Table Of ContentsNext Page

Evaluation of simtag and nwheat in Simulating Wheat Phenology in Southeastern Australia

M. Stapper and J.M. Lilley

CSIRO Plant Industry, Canberra, ACT.

Abstract

Many wheat phenology models have been developed but they are usually only applicable for target region and conditions. Thorough validation of phenology models using data sets of varieties grown across a wide range of environments is rare. This study presents an evaluation of SIMTAG and APSIM's NWHEAT models using data from 15 genotypes across 42 sowing dates at 12 locations. SIMTAG was the more accurate predictor of phenology with root mean square error (RMSE) for anthesis date between 3.2 and 11.5 days with an average of 4.8. NWHEAT (based on CERES-Wheat) had a RMSE for anthesis between 7.6 and 18.4 days, and could not be made to better fit observations over the wide range of data for either spring or (semi-) winter wheats, as it only has two phenology input parameters. Differences in simulation between the wide and standard sowing date ranges are shown. Simulation models need systematic testing to cover intended applications.

Key words

Crop models, crop development, SIMTAG, APSIM-Wheat, CERES-Wheat.

Introduction

Genotypic differences in maturity are important in the adaptation of crops to climate. Differences of a few days in flowering can have a large effect on yield of dryland crops within and between sowing dates. Crop simulation models have improved steadily over the past three decades aided by the rapid improvement of computer technology. This has lead, for example, to the development of APSIM that is a modelling framework enabling simulation of cropping systems (5). Another integrated package is DSSAT developed by the International Benchmark Sites Network for Agrotechnology Transfer that contains, amongst others, the CERES models (15). Crop models thus enable the evaluation of the effects of different environmental and management factors on crop performance both independently and as they interact together. This enables the dynamic behaviour of cropping systems over time to be both described and studied theoretically.

Properly validated models provide a method for interpolation, extrapolation and prediction of crop results under conditions that were not measured or could not be measured in the real system. Many wheat phenology models have been developed, but they have rarely been validated for a genotype over a wide range of sowing dates and across environments. They are, therefore, usually only applicable for standard sowing dates in the region for which they were developed. Some models have been successfully applied across environments, but usually with different varieties for different regions, for example CERES-Wheat (7). Meinke (6), working with APSIM, identified the need to improve wheat phenology modelling and showed the impact of erroneous phenology simulation on overall model performance. However, a numeric validation was not given. Accuracy of wheat phenology simulation has been reported to vary (e.g. 3, 8). Reported differences between observed and calculated stages of development in those studies had root mean square errors (RMSE) of 3.3 to 22.3 days (see models). No studies were found with a comprehensive evaluation of genotypes across sowing dates and agro-ecological zones because the required phenology data sets for genotypes are scarce. Calibration of new varieties in models, therefore, often only mimics the developmental responses for local conditions and standard sowing dates as it substitutes environmental effects as genetic effects through numeric variety input parameters. This study evaluates the phenology simulation of two different modelling approaches with a comprehensive data set for southeastern Australia using SIMTAG (12) and CERES-Wheat (10)/NWHEAT (2).

Field data

This study used data sets from Ginninderra (ACT) and the following locations in New South Wales: Murrumbateman, Wagga Wagga, Stockinbingal, sites around Temora, Moombooldool, Griffith and Goolgowi from east to west in the south (~350 km, ca. 35 oS, altitude 570 to 120 m.), Condobolin in the centre and Narrabri (30 oS) in the north. Crops sown on the same date in locations along the southern transect develop under similar photoperiod but different temperatures resulting in a one month spread in anthesis date. Long term minimum/maximum temperatures for the coldest month July are -0.2/11.1 and 2.6/14.4 oC for the extreme locations Ginninderra and Griffith, respectively, with difference in average daily temperature of about 3 oC over the season. Crop observations were made using the Decimal Code (DC; 17). There were 42 sowing dates selected between late March and late September each with at least three genotypes. This resulted in 304 observed anthesis dates for 15 genotypes in 11 seasons, with 12 to 27 sowings per genotype. Data for other stages were less frequently observed, for example, first node stage in 63% and flag leaf stage in 72% of crops. The dates for floral initiation, double ridge or terminal spikelet stages were determined by microscopic inspection for some 40% of crops. Genotypes used were, from early to late, Sonalika, Yecora70, Kulin, Dollarbird, Mexipak, Vulcan, Hartog, Janz, Banks, Kite, Egret, Oxley, Osprey, Olympic and UQ189. The senior author made some 90% of the observations (Stapper, unpublished; 13, 14) with the remainder being unpublished data supplied by others.

