bioRxiv preprint doi: https://doi.org/10.1101/2021.03.13.435242; this version posted March 13, 2021.

The copyright holder for this preprint

(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is
made available under aCC-BY-NC-ND 4.0 International license.

1 Running title: Modelling for wheat bulb fly tolerance

3 Development of a pest threshold decision support system for minimising damage to winter wheat
4 from wheat bulb fly, Delia coarctata

6 Daniel J Leybourne 1,2,‡, Kate E Storer 1,‡, Pete Berry 1, Steve Ellis 1

7
1
8 ADAS, High Mowthorpe, Duggleby, Malton, North Yorkshire. YO18 9BP
2
9 Zoological Biodiversity, Institute of Geobotany, Leibniz University of Hannover, Hannover, Germany
‡
10 These authors contributed equally

11
 bioRxiv preprint doi: https://doi.org/10.1101/2021.03.13.435242; this version posted March 13, 2021. The copyright holder for this preprint
(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is
made available under aCC-BY-NC-ND 4.0 International license.

12 Graphical Abstract

13 In this article we describe two predictive models that can be used for the integrated management of
14 wheat bulb fly. Our first model is a pest level prediction model and our second model predicts the
15 number of shoots a winter wheat crop will achieve by the terminal spikelet developmental stage. We
16 revise and update current wheat bulb fly damage thresholds and combine this with our two models
17 to devise a tolerance-based decision support system that can be used to minimise the risk of crop
18 damage by wheat bulb fly.

19

20
 bioRxiv preprint doi: https://doi.org/10.1101/2021.03.13.435242; this version posted March 13, 2021. The copyright holder for this preprint
(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is
made available under aCC-BY-NC-ND 4.0 International license.

21 Summary

22 Wheat bulb fly, Delia coarctata, is an important pest of winter wheat in the UK, causing significant
23 damage of up to 4 t ha-1. Accepted population thresholds for D. coarctata are 250 eggs m-2 for crops
24 sown up to the end of October and 100 eggs m -2 for crops sown from November. Fields with
25 populations of D. coarctata that exceed the thresholds are at higher risk of experiencing economically
26 damaging pest infestations. In the UK, recent withdrawal of insecticides means that only a seed
27 treatment is available for chemical control of D. coarctata, however this is only effective for late-sown
28 crops (November onwards) and accurate estimations of annual population levels are required to
29 ensure a seed treatment is applied if needed. As a result of the lack of post-drilling control strategies,
30 the management of D. coarctata is becoming increasingly reliant on non-chemical methods of
31 control. Control strategies that are effective in managing similar stem-boring pests of wheat include
32 sowing earlier and using higher seed rates to produce crops with more shoots and greater tolerance
33 to shoot damage.

34 In this study we develop two predictive models that can be used for integrated D. coarctata
35 management. The first is an updated pest level prediction model that predicts D. coarctata
36 populations from meteorological parameters with a predictive accuracy of 70%, which represents a
37 significant improvement on the previous D. coarctata population prediction model. Our second model
38 predicts the maximum number of shoots for a winter wheat crop that would be expected at the
39 terminal spikelet development stage. This shoot number model uses information about the thermal
40 time from plant emergence to terminal spikelet, leaf phyllochron length, plant population, and sowing
41 date to predict the degree of tolerance a crop will have against D. coarctata. The shoot number
42 model was calibrated against data collected from five field experiments and tested against data from
43 four experiments. Model testing demonstrated that the shoot number model has a predictive
44 accuracy of 70%. A decision support system using these two models for the sustainable
45 management of D. coarcata risk is described.

46 Keywords

47 Agronomy, Crop Modelling, Crop Tolerance, Insect Pests, Integrated Pest Management, Modelling,
48 Pest Forecasting

49
 bioRxiv preprint doi: https://doi.org/10.1101/2021.03.13.435242; this version posted March 13, 2021. The copyright holder for this preprint
(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is
made available under aCC-BY-NC-ND 4.0 International license.

50 Introduction

51 The wheat bulb fly, Delia coarctata (Fallén), is an important herbivorous insect of wheat in the UK.
52 Significant economic damage to winter wheat crops is caused by D. coarctata larvae between
53 January and April (Fig. 1) when larvae infest the developing shoots of cereal crops, causing shoot
54 discolouration and stunting (‘deadhearts’). Economic damage can vary between years, reaching 4 t
55 h-1 in years of significant infestation (Rogers et al., 2014). D coarctata larvae feed until late spring
56 before pupating at the base of the plant, upon emergence adult D. coarctata feed on saprophytic
57 fungi present on plant tissue (Jones, 1970) and reproduce before migrating to adjacent fields where
58 oviposition occurs on the bare soil (Bardner et a., 1977).

59 The level of D. coarctata risk fluctuates yearly (Young & Cochrane, 1993). Previous studies have
60 indicated that this correlates with January temperature, July temperature, and August rain days and
61 can be affected further by the previous crop grown in the rotation and the date of harvest (Young &
62 Cochrane, 1993). In the UK, recent withdrawal of insecticides means that only a seed treatment
63 (Signal 300 ES 300 g/l cypermerthrin, UPL Europe Ltd) is available for chemical control of D.
64 coarctata. This seed treatment is only effective for late-sown crops (November onwards) as the
65 active ingredient is insufficiently persistent for crops sown at the more conventional drilling dates of
66 September and October. Therefore, alternative non-chemical means of D. coarctata control, that are
67 capable of managing D. coarctata in crops sown at conventional drilling dates, are becoming
68 increasingly desirable.

69
70 Fig. 1: Illustrative example of the annual life cycle of D. coarctata – this image was created with BioRender.com
 bioRxiv preprint doi: https://doi.org/10.1101/2021.03.13.435242; this version posted March 13, 2021. The copyright holder for this preprint
(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is
made available under aCC-BY-NC-ND 4.0 International license.

71 One option for the non-chemical control of D. coarctata is to adjust crop management practices in
72 order to produce a wheat crop that is able to tolerate D. coarctata damage. Crop tolerance, or the
73 economic injury level, can be broadly defined as the amount of pest damage that a crop can
74 withstand before an economic consequence is observed (Stern et al., 1959). The most recent D.
75 coarctata pest thresholds of 250 or 100 eggs m-2, for crops sown before the end of October and from
76 November, respectively, were devised c. 60 years ago by Gough et al (1961) and only take account
77 of pest abundance with no consideration of crop tolerance. Adjusting various agronomic factors,
78 such as sowing date and seed rate, has been used to successfully achieve tolerance against other
79 stem-boring pests of wheat, including the gout fly, Chlorops pumilionis (Bjerkander) (Bryson et al.,
80 2005) and wheat saw fly, Cephus cinctus Norton (Beres et al., 2011). Modifying agronomic practices
81 to achieve crop tolerance represents a potential method of non-chemical D. coarcata control, but to
82 achieve this accurate D. coarctata thresholds are required.

