Review Copy

0 downloads 0 Views 887KB Size Report
300 on each calendar day, a specific seedbank fraction loses its dormant ... programming skills and sufficient level of expertise to develop its heuristics. .... CALVO HARO RM, GONZALEZ-ANDUJAR JL & PEREZ S (1994) Manual de modelos ...... 3; H aj Se ye d. H ad i & G on za le z-An du ja r, 2. 00. 9. M atric aria p erfo ra ta.
Weed Research

Predicting field weed emergence with empirical models and soft computing techniques

Journal: Manuscript ID

WRE-2015-0204.R1 Review Paper

Re

Manuscript Type:

Weed Research

Date Submitted by the Author:

Complete List of Authors:

n/a

Keywords:

ew

vi

Gonzalez-Andujar, Jose Luis; Instituto de Agricultura Sostenible (CSIC), Crop Protection; Chantre, Guillermo; UNIVERSIDAD NACIONAL DEL SUR, AGRONOMIA; Morvillo, Claudia; University of Buenos Aires. Faculty of Agronomy, Crop production Blanco, Anibal; UNIVERSIDAD NACIONAL DEL SUR, Planta Piloto de Ingeniería Química Forcella, Frank; USDA-ARS, North Central Soil Conservation Research Lab.

py

Co

Artificial neural networks, genetic algorithms, predictive modelling, nonlinear regression, degree days, weed control

Page 1 of 40

Weed Research

1

Predicting field weed emergence with empirical models and soft computing

2

techniques

3

J. L. Gonzalez-Andujar1, G. R. Chantre2, C. Morvillo1,3, A. M. Blanco4, F. Forcella5 1

4 5 6 7 8 9 10

12

Instituto de Agricultura Sostenible (CSIC), Aptdo. 4084, 14080 Córdoba, Spain Departamento de Agronomía/CERZOS, Universidad Nacional del Sur/CONICET, Bahía

Blanca, Buenos Aires, 8000, Argentina 3

Departamento de Producción Vegetal, Facultad de Agronomía, Universidad de Buenos

Aires, Av. San Martín 4453, (1417), Buenos Aires, Argentina 4

Planta Piloto de Ingeniería Química, Universidad Nacional del Sur/CONICET, Bahía

Blanca, Buenos Aires 8000, Argentina. 5

Re

11

2

USDA-ARS North Central Soil Conservation Research Laboratory, Morris, MN 56267,

USA

14

ew

vi

13

Summary

16

Seedling emergence is one of the most important phenological processes that influences

17

the success of weed species. Therefore, predicting weed emergence timing plays a critical

18

role in scheduling weed management measures. Important efforts have been made in the

19

attempt to develop models to predict seedling emergence patterns for weed species under

20

field conditions. Empirical emergence models have been the most common tools used for

21

this purpose. They are based mainly on the use of temperature, soil moisture and light. In

22

this review, we present the more popular empirical models, highlight some statistical and

23

biological limitations that could affect their predictive accuracy and, finally, we present a

24

new generation of modelling approaches to tackle the problems of conventional empirical

25

models, focusing mainly on soft computing techniques. We hope that this review will

26

inspire weed modellers and that it will serve as a basis for discussion and as a frame of

27

reference when we proceed to advance the modelling of field weed emergence.

28

Key words: Artificial neural networks, genetic algorithms, predictive modelling, nonlinear

29

regression, weed control, degree days.

py

Co

15

30 31 1

Weed Research

32 33

Page 2 of 40

Introduction Seedling emergence is one of the most important phenological processes that influences the success of weed species (Forcella et al., 2000) and, therefore, predicting

35

weed emergence timing plays a critical role in scheduling weed management measures

36

(Ghersa, 2000). For example, from the chemical control perspective, many POST

37

herbicides will not control weeds that have not yet emerged at the time of application.

38

Conversely, if control is applied too late, early emerged weeds may be too large for

39

adequate control (Dalley et al., 2004). Moreover, if weeds are controlled too late, potential

40

crop yield can be reduced. Organic (ecological) farming practices using mechanical weed

41

control techniques also would benefit from weed emergence predictive tools to better

42

define intervention times (Oriade & Forcella 1999; Forcella et al. 2012).

43

Re

34

Since the early 1990´s important efforts have been made in the attempt to develop models to predict seedling emergence patterns for several weed species (e.g. Dorado et al.,

45

2009; Izquierdo et al., 2013; Werle et al., 2014; Zambrano-Navea et al., 2013) as a

46

fundamental step in the development of IWM strategies. These efforts vary from empirical

47

weed emergence models to more complex mechanistic models.

ew

48

vi

44

Although mechanistic or reductionist models can help to provide a close description

Co

of the basic ecophysiological processes underlying weed emergence (i.e., seed dormancy,

50

germination and pre-emergence shoot growth) (Batlla et al., 2004; Vleeshouwers &

51

Kropff, 2000; Gardarin et al., 2012), they usually require a large amount of often

52

unavailable or difficult to gather experimental information to be developed, calibrated and

53

finally validated (Grundy, 2003). Conversely, empirical weather-based models aim to

54

identify the relation between soil environmental variables and emergence data collected

55

under field conditions with the final objective to provide real-time information on weed

56

management programs.

57

py

49

Empirical models simply describe the general shape of the data set to which the

58

researcher is trying to fit a curve. They provide a simpler framework for practical support

59

decisions regarding optimal time for intervention. The most common means of analyzing

60

such relationships is some form of regression, which typically is nonlinear.

61 62

In this review we have restricted our attention to empirical models as they are the most common models used for modelling weed emergence under field conditions. We 2

Page 3 of 40

Weed Research

63

identify some statistical and biological limitations that could affect their predictive

64

accuracy and, finally, we present briefly a new generation of algorithm modelling

65

approaches to overcome their limitations, focusing mainly on soft computing techniques.

66

Predicting weed emergence

67

Empirical emergence models are tools for predicting the percentage of total weed emergence occurrence based mainly on the use of temperature and soil moisture (Forcella

69

et al., 2000; Grundy, 2003;) and, more recently, light (Royo-Esnal et al. 2015). They are

70

based on the thermal or hydrothermal time concepts (Gummerson, 1986; Roman et al.,

71

2000; Martinson et al., 2007), which assume that seeds need to accumulate a certain

72

amount of growing degree days (GDD) before completing germination and emergence.

73

They link emergence with GDD and can be classified, in a general way, as thermal-time,

74

hydrothermal-time, and photohydrothermal-time models. In thermal time models (TT),

75

daily mean soil temperature (eventually air temperature) is accumulated above a specific

76

threshold during the cropping season until weed emergence is completed. Thermal time

77

accumulation can start at various times depending upon specific situations. These dates

78

could be the last soil tillage event or burndown herbicide application; crop sowing; or an

79

arbitrary date such as 1 January for summer-growing weeds in cold climates of the

80

Northern Hemisphere, or winter-growing weeds in warm climates of the Southern

81

Hemisphere.

ew

vi

Co

A more complex approach includes the integration of soil water potential with soil

py

82

Re

68

83

temperature into hydrothermal-time (HTT). These models can be better at predicting

84

emergence than TT models (McGiffen et al., 2008), as they include soil water availability,

85

which is an important factor for seed germination and shoot growth. In these models, GDD

86

are accumulated when the daily average of soil water potential and soil temperature are

87

above specified threshold values below which seedlings cannot emerge (Gummerson,

88

1986). If experimental soil temperature and water potential data are missing, the weed

89

modeler can easily and accurately estimate such variables using free access software (e.g.,

90

Soil Temperature and Moisture Model (STM2); Spokas and Forcella, 2009). In

91

photohydrothermal-time models (PhHTT), photoperiod is used to modify TT based upon

92

daylength (Royo-Esnal et al. 2015). The rationale is that with longer day lengths, soil is

93

irradiated longer and accumulates more heat than with shorter day lengths, even when

94

maximum and minimum air temperatures are identical between days with long and short 3

Weed Research

Page 4 of 40

95

day lengths. The output for these models is percentage of the total annual emergence of a

96

given species. A typical non-linear regression model (NLR) for weed emergence is like the

97 98

following Y=f(x, φ)+ε

99

(1)

where ε ≈ N(0,σ2), Y is cumulative emergence, x is cumulative GDD (TT, HTT, or

101

PhHTT) and φ are specific function parameters. The term f(x, φ) is a non-linear S-shaped

102

function. S-shaped curves start at some fixed point, then increase up to an inflection point,

103

after which the slope of the curve decreases and the curve approaches the upper asymptote

104

(Ratkowsky, 1983) (Fig. 1). Different parametric S-type models have been commonly

105

fitted to emergence data (e.g. Schutte et al., 2008; Dorado et al., 2009) and, among these,

106

the Logistic, Weibull and Gompertz have been used most widely in predicting weed

107

emergence (Table S1).

