Soybean Phenology Simulation in the North-Central

0 downloads 0 Views 1012KB Size Report
Sep 15, 2011 - The degree of bias from congruence was esti- ... observed DAE to R1 and to R3.5 were not quite as congruent ..... ASA, CSSA, and SSSA,.
Jessica Torrion, Tri Deri Setiyono, Kenneth Cassman, and James Specht* ABSTRACT

Forecast simulation of soybean [Glycine max (L.) Merr.] vegetative stem node (Vn) and reproductive (Rn) phases can be used to improve timing of agronomic practices to optimize input use efficiency. A soybean model (SoySim) was used to predict soybean phenological stages in researcher and producer fields and to evaluate sources of variance. Evaluation of SoySim phenology simulation involved 21 and 22 Nebraska soybean producers in 2009 and 2010, respectively, with cultivars of various maturity groups (MGs 2.5–3.8) and planting dates (21 April–11 June). SoySim evaluation in research fields was conducted in Nebraska, Iowa, and Indiana with cultivar MGs 0.8 to 4.2 planted 27 April to 17 June. The combined root mean square error (RMSE) in research fields for V1 (first-node), R3.5 (mid-pod), R5 (beginning seed), and R7 (beginning maturity) was 3.6 d. In producer fields, Vn and Rn stages RMSE was 5.5 and 5.7 d, respectively. With few exceptions, the simulated and observed Vn or Rn stage data clusters in the research and producer fields did not differ (α = 0.05). Aside from subjectivity in most Rn calls that required quantifying pod and seed sizes, and pod maturity, variance in simulated and observed Vn or Rn in producer fields was related to the reliability of VE (emergence) calls, or the use of planting dates instead of VE dates in some producer fields. For no-till to minimum-till fields, the accuracy and precision of SoySim forecast simulation requires a reliable VE date call for scheduling soybean growth-stage dependent critical farming decisions.

S

oybean vegetative phenology is typically described with an incremental ordinal numbers assigned to consecutive main stem nodes (Vn), beginning with the cotyledonary node, VC (= V0). Upon onset of reproductive development, reproductive phenology is described with incremental ordinal numbers assigned to the beginning and end of the floral, pod, seed, and maturity phases (Rn). This now universally used soybean staging system was described by Fehr and Caviness (1977). Knowing when a soybean stage has just occurred, or better yet, knowing the expected calendar date of its occurrence in the future, can be used to improve the timing of a crop management operation, such that it occurs at the developmental stage considered to be critical for enhancing seed yield or for mitigating pest-mediated yield loss. For example, chemical control of soybean insects or diseases is often best achieved when the application is timed to coincide with a specific phenological stage (Esker et al., 2010). The effectiveness of such phenologytimed pest control measures is well documented (Table 1). The most critical stage for controlling insect pests and diseases typically occurs during early reproductive phases (Krupke et al., 2009). Mueller et al. (2009) conducted studies in Paraguay, Zimbabwe, and the United States (Florida and Georgia) on the control of soybean rust (Phakopsora pachyrhizi). They reported that fungicide application at stage R3 Univ. of Nebraska-Lincoln, Dep. of Agronomy & Horticulture, P.O. Box 830915, Lincoln, NE 68583. Received 9 May 2011. *Corresponding author ([email protected]). Published in Agron. J. 103:1661–1667 (2011) Posted online 15 Sep 2011 doi:10.2134/agronj2011.0141 Copyright © 2011 by the American Society of Agronomy, 5585 Guilford Road, Madison, WI 53711. All rights reserved. No part of this periodical may be reproduced or transmitted in any form or by any means, electronic or mechanical, including photocopying, recording, or any information storage and retrieval system, without permission in writing from the publisher.

(beginning pod) was more effective than an earlier application at R1 (beginning bloom), or a later application at R5 (beginning seed). Results from a 3-yr study in Iowa on the control of brown spot (Septora glycines), downy mildew (Peronospora manshurica), cercospora leaf blight (Cercospora kikuchii), and frogeye leaf spot (C. sojina) supported these findings; application of fungicide at stage R3 resulted in higher yield than application at R1, at R2 (full bloom), at R4 (full pod), or at R5 (Mueller and Robertson, 2010). For bean leaf beetle (Cerotoma trifurcata), stage R5 is considered the most critical stage for control of this pest to avoid losses in both seed yield and quality (Krupke et al., 2009). Soybean pod development (R3–R4) and seed-filling (R5–R6 [full seed]) periods are the most water-stress sensitive developmental phases in terms of yield reduction (Purcell and Specht, 2004). For irrigated soybean producers, the ability to project the calendar dates of successive reproductive stages is critical for implementing Rn-stage-based irrigation scheduling. Specht et al. (1989) noted that there was no need to irrigate soybean in Nebraska until stage R3 on silty clay loam soils, or silt loam soils, that were normally recharged to field capacity soil water content during the winter–spring fallow period, or recharged just before, or after planting. These authors noted that irrigation deferred until R3 did not reduce seed yield relative to the control (full season irrigated treatment), even in years when rainfall was sparse between the dates of planting and R3. The defer-to-R3 irrigation strategy has the potential to reduce both water and energy use because the risk of yield reduction from not applying irrigation before R3 is quite low, and actually less than the probability of a rainfall event providing sufficient Abbreviations: DAE, days after emergence; MG, cultivar maturity group; R1, beginning bloom; R2, full bloom; R3, beginning pod, R3.5, mid-pod; R4, full pod; R5, beginning seed; R6, full seed; R7, beginning maturity; RMSE, root mean square error; Rn, reproductive floral, pod, seed, or maturity stage; SoySim, soybean simulation crop model; VE, emergence; Vn, vegetative stage, where n corresponds to the ordinal number of the given main stem node.