83 For stem-boring insects, crop tolerance can be achieved by growing a crop with a greater number of
84 shoots than those required to achieve an economically viable yield, this allows some shoots to be
85 lost to insect herbivory without incurring a yield penalty. It has been shown that UK wheat crops
86 require a minimum of 400 to 450 fertile shoots (ears) m-2 to achieve a typical commercial grain yield
87 (Spink et al., 2000). Wheat crops typically produce more than 1,000 shoots m-2 (and up to 1,600
88 shoots m-2) by the start of stem extension in March (Berry et al., 2003). From March to May there is
89 a decline in shoot numbers as the weakest shoots die, leaving a final shoot number of 400 to 700
90 fertile shoots m-2. In most cases, the majority of the shoots are produced during the autumn months,
91 and are therefore present when D. coarcata larvae infest plant tissue (January - April; Fig. 1). The
92 main crop management methods that can be used to increase maximum shoot number in spring are
93 to sow early in autumn, allowing more time for extra tillers to develop, and/or to sow a high rate of
94 seeds so that more plants establish, resulting in more shoots m-2 (Spink et al., 2000). Whilst it is well
95 understood that earlier sowing and higher seed rates usually result in more shoots m -2 (Spink et al.,
96 2000; Darwinkel et al., 1977), reliable methods for quantifying how many more shoots, or how
97 tolerant a crop will be to to D. coarcata, do not exist. It should also be recognised that manipulating
98 crop management to produce more shoots often has a practical or economic cost. For example,
99 early sown crops are more prone to lodging and it is not possible to sow early following late harvested
100 root crops. In addition, adverse weather may prevent early sowing and higher seed rates will
101 increase seed costs.

102 Annual D. coarctata pest levels are predicted through soil sampling surveys carried out in September
103 and October, with D. coarctata eggs extracted and counted from the sampled soil. This process is
104 time-intensive, the assessment of one sample can take up to three hours (Ramsden, et al., 2017),
105 and requires a suite of bulky equipment (Ramsden et al., 2017; Salt & Hollick, 1944) which are often
106 only available in specialist analytical laboratories. An efficient alternative means of predicting D.
107 coarctata pest levels is through predictive modelling (Young & Cochrane, 1993). The Young &
108 Cochrane D. coarctata population level prediction model is based on egg counts for East Anglia, UK
 bioRxiv preprint doi: https://doi.org/10.1101/2021.03.13.435242; this version posted March 13, 2021. The copyright holder for this preprint
(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is
made available under aCC-BY-NC-ND 4.0 International license.

109 between 1952 and 1991 (Young & Cochrane, 1993). The Young & Cochrane model predicts D.
110 coarctata egg numbers using a range of meteorological parameters, including the departure from
111 the long-term average for rainfall during October of the preceding year, January air temperature,
112 January soil temperature, and July air temperature. Meteorological parameters used in the Young &
113 Cochrane (1993) model were selected based on the hypothesis that they influence either the
114 reproductive development or oviposition of the emerging D. coarctata generation. The model
115 developed by Young & Cochrane has a predictive accuracy of 59%, although this is a satisfactory
116 level of prediction for a pest prediction model (Yonow et al., 2004) there is scope to refine the model
117 to further increase its accuracy and reliability. Accurate and reliable prediction of annual D. coarctata
118 pest levels before crops are sown is essential if agronomic practices are going to be adjusted to
119 achieve successful D. coarctata control. Accurate estimates of plant populations/shoot numbers are
120 also required if practices such as early sowing or sowing at a higher seed rate are going to be used
121 to improve crop tolerance to D. coarctata, as has been the case for similar wheat stem-borers (Beres
122 et al., 2011; Bryson et al., 2005).

123 Here, we develop two predictive models that can be used for integrated D. coarctata control. The
124 first is an updated pest level prediction model that estimates D. coarctata populations from
125 meteorological parameters with a greater predictive accuracy than the previous model developed by
126 Young and Cochrane (1993). The second is a model that predicts the maximum number of shoots
127 for a winter wheat crop just prior to the start of stem extension based on target plant population and
128 sowing date; this will be particularly valuable when trying to optimise the target plant population and
129 sowing date to produce crops which are able to tolerate D. coarctata infestation. Finally, using data
130 extracted from the literature we produce a revised calculation of D. coarctata threshold levels and
131 use this to develop a sustainable decision support system for minimising the risk of economic crop
132 damage by D. coarctata. .

133 Methods
134 Modelling to predict D. coarctata egg numbers

135 Sources of D. coarctata egg numbers and meteorological data

136 Delia coarctata egg number data were extracted from two sources. Historic data from East Anglia
137 (1952 – 1991) were extracted from Young & Cochrane (1993) and combined with the results from
138 the AHDB Autumn survey of D. coarctata incidence from northern England (2005 – 2019) and East
139 Anglia (2008 – 2019). Up to 30 fields were sampled in September or October of each survey year in
140 areas prone to D. coarctata infestation, with c. 15 in eastern England and 15 in northern England.
141 Samples were taken in September or October once egg laying was complete (usually early
142 September; Fig. 1). For each field sampled, either 32 cores (each of 7.2 cm diameter) or 20 cores
143 (each of 10 cm diameter) were taken to cultivation depth. Fields were sampled in a standard ‘W’
144 sampling pattern across the direction of cultivation. D. coarctata eggs were extracted following soil
 bioRxiv preprint doi: https://doi.org/10.1101/2021.03.13.435242; this version posted March 13, 2021. The copyright holder for this preprint
(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is
made available under aCC-BY-NC-ND 4.0 International license.

145 washing (Salt & Hollick, 1944) and flotation in saturated magnesium sulphate. Egg numbers were
146 expressed as number of eggs m -². Meteorological data were extracted from UK meteorological office
147 data for each region (www.metoffice.gov.uk/research/climate/maps-and-data/uk-and-regional-
148 series). Meteorological data extracted included minimum, mean, and maximum temperature, rain
149 days, rainfall amount (mm), sunshine days, and air frost days. From the extracted data, the deviation
150 from the long-term average was calculated. Long-term averages of 30-year periods were used, the
151 only deviation for this was for air frost for which reporting did not commence until 1960. The UK
152 meteorological office categorises seasons into: Winter (preceding December – February), spring
153 (March – May), summer (June – August), autumn (September – November). For each year of egg
154 collection, the meteorological data included in the model were the autumn of the preceding calendar
155 year, winter (starting in December of the preceding year), current spring, and current summer. For
156 each season the meteorological parameters included were: minimum temperature, mean
157 temperature, maximum temperature, the number of sun days, the number of rain days, and the
158 amount of rainfall. The number of air frost days was included for the winter, spring, and preceding
159 autumn seasons only.

160 Modelling approach – developing the pest level prediction models

161 Data modelling for the pest level prediction model was carried out in R v.3.6.1 with additional
162 package ggplot 2 v.3.2.1 (Wickham, 2016) used for data visualisation. Linear regressions were used
163 to build all models and backwards stepwise model selection was employed to arrive at the final
164 predictive models (Marill & Green, 1963). At each simplification stage analysis of variance was
165 carried out to ensure model simplification was justified and did not significantly affect model structure.
166 Model residuals were observed at each stage.

