FORSK, for performing the fieldwork. (TURBO). In: Proceedings of the 13 th EWRS Symposium . June 19 23, 2005. Bari, Italy. Gerowitt, B., & Heitefuss, R. (1990). Weed economic thresholds in cereals in the Federal Republic of Germany. Crop References Protection , 9 , 323 331. Heisel, T., Christensen, S., & Walter, A. M. (1999). Whole-field Andreasen, C., Rudemo, M., & Sevestre, S. (1997). Assessment experiments with site-specific weed management. In J V of weed density at an early stage by use of image processing. Stafford (Ed.), Proceedings of the Second European Conference Weed Research , 37,5 18. on Precision Agriculture Part 2 (pp. 759 768). Sheffield Anonymous. (2004). Action plan on reducing risk connected to Academic Press. the use of pesticides (2004 2008). Adopted by the Ministry Minitab Inc. (2003). MINITAB Statistical Software, Release 14 of Agriculture (of Norway) February 16, 2004. 14 pp. for Windows, State College . Pennsylvania. Barroso, J., Fernandez-Quintanilla, C., Maxwell, B. D., & Rew, L. Nordmeyer, H., Zuk, A., & Ha ¨ usler, A. (2003). Experiences of J. (2004). Simulating the effects of weed spatial pattern and site-specific weed control in winter cereals. In J Stafford, & A resolution of mapping and spraying on economics of site- Werner (Eds.), Proceedings of Precision Agriculture 2003, peer- specific management. Weed Research , 44 , 460 468. reviewed papers . Wageningen Academic Publishers. Bjugstad, N. (2000). Precise and secure application of pesticides. Nordmeyer, H. & Zwerger, P. (2005). Long-term investigations in Kompendium i kursene TLT100 og TLT200. Institutt for th tekniske fag, Norges Landbrukshøyskole. 175 pp. (In Nor- precision weed control. In: Proceedings of the 13 EWRS wegian.) Symposium . June 19 23, 2005. Bari, Italy. Black, I. D., & Dyson, C. B. (1993). An economic threshold Onyango, C. M., & Marchant, J. A. (2003). Segmentation of row model for spraying herbicides in cereals. Weed Research , 33 , crop plants from weeds using colour and morphology. 279 290. Computers and Electronics in Agriculture , 39 , 141 155. Patch spraying of weeds in spring cereals 221 Paice, M. E. R., Day, W., Rew, L. J., & Howard, A. (1998). A Søgaard, H. T. (2005). Weed classification by active shape models. stochastic simulation model for evaluating the concept of Biosystems Engineering , 91 , 271 281. patch spraying. Weed Research , 38 , 373 388. Tian, L. F., & Slaughter, D. C. (1998). Environmentally adaptive Pe ´rez, A. J., Lo ´ pez, F., Benlloch, J. V., & Christensen, S. (2000). segmentation algorithm for outdoor image segmentation. Colour and shape analysis techniques for weed detection in Computers and Electronics in Agriculture , 21 , 153 168. cereal fields. Computers and Electronics in Agriculture , 25 , Wallinga, J., Groeneveld, R. M. W., & Lotz, L. A. P. (1998). 197 212. Measures that describe weed spatial patterns at different Stafford, J. V., & Miller, P. C. H. (1993). Spatially selective levels of resolution and their applications for patch spraying application of herbicide to cereal crops. Computers and of weeds. Weed Research , 38 , 351 359. Electronics in Agriculture , 9 , 217 229. Wilkerson, G. G., Wiles, L. J., & Bennett, A. C. (2002). Weed Steward, B. L., & Tian, L. F. (1999). Machine-vision weed management decision models: pitfalls, perceptions, and density estimation for real-time, outdoor lighting conditions. possibilities of the economic threshold approach. Weed Transactions of the ASAE (Am. Soc. Agric. Eng.) , 42 , 1897 Science , 50 , 411 424. Zadoks, J. C., Chang, T. T., & Konzak, G. F. (1974). A decimal Søgaard, H. T., & Olsen, H. J. (2003). Determination of crop rows code for the growth stages of cereals. Weed Research , 14 , by image analysis without segmentation. Computers and 415 421. Electronics in Agriculture , 38 , 141 158.

Journal

Acta Agriculturae Scandinavica Section B – Taylor & Francis

Published: Sep 1, 2007

Keywords: Herbicide reduction; precision farming; site-specific weed management

There are no references for this article.