A g ro n o my J o u r n a l  •   Vo l u m e 10 3 , I s s u e 6  •  2 011

1661

Crop Economics, Production & Management

Soybean Phenology Simulation in the North-Central United States

Table 1. Critical soybean vegetative (Vn) and reproductive (Rn) stage-based agronomic practices to avoid yield loss and enhance farm input use efficiency. Management Weeds, pests, and diseases   Early weed removal   Late weed removal   Herbicide application (Glyphosate)   Bean leaf beetle†   Blister beetle complex   Mexican bean beetle   Japanese beetle   Green cloverworm   Grasshopper complex   Woolybear or saltmarsh caterpillar   Potato leafhopper          

Soybean stage

Source

V1-V4 Acker et al. (1993) Gustafson et al. (2006) R1 Acker et al. (1993) V2-V6 Vangessel et al. (2001) R5 Krupke et al. (2009) Pre-R1,40% defoliation; R1-R5,15% defoliation; R6-harvest,25% defoliation

V1 to pre-R1, 2 leafhopper per plant; R1-R2, 1 leafhopper per trifoliolate; R3, 2 leafhopper per trifoliolate

Aphids scouting Aphids spraying Stem rot Fungicide application (soybean rust) Fungicide application

  Bean pod mottle virus Irrigation   Irrigation initiation   Irrigation termination

R3 R4-R5 R1-R3 R3 R1-R5

Vangessel et al. (2001) Mueller et al. (2009) Backman et al. (1979)

R1

Ren et al. (1997)

R3 R7

Specht et al. (1989) Wright et al. (2006)

† Scientific names: bean leaf beetle, Cerotoma trifurcata; blister beetle, Epicauta spp.; Mexican bean beetle, Epilachna varivestris; Japanese beetle, Popillia japonica; Green cloverworm, Plathypena scabrai; grasshopper, Melanoplus spp.; wooly bear or saltmarsh caterpillar, Estigmene acrea; potato leafhopper, Empoasca fabae.

amount of water. Knowledge of the date of R3 would thus be useful for irrigation scheduling purposes. Another phenology-based application in soybean irrigation management is determining when the last irrigation should be applied. Wright et al. (2006) noted that irrigation was not needed after the date when a soybean crop attains stage R7 (beginning maturity), otherwise known by crop physiologists as the physiological maturity stage. In fact, the last irrigation should be scheduled on a calendar date before stage R7 in a manner that, based on projected crop water use, would allow the crop to deplete soil water to the soil water depletion threshold (for triggering an irrigation event) on or just after the calendar date of R7. However, it is difficult for producers to mentally project a calendar date for this R7 stage, which in turn makes the last pre-R7 irrigation difficult to schedule. Monitoring soybean phenology is time-consuming because it typically requires biweekly producer observation of randomly chosen soybean plants present at several representative locations in each field. Although self-tracking of crop phenology provides a reliably accurate assessment of any current crop stage, some Rn stages are notoriously difficult to quantify by producers or consultants who may lack the experience or training. In any event, projection of the calendar dates of Vn and Rn stages is of greater value to producers than just knowing the “today’s” current developmental stage, because a stage forecast provides lead time to schedule a field operation that may be most effective when precisely applied at a critical phenological stage. Recent advances in the understanding of soybean growth and development have provided new approaches for robust projection of phenological stages. Setiyono et al. (2007) developed a new approach to simulate soybean phenology that moves away 1662