167 Seasonal and monthly models for 1952 – 2019

168 An initial seasonal model was developed using the following meteorological factors on a seasonal
169 basis: minimum air temperature, mean air temperature, maximum air temperature, rain days, rainfall,
170 and sun days. The model was refined through backwards stepwise model selection until the final
171 seasonal model was produced. Based on this final seasonal model an initial monthly model was
172 developed that included monthly inputs for all the parameters included in the final seasonal model
173 (e.g. the final seasonal model included summer minimum temperature, therefore the initial monthly
174 model included minimum temperature for June, July, and August). The monthly model was simplified
175 through backward stepwise model selection until the final monthly model was produced.

176 Seasonal and monthly models for 1971 – 2019 (air frost models)

177 In order to allow air frost to be included in the model a subdataset was developed comprising all
178 observations from 1971 – 2019. The 1971 – 2019 period used in this subdataset was to allow the
179 long-term average to be calculated for a minimum period of ten-years. Seasonal and monthly models
180 were simplified as described above and the final model was used to predict egg numbers.
 bioRxiv preprint doi: https://doi.org/10.1101/2021.03.13.435242; this version posted March 13, 2021. The copyright holder for this preprint
(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is
made available under aCC-BY-NC-ND 4.0 International license.

181 The final monthly model was validated following a similar procedure to the one employed by Young
182 & Cochrane (1993): four five-year periods were removed from the dataset from which the model was
183 developed and observations were made of how this affected the model predictions. Data were
184 removed from the first five years (validation 1: years 1971 – 1975 removed), the last five years
185 (validation 2: years 2015 – 2019), the five years with the highest recorded D. coarctata egg numbers
186 (validation 3: years 1978, 1984, 1985, 1986, 2010), and the five years with the lowest recorded D.
187 coarctata egg numbers (validation 4: years, 2005, 2006, 2007, 2014, 2017). A further four validations
188 (validations 5-8) were carried out by randomly removing five one-year periods from the dataset from
189 which the model was developed.

190 Soil samples were taken from 30 sites in England (divided into 15 northern and 15 eastern sites) in
191 September and October 2020 and the number of D. coarctata eggs per sample were determined;
192 these samples were used to independently test the prediction model.

193 Shoot number model

194 Model of potential shoot number

195 Data modelling for the shoot number prediction model was carried out in Microsoft Excel and
196 involved two separate processes. First, the potential shoot number of a single wheat plant growing
197 in isolation (i.e. without any competition from neighbouring plants and unlimited resources) was
198 determined. Secondly, this calculation was calibrated using field data to account for plant competition
199 and environmental factors that might limit shoot production.

200 Building a thermal time-based model for shoot production

201 Published principles of wheat shoot development were used to develop a thermal time-based model
202 for shoot production (Klepper et al., 1984). The thermal duration between sowing and plant
203 emergence was taken as 150oCd (Sylvester-Bradley et al., 1998). The end of tillering generally
204 coincides with the start of stem extension and the formation of the terminal spikelet within the
205 developing ear. The thermal time between sowing date and terminal spikelet production for early
206 and late sown winter wheat crops was reported in Kirby et al. (1999). Data from Kirby et al. (1999)
207 was used to estimate the effect of sowing date (1st September to 8th November) on the thermal
208 duration between sowing and terminal spikelet, assuming that it decreased linearly with time of
209 sowing. This meant that the thermal time from sowing to terminal spikelet decreased by 9oC for each
210 day that sowing was delayed after 1st September from a value on 1st September of 1582oCd. The
211 thermal time from sowing to plant emergence was then subtracted from this value, leaving the total
212 thermal time available for leaf and shoot production.

213 The thermal time between emergence of successive leaves (phyllochron) decreases the later crops
214 are sown (Equation 1) where d = the number of days after 1st September when the crop was sown
215 (Kirby et al., 1985).

216 𝑃ℎ𝑦𝑙𝑙𝑜𝑐ℎ𝑟𝑜𝑛 𝑙𝑒𝑛𝑔𝑡ℎ = −0.5383 × 𝑑 + 140.3

 bioRxiv preprint doi: https://doi.org/10.1101/2021.03.13.435242; this version posted March 13, 2021. The copyright holder for this preprint
(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is
made available under aCC-BY-NC-ND 4.0 International license.

217 Equation 1

218 The phyllochron length, along with the thermal time between plant emergence and terminal spikelet,
219 was then used to estimate the number of leaves and shoots that could be produced over time for an
220 individual plant. The following assumptions were made based on Klepper et al., (1984):

221 • Three phyllochrons after plant emergence: The first primary shoot emerges from the axil of
222 the 1st leaf on the main shoot
223 • Four phyllochrons after plant emergence: The second primary shoot emerges from the axil
224 of the second leaf of the main shoot
225 • Five phyllochrons after plant emergence: The third primary shoot emerges from the axil of
226 the third leaf of the main shoot. The first secondary shoot emerges from the axil of the first
227 leaf of the first primary shoot
228 • Six phyllochrons after plant emergence: The fourth primary shoot emerges from the axil of
229 the fourth leaf of the main shoot. The second secondary shoot emerges from the axil of the
230 second leaf of the first primary shoot. The third secondary shoot emerges from the axil of
231 the first leaf of the second primary shoot.

232 Calibrating and testing the shoot number model

233 Data from five winter wheat trials (Table S1) were used to calibrate the shoot number model and
234 data from three winter wheat trials (Table S2), alongside data from Spink et al., 2000, were used to
235 test the shoot number model. Each of the field trials (Table S1; S2) used the same winter what variety
236 (Evolution) and included a range of seed rates (40, 80, 160, 320, 480, and 640 seeds m-2). All trials
237 were sown at either ‘standard’ or ‘late’ timings for the region, and were treated with an insecticide to
238 control for D. coarctata (Tables S1; S2). The experiment at Huggate 2016 was not treated with
239 insecticide and assessments showed that less than 1% of shoots were infested with D. coarctata at
240 the start of stem extension. Each experiment was arranged in a fully randomised block design with
241 either three or four replicates of each treatment (Tables S1; S2). Experimental plots were 2m x 12m
242 and were drilled using an Ojyard experimental plot drill. Plant number and shoot number were
243 measured at the start of stem extension (BBCH Growth Stage 31 (GS31)) in each experimental plot
244 by counting all plants and shoots within a 0.7m x 0.7m quadrat. Field trial data (Table S1) were
245 analysed in Genstat (v-14) using a one-way ANOVA. Standard error of the difference (SED) values
246 are reported alongside p-values where relevant. Data used to calibrate the model were not used for
247 model testing. The model was tested against three field trials (Table S2) and data extracted from a
248 previous seed rate and sowing date winter wheat experiment (Spink et al., 2000). Multiple linear
249 regression analysis, with Experiment included as ‘Group’, was used to compare the predicted shoot
250 numbers against the model predictions using Genstat (v-14).

251 Results

252 The D. coarctata pest level prediction model

 bioRxiv preprint doi: https://doi.org/10.1101/2021.03.13.435242; this version posted March 13, 2021. The copyright holder for this preprint
(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is
made available under aCC-BY-NC-ND 4.0 International license.