Models

Maturity differences in crops like wheat are simulated with genetic input parameters representing the calibrated response to temperature, photoperiod and, for some genotypes, vernalisation in the timing of one or more visible (eg. anthesis) and invisible (eg. terminal spikelet) phenological stages. The modelling approaches evaluated here are part of the wheat models SIMTAG (12) and CERES-Wheat (10). APSIM-NWHEAT (2, 9) was developed from CERES-Wheat, and NWHEAT was used here to evaluate the widely adopted CERES-Wheat phenology as the crop development subroutine was not changed (B.A. Keating, pers. comm.). Both models use a daily time step with calculated daylength (including civil twilight) and measured maximum and minimum temperatures as inputs. Stages prior to grain filling are evaluated, as these are not influenced by the growing status of crops in these models, apart from emergence that was covered by provision of sufficient moisture at sowing.

simtag - The model uses an exponential function of photoperiod above a critical value and a photoperiod sensitivity parameter as proposed by Angus et al. (1). This results in a photoperiod factor (PF) with values between 0 and 1. Temperature influences development using a thermal-time concept with stage-dependent base temperatures and a maximum effective mean temperature for development of 20 oC (upper limit). Base temperatures are 2.5 oC for germination to emergence and 3 oC for emergence to anthesis. Daily thermal time (DTT) is calculated by subtracting base temperature from mean temperature. DTT is based on 3-hour steps between maximum and minimum temperatures when the minimum temperature is below the base temperature. DTT is zero if the maximum temperature is lower than the base. DTT is multiplied with PF resulting in the daily photothermal time (PTT). Treatment of vernalisation in SIMTAG was similar to that in CERES-Wheat leading to a vernalisation factor (VF). The minimum value of PF and VF multiplied with DTT gives the daily PTT. Local genotypes, however, could not be calibrated successfully and the calculation of vernalisation was changed towards the method used in AFRCWheat (16) but simulations were never good. Vernalising temperatures in Australia appeared higher than in Europe/North America.

The model calculates phyllochron (leaf appearance interval), floral initiation (which determines leaf number) and anthesis independently, with double ridge, terminal spikelet, first node, flag leaf (emergence of last leaf) and spike emergence as stages dependent on floral initiation. Stages of development occur, with the exception of the flag leaf stage, when the required cumulative PTT for that stage is reached. Development of crops emerging when daylength is decreasing is slowed by a factor depending on temperature and latitude. Critical photoperiod, photoperiod sensitivity, vernalisation base, vernalisation sensitivity and values of PTT (representing intrinsic earliness) from emergence to both floral initiation and anthesis are six input parameters for each genotype. The two vernalisation parameters can be set to eliminate vernalisation as a factor in development.

SIMTAG had a RMSE of 3.3 days (n=33) for anthesis and 5.6 for terminal spikelet in the mediterranean climate of Syria where it was developed (12). A RMSE of 3.8 days (n=25) with a bias (=computed-observed) of –1 day was determined for anthesis in a validation with local data for the application in southern New South Wales of Fischer et al. (4).

nwheat - Photoperiod influences development in NWHEAT (=CERES-Wheat) through a power equation determining PF based on the difference between 20 hours and current daylength with photoperiod sensitivity as a variety specific input parameter (cvpar2). DTT is calculated from maximum and minimum temperatures using a base temperature of 0 oC for the whole season and no upper limit. Plant crown temperature is derived from screen temperatures and used to calculate DTT when maximum or minimum is below 0 oC. Vernalisation days (VD) are derived from temperature when the minimum temperature is less than 15 oC and is expressed as a fraction of the day and accumulated as CUMVD. VF is calculated from CUMVD and the vernalisation sensitivity input parameter (cvpar1). The minimum value of PF and VF is then multiplied with DTT to determine the daily PTT.

In this model photoperiod affects development only during the stage from emergence to terminal spikelet (DC30) which is set for all genotypes as a PTT of 400 oCd. Leaf appearance is based on the phyllochron (leaf appearance interval) supplied as an input parameter which is commonly taken as 95 oCd (15). It takes three phyllochrons to reach the flag leaf stage after DC30 is attained, and from flag leaf a further two phyllochrons to reach anthesis. This approach makes anthesis dependent on the accurate calculation of previous stages. Photoperiod and vernalisation sensitivities are the two input parameters affecting phenological development in this model.