Logistic:

! = #/%1 + ()*+−-+) − *))/

111

Gompertz: ! = # ()*%−()*+−-+) − *))/

113

(2)

py

Co

110

112

ew

Figure 1 near here

108 109

vi

Re

100

(3)

114 115

Weibull ! = #%1 − ()*+−+-+) − *))1 )/

116

(4)

In these expressions, ! and ) as in eqn 1, - is the slope parameter (emergence rate),

117 118

* represents the inflection point on the ) axis, 5 is a shape factor that determines the

119

skewness and kurtosis of the distribution, and # is the maximum emergence fraction of the

120

model.

4

Page 5 of 40

Weed Research

121

Gompertz and Logistic are very similar functions. Both are special cases of a generalized

122

logistic model (Calvo et al., 1994). The main differences between the Gompertz and

123

Logistic functions is that the Gompertz function approaches the asymptote much more

124

gradually, which often matches observations of late-emerging seedlings. The Weibull

125

model is a more flexible function than the Logistic and Gompertz. It can acquire the

126

characteristics of other types of distributions based on the value of the shape parameter c

127

(eqn 4). It typically has been used for modelling seedling emergence patterns (Table S1).

128

Comparison amongst models has been performed (e.g. Dorado et al., 2009). However, a clearly general favored model has not emerged. Even the same species (e.g.

130

Chenopodium album and Avena fatua) fitted different models best under diverse crops (see

131

Table S1).

132

134

Statistical limitations

135

ew

Model limitations

vi

133

Re

129

The aim of NLR models is to predict weed emergence easily and provide a practical tool to make informed management decisions. In this sense, NLR fitting presents several

137

statistical limitations that could affect their predictive accuracy (Cao et al., 2011; Onofri et

138

al., 2010). Some of these issues are briefly presented bellow:

139

-Fitting nonlinear models requires good initial parameter estimates. Poor initial estimates

140

could lead to wrong solutions or may yield no solutions at all (Holmström & Petersson,

141

2002). The algorithm used to find the solutions of nonlinear equations may incorrectly

142

pinpoint a local minima and thus only find local optima and, therefore, resulting parameter

143

estimations are biased.

144

-The parameters are estimated by NLR fitting routine using algorithms such Marquardt-

145

Levenberg (ML) or Gauss–Newton (GN) (Ratkowsky, 1983). There is a wide range of

146

statistical software to fit NLR models, which likely is the cause of NLR’s popularity.

147

Choice of the appropriate fitting algorithm is very important. For instance, ML is more

148

robust than GN, which means that in many cases it finds a solution even if it starts very far

149

from the global minimum. Unfortunately, the algorithms used in the fitting procedures may

py

Co

136

5

Weed Research

150

not be selected adequately and users often rely on their statistical software´s default

151

options.

152

- Observations are censored. When we make periodic seedling counts, the “new

153

emergence” events could have occurred at any time in the interval between the last and

154

current counts. Neglecting the existence of censored data may lead to biased results.

155

- Observed cumulative emergence values obtained from consecutive monitoring are not

156

statistically independent, resulting in positive autocorrelation of the residuals and,

157

therefore, which leads to erroneous predictions.

Page 6 of 40

158

160

Biological/ecological limitations

Re

159

Because TT, HTT, and PhHTT models are developed based on environmental conditions, they might be used to predict weed emergence across different years and

162

geographical regions. Nevertheless, validation results show that empirical models may not

163

be accurate if environmental conditions vary significantly from the original conditions in

164

which the experiment was conducted (e.g., Izquierdo et al., 2013). Many reasons can

165

explain these differences. For instance, different soil management (e.g., cultivation

166

operation) may reduce the accuracy of the model by varying the vertical distribution of the

167

seeds within the soil profile (Grundy et al., 1996). As a result, seeds will have different

168

temperature, soil moisture and light conditions, and the rate of dormancy release and

169

germination will be different (Cao et al., 2011). In addition, deeply buried seeds need more

170

time to emergence than seeds near the soil surface, thereby delaying and/or extending the

171

emergence flush. Also, model accuracy may be limited by possible population

172

differentiation at diverse geographic locations as a result of different selection pressures in

173

different environmental conditions (Dorado et al., 2009).

ew

vi

161

py

Co

174

In addition, the use of microclimate indices (TT, HTT, PhHTT) as single

175

explanatory variables is based on the following assumptions: (i) the seedling emergence

176

process is considered as “a whole” without proper discrimination between seed dormancy,

177

germination and post-germination growth subprocesses, and (ii) population-based thermal

178

(Tb = base temperature) and hydric (Ψb = base water potential) parameters are assumed

179

independent of soil microclimatic variables (soil water potential and soil temperature,

180

respectively). As suggested by Dahal and Bradford (1994) and Kebreab and Murdoch 6

Page 7 of 40

Weed Research

181

(1999) interactions between Tb and Ψ, as well as between Ψb and T are to be expected,

182

mainly when estimated parameters are not constant within the population (Bradford 2002).

183 184 185

Further developments Recently, a new generation of modelling approaches is beginning to tackle the statistical problems of conventional NLR models. For instance, Davis et al. (2013) used

187

nonlinear mixed-effect models, which can reduce the number of restrictive assumptions

188

presented by NLR models. Onofri et al. (2010; 2011; 2014) presented new developments

189

that maintain the positive features of parametric approaches; and Cao et al. (2013)

190

described a non-parametric approach as a very flexible tool to model weed emergence.

191

Other new conceptual approaches come from the algorithmic modelling culture

Re

186

(sensu Breiman, 2001). Algorithmic modelling departs from the concept of an “ideal

193

system” described by complete-and-precise information, and heads toward a real, uncertain

194

and complex system where precise and deterministic models hardly apply. As highlighted

195

by Bonissone (1997) “as we attempt to solve real-world problems we realize that they are

196

typically ill-defined systems, difficult to model and with large-scale solution spaces.” In

197

this context, Soft Computing Techniques (also called computational intelligence) are

198

capable of dealing with complex systems because it does not require strict mathematical

199

definitions. Unlike conventional computing (i.e. hard computing), Soft computing is

200

tolerant of uncertainty, imprecision and partial truth providing a better rapport with reality

201

(Das et al., 2013). Among Soft Computing Techniques (SCTs), Artificial Neural Networks

202

(ANNs) (Chantre et al., 2012; 2014) and Genetic Algorithms (GAs) (Haj Seyed Hadi &

203

Gonzalez-Andujar 2009; Blanco et al., 2014) have been proposed as promising tools for

204

weed emergence prediction.