from use of multiple cultivar-specific input parameters—which are frequently not available or difficult to measure—to a more generic parameterization based on MG. A SoySim model prediction of the calendar dates of key phenological stages requires only two initial inputs: date of planting (or emergence, VE) and MG of the planted cultivar. In addition to these crop-specific inputs, the model uses easily available site-specific inputs such as latitude (used in estimating daylengths) and daily maximum and minimum air temperatures to simulate soybean phenology. To simulate other agronomic data, for example, leaf growth and seed yield, SoySim requires daily inputs of solar radiation, relative humidity, and ambient CO2 concentration, in addition to the above weather inputs (Setiyono et al., 2010). Although SoySim model output with respect to vegetative and reproductive dry matter by plant part has been validated relative to soybean dry matter and seed yield data collected at four field sites across the North-Central United States (Setiyono et al., 2010), where MG 2.0 to 4.0 cultivars are grown (Scott and Aldrich, 1970), the reliability and accuracy of SoySim predictions of phenological stages has not yet been rigorously evaluated. In the research reported here, the performance of SoySim in predicting the calendar date occurrence of Vn and Rn stages was examined in soybean crops grown for two or more years in many Nebraska, Iowa, and Indiana experimental research plots, and for 2 yr in 21 to 22 Nebraska producer fields. Our objectives were to: (i) to compare the performance of SoySim in predicting the calendar dates of Vn and Rn stage occurrence with the calendar dates of the observed occurrence of those stages in the research plots or fields, as ascertained by researchers or producers, and (ii) to identify the possible reasons for any variance between the Agronomy Journal  •  Volume 103, Issue 6  •  2011

Fig. 1. Locations of the Nebraska, Iowa, and Indiana research fields and the 21 (2009) and 22 (2010) Nebraska producer fields (see symbol legend) that were used to evaluate the reliability and accuracy of the SoySim model for prediction of soybean vegetative stage (Vn) and reproductive floral, pod, seed, or maturity stage (Rn).

SoySim-predicted and researcher- or producer-observed dates of the various soybean phenological stages. MATERIALS AND METHODS Soybean phenology was monitored in several field experiments conducted by agronomists at research facilities located in Nebraska, Iowa, and Indiana, and in soybean producer fields in Nebraska (Fig. 1). The producer farms were scattered across the eastern third of Nebraska where much of the state’s soybean crop is found. These included mostly irrigated, but also some rainfed, fields. The fields of the collaborating producers were representative of a wide range of Nebraska production systems and crop management practices (total fields: 21 and 22 in 2009 and 2010, respectively). The producer field sites had elevations ranging from 296 to 707 m above sea level (asl). The cultivars in those fields had MGs ranging from 2.5 to 3.8, and were planted with inter-row spacings ranging from 0.19 to 0.76 m. The research field sites in Nebraska included two surface-dripirrigated experiments; one at Lincoln in 2009–2010 (Ln NE, 09–10) at geographic position 40°50˝ N, 96°39˝ W, 357 m asl, and the other at Mead in 2010 (Md NE, 10) at geographic position 41°8˝ N, 96°29˝ W, 370 m asl, plus two sprinkler-irrigated experiments; one at Lincoln in 1999–2005 (Ln NE, 99–05) at geographic position 40°49˝ N, –96°24˝ W, 357 m asl, and the other in Lincoln from 2004 to 2005 (Ln NE, 04–05) at geographic position 40°51˝ N, 96°45˝ W, 357 m asl. The observed dates of occurrence of the Vn and Rn stage in the first two experiments (2009–2010) were collected by the senior author of this paper and in the other two experiments by Setiyono et al. (2007)

and Bastidas et al. (2008), respectively. The research plots at nonNebraska sites included an irrigated planting date experiment conducted by De Bruin and Pedersen (2008) at Whiting, IA in 2004–2006 (Wh IA, 04–06) at geographic position 42°08˝ N, 96°09˝ W, 323 m asl, and a rainfed planting date experiment conducted by Robinson et al. (2009) at West Lafayette, IN in 2006–2007 (WL, IN 06–07) at geographic position 41°29˝ N, 86°59˝ W, 190 m asl. At all research sites, best management practices were used to ensure that nutrients, weeds, insects or disease did not impact crop growth and development. The cultivars in the research test sites encompassed a MG range of 0.8 to 4.2. Soils at each research or producer site were typical of the soils (largely silt loam to silty clay loam) on which soybean is produced in each State. Soybean was always planted after a previous corn (Zea mays L.) crop, except for one Nebraska producer who used a continuous soybean cropping system. At the research experiment sites, fields were fall-plowed after maize harvest and then field-cultivated twice in spring each year. However, all Nebraska producers practiced either no-till or minimum-till after the prior crop. Planting dates in the research fields ranged from 27 April to 17 June, and from 21 April to 11 June in producer fields. At each site, soybean phenological stages were evaluated using a system developed and described by Fehr et al. (1971). Phenological stages in research experiments were scored twice each week per cultivar and replicate. Five of the 21 or 22 producer fields that were close to Lincoln, NE were stagescored weekly by the senior author, whereas the more distant producer fields were stage-scored in mid-June (mid-vegetative phase), early July (early reproductive phase), and mid-August