253 In order to improve on a previous D. coarctata population prediction model (the Young & Cochrane
254 1993 model), open-access meteorological data (published by the UK Meteorological Office) were
255 extracted on a seasonal and monthly basis and incorporated into two linear models. In order to
256 identify which seasons are most important in determining D. coarctata egg numbers an initial model
257 was developed on a seasonal basis. This model (1952 – 2019 seasonal model; Fig. S1A) indicated
258 that the number of preceding autumn rain days, the minimum winter temperature, spring mean
259 temperature, spring maximum temperature, spring rainfall, and summer minimum temperature are
260 the most important meteorological parameters affecting D. coarctata egg numbers on a seasonal
261 basis (adjusted R2 = 0.49; F6,60 = 11.54; p = <0.001). The seasonal meteorological parameters listed
262 above were used as inputs for the predictive model. Model predictions versus observations for the
263 seasonal 1952 - 2019 model, including model predictions for 1992 – 2004, are displayed in Fig. S1A.

264 In order to identify the months which influence D. coarctata egg numbers the meteorological
265 parameters included in the seasonal 1952 – 2019 model were assessed on a monthly basis for the
266 relevant seasons: preceding September – preceding November for preceding autumn season,
267 preceding December – February for winter season, and June – August for summer season. Linear
268 regression modelling indicated that the most important monthly meteorological parameters for
269 predicting D. coarctata egg numbers were preceding October rain days, preceding December
270 minimum temperature, January minimum temperature, March mean temperature, May mean
271 temperature, March maximum temperature, May maximum temperature, May rainfall, June minimum
272 temperature, July minimum temperature, and August minimum temperature (F11,55 = 6.66; p =
273 <0.001; adjusted R2 = 0.49). The monthly meteorological parameters listed above were used as
274 inputs for the predictive model. The predictions of this model, compared with the observed values,
275 are shown in Fig. 2A.

276
 bioRxiv preprint doi: https://doi.org/10.1101/2021.03.13.435242; this version posted March 13, 2021. The copyright holder for this preprint
(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is
made available under aCC-BY-NC-ND 4.0 International license.

277
278 Fig. 2: Predicted and observed D. coarctata egg numbers. A) 1952 – 2019 monthly model, B) 1971 – 2019 monthly model.
279 Predictions (grey) are plotted alongside the mean observed value (black) and divided into the two regions, north (triangle;
280 dashed line) and east (circle; solid line). For clarity, trendlines are included for the observed values only.

281 Introducing air frost into the pest prediction model increases predictive power

282 One key factor that might influence D. coarctata egg numbers is air frost. The UK meteorological
283 office began long-term monitoring of air frost in 1960. A seasonal model was produced using data
284 from the 1971 – 2019 subdataset. This comprised of all meteorological parameters described above
285 as well as the departure from long-term average for preceding autumn, winter, and spring, air frost
286 days. Following model simplification, this air frost seasonal model had a higher predictive power
287 compared with the previously developed models (adjusted R2 = 0.59, F9,38 = 842; p = <0.001) and
288 the final meteorological parameters included departure from long-term average for: preceding
289 autumn rain days, preceding autumn sun days, winter mean temperature, winter air frost days, spring
290 maximum temperature, spring rainfall, summer minimum temperature, summer mean temperature,
291 and summer maximum temperature. These seasonal meteorological parameters were used as
292 inputs for the predictive model. The predictions of this model, compared with the observed values,
293 are shown in Fig. S1B.

294 A refined 1971 – 2019 monthly model was created using monthly average data for the seasons
295 shown to be important in the seasonal model, using the same approach described above for the
296 1952-2019 model. The meteorological inputs for this refined monthly model were the departure from
 bioRxiv preprint doi: https://doi.org/10.1101/2021.03.13.435242; this version posted March 13, 2021. The copyright holder for this preprint
(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is
made available under aCC-BY-NC-ND 4.0 International license.

297 long-term average for: preceding September sun days, preceding October rain days, January mean
298 temperature, January frost, April maximum temperature, May maximum temperature, April rainfall,
299 and July minimum temperature. This model had the highest predictive power (adjusted R2 = 0.70,
300 F8,39 = 14.88; p = <0.001). The monthly meteorological parameters listed above were used as inputs
301 for the predictive model. The predictions of this model, compared with the observed values, are
302 shown in Fig. 2B. On average, the predicted values only deviated from the observed values by -9%
303 (median = -2%; range = -155% to +50%).

304 Validating the final pest prediction model

305 The 1971 – 2019 monthly model was validated by removing a series of years from the model, re-
306 running the model, and observing the effect the removal of these years had on the ability of the
307 model to predict D. coarctata egg numbers for all years (1971 – 2019), similar to the validation
308 process deployed by Young & Cochrane (1993). Eight validation models were developed in total
309 (Fig. S2). Validations had no significant detrimental effect on the predictive power of the models;
310 average deviation from the predictive values of the full model were: -20.20% (validation 1), +0.05%
311 (validation 2), +0.58% (validation 3), -17.10% (validation 4), +2.72% (validation 5), -1.28% (validation
312 6), +2.70% (validation 7), -1.94% (validation 8). The relationship between the observed and
313 predicted values of the final 1971 – 2019 monthly model is shown in Fig. S3.

314 Testing the D. coarctata prediction model

315 The model was used to predict mean D. coarctata egg numbers for each region in 2020. The model
316 predictions were then compared with the mean egg counts per region obtained by soil sampling. The
317 model predicted a mean D. coarctata egg number of 60 eggs m-2 for eastern England and 107 eggs
318 m-2 northern England. Soil sampling (15 sites per region) indicated that the observed regional risk
319 was 111 eggs m-2 for northern England, and 173 eggs m-2 for eastern England. Higher observed
320 values for eastern England than were estimated were mainly driven by three sites with very high
321 counts of 1000, 850, and 404 eggs m-2.

322 Predicting shoot production number for individual plants growing in isolation

323 The shoot production model estimated the shoot production of a single plant grown in isolation at a
324 range of sowing dates between 1st September and 30th December (Fig. 3). The model predicted
325 that a single plant sown on 1st September has the potential to produce 38 shoots by terminal spikelet
326 (which approximates to the start of stem extension), whereas at the other extreme a single plant
327 sown in mid-November would only produce ten shoots.

328
 bioRxiv preprint doi: https://doi.org/10.1101/2021.03.13.435242; this version posted March 13, 2021. The copyright holder for this preprint
(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is
made available under aCC-BY-NC-ND 4.0 International license.

329
330 Fig. 3: Predicted number of shoots per plant produced by terminal spikelet (which approximates to the start of stem
331 extension), for a given sowing date when grown in the absence of competition and environmental stress factors.