205

Artificial neural networks (ANNs)

ew

py

Co

206

vi

192

Artificial neural networks are machine learning computer techniques that provide a

207

practical and flexible modelling framework for multivariate nonlinear mapping (Lek &

208

Guégan 1999). Such models are inspired by the operation of biological neural networks,

209

specifically the animal brain. ANNs are generally represented as a system of

210

interconnected processing units (also called nodes or neurons), which exchange signals

211

(i.e., information) between each other. The connections have numeric weights that are 7

Weed Research

Page 8 of 40

212

adjusted during the training process using a given learning algorithm. Therefore, an ANN

213

model is characterized by (1) its architecture (i.e. the pattern of connections among nodes),

214

(2) its method of determining the weights on the connections (i.e., learning or training

215

process), and (3) its activation functions (i.e. mathematical functions that process input

216

data). Briefly, a feed-forward neural network (also called multilayer perceptron) consists of

217

input variables (xn), output variables (Yn) and a given number of hidden layers containing

218

n nodes (for further details on ANN architectures the reader is referred to Fausett (1994)).

219

A basic three layer feed-forward ANN was implemented successfully by Chantre et

220

al. (2014) for Avena fatua field emergence prediction in different temperate regions of the

221

United States, Canada and South Australia (Fig. 2).

Re

Figure 2 near here

222 223

As observed in Fig. 2, each of the input nodes of the model receives a given input

vi

variable (note that cumulative thermal-time and hydro-time are considered independent

225

variables) and broadcasts it to each one of the hidden neurons. Hidden neurons compute

226

their activation functions and broadcast their results (z1, z2, z3) to the single output neuron

227

that finally produces the response of the network (Y). The output signal of each hidden

228

neuron (zj) is calculated as:

231 232

j=1, …, 3

while the output of the network is given by:   y = f  ∑ w jz j + w 0   j=1,3 

py

230

  z j = f  ∑ vij x i + v 0j   i =1,2 

Co

229

ew

224

(5)

(6)

In Eqns (5) and (6), f(.) is the activation function of the network, xi is the

233

corresponding input variable, vij are the weights of the connections between the input and

234

hidden neurons, and v0j is the bias of hidden neuron j. Similarly, wj represents the weights

235

of the connections between hidden and output neurons and w0 is the bias of the output

236

neuron.

237

From a statistical point of view, ANNs have many positive features that should be

238

highlighted. They: (i) effectively implement a wide range of nonlinear mapping problems

239

due to their excellent potential for data interpolation, (ii) admit multivariate input/output

240

mapping, (iii) have no requirements for a given a priori shape fitting function (i.e., ANNs

241

can easily generate an outcome in the form of a polynomial function of degree n), and (iv) 8

Page 9 of 40

Weed Research

242

for certain cases involving complex emergence patterns, ANNs require fewer adjustable

243

parameters than conventional multivariate techniques to obtain a similar parsimony.

244

Among the negative aspects of ANNs implementation we might cite the following:

245

First, they have very low extrapolation capacity. Thus to adequately optimize a given

246

objective function (e.g., error minimization), a wide range of observed scenarios is needed

247

to exploit ANN interpolation capability. Second, ANNs employ a “black-box” model

248

approach with limited explanatory power from a mechanistic biological perspective. Third,

249

it is usually required trial and error to identify the network that renders the largest

250

parsimony.

251

Although ANNs have been used intensively to solve highly complex non-linear mapping problems in agronomic and biological systems (Saberali et al., 2007; Fortin et al.,

253

2010; Dai et al., 2011), their application for modelling weed emergence remains largely

254

unexplored. Recent studies performed by Chantre et al. (2012) and Chantre et al. (2014)

255

have shown that ANNs are able to outperform conventional NLR models. As discussed by

256

Chantre et al. (2014), the adoption of two independent input variables (i.e., thermal-time

257

and hydro-time) instead of a single explanatory variable (i.e., hydrothermal-time) weighted

258

differently within the neural network which produced better results. This possibly was due

259

to a more intimate discrimination of the germination and post-germination growth

260

processes. In other words, the use of two (or more) independent variables instead of a

261

single aggregated variable enhances the statistical power of ANNs compared to NLR.

262

Thus, additional variables such as (i) seed burial depth (tillage system), (ii) dispersal or

263

seeding date, (iii) photoperiod, or (iv) seedbank dormancy level might be included.

264

However, overparameterization my occur by increasing the number of input variables thus

265

generating less parsimonious models.

ew

vi

Re

252

py

Co

266 267 268

Genetic Algorithms (GAs) Genetic algorithms are stochastic optimization techniques based on the evolution of

269

sets of potential solutions (also called individuals or phenotypes) following natural

270

selection rules (i.e., selection, crossover, mutation). Each individual has a set of specific

271

properties (chromosomes in a biological sense), which can suffer crossover and mutation.

272

Basically, from an initial population of randomly generated individuals the GA algorithm

273

iteratively performs a stochastic search in which successive generations (i.e., populations)

274

are obtained. The fitness (i.e. the value of the objective function in the optimization

275

problem) of every individual in successive populations is evaluated, and the evolutionary 9

Weed Research

Page 10 of 40

276

process proceeds until the best fitness value (i.e. solution) is obtained. These techniques

277

have demonstrated good performance in non-linear unconstrained models (Michalewicz,

278

1996; Rangaiah, 2010; Blanco et al., 2014). In fact, GAs show a good balance between

279

exploration and exploitation of the search space increasing the probability of convergence

280

to global optima (i.e., expected solutions) instead of reaching local minima (Blanco et al.,

281

2014). In addition, stochastic techniques only use objective function (e.g., by minimizing

282

an error function) values in the calculation compared to deterministic algorithms which

283

also require calculation of derivatives). This feature facilitates their computational

284

implementation and robustness of the search, due to a higher convergence speed. Either,

285

GA or ANN approaches can be implemented with average computing time of 15 to 20

286

minutes (personal observation). For more information on GAs the interested reader is referred to Michalewicz

Re

287

(1996). Based on the mentioned strengths, Haj Seyed Hadi and Gonzalez-Andujar (2009)

289

suggested that GAs are more appropriate to deal with ill-conditioned optimization

290

problems than the deterministic optimization algorithms usually used in the NLR

291

approach. The authors compared GAs with NLR for fitting emergence data of six weed

292

species and found that GAs often provided a better fit than NLR.

ew

vi

288

293

Recently, Blanco et al. (2014) developed an emergence model for Avena fatua

295

based on the disaggregation of seed dormancy release (i.e., after-ripening process) and

296

germination/pre-emergence growth processes. Logistic functions (eq. 2) were adopted to

297

model separately (i) seedbank dormancy release as a function of after-ripening thermal-

298

time accumulation, and (ii) germination/pre-emergence growth as a function of

299

hydrothermal-time accumulation (Fig. 3A). In such work, a GA optimization approach was

300

implemented for parameter estimation. The logic of the model illustrated in Fig. 3A is that

301

on each calendar day, a specific seedbank fraction loses its dormant condition due to after-

302

ripening thermal-time accumulation (θAT). The accumulated dormancy release function

303

(AcDr) is obtained by integrating such fractions over time following a logistic distribution

304

(Fig. 3A). Thereafter, each specific non-dormant fraction undergoes the germination/post-

305

germination process by accumulating hydrothermal-time (θHTG). Finally, accumulated

306

emergence (AcG) is obtained by integrating each day the germinated/post-germinated

307

fractions corresponding to the different portions of the seedbank that lost dormancy during

308

the previous days.

py

Co

294

309 10

Page 11 of 40

Weed Research

Figure 3 near here

310 311 312

The scheme of the model (i.e. Genetic Algorithm block) is shown in Fig. 3B. Model input consist of the soil microclimatic data (daily mean soil temperature and soil

314

water potential). Thereafter, θAT and θHTG (cumulative time variables) constitute the input

315

of the dormancy release (AcDr) and accumulated emergence (AcG) logistic functions,

316

respectively. In the present example, the solutions are represented by individuals, each one

317

constituted by the parameters of the logistic functions = adr, ag, pdr, pg) (see Fig. 3). The

318

GA algorithm performs selection, crossover and mutation operations (Forrest, 1993) on

319

such individuals (i.e., optimization variables) from a randomly selected initial population

320

along a pre-specified number of generations or until some convergence criterion is met.