Asseng et al. (3) reported a RMSE of 4 days for the calibration of NWHEAT with local crop observations from Western Australia. Porter et al. (8) compared CERES-Wheat with other models for two varieties grown in New Zealand. The RMSE’s were 8.2 and 22.3 days for anthesis, and 9.9 and 14.1 days for terminal spikelet. This compared with RMSE of 4.5 and 6.5 days for anthesis using model AFRCWheat.

Results and discussion

A selection of predicted development is presented in Table 1. The RMSE in anthesis simulation with SIMTAG for the 15 genotypes varied between 3.2 and 11.5 days for all data and between 2.0 and 5.7 days for only standard sowings. For half the data the absolute error with observed dates was within two days, probably equivalent to (good) field observation error. There were no significant differences between early and late sowing dates in bias. The mean date of emergence was 20 June and the bias of simulated anthesis dates for crops emerging before and after this date was –0.4 and 1.1 day, respectively. Late sowings, therefore, tend to be over-predicted. This could indicate that warmer temperatures are increasingly effective as daylength increases (ie. higher PF). However, this trend is not consistent across all genotypes as Vulcan and Dollarbird, for example, become increasingly earlier than Janz at late sowings. Therefore, another possibility is changing photoperiod sensitivities for different phases of pre-anthesis development as found by Slafer and Rawson (11). These maturity differences between genotypes are difficult to quantify as different genotype combinations were present across sowing dates and locations. Each genotype is calibrated separately and computed anthesis dates for genotypes in a given sowing date are not always representative of observed maturity differences. For example, the observed/calculated differences between Vulcan and Kulin were 4.0/6.0, 3.1/3.1, 5.0/3.3 and 7.0/4.7 days for different trials and 4.6/4.2 days on average, while those between Janz and Kulin were 6.1/6.9, 4.4/4.4, 5.3/4.8 and 8.0/7.0 days for different trials and 5.9/5.5 days on average. Another indication that Vulcan might have a changing photoperiod sensitivity over pre-anthesis phases.

Table 1: Evaluation of SIMTAG and NWHEAT in calculating for wheat genotypes the number of days from sowing to start of stem elongation (DC30), flag leaf stage and anthesis with root mean square error (RMSE) and bias (=computed-observed) for all sowing dates, and anthesis for only standard sowing dates for the genotype.

Table 1 shows that simulations with NWHEAT (=CERES-Wheat) were poor for all and slightly better for standard sowing dates. This is not surprising since the model uses only two input parameters for phenology and does not have one to indicate intrinsic earliness. Genotypes can be calibrated for a particular sowing period, but the resulting parameters may result in erroneous simulations for different sowing dates or agro-ecological zones, as shown in Table 1 by greater changes in anthesis for the two groups compared to SIMTAG. NWHEAT input parameters were used for the first three genotypes (2, 3). The table does not show the result of NWHEAT simulations for all 15 genotypes, because of the poor performance of those spring and winter types listed, and modelling analysis as described below. Hence other genotypes were not calibrated as this is a very slow process with NWHEAT. Osprey, the genotype most sensitive to vernalisation, had the best fit for all sowings in NWHEAT (Table 1). However, it was still 18 days too early for March sowing and vernalisation calibration could not further slow crop development for early sowing without delaying development of late sown crops too much. Vernalisation sensitivity parameter (cvpar1) is 0.5 for spring wheats (15), but is used in NWHEAT to fit all genotypes. For example, Kulin was calibrated by Asseng et al. (3) with a cvpar1 of 1.6 to fit available (standard) sowing dates. However, NWHEAT simulations for Kulin sowing dates here resulted in anthesis being 30 days too early for April sowing and 23 days too late for September sowing. This can not be fixed by a ‘vernalisation’ factor.