332 Calibrating the shoot production number model

333 It was recognised that achieving a shoot population of 38 shoots per plant (Fig. 3) was unrealistic
334 under field conditions, primarily due to competition between shoots for limited resources but also
335 due to stress factors (e.g. soil capping, pests, disease). Shoot number data from specific treatments
336 in a series of field trials (Table S1) were used to calibrate the model to ensure that it provided a
337 realistic estimation for the number of shoots that could be expected in field conditions. The ratio of
338 observed to predicted shoots per plant was negatively related to the observed plants m-2 (Fig. 4).
339 This was because the shoot number model predicted the highest potential number of shoots m -2 for
340 high plant populations, but these populations also had the greatest competition between shoots
341 resulting in the lowest ratio of observed to predicted shoots per plant. The relationship between the
342 observed plants m-2 and ratio of observed to predicted shoots per plant described in Fig. 4 was used
343 to calibrate the potential shoot number model for field conditions.

344
 bioRxiv preprint doi: https://doi.org/10.1101/2021.03.13.435242; this version posted March 13, 2021. The copyright holder for this preprint
(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is
made available under aCC-BY-NC-ND 4.0 International license.

345
346 Fig. 4: The ratio of observed to predicted shoots per plant at the beginning of stem extension plotted against measured
347 plants m-2 for crops sown in the first week of October.

348 The calibrated shoot number model predicts a decline in maximum shoot number due to reduced
349 plant population and delayed sowing date, as is generally observed in practice (Fig. 5). A crop sown
350 at the end of September with a plant population of 200 plants m -2 is predicted to produce a maximum
351 shoot number of approximately 1071 shoots m-2. The model can be used to estimate the minimum
352 plants m-2 required to achieve 500 shoots m-2 (the minimum number of shoots required to achieve a
353 typical commercial UK wheat yield). To achieve 500 shoots m-2 by terminal spikelet the model
354 estimates that a minimum of 56 plants m-2 for late September sowing, 91 plants m-2 for mid-October
355 sowing, 163 plants m-2 for late October sowing, and 418 plants m-2 for mid-November sowing are
356 required. These plant populations are similar, or slightly greater, than estimates of the economic
357 optimum plant density reported by Spink et al. (2000), which provides confidence that the shoot
358 number model is giving plausible predictions. Although the number of shoots can further increase
359 between terminal spikelet stage and harvest, the damage caused by D. coarctata occurs prior to
360 terminal spikelet (Fig. 1). Therefore, shoot production up to terminal spikelet was considered most
361 appropriate for developing a D. coarctata tolerance scheme. The shoot production model has been
362 used to quantify the increase in maximum shoot number by terminal spikelet as a result of sowing
363 earlier and establishing a higher plant population (Fig 6). Sowing earlier generally results in a larger
364 increase in shoots m-2 compared with increases in plant population. This information can be used to
365 help estimate changes in sowing date and seed rate to minimise the risk of yield loss to D. coarcata.

366
 bioRxiv preprint doi: https://doi.org/10.1101/2021.03.13.435242; this version posted March 13, 2021. The copyright holder for this preprint
(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is
made available under aCC-BY-NC-ND 4.0 International license.

367
368 Fig. 5: Number of shoots m-2 predicted by the shoot number model after calibration for field conditions for crops sown on
369 different sowing dates after 1st September with plant populations of 100, 200, and 400 plants m-2.

370
371 Fig. 6: Effect of changes in sowing date and plant population on the predicted number of shoots m-2.

372 Testing the shoot number prediction model

373 Data from three winter wheat field experiments (Table S2) were combined with data from other field
374 experiments (Spink et al., 2000) to test the predictive power of the model. (Fig. 7). The number of
375 shoots m-2 at GS31 were measured in each of the field trials (Table S2), which approximates to the
376 timing of terminal spikelet production. Multiple linear regression analysis showed that a single best
 bioRxiv preprint doi: https://doi.org/10.1101/2021.03.13.435242; this version posted March 13, 2021. The copyright holder for this preprint
(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is
made available under aCC-BY-NC-ND 4.0 International license.

377 fit line had an R2 value of 0.70 (p = <0.001) (Fig. 7). Fitting separate best fit lines to each experimental
378 site with the same slope but different y axis intercepts gave an R 2 value of 0.83 (p = <0.001). Across
379 all the experimental sites the single best fit line was slightly greater than the 1:1 relationship.

380
381 Fig. 7: Model predicted shoots m-2 plotted against the observed shoots m-2 for data measured in the experiments listed in
382 Table S2 and Spink et al., 2000. Solid black line represents the linear regression fitted to all data (R2 = 0.70). The dashed
383 line represents a 1:1 relationship.

384 Revising the economic injury level of wheat to D. coarctata

385 The following factors determine how much damage a wheat crop can sustain from a stem-boring
386 insect before the damage becomes economically damaging, and can be used to provide a more
387 comprehensive estimation of economic thresholds for D. coarctata:

388 1. The number of shoots a larva can destroy

389 2. The minimum number of fertile shoots a crop requires to achieve a yield potential
390 3. The maximum number of shoots a crop is expected to produce in winter

391 These factors can be used to revise the D. coarctata threshold using Equation 2.

(𝑆𝑁 − 𝑆𝑁𝑀𝐼𝑁 )/𝑆𝑁𝐾𝐼𝐿𝐿

392 𝐸𝐼𝐿 =
𝐸𝑔𝑔 𝑉𝑖𝑎𝑏𝑖𝑙𝑖𝑡𝑦 𝑥 100

393 Equation 2: Economic Injury Level (EIL) equation used to estimate wheat tolerance against D. coarctata. SN = the number
394 of shoots per m-2 in winter, SNMIN = the minimum number of fertile shoots per m-2 required to achieve a yield potential,
395 SNKILL = the number of shoots killed by an individual larva, and Egg Viability the proportion of eggs that develop into larva.

396 Using values from the literature the parameters required for the equation can be estimated: Egg
397 viability is estimated at 56% (Ryan, 1973A; Raw, 1967; Gough, 1947), SNKILL is estimated at four
398 shoots destroyed per larva (Young & Ellis, 1996; Ryan, 1975; Ryan, 1973B), SN is estimated at 1000
 bioRxiv preprint doi: https://doi.org/10.1101/2021.03.13.435242; this version posted March 13, 2021. The copyright holder for this preprint
(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is
made available under aCC-BY-NC-ND 4.0 International license.

399 shoots m-2 (Sylvester-Bradley et al. 1998), and SNMIN is estimated at 400-450 shoots m-2 (Spink et
400 al., 2000). Using these assumed values, but increasing the SNMIN to 500 shoots m-2 to allow a modest
401 degree of insurance against achieving too few shoots, an updated threshold for D. coarctata in winter
402 wheat is presented in Table 1 and can be used as the foundation to develop targeted agronomic
403 approaches for minimising the risk of economic damage by D. coarctata control using crop tolerance.