321

The main output of the model is the accumulated emergence function, which is obtained by

322

integrating the daily germinated fraction of A. fatua corresponding to each non-dormant

323

seed bank fraction which, in turn is finally represented by the cumulative dormancy release

324

curve (Fig 3A).

326

ew

325

vi

Re

313

As discussed by Blanco et al. (2014), the GA optimization approach provided enough flexibility to closely represent complex A. fatua emergence patterns in the semiarid

328

region of Argentina. In this case, the possibility to quantify separately dormancy release

329

requirements (after-ripening thermal-time accumulation) from hydrothermal-time needs for

330

germination and emergence clearly outperformed the prediction accuracy of previously

331

developed NLR and ANN models. Although the model proposed by Blanco et al. (2014) is

332

rather simple in terms of the equations used (logistics) and parameters interpretation, a

333

negative aspect of such a modelling approach is the fact that it requires moderate

334

programming skills and sufficient level of expertise to develop its heuristics.

py

Co

327

335 336 337

Conclusion Models based on SCTs can provide enough flexibility to represent weed seedling

338

emergence patterns better than NLR models and, therefore, offer more reliable results. An

339

important feature of these new modelling alternatives is the lack of substantial conceptual

340

drawbacks observed in NLR models (Cao et al., 2011), such as the requirement for initial

341

parameter estimates to start the optimization process that conditions the final solution.

342

Failure to provide an accurate prediction has practical consequences for weed emergence

343

models. Economic losses may occur if control measures do not coincide with the seasonal 11

Weed Research

Page 12 of 40

344

dynamics of the weed populations, not only in terms of crop/weed competition, but harvest

345

efficiencies, and production of seeds by escaped weeds as well.

346

From a practical point of view, the implementation of SCT models for use by farmers and technicians can be done effectively. A good example of a successful

348

implementation is the AVEFA-Bordenave model developed by Blanco et al. (2014), which

349

was described previously. Output from this model can accessed freely on line at

350

http://www.meteobahia.com.ar/productoavefa.php. Extension of this system to multiple

351

sites within a region as well as multiple regions can be envisioned easily. The

352

implementation of these techniques can also be seen as sub-models integrated within

353

operational planning of herbicide-based (Lodovichi et al. 2013) or strategic IWM programs

354

(Beltran et al. 2012).

355

Re

347

Despite their advantages, some caveats of ANNs and GAs should be mentioned: (i) To obtain reliable model predictions a good pool of input data (i.e., series of cumulative

357

field emergence data with their corresponding soil microclimatic data) are required. These

358

should be representative of a wide range of observed field scenarios and to take advantage

359

of the maximum interpolation capacity of both ANNs and GAs during the training process;

360

(ii) Although SCTs are relatively simple to develop and implement, they require a

361

minimum degree of familiarity with specific programming platforms. For example, the

362

MatLab environment (MathWorks, Inc.) provides well developed toolboxes for modeling

363

ANNs and implementing GA optimization. Nevertheless, SCT models are a valuable

364

alternative to the classical parametric non-linear regression models to describe and predict

365

weed emergence. However, further research is required to establish more generally the

366

usefulness of these approaches.

ew

vi

356

py

Co

367 368

Acknowledgments

369

J. L. Gonzalez-Andujar was supported by FEDER (European Regional Development

370

Funds) and the Spanish Ministry of Economy and Competitiveness funds (Projects

371

AGL2012-33736). G. R. Chantre was supported by grants from Consejo Nacional de

372

Investigaciones Cientificas y Tecnicas (CONICET) and Universidad Nacional del Sur

373

(PGI-UNS).

374 375

References 12

Page 13 of 40

Weed Research

BATLLA D, KRUK BC & BENECH-ARNOLD RL (2004) Modelling changes in

377

dormancy in weed soil seed banks: Implications for the prediction of weed emergence. In:

378

Handbook of seed physiology. Applications to agriculture (eds. RL BENECH-ARNOLD &

379

RA SANCHEZ), 245-264. Haworth Press Inc, New York, USA.

380

BELTRAN JC, PANNELL DJ, DOOLE GJ & WHITE B (2012) A bioeconomic model for

381

analysis of integrated weed management strategies for annual barnyardgrass (Echinochloa

382

crus-galli complex) in Philippine rice farming systems. Agricultural Systems 112, 1-10.

383

BLANCO AM, CHANTRE GR, LODOVICHI MV et al. (2014) Modeling seed dormancy

384

release and germination for predicting Avena fatua L. field emergence: A genetic

385

algorithm approach. Ecological Modelling 272, 293-300.

386

BONISSONE PP (1997) Soft computing: the convergence of emerging reasoning

387

technologies. Soft Computing 1, 6-18.

388

BRADFORD KJ (2002) Applications of hydrothermal time to quantifying and modeling

389

seed germination and dormancy. Weed Science 50, 248-260.

390

BREIMAN L (2001) Statistical modeling: The two cultures (with comments and a

391

rejoinder by the author). Statistical Science 16, 199-231.

392

CALVO HARO RM, GONZALEZ-ANDUJAR JL & PEREZ S (1994) Manual de modelos

393

no lineales en los ámbitos agronómico, ganadero y forestal. INIA, Madrid, Spain.

394

CAO R, FRANCISCO-FERNANDEZ M, ANAND A, BASTIDA F & GONZALEZ-

395

ANDUJAR JL (2011) Computing statistical indices for hydrothermal times using weed

396

emergence data. The Journal of Agricultural Science 149, 701-712.

397

CAO R, FRANCISCO-FERNÁNDEZ M, ANAND A, BASTIDA F & GONZALEZ-

398

ANDUJAR JL (2013) Modeling Bromus diandrus Seedling Emergence Using

399

Nonparametric Estimation. Journal of Agricultural, Biological, and Environmental

400

Statistics 18, 64-86.

401

CHANTRE G, BLANCO A, FORCELLA F, VAN ACKER R, SABBATINI M &

402

GONZALEZ-ANDUJAR JL (2014) A comparative study between non-linear regression

403

and artificial neural network approaches for modelling wild oat (Avena fatua) field

404

emergence. The Journal of Agricultural Science 152, 254-262.

ew

vi

Re

376

py

Co

13

Weed Research

Page 14 of 40

CHANTRE GR, BLANCO AM, LODOVICHI MV et al. (2012) Modeling Avena fatua

406

seedling emergence dynamics: an artificial neural network approach. Computers and

407

Electronics in Agriculture 88, 95-102.

408

DAHAL P & BRADFORD KJ (1994) Hydrothermal time analysis of tomato seed

409

germination at suboptimal temperature and reduced water potential. Seed Science Research

410

4, 71-80.

411

DAI X, HUO Z & WANG H (2011) Simulation for response of crop yield to soil moisture

412

and salinity with artificial neural network. Field Crops Research 121, 441-449.

413

DALLEY CD, KELLS JJ & RENNER KA (2004) Effect of Glyphosate Application

414

Timing and Row Spacing on Corn (Zea mays) and Soybean (Glycine max) Yields. Weed

415

Technology 18, 165-176.

416

DAS SK, KUMAR A, DAS B & BURNWAL A (2013) On soft computing techniques in

417

various areas. International Journal of Information Technology and Computer Science 3,

418

59-68.

419

DAVIS AS, CLAY S, CARDINA J et al. (2013) Seed burial physical environment

420

explains departures from regional hydrothermal model of giant ragweed (Ambrosia trifida)

421

seedling emergence in US Midwest. Weed Science 61, 415-421.

422

DORADO J, SOUSA E, CALHA I, GONZALEZ-ANDUJAR JL & FERNANDEZ-

423

QUINTANILLA C (2009) Predicting weed emergence in maize crops under two

424

contrasting climatic conditions. Weed Research 49, 251-260.

425

FAUSETT LV (1994) Fundamentals of neural networks: architectures, algorithms, and

426

applications (ed. LV FAUSETT). Prentice-Hall, Englewood Cliffs, New Jersey, USA.