Following is an analysis of the NWHEAT photoperiod sensitivity parameter (cvpar2) with values incrementing over the recommended range from 1 to 5 for genotype UQ189 on 18 sowing dates from late March to late September with flowering between late September and early December. Cvpar1 was set at 0.5 as it is a true spring wheat that does not respond to seed vernalisation (1). Simulations show that UQ189 can not be described by a single value of cvpar2, as the best fit moves from a cvpar2 of 5 for early sowings to 3 for late sowings. A cvpar2 of 4.5 gave the best fit for all with a bias of 0.8 day, but SIMTAG was better (Table 1). Here flag leaf stage was predicted poorly by SIMTAG as 1.5 leaves less than observed (bias –12 d) were simulated for this photoperiod sensitive genotype. Only the vegetative period is sensitive to photoperiod in NWHEAT and this probably contributes to the errors since Slafer and Rawson (11) showed that the whole period is sensitive.

Conclusions

Phenology models need functions describing intrinsic earliness and delay in development with declining daylength such as used in SIMTAG in addition to those describing sensitivities to photoperiod and vernalisation, to enable accurate simulation over a wide range of sowing dates and locations. Wheat is sensitive to photoperiod during the whole pre-anthesis period rather than only the vegetative phase, but a changing sensitivity to photoperiod during pre-anthesis phases could be introduced to improve simulation for many genotypes. Anthesis needs to be calculated from emergence independently rather than by the cumulative stage-upon-stage (error-upon-error) approach used in CERES-Wheat/NWHEAT. Vernalisation simulation needs improvement. Simulation models need systematic testing during development and before applications by a third party, especially in integrated software packages such as APSIM and DSSAT. Detailed phenology data are required for a set of geno-types differing in maturity and sown at various dates in locations covering main agro-ecological zones.

Acknowledgments

We thank Dr J.F. Angus, Dr A.G. Condon, Dr F.W. Ellison, Dr R.A. Fischer and Dr A.F. van Herwaarden for use of unpublished crop observations.

References

1. Angus, J.F., Mackenzie, D.H., Morton, R. and Schafer, C.A. 1980. Field Crops Res., 4, 269-283.

2. APSRU, 1999. APSIM NWHEAT Module.
http://www.apsim-help.tag.csiro.au/APSIM_docs/modules/nwheat_science.pdf

3. Asseng, S., Keating, B.A., Fillery, I.R.P., Gregory, P.J., Bowden, J.W., Turner, N.C., Palta, J.A. and Abrecht, D.G. 1998. Field Crops Res., 7, 163-179.

4. Fischer, R.A., Armstrong, J.S. and Stapper, M. 1990. Agric. Systems, 33, 215-240.

5. McCown, R.L., Hammer, G.L., Hargreaves, J.N.G., Holtzworth, D.P. and Freebairn, D.M. 1996. Agric. Syst., 50, 255-271.

6. Meinke, H. 1996. Improving wheat simulation capabilities in Australia from a cropping systems perspective (PhD thesis, Wageningen Univ.)

7. Otter-Nacke, S., Godwin, D.C. and Ritchie, J.T. 1986. Testing and Validating the CERES-Wheat Model in Diverse Environments. AGRISTARS YM-15-00407.

8. Porter, J.R., Jamieson, P.D. and Wilson, D.R. 1993. Field Crops Res., 33, 131-157.

9. Probert, M.E., Keating, B.A., Thompson, J.P. and Parton, W.J. 1995. Aust. J. Exp. Agric., 35:941-950.

10. Ritchie, J.T., Godwin, D.C. and Otter, S. 1985. CERES-Wheat: A User-Oriented Wheat Yield Model. AGRISTARS YM-U3-04442-JSC-18892.

11. Slafer, G.A. and Rawson, H.M. 1994. Aust. J. Plant Physiol., 21, 393-426.

12. Stapper, M. 1984. SIMTAG: A Simulation Model of Wheat Genotypes. Model Documentation (Univ. of New England, Armidale).

13. Stapper, M. 2001. This Proceedings.

14. Stapper, M. and Fischer, R.A. 1990. Aust. J. Agric. Res., 41, 997-1019.

15. Tsuji, G.Y., Jones, J.W. and Balas, S. (eds) 1994. DSSAT v3 (Univ. of Hawaii, Honolulu, Hawaii).

16. Weir, A.H., Bragg, P.L., Porter, J.R. and Rayner, J.H. 1984. J. Agric. Sci., Camb., 102, 371-382.

17. Zadoks, J.S., Chang, T.T. and Konzak, C.F. 1974. Weed Res., 14, 415-421.

Top Of PageNext Page

...

Search Site