Agronomy Journal  •  Volume 103, Issue 6  •  2011

1663

(late seed-filling). All Vn or Rn scoring was accomplished by the evaluation of 10 contiguous plants growing in two replicate rows that on each visit were randomly selected from a larger pre-identified uniform area in each producer field. A different group of plants were scored at each scoring date. The soybean Vn and Rn stages in the research experiments were simulated using the SoySim crop model with the following inputs: maximum and minimum air temperatures, and latitude to estimate daylengths obtained from the High Plains Regional Climate Center weather database (http://www.hprcc. unl.edu) for the Nebraska and Iowa sites, and from the Indiana State Climate Office database (http://iclimate.org/index.asp) for the Indiana site. Cultivar MG and VE dates were the only other crop-related inputs required in SoySim (Setiyono et al., 2010). However, for some producer fields where VE dates were not provided or not known, the planting date had to be used as an alternative input in the SoySim model simulation. The calendar dates of soybean R7 in 2009 were mostly reported by the collaborating producers, after the senior author provided them with detailed instructions and training, for example, photos, sketches from Fehr and Caviness (1977). In 2010, the date of R7 occurrence was determined mostly by the senior author based on timed visits to nearby soybean fields, or based on cues provided to the senior author in telephone calls with the producers (e.g., yellowing of leaves, yellowing of few or one pods). Simulated and observed phenological stage values were compared based on their respective occurrence, that is, days after emergence (DAE). Emergence was considered to have occurred on the calendar date when 50% of the hypocotyls of the seedlings had pulled cotyledons just free of the soil surface (Fehr et al., 1971). The degree of congruence between the simulated and observed pairs of DAE values for the given soybean phenological stages was evaluated by the RMSE parameter as described by Janssen and Heuberger (1995), and calculated as:

= RMSE

∑ (s − o ) i

i

2

n

where s is the simulated day of occurrence and o is the observed day of occurrence of ith V or R stage, and n is the number of simulated and observed data pairs for those stages. The smaller the RMSE, the greater the degree of congruence between the simulated and observed data pairs. Means for the simulated DAE and observed DAE for a given stage were also computed and subjected to a paired t test to assess the significance of the simulated-observed difference in DAE. The degree of bias from congruence was estimated with linear regression of simulated DAE on observed DAE using the equation, y = bx + a, where b is the regression coefficient (i.e., slope) and a is the regression intercept. With no bias, the regression coefficient should equal to unity and intercept should be zero. This null hypothesis was evaluated by a t test of the observed b value with b = 1.0, and was conducted on specific clusters of simulated vs. observed data pairs for each Vn and Rn. RESULTS AND DISCUSSION In the research experiments conducted in six site-year combinations in the North-Central United States (Fig. 1), the dates of the observed phenological stages were closely correlated with the dates of the corresponding stages simulated by SoySim (Fig. 2). The RMSE for this comparison of n = 497 data pairs 1664

was 3.6 d. The RMSE value for some individual stage data pairs was larger, that is, 5.3 d for R5 and 4.2 d for R3.5 (mid-pod stage). In contrast, the R1 and R7 stages had near-identical respective RMSE values of 3.2 and 3.1 d, whereas the V1 (firstnode stage) had the smallest RMSE (i.e., 1.7 d). Stage V1 was the only in-common vegetative stage reported in the six site-year combinations in research fields. The mean values for simulated and observed DAE to V1 dates were nearly identical, that is, 12.4 and 11.9 d, respectively (see coordinate positioned cross-hair shown in Fig. 2). The regression of the simulated V1 DAE on the observed V1 DAE revealed a slope that was significantly less than unity (i.e., 0.78) and an intercept greater than zero (i.e., 3.15), suggesting a tendency for the simulated DAE to V1 to be inversely earlier or later than the corresponding observed DAE when the latter deviated from its mean value. Still, the paired data were tightly clustered as the low RMSE value of 1.7 d indicates. The means for the simulated and observed DAE to R7 (Fig. 2) also were nearly identical (i.e., 112.1 and 113.3). The near-unity regression coefficient, near-zero intercept, and the high coefficient of determination (r 2 = 0.94), indicated that the simulated and observed R7 DAE data pairs were highly congruent over a wide range of R7 DAE values. The simulated and observed DAE to R1 and to R3.5 were not quite as congruent as those for R7. Still, the DAE means, regression coefficients, and intercepts for those two stages did not differ significantly from their null hypothesis expected values. The simulated and observed DAE to R5 were not as congruent as those of the other R stages (Fig. 2). The mean values to simulated and observed DAE to R5 did not fall on the 1:1 line (i.e., simulated DAE was earlier than observed DAE). Moreover, the regression coefficient was significantly less than unity (i.e., 0.80) and the intercept was much greater than zero (i.e., 12.94)—see regression line in black on R5 stage data clusters in Fig. 2. This bias might be due to the difficulty in accurately quantifying, via visual assessment, the size of a 3-mm seed, which when present in at least one cavity of a pod at one of the four uppermost stem nodes constitutes the definition of stage R5 as described by Fehr et al. (1971). Moreover, observers using an English (rather than metric) ruler would likely use the nearest English unit of 1/8 inch, which is actually 3.2 mm, which could result in a slightly biased later call of R5. In general, the R1 and R7 stages were easy for observers to call, because the presence of just one flower on a plant defines R1 for that plant, and because just one mature brown pod on a plant defines R7 for that plant. These “yes” or “no” plant calls are easily summed and the Rn stage is easily computed as an average of (say) two 10-plant replications (e.g., if 18 of 20 plants are R7, the call would be R6.9). On the other hand, calling the dates of R3.5 and R5 requires the observer to make a quantitative assessment of pod length or seed diameter, respectively. The potential intrinsic error in the more subjective R3.5 and R5 calls is notable in terms of the degree to which the mean DAE for the paired simulated and observed stages did not fall in the 1:1 line. An identical deviation of 1.8 d was detected for both R3.5 and R5. The mean DAE for simulated and observed R1 did not deviate from the 1:1 line (i.e., 0.0 d), whereas R7 deviated by only 1.2 d from 1:1 line, implying minimal intrinsic errors for the R1 Agronomy Journal  •  Volume 103, Issue 6  •  2011