427

FORCELLA F (2000) Rotary Hoeing Substitutes for Two-Thirds Rate of Soil-Applied

428

Herbicide. Weed Technology 14, 298-303.

429

FORCELLA F (2012) Air-propelled abrasive grit for postemergence in-row weed control

430

in field corn. Weed Technology 26, 161-164.

431

FORCELLA F, BENECH-ARNOLD R, SANCHEZ R & GHERSA CM (2000) Modeling

432

seedling emergence. Field Crops Research 67, 123-139.

ew

vi

Re

405

py

Co

14

Page 15 of 40

Weed Research

FORREST S (1993) Genetic algorithms: principles of natural selection applied to

434

computation. Science 261, 872-878.

435

FORTIN JG, ANCTIL F, PARENT L-É & BOLINDER MA (2010) A neural network

436

experiment on the site-specific simulation of potato tuber growth in Eastern Canada.

437

Computers and Electronics in Agriculture 73, 126-132.

438

GARDARIN A, DÜRR C & COLBACH N (2012) Modeling the dynamics and emergence

439

of a multispecies weed seed bank with species traits. Ecological Modelling 240, 123-138.

440

GHERSA CM (2000) Plant phenology and the management of crop-weed interactions.

441

Field Crops Research 67, 91-93.

442

GRUNDY A (2003) Predicting weed emergence: a review of approaches and future

443

challenges. Weed Research 43, 1-11.

444

GRUNDY A, MEAD A & BOND W (1996) Modelling the effect of weed_seed

445

distribution in the soil profile on seedling emergence. Weed Research 36, 375-384.

446

GUMMERSON R (1986) The effect of constant temperatures and osmotic potentials on

447

the germination of sugar beet. Journal of Experimental Botany 37, 729-741.

448

HAJ SEYED HADI MR & GONZALEZ-ANDUJAR JL (2009) Comparison of fitting

449

weed seedling emergence models with nonlinear regression and genetic algorithm.

450

Computers and Electronics in Agriculture 65, 19-25.

451

HOLMSTRÖM K & PETERSSON J (2002) A review of the parameter estimation

452

problem of fitting positive exponential sums to empirical data. Applied Mathematics and

453

Computation 126, 31-61.

454

IZQUIERDO J, BASTIDA F, LEZAUN J, SÁNCHEZ DEL ARCO M & GONZALEZ-

455

ANDUJAR JL (2013) Development and evaluation of a model for predicting Lolium

456

rigidum emergence in winter cereal crops in the Mediterranean area. Weed Research 53,

457

269-278.

458

KEBREAB E & MURDOCH A (1999) Modelling the effects of water stress and

459

temperature on germination rate of Orobanche aegyptiaca seeds. Journal of Experimental

460

Botany 50, 655-664.

ew

vi

Re

433

py

Co

15

Weed Research

Page 16 of 40

LEK S & GUÉGAN J-F (1999) Artificial neural networks as a tool in ecological

462

modelling, an introduction. Ecological Modelling 120, 65-73.

463

LODOVICHI MV, BLANCO AM, CHANTRE GR et al. (2013) Operational planning of

464

herbicide-based weed management. Agricultural Systems 121, 117-129.

465

MARTINSON K, DURGAN B, FORCELLA F, WIERSMA J, SPOKAS K & ARCHER D

466

(2007) An emergence model for wild oat (Avena fatua). Weed Science 55, 584-591.

467

MCGIFFEN M, SPOKAS K, FORCELLA F, ARCHER D, POPPE S & FIGUEROA R

468

(2008) Emergence Prediction of Common Groundsel (Senecio vulgaris). Weed Science 56,

469

58-65.

470

MICHALEWICZ Z (1996) Genetic algorithms+ data structures= evolution programs (ed.

471

Z MICHALEWICZ). Springer Science & Business Media, Berlin, Germany.

472

ONOFRI A, GRESTA F & TEI F (2010) A new method for the analysis of germination

473

and emergence data of weed species. Weed Research 50, 187-198.

474

ONOFRI A, MESGARAN M, TEI F & COUSENS R (2011) The cure model: an improved

475

way to describe seed germination? Weed Research 51, 516-524.

476

ONOFRI A, MESGARAN M, NEVE P & COUSENS R (2014) Experimental design and

477

parameter estimation for threshold models in seed germination. Weed Research 54, 425-

478

435.

479

ORIADE C & FORCELLA F (1999) Maximizing efficacy and economics of mechanical

480

weed control in row crops through forecasts of weed emergence. Journal of Crop

481

Production 2, 189-205.

482

RANGAIAH GP (2010) Stochastic global optimization: techniques and applications in

483

chemical engineering (ed. GP RANGAIAH). World Scientific, Singapore.

484

RATKOWSKY D (1983) Nonlinear regression analysis.(ed. D RATKOWSKY). Marcel

485

Decker Inc., New York, USA.

486

ROMAN ES, MURPHY SD & SWANTON CJ (2000) Simulation of Chenopodium album

487

seedling emergence. Weed Science 48, 217-224.

ew

vi

Re

461

py

Co

16

Page 17 of 40

Weed Research

ROYO-ESNAL A, NECAJEVA J, TORRA J, RECASENS J & GESCH RW (2015)

489

Emergence of field pennycress (Thlaspi arvense L.): Comparison of two accessions and

490

modelling. Industrial Crops and Products 66, 161-169.

491

SABERALI S, NOORI S, KHAZAEI J & HEJAZI A (2007) Artificial neural network

492

modelling of common lambsquarters biomass production response to corn population and

493

planting pattern. Pakistan Journal of Biological Sciences 10, 326-334.

494

SCHUTTE BJ, REGNIER EE, HARRISON SK, SCHMOLL JT, SPOKAS K &

495

FORCELLA F (2008) A hydrothermal seedling emergence model for giant ragweed

496

(Ambrosia trifida). Weed Science 56, 555-560.

497

SPOKAS K & FORCELLA F (2009) Software tools for weed seed germination modeling.

498

Weed Science 57, 216-227.

499

VLEESHOUWERS L & KROPFF M (2000) Modelling field emergence patterns in arable

500

weeds. New Phytologist 148, 445-457.

501

WERLE R, SANDELL LD, BUHLER DD, HARTZLER RG & LINDQUIST JL (2014)

502

Predicting Emergence of 23 Summer Annual Weed Species. Weed Science 62, 267-279.

503

ZAMBRANO-NAVEA C, BASTIDA F & GONZALEZ-ANDUJAR JL (2013) A

504

hydrothermal seedling emergence model for Conyza bonariensis. Weed Research 53, 213-

505

220.

ew

vi

Re

488

py

Co

506 507

Figure captions

508

Figure 1. Typical S-type model fitted to emergence data.

509 510

Figure 2. A three layer feed-forward ANN. The network has two inputs thermal-time (ϴT=

511

(T - Tb)tg) and hydro-time (ϴH = (ψ - ψb)tg), three neurons in the hidden layer and a single

512

output variable (y = cumulative emergence). The term f(.) = the activation function of the

513

network; vij = connection weights between input-hidden layer neurons and wj = connection

514

weights between hidden and output layer neurons; v0j = bias on hidden neuron j, w0 =

515

output neuronbias (adapted from Chantre et al., 2014).