404 Table 1: Minimum shoot number at GS31 needed to tolerate different levels of D. coarctata damage

405 Egg count (eggs m-2) Shoot number m-2 required to

406 tolerate D. coarctata damage

407 125 720

408
250 940
409
500 1380
410

411 750 1820

412

413 Calculations assume 500 shoots m-2 is the minimum required to achieve a typical UK commercial wheat yield.

414 Discussion

415 Predicting D. coarctata risk to inform control measures

416 Delia coarctata larva infest cereal shoots between January and April, where they can cause
417 devastating crop damage resulting in yield losses of up to 4 t ha-1 (Rogers et al., 2014). Depending
418 on the sowing date, current thresholds indicate that a pest pressure of 250 eggs m -2 can cause
419 significant crop damage for crops drilled before October, with a pressure of 100 eggs m-2 causing
420 significant crop damage for crops drilled from November (Gough, 1961). Chemical-based options for
421 D. coarctata control in the UK are limited to a pre-drilling seed treatment. This is only effective for
422 late sown crops (November onwards) and is insufficiently persistent for earlier sowings. Therefore,
423 decisions on whether to apply a seed treatment need to be made before sowing and should take
424 into account the risk of D. coarctata infestation. Currently, the most accurate means of determining
425 D. coarctata risk requires soil extraction and manual egg counting after the previous crop has been
426 harvested, this is both time- and cost-intensive (Ramsden et al., 2017). Predicting pest levels and
427 risk is an important Integrated Pest Management (IPM) tool and represents a cost-effective means
428 of estimating annual risk for a wide range of important insect pests (Herms, 2004). The pest level
429 prediction model developed in this study provides an alternative to soil sampling and has the
430 significant advantages that it is less time consuming, less arduous, and provides an earlier estimate
431 of pest levels, which is crucial when deciding whether to treat seed. Similar prediction models have
432 been developed for a range of other agriculturally important insect pests, including the tea leaf roller,
 bioRxiv preprint doi: https://doi.org/10.1101/2021.03.13.435242; this version posted March 13, 2021. The copyright holder for this preprint
(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is
made available under aCC-BY-NC-ND 4.0 International license.

433 Caloptilia theivora Walsingham (Satake et al., 2005), leafhoppers, whiteflies, and thrips (Arya et al.,
434 2015). Pest level and risk prediction models can predict when a phenological event has occurred
435 that increases the in-season crop risk (such as pest egg hatch, development, or emergence of an
436 additional generation) to help time the application of pest management strategies (Satake et al.,
437 2005; Herms, 2004; Milonas et al., 2001). Alternatively, models can predict annual risk by estimating
438 pest population densities (Arya et al., 2015). As D. coarctata can currently only be controlled
439 chemically using a seed treatment, it is important that any pest level prediction model developed can
440 accurately predict pest populations before sowing, in time for sowing date and seed rates to be
441 adjusted.

442 The D. coarctata pest level prediction model developed here is based on the Young & Cochrane
443 model (Young & Cochrane, 1993) but incorporates a wider range of meteorological parameters than
444 the original. The meteorological parameters included in the Young & Cochrane model were selected
445 to include the factors hypothesised to have the greatest influence on D, coarctata biology and
446 phenology (Young & Cochrane, 1993). The Young & Cochrane model had a predictive accuracy of
447 59%, limiting its uptake as a decision support tool for wheat growers. When developing our models,
448 we included additional meteorological factors available from open-access data sources. The benefits
449 of building the models using open-access meteorological data are that the model inputs are
450 standardised across regions and freely available. The meteorological parameters included in the
451 models were departure from the long-term average for: minimum, mean, and maximum temperature,
452 rainfall, the number of rain days (days with rainfall > 0.2 ml), the number of sun days, and days of
453 air frost. Our final model (1971 – 2019 monthly model) had a predictive accuracy of 70%, an 11%
454 increase in accuracy when compared with the Young & Cochrane model. This is a significant
455 improvement of the potential for using predictive modelling to estimate D. coarctata risk.
456 Furthermore, as the model can be run prior to sowing in August/September, it is a possible alternative
457 to soil sampling, and it can be used to target seasons and regions where soil sampling should be
458 focussed.

459 During model testing the model performed well and predicted regional D. coarctata risk accurately
460 for the northern region; however, the level of risk predicted for the eastern region was lower than
461 observed. Further model development and refinement (through the potential inclusion of model
462 moderators such as soil type and previous crop) would enable a more robust and dynamic model to
463 be developed. Soil type is likely an important factor to consider in future model development, as the
464 three 2020 test sites with the highest D. coarctata counts were associated with clay soils, indicating
465 that soil type might influence D. coarctata oviposition preference. Furthermore, the previous crop in
466 the rotation has been reported to affect D. coarctata oviposition (Young & Cochrane, 1993; Gough,
467 1946). Therefore, including these two factors as components in subsequent model improvements
468 represents the logical next step in future model development. The predictive accuracy of our model
469 is similar to the accuracy of other pest prediction models that use meteorological data to predict
470 seasonal risk. Including models that predict the population dynamics of the Queensland fruit fly,
 bioRxiv preprint doi: https://doi.org/10.1101/2021.03.13.435242; this version posted March 13, 2021. The copyright holder for this preprint
(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is
made available under aCC-BY-NC-ND 4.0 International license.

471 Bactrocera tryoni Froggatt (R2 = 0.28 and 0.32; Yonow et al., 2004), annual populations of the
472 whitetail, Paronychiurus kimi (Lee) (R2 = 0.79; Choi & Ryoo, 2003), and the abundance of two
473 planthopper species, Helicoverpa spp., (R2 = 0.84 – 0.96; Zalucki & Furlong, 2005). Therefore, our
474 model represents a potential useful component of a sustainable IPM strategy.

475 Improving tolerance to D. coarctata through shoot number prediction

476 Non-chemical methods for D. coarctata control are becoming increasingly desirable, from both an
477 agronomic and an environmental perspective. Insecticidal sprays are no longer approved for D.
478 coarctata control in the UK and a growing body of evidence indicates that pesticides can have far-
479 reaching environmental consequences (Leather, 2018), resulting in a need to develop non-chemical
480 pest management methods. Cultural control can be an effective means of limiting pest damage
481 (Glen, 2000) and can involve the adjustment of both pre-drilling and in-season agronomic practices.
482 Non-chemical control of similar stem-boring pests of wheat can be effectively achieved through
483 adjustments to pre-drilling agronomic practices, such as increased seed rates and earlier sowings
484 (Glen, 2000). These practices have the potential to increase crop shoot numbers, and therefore
485 improve crop tolerance to herbivorous insects (Wenda-Pieskik et al., 2017; Beres et al., 2011; Bryson
486 et al., 2005).

487 Higher seed rates have been exploited as an IPM strategy to confer tolerance in wheat against other
488 stem-boring pests, including the wheat stem sawfly, Ce. cinctus (Beres et al., 2011) and the gout fly,
489 Ch. pumilionis (Bryson et al., 2005). Earlier drilling has also been reported to confer tolerance against
490 cereal leaf beetles, Oulema spp. (Wenda-Pieskik et al., 2017). Together, this showcases the
491 potential to achieve cultural control of herbivorous insect pests through the adjustment of pre-drilling
492 agronomic practices, including sowing date and seed rate adjustments. Therefore, accurate means
493 of predicting crop shoot numbers will be an important component of any IPM strategy based on these
494 methods. The shoot number prediction model we developed in this study can be used in conjunction
495 with our revised D. coarctata thresholds (Table 1) as a component of an integrative D. coarctata risk
496 management system.