Fig. 2. Days after emergence (DAE) to the occurrence of the observed (x axis) and simulated (y axis) soybean phenological stages of V1, R1, R3.5, R5, and R7 for the various cultivars (MG 0.8–4.2) and varied April, May, or June planting dates used in the research fields at six site-year combinations in three North-Central states Nebraska, Iowa, and Indiana (see symbol legend). The linear regression lines, along with equations for the associated slope and intercept parameters (an asterisk denotes statistical significance at α = 0.05), plus coefficient of determination, are shown for each soybean phenological stage. The cross-hairs represent the means of each (stage) clusters of the coordinate data pairs. The root mean square error (RMSE) for the entire set of simulated vs. observed data pairs (n = 497) was 3.6 d.

and R7 stages (i.e., again the yes or no calls) compared with the R3.5 and R5 stages (see the cross-hair positions in Fig. 2). In the 2009 and 2010 producer fields, six vegetative stages were targeted for comparison of the simulated DAE and observed DAE. In all six cases, mean value for the simulated and observed DAE pairs Vn fell below the 1:1 line, indicating that, on average, the stages had a simulated date of occurrence of 1.2 to 3.2 d earlier than the corresponding observed date of occurrence (Fig. 3a, cross-hairs). Bastidas et al. (2008) reported that from the V1 stage to the Vn stage that was coincident with the R5 stage, a new node appeared every 3.7 d. This phyllochron was also observed at the Lincoln, NE (Ln NE 09–10) research site. The V1 to Vn-at-R5 phyllochron was not directly evaluated in producer soybean fields. However, the length in days between V3 and V4, and between V7 and to V8, were a respective 3.9 and 4.0 d (Fig. 3a), which were close to the 3.7 d reported by Bastidas et al. (2008). For the collective set of simulated and observed data pairs for the six Vn stages (n = 134), the RMSE was 5.5 d for the producer fields (Fig. 3a), which was larger than the RMSE (i.e., 3.2–3.6 d) reported by Setiyono et al. (2007) in the SoySim calibration experiment. In producer soybean fields, emergence dates were not reported for 16% of site-year combinations, so planting date had to be used as a replacement input in SoySim phenology simulation. Bastidas et al. (2008) noted that because of the sensitivity of soybean germination and pre-emerged seedling development to soil temperature, dates of emergence were more reliable indicators of the subsequent dates of V1 occurrence than

Fig. 3. Days after emergence (DAE) to the occurrence of the observed (x axis) and simulated (y axis) soybean phenological stages of V3, V4, V7, V8, V11, and V15 (panel a), and stages of R1, R3, R5, and R7 (panel b) for the various cultivars and varied April, May, or June planting dates used in the irrigated or rainfed experiments conducted in 21 or 22 NE producer fields in 2009 (circles) and 2010 (squares), respectively. The soybean fields evaluated each year involved cultivars of varying maturity (MG 2.5– 3.8) planted into no-till or minimum-till seedbeds. The linear regression lines, along with the equations for the associated slope and intercept parameters (an asterisk denotes statistical significance at α = 0.05), plus coefficient of determination are shown for each soybean phenological stage. The cross-hairs represent the means of each (stage) clusters of the coordinate pairs. The root mean square error (RMSE) for the entire set of simulated vs. observed Vn stage data pairs (n = 134) was 5.5 d, and for Rn stage data pairs (n = 115) was 5.7 d.

were dates of planting, relative to the constant Vn phyllochron of 3.7 d beginning V1 to Vn coincident with stage R5. The SoySim model projection of emergence date may be too early when planting date is the input, because extensive residue amounts in