516 17

Weed Research

Page 18 of 40

517

Figure 3. Genetic Algorithm based model for Avena fatua field emergence prediction

518

(adapted from Blanco et al., 2014). (A) Schematic representation of the model showing

519

that accumulated dormancy release (dashed line) is obtained by integrating daily dormancy

520

release (bars). Dotted lines represent accumulated germination/pre-emergence growth of

521

each non-dormant fraction. Note that as time increases, the dotted line reaches the height of

522

the bar, meaning that each non-dormant seedbank fraction germinates during the following

523

days according to a specific distribution. Accumulated emergence (solid line) is obtained

524

by integrating each day the germinated/post-germinated fraction corresponding to each

525

non-dormant seedbank fraction. (B) Genetic algorithm based parameter estimation scheme.

ew

vi

Re py

Co 18

Weed Research

Em ergence (% )

Page 19 of 40

0

Re

100 90 80 70 60 50 40 30 20 10 0

300

600

900

1200

1500

1800

ew

vi

Soil Therm al Tim e (C)

Figure 1. Typical S-type model fitted to emergence data

2100

2400

py

Co

Weed Research

Page 20 of 40

v01

v11

f z1

w1

ϴT

v02 v12 v21

z2

w0

w2

f

f

y

v22

ϴH

w3

v13

z3

v23 v03

f

vi

Re Figure 2. A three layer feed-forward ANN. The network has two inputs (thermal-

ew

time= ϴT and hydro-time=ϴH), three neurons in the hidden layer and a single output variable (y = cumulative emergence) (adapted from Chantre et al., 2014).

py

Co

Page 21 of 40

Weed Research

1

ew

vi

Re py

Co

2 3 4 5

1

Weed Research

Page 22 of 40

6

Figure 3. Genetic algorithm based model for Avena fatua field emergence prediction

7

(adapted from Blanco et al.,2014). (A) Schematic representation of the model showing

8

that accumulated dormancy release (dashed line) is obtained by integrating daily

9

dormancy release (bars). Dotted lines represent accumulated germination/pre-emergence

10

growth of each non-dormant fraction. Note that as time increases, the dotted line reaches

11

the height of the bar, meaning that each non-dormant seedbank fraction germinates

12

during the following days according to a specific distribution. Accumulated emergence

13

(solid line) is obtained by integrating each day the germinated/post-germinated fraction

14

corresponding to each non-dormant seed bank fraction. (B) Genetic algorithm based

15

parameter estimation scheme.

16

ew

vi

Re py

Co 2

Page 23 of 40

Weed Research

Table S1. Empirical and soft computing based models for weed emergence published from 1996 to 2015. This list does not pretend to be exhaustive, but only gives a sample of the multitude of empirical weed emergence models developed in the last 20 years. Abbreviations: RH: Chapman-Richards; CSP: Cubic Splines; POL: Polynomial; NLME: Nonlinear mixed-effect models. !

ew

vi

Re py

Co

X

X

Apocynum androsaemifolium L.

turf Turf

X

Datura stramonium L.

X

X

X

X

X

X

X

Digitaria sanguinalis (L.) Scop.

Echinocloa colonum (L.) Link

Echinocloa cruss-galli (L.) Beauv.

Echinochloa ssp.

Eleusine indica (L.) Gaertn.

Eriochloa villosa (Thunb.) Kunth

X

X

X

X

X

Galium spp

Galium spurium L.

Galium tricornutum Dandy

Heliotropium europaeum L.,

Hibiscus trionium L.

X

X

Galinsoga parviflora Cav.

X

X

Galinsoga ciliate Raf. Blake

Galium aparine L.

X

Fallopia convolvulus (K.) A. Löve

X

X

X

Digitaria ischaemum (Schreb.) Schreb. ex Muhl.

X

X

X

X

X

X

X

X

Datura ferox L.

X

X

Cyperus rotundus L.

RH-CSP

NLME

RH-CSP

NLME

X

turf

Cucumber

winter cereals

winter cereals

winter cereals

winter cereals; No indicated

No indicated

No indicated

No indicated

turf

maize, soybean, turf

Rice

corn, soybean,rice, cotton, turf

Maize, Soybean

Maize; soybean, turf

Maize; Soybean

Cucumber

Rice

sunflower; cucumber

No indicated

Cyperus difformis L.

RH-CSP

Cucumber

Wheat

No indicated

turf; cucumber

corn/Soybean, No indicated

turf

No indicated

cereals

winter cereals, No indicated

cereals

cereals, canola

Garcia et al., 2013; Cao et al., 2013

Haj Seyed Hadi & Gonzalez-Andujar, 2009; Leguizamon et al., 2005

Page et al., 2006; Bullied et al., 2003

Martinson et al., 2007; Gonzalez-Diaz et al., 2007

Graziani & Steinmaus, 2009 Chantre et al., 2014; Blanco et al., 2014; Chantre et al., 2012;

Webster & Cardina, 1999

Wu et al., 2013

Blackshaw, 2003

Schutte et al., 2008; Werle et al., 2014

Davis et al., 2013; Schutte et al., 2012;

Myers et al., 2004; Werle et al. 2014

Otto et al., 2007

Co

py

Werle et al., 2014

Tursun et al., 2015

Royo-Esnal, 2010

Royo-Esnal, 2010

Royo-Esnal, 2010

Royo-Esnal, 2010; Benjamin et al., 2010

Otto et al., 2007

Otto et al., 2007

Otto et al., 2007

Werle et al., 2014

Leguizamon et al. 2009; Masin et al., 2005

Lundy et al., 2014

Dorado et al., 2009; Werle et al., 2014

Masin et al., 2014; Bagavathiannan et al, 2011;

Leguizamon et al. 2009

Leguizamon et al. 2009; Masin et al., 2005;Myers et al., 2004

Masin et al., 2014; Masin et al., 2010;

Fidanza et al., 1996

Werle et al., 2014

Dorado et al., 2009; Vitta & Satorre, 1999

Tursun et al., 2015

Lundy et al., 2014

Satorre et al., 1996; Tursun et al., 2015

Zambrano-Navea et al., 2013

Tursun et al., 2015

Donald, 2000

Otto et al., 2007

Roman et al., 2000; Werle et al., 2014; Tursun et al., 2015

Bullied et al., 2003;Grundy & Mead, 2000;

Otto et al., 2007;Myers et al., 2004; Leblanc et al., 2004;

Grundy et al., 2010;Masin et al., 2010;

Werle et al., 2014 Masin et al., 2014; Schutte et al. 2014: Masin et al., 2012;

Otto et al., 2007; Grundy & Mead, 2000

Haj Seyed Hadi & Gonzalez-Andujar, 2009

w

vie

No indicated

non crop

Blueberry

no indicated

X

X

No indicated; corn/soybean, turf

corn, soybean;turf

No indicated

Werle et al., 2014

Masin et al., 2010; Haj Seyed Hadi & Gonzalez-Andujar, 2009

Vitta & Satorre, 1999

Myers et al., 2004

Haj Seyed Hadi & Gonzalez-Andujar, 2009

Bullied et al., 2003; Werle et al. 2014; Tursun et al., 2015

turf

X

X

X

Dorado et al., 2009; Myers et al., 2004; Werle et al., 2014 Ekeleme et al., 2005 Benjamin et al., 2010; Otto et al., 2007

canola; no indicated, cucumber

X

X

X

Reference Masin et al., 2014;Masin et al., 2010;

Maize; cereals, canola,

Soybean

corn, Soybean

cereals

Re X

X

X

wheat; No indicated

corn, cassava

Maize; Soybean; turf

Nonparametric ANNs GAs Crop/s

Cynodon dactylon L. Pers.

RH-CSP

RH-CSP

RH

Other

Conyza bonariensis (L.) Cronquist

X

X

Convolvulus arvensis L.

X

X

X

X

X

X

X

Cirsium arvense ( L.) Scop.

X

X

Chenopodium polyspermum L.

X

Chenopodium album L.

X

Capsella bursa-pastoris L. (Medicurs)

Cenchrus spinifex Cav.

X

X

Bromus spp.

Bromus diandrus Roth

Avena sterilis L.

X

Avena fatua L.

X

X

X

Apocynum cannabinum L.

X

X

X

Arundo donax L.

X

X

Ambrosia trifida L.

Androsace septentrionalis L.

X

Ambrosia artemisiifolia L

X

X

Amaranthus rudis Sauer

Amaranthus spp.

X

X

Amaranthus retroflexus L.