497 Our shoot number model has an overall predictive accuracy of 70% and the predicted values were
498 close to the 1:1 relationship across all the experiments the model was tested against. For some sites
499 the model made more accurate predictions (e.g. Foxholes, 2017), although for others a weaker
500 relationship was observed (e.g. Bardwell, 2017). This contrast in accuracy could be related to
501 specific environmental factors that limited shoot production such as soil capping or water logging
502 (Robertson et al., 2009). Our revised thresholds for D. coarctata in winter wheat demonstrate that
503 the current D. coarctata pest threshold of 250 eggs m-2 (Gough et al., 1961) is too simplistic and for
504 many crops this likely represents either an overestimation of the potential pest damage, an
505 underestimation of the amount of damage that can be tolerated by a winter wheat crop, or a
506 combination of both factors. Sensitivity analysis (achieved by adjusting each parameter used in
507 Equation 2 from its likely minimum value to its maximum value; Fig. 8) demonstrates that thresholds
 bioRxiv preprint doi: https://doi.org/10.1101/2021.03.13.435242; this version posted March 13, 2021. The copyright holder for this preprint
(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is
made available under aCC-BY-NC-ND 4.0 International license.

508 for D. coarctata could be smaller or greater than 250 eggs m-2, and that the thresholds are particularly
509 sensitive to the number of shoots a crop will produce. Combining the shoot prediction model with the
510 revised D. coarctata thresholds will facilitate the production of a winter wheat crop that is capable of
511 tolerating the predicted level of D. coarctata damage through adjustments to seed rate, sowing date,
512 or both. This approach is similar to those devised by Bryson et al., (2005), Beres et al., (2011), and
513 Wenda-Pieskik et al., (2017). Although, increasing shoot number would need to be considered
514 alongside higher crop production costs and the increased risk of crop lodging (Berry & Spink, 2012).

515
516 Fig. 8: Sensitivity analysis for the economic injury level (EIL) of D. coarctata eggs m -2 for the minimum and maximum
517 ranges of each of the parameters used to calculate the EIL. Ranges for each parameter (Low – Default – High): Egg
518 viability = 30%, 56%, 70%; maximum shoots = 600, 1000, 1600, minimum ears = 400, 500, 600; shoots killed per larvae =
519 3, 4, 5.

520 The two models we have developed, pest level prediction and crop tolerance prediction (via shoot
521 number estimation), have the potential to represent central components of an IPM strategy for D.
522 coarctata control. Although the models have been tested and validated as part of this research,
523 further optimisation is required. For the pest level prediction model this could involve the inclusion of
524 model moderators (soil type, previous crop) to adjust the value using bespoke farm-specific traits.
525 For the shoot number prediction model, this could include the introduction of crop variety and site
526 factors as variables to estimate shoot number using more detailed agronomic factors. The revised
527 threshold scheme will also require experimental validation.

528 A new D. coarctata IPM strategy

 bioRxiv preprint doi: https://doi.org/10.1101/2021.03.13.435242; this version posted March 13, 2021. The copyright holder for this preprint
(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is
made available under aCC-BY-NC-ND 4.0 International license.

529 In Fig. 9 we outline a D. coarctata IPM scenario based on the pest level prediction model, the revised
530 threshold level, and the shoot number prediction model. We believe that this strategy would facilitate
531 non-chemical risk-based control of D. coarctata and would comprise the following steps:

532 1. A seasonal estimation of D. coarctata risk per region to advise on the potential level of control
533 required and to enable targeted soil sampling in high-risk regions
534 2. Use of the revised thresholds to compare predicted D. coarctata risk with the minimum
535 number of shoots required to tolerate pest damage while obtaining a viable crop yield
536 3. Utilisation of the shoot number prediction model to estimate the number of shoots that will be
537 produced for the planned sowing date and sowing rate.
538 4. Subtraction of the estimated number of shoots required to achieve D. coarctata tolerance at
539 the predicted level of D. coarctata risk from the estimated number of shoots expected with
540 the planned agronomic practice: A positive value indicates that D. coarctata damage can be
541 tolerated naturally, a negative value indicates that additional crop protection or risk-mitigation
542 steps are required, e.g. earlier drilling, higher seed rate, seed treatment (if sown late).
543 5. The shoot number prediction model can then be used to estimate the minimum plant
544 population required to achieve the minimum shoot number for a given sowing date, or the
545 latest sowing date that could be achieved for a given target plant population.

546
547 Fig. 9: An IPM flow-chart for sustainable management of D. coarctata through the optimisation of pest level and wheat
548 shoot number prediction and crop tolerance – this image was created with BioRender.com
 bioRxiv preprint doi: https://doi.org/10.1101/2021.03.13.435242; this version posted March 13, 2021. The copyright holder for this preprint
(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is
made available under aCC-BY-NC-ND 4.0 International license.

549 This prescriptive pest management scheme will provide a framework for sustainable D. coarctata
550 management. Where the framework indicates that additional crop protection steps are required this
551 could be achieved by adjusting sowing date and/or target plant population to produce a crop with
552 sufficient shoots to tolerate the pest and still achieve potential yield. During seasons of high risk, the
553 option of combining manipulation of sowing date and/or plant population with seed treatment could
554 be employed for late sown crops (November onwards).

555 Acknowledgements

556 The authors gratefully acknowledge the Agriculture and Horticulture Development Board (AHDB)
557 Cereals and Oilseeds for funding this project and would like to thank David Lunn, Josh Humphrey,
558 and Andrew Moore carrying out the 2020 D. coarctata surveys and David Lunn, Tom Whiteside, and
559 Nicola Rochford for managing the field trials.

560 Conflict of interests

561 The authors have no conflict of interests to declare

 bioRxiv preprint doi: https://doi.org/10.1101/2021.03.13.435242; this version posted March 13, 2021. The copyright holder for this preprint
(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is
made available under aCC-BY-NC-ND 4.0 International license.

562 References

563 Arya P., Paul RK., Kumar A., Singh KN., Sivaramne N. & Chaudhary P. (2015). Predicting pest
564 populations using weather variables: an arimax time series framework. International journal of
565 agricultural and statistical sciences, 11, pp. 381-386

566 Bardner R., Fletcher KE. & Jones MG. (1977). The activity and distribution of wheat bulb fly (Delia
567 coarctata) (Diptera:Anthomyiidae) in cereal crops over fallow land. Annals of applied biology, 86, pp.
568 329-337

569 Beres BL., Cárcamo HA., Yang R-C. & Spaner DM. (2011). Integrating spring wheat sowing density
570 with variety selection to manage wheat stem sawfly. Agronomy journal, 103, pp. 1755-1764

571 Berry PM., Spink JH, Foulkes MJ. & Wade A. (2003). Quantifying the contributions and losses of dry
572 matter from non-surviving shoots in four cultivars of winter wheat. Field crops research, 80, pp. 111-
573 121

574 Berry PM. & Spink J. (2012). Predicting yield losses caused by lodging in wheat. Field crops
575 research, 137, pp. 19-26

576 Bryson RJ., Alford J. & Oakley J. (2005). Development of guidelines for improved control of gout fly
577 (Chlorops pumilionis) in winter wheat. HGCA project report no. 372