Agronomy Journal  •  Volume 103, Issue 6  •  2011

1665

no-till or minimum-till soybean production systems can slow soil warming in the spring (producer fields in this study were all no-till or minimum-till systems). It has been reported that the greater the amount of residue on the soil surface, the smaller the diurnal range in soil temperature (Potter, 2004), which can delay emergence via a lower accumulated growing degree days (GDD) for the germinating seed. Evidence of the delayed soybean emergence has been reported in northern United States in no-till systems relative to the conventional tillage planted at the same depths (i.e., 50 mm). The delayed emergence was due to the significantly lower soil temperature in no-till plots (i.e., 1.4–2.1°C) relative to the conventional tillage plots (Sharratt and Gesch, 2008). Thus, if SoySim is provided with planting dates for no-till soil conditions, its estimates of VE dates may be earlier than actual emergence dates, resulting in earlier than the expected dates for the subsequently projected Vn stages (as seen in Fig. 3a). Because emergence date is used by SoySim as the start date for SoySim projections of calendar dates for Vn stages, it appears that use of emergence date, rather than planting date, gives more accurate SoySim forecast of subsequent Vn stages. An anecdotal case of the foregoing scenario was observed in one of the Nebraska producer fields. In that 2010 field, the prior maize crop residues had been floated away from an area of the field by a substantive rainfall event just after planting. The plants in residue-free area had an earlier V1 occurrence, which resulted in a 1.5 node advantage over plants in an adjacent undisturbed maize residue portion of the same field. This advantage was presumed to have resulted from the warmer soil temperatures in the residue-free area, thereby resulting in more rapid germination, earlier emergence, and an earlier start on plant node generation (Bastidas et al., 2008). Moreover, the absence of surface crop residue also removed a physical barrier to soybean emergence. The SoySim model is not currently programmed to account for the cooler soil temperatures in no-till or minimumtill conditions that may slow down germination and emergence, because it uses only air (not soil) temperature as an input. In the 2009 and 2010 producer fields, the DAE means for the simulated and observed Rn stage clusters (cross-hairs in Fig. 3b) also fell below the 1:1 line except for the means at R5. However, respective stage DAE means for R1, R3, R5, and R7 were not significantly different from their null hypothesis expected values. Like the Vn stage data calls (Fig. 3a), SoySim was generating Rn call dates that were somewhat earlier than the observed Rn call dates (Fig. 3b). In the case of stages R1, R3, R5, and R7, the SoySim calls were a respective 2.1, 3.7, 0.7, and 4.6 d earlier than the observed calls. At all four Rn stages, regression coefficients were respectively less than unity, though only those for R3 and R7 were significantly less. The collective pairs of simulated and observed Rn DAE values (n = 115) had a RMSE of 5.7 d (Fig. 3b), which was just 0.20 d larger than that of the Vn DAE data pairs (Fig. 3a). As noted previously with the delayed emergence in no-till to minimum-till, inputting a VE date (instead of planting date, if VE had been observed) into SoySim would likely have generated a better simulation fit to the observed soybean Vn and Rn stages in the 2009–2010 producer fields. Still, from a crop scouting standpoint, the projection of erroneous “premature” calendar dates for the Rn stages would be more acceptable than the projection of “tardy” ones, except for the R7 stage. 1666

There is limited literature relative to scientific evaluation reporting the congruence between a crop model simulated growth and development stages with those actually observed in producer fields. Mercau et al. (2007) reported that the comparison of CROPGRO–Soybean model simulated and observed durations of the three developmental phases of R1-R5, R5-R7, and R1-R7, had RMSE values of a respective 9, 10, and 10 d in soybean fields located in the Argentine Pampas. Another comparison involving OilCrop-Sun model simulated and observed yes or no calls of sunflower (Helianthus annuus L.) anthesis date in Argentina producer fields, had a RMSE of 5 d (Grassini et al., 2009). These reported values would suggest that in the Nebraska producer fields, RMSE values of 5.5 and 5.7 d for Vn and for Rn stage calls, respectively, were not unusually high nor low. The lower RMSE in the research fields of this study (3.6 d) was assumed to be related to researchers making more reliable VE calls in the fall-plowed field site-year combinations compared to the VE calls in the producer no-till or minimum-till soybean fields, which enabled SoySim to more reliably predict soybean growth stage and development in the former. Implementing management practices based on soybean phenological stage is difficult to achieve in practice because some soybean reproductive stages are notoriously difficult to call. This is particularly true for stage R3, R4, R5, and R6, which are not easily quantifiable calls, primarily because one must first evaluate pod length or the degree to which a pod cavity is filled by a seed, then make that call after examining all pods within the uppermost four nodes of at least 10 to 20 plants as recommended by Fehr and Caviness (1977). The availability of a user-friendly crop model such as SoySim (Setiyono et al., 2010) would help producers or consultants to not only track and project the crop stages of each field, but also project the dates of occurrence of those stages, thereby facilitating the adoption of just-in-time phenology based farming decisions. Producers and consultants in the North-Central United States soybean-growing region can use the SoySim model itself (Setiyono et al., 2010) available at http:// www.soysim.unl.edu/or the website known as UNL SoyWater (Specht et al., 2010), which is available at http://www.hprcc3. unl.edu/soywater/, to predict the calendar dates of the Vn and Rn phenological stages in their soybean fields. SUMMARY Simulated and observed dates of soybean phenological stages at experimental research fields in the North-Central United States were reasonably congruent, judging by the low RMSE. This was likely attributable to the more reliable calls of the date of emergence in the research fields. Use of emergence date as input to SoySim model simulations gives more reliable projections of the calendar dates of Vn and Rn occurrence. Higher RMSE for respective Vn and Rn stages at soybean producer fields were likely the result of subjective error in calling the date of emergence, or using planting date as a substitute input to SoySim when the emergence date was not reported. The simulated dates of Vn or Rn in the producer fields were somewhat earlier than the observed dates, perhaps because of cooler spring soil temperatures in the no-till or minimum-till fields that slowed down seedling development (pre- and postemergence). At present, SoySim is not programmed for input of soil temperatures. It also appears that some of the differential between the simulated Agronomy Journal  •  Volume 103, Issue 6  •  2011