X

X

X

Amaranthus quitensis Kunth

X

X

Amaranthus hibridus L.

Amaranthus blitoides S. Wats

X

LOGISTIC

X

X

Ageratum conyzoides L.

X

GOMPERTZ

Alopecurus myosuroides L

X

WEIBULL

Abutilon teophrasti Medik.

Species

Weed Research

Page 24 of 40

X

Kochia scoparia (L.) Schrad.

X

Portulaca oleracea L.

X

Solanum ptychanthum Dunal

X

Xanthium strumarium L.

X

X

X

Veronica hederifolia L.

Veronica persica Poir.

X

X

X

X

Maize; Soybean; turf

No indicated

No indicated

Maize, Soybean

Cucumber

canola, No indicated

X

X

X

Thlaspi arvense L.

Tribulus terrestris L. Urochloa platyphylla (Munro ex C.Wright) R.D.Webster

Maize; soybean

turf

No indicated

turf, corn, soybean

Wheat; No indicated

X

X

Cucumber

turf

X

X

X

X

Sorghum halepense (L.) Pers

X

Stellaria media (L.) Vill.

X

Sorghum bicolor (L.)

Sonchus oleraceus L.

X

RH-CSP

canola No indicated

X

RH-CSP

X

X

X

Sisymbrium irio L.

Solanum nigrum L.

Sinapis arvensis L.

X

Maize, turf, canola

Maize; turf

Sicyos angulatus L.

X

X

X

X

X

Turf, corn, soybean

Maize, Soyberan

Bullied et al., 2003; Werle et al., 2014

Masin et al., 2010; Masin et al., 2005a

Masin et al., 2010; Werle et al., 2014

Masin et al., 2005;Myers et al., 2004

Leguizamon et al. 2009

Myers et al., 2004;Werle et al., 2014

McGiffen et al., 2008

Harris et al., 1998

Tursun et al., 2015

Werle et al., 2014

Bullied et al., 2003

Co

Werle et al., 2014

Dorado et al., 2009; Norsworthy & Oliveira, 2007;

Otto et al., 2007

Grundy & Mead, 2000

Leguizamon et al. 2009

Tursun et al., 2015

Bullied et al., 2003; Werle et al., 2014;Royo-Esnal et al., 2015

Benjamin et al., 2010; Grundy et al., 2010; Grundy & Mead, 2000

Leguizamon et al. 2009;Dorado et al., 2009

Masin et al., 2012;Masin et al., 2010;

Werle et al., 2014

Otto et al., 2007

Werle et al., 2014; Myers et al., 2004

Tursun et al., 2015

Ray et al., 2005

Bullied et al., 2003

Werle et al., 2014

py

Otto et al., 2007; Calado et al., 2009; Tursun et al., 2015; Royo-Esnal et al., 2015

w

vie

corn, Soybean; turf

berries,ornmaentals, vegetables

forages, cereals, mint,

non crop

Cucumber

Re

No indicated

canola

Izquierdo et al., 2009

Werle et al., 2014

Grundy & Mead, 2000 Eizenberg et al. 2012

Izquierdo et al.,2013; Haj Seyed Hadi & Gonzalez-Andujar, 2009

Otto et al., 2007

Blackshaw et al. 2002

Schwinghamer & Van Acker, 2008; Werle et al., 2014

Werle et al., 2014

Haj Seyed Hadi & Gonzalez-Andujar, 2009

Ekeleme et al., 2003

Weed Research

No indicated; cucumber; cereal

cereals

turf

Sunflower

No indicated

cereals

No indicated

No indicated

Grassland; turf

turf

Setaria viridis (L.) Beauv.

X

X

RH-CSP

corn, cassava cereals

Setaria pumila (Poir.)

X

Setaria geniculata (Lam) Beauv.

X

X

RH-CSP-POL

X

X

X

X

Setaria faberi Herrm.

X

X

X

X

X

X

X

X

Setaria glauca (L.) Beauv.

X

Senecio vulgaris L.

Ranunculus repens L.

X

Polygonum pensylvanicum L.

X

X

Polygonum convolvulus L.

Polygonum aviculare L.

X

X

X

Papaver rhoeas L.

X

X

Matricaria perforata Merat

Orobanche cumana wallr.

X

Panicum dichotomiflorum Michx.

X

X

Lolium rigidum L.

X

Lamium purpureum L.

Lamium amplexicaule L.

X

X

Ipomoea hederacea Jacq.

Ipomoea purpurea (L.) Roth

Imperata cylindrica (L.) Beauv.

Page 25 of 40

Weed Research

REFERENCES BAGAVATHIANNAN MV, NORSWORTHY JK, SMITH KL & BURGOS N (2011) Seedbank size and emergence pattern of barnyardgrass (Echinochloa crus-galli) in Arkansas. Weed Science 59, 359-365. BENJAMIN LR, MILNE AE, PARSONS DJ & LUTMAN PJ (2010) A model to simulate yield losses in winter wheat caused by weeds, for use in a weed management decision support system. Crop Protection 29, 1264-1273. BLACKSHAW RE (2003) Soil temperature and soil water effects on pygmyflower (Androsace septentrionalis) emergence. Weed Science 51, 592-595.

Re

BLACKSHAW RE, BRANDT RN & ENTZ T (2002) Soil temperature and soil water effects on henbit emergence. Weed Science 50, 494-497.

vi

BLANCO AM, CHANTRE GR, LODOVICHI MV et al. (2014) Modeling seed dormancy release and germination for predicting Avena fatua L. field emergence: A genetic algorithm

ew

approach. Ecological Modelling 272, 293-300.

BULLIED WJ, MARGINET AM & VAN ACKER RC (2003) Conventional-and conservationtillage systems influence emergence periodicity of annual weed species in canola. Weed

Co

Science 51, 886-897.

CALADO JM, BASCH G & DE CARVALHO M (2009) Weed emergence as influenced by soil

py

moisture and air temperature. Journal of Pest Science 82, 81-88.

CAO R, FRANCISCO-FERNÁNDEZ M, ANAND A, BASTIDA F & GONZÁLEZ-ANDÚJAR JL (2013) Modeling Bromus diandrus Seedling Emergence Using Nonparametric Estimation. Journal of Agricultural, Biological, and Environmental Statistics 18, 64-86. CHANTRE G, BLANCO A, FORCELLA F, VAN ACKER R, SABBATINI M & GONZALEZANDUJAR J (2014) A comparative study between non-linear regression and artificial neural network approaches for modelling wild oat (Avena fatua) field emergence. The Journal of Agricultural Science 152, 254-262. CHANTRE GR, BLANCO AM, LODOVICHI MV et al. (2012) Modeling Avena fatua seedling emergence dynamics: an artificial neural network approach. Computers and Electronics in Agriculture 88, 95-102.

Page 26 of 40

Page 27 of 40

Weed Research

DAVIS AS, CLAY S, CARDINA J et al. (2013) Seed burial physical environment explains departures from regional hydrothermal model of giant ragweed (Ambrosia trifida) seedling emergence in US Midwest. Weed Science 61, 415-421. DONALD WW (2000) A degree-day model of Cirsium arvense shoot emergence from adventitious root buds in spring. Weed Science 48, 333-341. DORADO

J,

SOUSA

E,

CALHA

I,

GONZÁLEZ!ANDÚJAR

J

&

FERNÁNDEZ!QUINTANILLA C (2009) Predicting weed emergence in maize crops under two contrasting climatic conditions. Weed Research 49, 251-260. EIZENBERG H, HERSHENHORN J, ACHDARI G & EPHRATH J (2012) A thermal time model for predicting parasitism of Orobanche cumana in irrigated sunflower—Field validation.

Re

Field Crops Research 137, 49-55.