578 Choi WI. & Ryoo MI. (2003). A matrix model for predicting seasonal fluctuations in field populations
579 of Paronychiurus kimi (Collembola: Onychiruidae). Ecological modelling, 162, pp. 259-265

580 Darwinkel A., ten Hag BA. & Kuizenga J. (1977). Effect of sowing date and seed rate on crop
581 development and grain production in winter wheat. Netherlands journal of agricultural science, 25,
582 pp. 83-94

583 Glen D. (2000). The effects of cultural measures on cereal pests and their role in integrated pest
584 management. Integrated pest management reviews, 5, pp. 25-40

585 Gough H. (1946). Studies on wheat bulb fly, Leptohylemyia coarctata, Fall: I. Biology. Bulletin of
586 entomological research, 37, pp. 251-271

587 Gough H. (1947). Studies on wheat bulb fly, Leptohylemyia coarctata, Fall: II. Numbers in relation to
588 crop damage. Bulletin of entomological research, 37, pp. 439-454

589 Gough H., Woods A., Maskell F. & Towler M. (1961). Field experiments on the control of wheat bulb
590 fly, Leptohylemyia coarctata (Fall.). Bulletin of entomological research, 52, pp. 621-634

591 Herms DA. (2004). Using degree-days and plant phenology to predict pest activity. In: V. Krischik &
592 J. Davidson, eds. IPM (Integrated Pest Management) of Midwest landscapes, pp. 49-59. Minnesota
593 Agricultural Experiment Station Publication SB-07645

594 Jones M. (1970). Observations on feeding and egg development of the wheat bulb fly Leptohylemyia
595 coarctata (Fall.). Bulletin of entomological research, 60, pp. 199-207
 bioRxiv preprint doi: https://doi.org/10.1101/2021.03.13.435242; this version posted March 13, 2021. The copyright holder for this preprint
(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is
made available under aCC-BY-NC-ND 4.0 International license.

596 Kirby EJM., Appelyard M. & Fellowes G. (1985). Effect of sowing date and variety on main shoot leaf
597 emergence and number of leaves of barley and wheat. Agronomie, 5, pp. 117-126

598 Kirby EJM., Spink JH., Frost DL., Sylvester-Bradley R., Scott RK., Foulkes MJ., Clare RW. & Evans
599 EJ. (1999). A study of wheat development in the field: Analysis by phases. European journal of
600 agronomy, 11, pp. 63-82

601 Klepper B., Belford RK. & Richman RW. (1984). Root and shoot development in winter wheat.
602 Agronomy journal, 76, pp. 117-122

603 Leather SR. (2018). “Ecological Armageddon” – more evidence for the drastic decline in insect
604 numbers. Annals of applied biology, 172, pp. 1-3

605 Marill T. & Green DM. (1963). On the effectiveness of receptors in recognition systems, IEEE
606 Transactions on information theory, 9, pp. 11-17

607 Milonas PG., Savopoulou-Soultani M. & Stavridis DG. (2001). Day-degree models for predicting the
608 generation time and flight activity of local populations of Lobesia botrana (Den. & Schiff.) (Lep.,
609 Tortricidae) in Greece. Journal of applied entomology, 125, pp. 515-518

610 Ramsden M., Kendall S., Ellis S. & Berry P. (2017). A review of economic thresholds for invertebrate
611 pests in UK arable crops. Crop protection, 96, pp. 30-43

612 Raw F. (1967). Some aspects of the wheat bulb fly problem. The fifth Barnes Memorial Lecture.
613 Annals of applied biology, 59, pp. 155-1733

614 Robertson D., Zhang H., Palta JA., Colmer T. & Turner NC. (2009). Waterlogging affects the growth,
615 development of tillers, and yield of wheat through a severe, but transient, N deficiency. Crop and
616 pasture science, 60, pp. 578-586

617 Rogers C., Guimarães R., Evans K. & Rogers S. (2014). Spatial and temporal analysis of wheat bulb
618 fly (Delia coarctata, Fallén) oviposition: consequences for pest population monitoring. Journal of pest
619 science, 88, pp. 75-86

620 Ryan M. (1973A). The natural mortality of wheat bulb fly eggs. The journal of applied ecology, 10,
621 pp. 869-874

622 Ryan M. (1973B). The natural mortality of wheat bulb fly larvae. The journal of applied ecology, 10,
623 pp. 875-879

624 Ryan M. (1975). The natural mortality of wheat bulb fly pupae, Leptohylemia coarctata (Fall.) (Dipt.,
625 Anthomyiidae). Plant pathology, 24, pp. 27-30

626 Salt G. & Hollick F. (1944). Studies of wireworm populations I. A census of wireworm in pasture.
627 Annals of applied biology, 31, pp. 52-64
 bioRxiv preprint doi: https://doi.org/10.1101/2021.03.13.435242; this version posted March 13, 2021. The copyright holder for this preprint
(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is
made available under aCC-BY-NC-ND 4.0 International license.

628 Satake A., Ohgushi T., Urano S. & Uchimura K. (2005). Modelling population dynamics of a tea pest
629 with temperature-dependent development: predicting emergence timing and potential damage.
630 Ecological research, 21, pp. 107-116

631 Spink JH., Semere T., Sparkes DL., Whaley JM., Foulkes MJ., Clare RW & Scott RK. (2000). Effect
632 of sowing date on the optimum plant density of winter wheat. Annals of applied biology, 137, pp.
633 179-188

634 Stern VM., Smith RF., van Den Bosch R. & Hagen KS. (1959). The integration of chemical and
635 biological control of the spotted alfalfa aphid: the integrated control concept. Hilgardia, 29, pp. 81-
636 101

637 Sylvester-Bradley R., Scott RK., Clare RW. & Duffield S. (1998). Assessments of wheat growth to
638 support its production and improvement. HGCA project report no. 151

639 Wenda-Pieskik A., Kazek M. & Piesik D. (2017). Cereal leaf beetles (Oulema spp., Coleoptera:
640 Chrysomelidae) control following various dates of wheat sowing. International journal of pest
641 management, 64, pp. 157–165

642 Wickham H. (2016). ggplot2: elegant graphics for data analysis. 2 ed. Springer International
643 Publishing

644 Yonow T., Zalucki M., Suthers R., Dominiak B., Maywald G., Maelzer D. & Kriticos D. (2004).
645 Modelling the population dynamics of the Queensland fruit fly, Bactrocera (Dacus) tryoni: a cohort-
646 based approach incorporating the effects of weather. Ecological modelling, 171, pp. 9-30

647 Young J. & Cochrane J. (1993). Changes in wheat bulb fly (Delia coarctata) populations in East
648 Anglia in relation to crop rotations, climate data and damage forecasting. Annals of applied biology,
649 122, pp. 485-498

650 Young J. & Ellis S. (1996). Impact of changes in arable agriculture on the biology and control of
651 wheat bulb fly. HGCA research review no. 33

652 Zalucki MP. & Furlong MJ. (2005). Forecasting Helicoverpa populations in Australia: A comparison
653 of regression based models and a bioclimatic based modelling approach. Insect science, 12, pp. 45-
654 56

655