and observed Rn dates was likely attributable to the difficulty of quantifying the metrics used to call the R3 to R7 stages, such as: pod length, seed size, and pod maturity. In any event, the results of this study indicate that SoySim projections of the calendar dates of Vn and Rn stages were reasonably reliable when RMSE value are compared with those reported in the literature. However, producers and consultants using this tool for crop-stage-specific farming decisions would be advised to determine as accurately as possible the soybean emergence date for use as input in the SoySim model. ACKNOWLEDGMENTS We thank the Water, Energy, and Agriculture Initiative (WEAI) at the University of Nebraska in joint funding with Nebraska Soybean Board, Nebraska Corn Board, Nebraska Public Power District, and University of Nebraska Agricultural Research Division. We are also grateful for the weather data support by the High Plains Regional Climate Center and the Nebraska soybean producers for their permission of use to their respective soybean fields and for providing us with the data input in the SoySim model. Thanks to Andrew Robinson for facilitating the use of IN site weather data. Last but not the least, much gratitude is extended to Aaron Hoagland for his critical assistance during field visits. REFERENCES Acker, R.C.V., C.J. Swanton, and S.F. Weise. 1993. [Glycine max (L.) Merr.] The critical period of weed control in soybean. Weed Sci. 41:194–200. Backman, P.A., R. Rodriguez-Kabana, J.M. Hammond, and D.L. Thurlow. 1979. Cultivar, environment, and fungicide effects on foliar disease losses in soybeans. Phytopathology 69:562–564. doi:10.1094/Phyto-69-562 Bastidas, A.M., T.D. Setiyono, A. Dobermann, K.G. Cassman, R.W. Elmore, G.L. Graef, and J.E. Specht. 2008. Soybean sowing date: The vegetative, reproductive, and agronomic impacts. Crop Sci. 48:727–740. doi:10.2135/cropsci2006.05.0292 De Bruin, J.L., and P. Pedersen. 2008. Soybean seed yield response to planting date and seeding rate in the Upper Midwest. Agron. J. 100:696–703. doi:10.2134/agronj2007.0115 Esker, P., E. Cullen, A. MacGuidwin, N. Koval, and J. Gaska. 2010. Soybean plant health topics. Available at http://www.plantpath.wisc.edu/soyhealth/ (accessed 20 June 2010; verified 10 Aug. 2011). Dep. of Agronomy, Entomology, and Plant Pathology, Univ. of Wisconsin, Madison. Fehr, W.R., and C.E. Caviness. 1977. Stages of soybean development. Spec. Rep. 80. Iowa Agric. Home Econ. Exp. Stn. Iowa State Univ., Ames. Fehr, W.R., C.E. Caviness, D.T. Burmood, and J.S. Pennington. 1971. Stage of development descriptions for soybeans, Glycine max (L.) Merrill. Crop Sci. 11:929–931. Grassini, P., A.J. Hall, and J.L. Mercau. 2009. Benchmarking sunflower water productivity in semiarid environments. Field Crops Res. 110:251–262. doi:10.1016/j.fcr.2008.09.006 Gustafson, T.C., S.Z. Knezevic, T.E. Hunt, and J.L. Lindquist. 2006. Earlyseason insect defoliation influences the critical time for weed removal in soybean. Weed Sci. 54:509–515. doi:10.1614/WS-05-071R.1 Janssen, P.H.M., and P.S.C. Heuberger. 1995. Calibration of process-oriented models. Ecol. Modell. 83:55–66. doi:10.1016/0304-3800(95)00084-9 Krupke, C.H., L.W. Bledsoe, and J.L. Obermeyer. 2009. Soybean insect control recommendations. Ext. Rep. E-77-W. Available at http://extension.