EKELEME F, FORCELLA F, ARCHER DW, CHIKOYE D & AKOBUNDU IO (2004)

vi

Simulation of shoot emergence pattern of cogongrass (Imperata cylindrica) in the humid tropics. Weed Science 52, 961-967.

ew

FIDANZA M, DERNOEDEN P & ZHANG M (1996) Degree-days for predicting smooth crabgrass emergence in cool-season turfgrasses. Crop Science 36, 990-996. GARCÍA AL, RECASENS J, FORCELLA F, TORRA J & ROYO-ESNAL A (2013)

Co

Hydrothermal emergence model for ripgut brome (Bromus diandrus). Weed Science 61, 146-153.

py

GONZÁLEZ-DÍAZ L, LEGUIZAMÓN E, FORCELLA F & GONZÁLEZ-ANDÚJAR J (2007) Short communication. Integration of emergence and population dynamic models for long term weed management using wild oat (Avena fatua L.) as an example. Spanish Journal of Agricultural Research 5, 199-203. GRAZIANI A & STEINMAUS SJ (2009) Hydrothermal and thermal time models for the invasive grass, Arundo donax. Aquatic Botany 90, 78-84. GRUNDY A, MEAD A, BOND W, CLARK G & BURSTON S (2011) The impact of herbicide management on long!term changes in the diversity and species composition of weed populations. Weed Research 51, 187-200. GRUNDY AC & MEAD A (2000) Modeling weed emergence as a function of meteorological records. Weed Science 48, 594-603.

Weed Research

Page 28 of 40

HADI MHS & GONZALEZ-ANDUJAR J (2009) Comparison of fitting weed seedling emergence models with nonlinear regression and genetic algorithm. Computers and Electronics in Agriculture 65, 19-25. HARRIS S, DOOHAN D, GORDON R & JENSEN K (1998) The effect of thermal time and soil water on emergence of Ranunculus repens. Weed Research 38, 405-412. IZQUIERDO

J,

BASTIDA

F,

LEZAÚN

J,

SÁNCHEZ

DEL

ARCO

M

&

GONZALEZ!ANDUJAR JL (2013) Development and evaluation of a model for predicting Lolium rigidum emergence in winter cereal crops in the Mediterranean area. Weed Research 53, 269-278. LAWSON AN, VAN ACKER RC & FRIESEN LF (2006) Emergence timing of volunteer canola

Re

in spring wheat fields in Manitoba. Weed Science 54, 873-882. LEBLANC ML, CLOUTIER DC, STEWART KA & HAMEL C (2003) The use of thermal time Science 51, 718-724. E,

FERNANDEZ!QUINTANILLA

ew

LEGUIZAMON

vi

to model common lambsquarters (Chenopodium album) seedling emergence in corn. Weed C,

BARROSO

J

&

GONZALEZ!ANDUJAR J (2005) Using thermal and hydrothermal time to model seedling emergence of Avena sterilis ssp. ludoviciana in Spain. Weed Research 45, 149-156.

Co

LEGUIZAMÓN ES, RODRIGUEZ N, RAINERO H et al. (2009) Modelling the emergence pattern of six summer annual weed grasses under no tillage systems in Argentina. Weed

py

Research 49, 98-106.

LUNDY M, HILL J, VAN KESSEL C et al. (2014) Site-specific, real-time temperatures improve the accuracy of weed emergence predictions in direct-seeded rice systems. Agricultural Systems 123, 12-21. MARTINSON K, DURGAN B, FORCELLA F, WIERSMA J, SPOKAS K & ARCHER D (2007) An emergence model for wild oat (Avena fatua). Weed Science 55, 584-591. MASIN R, LODDO D, BENVENUTI S, OTTO S & ZANIN G (2012) Modeling weed emergence in Italian maize fields. Weed Science 60, 254-259. MASIN R, LODDO D, BENVENUTI S, ZUIN MC, MACCHIA M & ZANIN G (2010) Temperature and water potential as parameters for modeling weed emergence in centralnorthern Italy. Weed Science 58, 216-222.

Page 29 of 40

Weed Research

MASIN R, LODDO D, GASPARINI V, OTTO S & ZANIN G (2014) Evaluation of Weed Emergence Model AlertInf for Maize in Soybean. Weed Science 62, 360-369. MASIN R, ZUIN MC, ARCHER DW, FORCELLA F & ZANIN G (2005) WeedTurf: a predictive model to aid control of annual summer weeds in turf. Weed Science 53, 193-201. MASIN R, ZUIN MC & ZANIN G (2005) Phenological observations on shrubs to predict weed emergence in turf. International Journal of Biometeorology 50, 23-32. MCGIFFEN M, SPOKAS K, FORCELLA F, ARCHER D, POPPE S & FIGUEROA R (2008) Emergence Prediction of Common Groundsel (Senecio vulgaris). Weed Science 56, 58-65. MYERS MW, CURRAN WS, VANGESSEL MJ et al. (2004) Predicting weed emergence for eight annual species in the northeastern United States. Weed Science 52, 913-919.

Re

OTTO S, ZUIN M, CHISTE G & ZANIN G (2007) A modelling approach using seedbank and soil properties to predict the relative weed density in organic fields of an Italian pre!alpine

vi

valley. Weed Research 47, 311-326.

PAGE ER, GALLAGHER RS, KEMANIAN AR, ZHANG H & FUERST EP (2006) Modeling 54, 838-846.

ew

site-specific wild oat (Avena fatua) emergence across a variable landscape. Weed Science RAY J, CREAMER R, SCHROEDER J & MURRAY L (2005) Moisture and temperature

Co

requirements for London rocket (Sisymbrium irio) emergence. Weed Science 53, 187-192. ROMAN ES, MURPHY SD & SWANTON CJ (2000) Simulation of Chenopodium album

py

seedling emergence. Weed Science 48, 217-224.

ROYO-ESNAL A, NECAJEVA J, TORRA J, RECASENS J & GESCH RW (2015) Emergence of field pennycress (Thlaspi arvense L.): Comparison of two accessions and modelling. Industrial Crops and Products 66, 161-169. ROYO-ESNAL A, TORRA J, CONESA JA, FORCELLA F & RECASENS J (2010) Modeling the emergence of three arable bedstraw (Galium) species. Weed Science 58, 10-15. SATORRE E, RIZZO F & ARIAS S (1996) The effect of temperature on sprouting and early establishment of Cynodon dactylon. Weed Research 36, 431-440. SCHUTTE B, TOMASEK B, DAVIS A et al. (2014) An investigation to enhance understanding of the stimulation of weed seedling emergence by soil disturbance. Weed Research 54, 1-12. SCHUTTE BJ, REGNIER EE & HARRISON SK (2012) Seed dormancy and adaptive seedling emergence timing in giant ragweed (Ambrosia trifida). Weed Science 60, 19-26.

Weed Research

SCHWINGHAMER TD & VAN ACKER RC (2008) Emergence Timing and Persistence of Kochia (Kochia scoparia). Weed Science 56, 37-41. TURSUN N, AKıNCı !E, "AHIN M & ULUDA# A (2015) Estimating Time of Weed Emergence in Cucumber (Cucumis sativus L.). Turkish Journal of Agriculture-Food Science and Technology 3, 271-278. VITTA J & SATORRE E (1999) Validation of a weed: crop competition model. Weed Research 39, 259-269. WEBSTER TM & CARDINA J (1999) Apocynum cannabinum seed germination and vegetative shoot emergence. Weed Science 47, 524-528. WERLE R, SANDELL LD, BUHLER DD, HARTZLER RG & LINDQUIST JL (2014) Predicting

Re

Emergence of 23 Summer Annual Weed Species. Weed Science 62, 267-279. WU L, BOYD NS, CUTLER GC & OLSON AR (2013) Spreading dogbane (Apocynum

vi

androsaemifolium) development in wild blueberry fields. Weed Science 61, 422-427. ZAMBRANO!NAVEA C, BASTIDA F & GONZALEZ!ANDUJAR JL (2013) A hydrothermal

ew

seedling emergence model for Conyza bonariensis. Weed Research 53, 213-220.

Page 30 of 40

py

Co