entm.purdue.edu/publications/E-77.pdf (accessed 20 June 2010; verified 10 Aug. 2011). Purdue Univ. Coop. Ext. Serv., West Lafayette, IN. Mercau, J.L., J.L. Dardanelli, D.J. Collino, J.M. Andriani, A. Irigoyen, and E.H. Satorre. 2007. Predicting on-farm soybean yields in the pampas using cropgro-soybean. Field Crops Res. 100:200–209. doi:10.1016/j. fcr.2006.07.006 Mueller, A.T., R.M. Miles, W. Morel, J.J. Marois, L.D. Wright, C.R. Kemerait, C. Levy, and L.G. Hartman. 2009. Effect of fungicide and timing of application on soybean rust severity and yield. Am. Phytopathol. Soc., St. Paul, MN. Mueller, D., and A. Robertson. 2010. Summary: Foliar fungicide on soybean in Iowa (2006–2009). Available at http://www.extension.iastate.edu/Crop News/2010/0707muellerandrobertson.htm (accessed 1 Nov 2010; verified 10 Aug. 2011) Iowa State Univ. Ext., Ames. Potter, B. 2004. SW Minnesota IPM STUFF 2004–3. SW Res. and Outreach Center. Available at http://swroc.cfans.umn.edu/prod/groups/ cfans/@pub/@cfans/@swroc/documents/asset/cfans_asset_247321. pdf (accessed 5 Nov 2010; verified 10 Aug. 2011) Univ. of Minnesota Ext. Serv., Lamberton. Purcell, L.C., and J.E. Specht. 2004. Physiological traits for ameliorating drought stress. p. 569–620. In H.R. Boerma and J.E. Specht (ed.) Soybeans: Improvement, production and uses. ASA, CSSA, and SSSA, Madison, WI. Ren, Q., T.W. Pfeiffer, and S.A. Ghabrial. 1997. Soybean mosaic virus incidence level and infection time: Interaction effects on soybean. Crop Sci. 37:1706–1711. doi:10.2135/cropsci1997.0011183X003700060005x Robinson, A.P., S.P. Conley, J.J. Volenec, and J.B. Santini. 2009. Analysis of high yielding, early-planted soybean in Indiana. Agron. J. 101:131–139. doi:10.2134/agronj2008.0014x Scott, W.O., and S.R. Aldrich. 1970. Modern soybean production. S and A Publ., Champaign, IL. Setiyono, T.D., K.G. Cassman, J.E. Specht, A. Dobermann, A. Weiss, H.S. Yang, S.P. Conley, A.P. Robinson, P. Pedersen, and J.L. Bruin. 2010. Simulation of soybean growth and yield in near-optimal growth conditions. Field Crops Res. 119:161–174. doi:10.1016/j.fcr.2010.07.007 Setiyono, T.D., A. Weiss, J. Specht, A.M. Bastidas, K.G. Cassman, and A. Dobermann. 2007. Understanding and modeling the effect of temperature and daylength on soybean phenology under high-yield conditions. Field Crops Res. 100:257–271. doi:10.1016/j.fcr.2006.07.011 Sharratt, B.S., and R.W. Gesch. 2008. Emergence of polymer-coated corn and soybean influenced by tillage and sowing date. Agron. J. 100:585–590. doi:10.2134/agronj2007.0158 Specht, J.E., R.W. Elmore, D.E. Eisenhauer, and N.W. Klocke. 1989. Growth stage scheduling criteria for sprinkler-irrigated soybeans. Irrig. Sci. 10:99–111. doi:10.1007/BF00265687 Specht, J.E., J.A. Torrion, T.D. Setiyono, K.G. Cassman, I. Suat, K. Hubbard, M. Schulski, J. Li, and W. Sorensen. 2010. Station-specific weather & crop water use: A decision-aid webpage to schedule irrigation in soybeans. Available at http://www.hprcc3.unl.edu/soywater/ (accessed 1 May 2010; verified 10 Aug. 2011) Agronomy and Horticulture, Univ. of Nebraska, Lincoln. Vangessel, M.J., A.O. Ayeni, and B.A. Majek. 2001. Glyphosate in double-crop no-till glyphosate-resistant soybean: Role of preplant applications and residual herbicides. Weed Technol. 15:703–713. doi:10.1614/0890-037X(2001)015[0703:GIDCNT]2.0.CO;2 Wright, J., D. Hicks, and S. Naeve. 2006. Predicting the last irrigation for corn and soybeans in Central Minnesota. Minnesota crop e-news. Available at http://www.extension.umn.edu/cropenews/2006/pdfs/06MNCN47. pdf (accessed 23 June 2010; verified 10 Aug. 2011). Ext. Serv. Univ. of Minnesota, St. Paul.

Agronomy Journal  •  Volume 103, Issue 6  •  2011

1667