search

APSIMInitiative / APSIM710

APSIM
https://www.apsim.info
29 stars 46 forks source link

Change to wheat phenology parameters #1473

Open ApsimBot opened 5 years ago

ApsimBot commented 5 years ago

Legacy Bug ID: 2044 Author: Dean Date: 2015-03-13 08:30:56 +0000 UTC

From ZviH:

Please treat this as a request to change the phenology parameter values for the 19 varieties in this study.

The document is a bit rough and ready as I rushed it to make the reference panel meeting deadline as I think it’s important to have these values in a current release for this year’s season. There are a number of reasons for this but the main one is that we don’t want to change mid-season for Yield Prophet users.

Cheers

Zvi

ApsimBot commented 5 years ago

Author: hol353 Date: 2015-03-13 08:32:00 +0000 UTC

Optimised APSIM Wheat Phenology parameters_20150312.pdf Size: 1417523

ApsimBot commented 5 years ago

Author: murphych Date: 2015-03-16 11:19:00 +0000 UTC

RP at its meeting of 13 March 2015, approved to review the Optimised Wheat paramenters. Scott Chapman was a suggested reviewer.

ApsimBot commented 5 years ago

Author: murphych Date: 2015-04-29 14:11:00 +0000 UTC

3.1_Scott Chapman_Wheat Parameters_APSIM Reviewers Report v1 3.doc Size: 1642496

ApsimBot commented 5 years ago

Author: murphych Date: 2015-04-30 15:34:00 +0000 UTC

Zheng and Chapman Check of Wheat model parameters April 2015.pdf Size: 1723012

ApsimBot commented 5 years ago

Author: hol353 Date: 2015-05-01 07:50:00 +0000 UTC

From Zvi H.

By necessity this is a rushed response and I don’t have time to address many of Scott’s comments so I will focus on what I think the essence of the problem in his critique.

I accept that the work will be strengthened when it is tested against an independent data set and we plan to do so over the next few months with data that we can trust. However, I reject most of Scott’s report because a) it is based on a data set that I am not in a position to critique and b) When we used the parameters in Zheng et al 2013 to predict our data (see attached) the simulations produced systematically worse predictions of observed data than those in APSIM v7.6 . So what we have is two sets of data that do not agree with each other, rather than two conflicting methods of parameterisation. For this reason it is unlikely that combining the data sets or cross validation will actually improve parameter estimates.

There are a number of reasons for preferring our data to that of Scott’s:

It contains 5442 observations vs 881

Observations are over a wide range of growth stages which avoids getting the right prediction of one stage at the expense of others

Observations cover the whole range of conditions (temperatures and daylegnth) likely to be encountered in commercial wheat crops in Australia

Observations are in the field under natural conditions and not with pre-vernalised seed or artificial lighting.

If, as I suspect, Scott is counting NVT observations among his 881 then I would reject those out of hand as would anyone who has seen these data.

Finally the proof of the improved phenology parameters is that using them has significantly improved APSIM Wheat’s accuracy in capturing observed yield differences in our Site by Variety by Time of Sowing experiments (see Appendix 8 attached): Using APSIM 7.6 the simulations accounted for 68% of the variation in observed yields with a residual error of 892 kg/ha. The slope of the regression was 0.904 (+/- 0.0296) and the intercept was 352 (+/- 89.2) kg/ha (Figure 1).

When the new phenology parameters were applied the simulation accounted for 71% of the variation in observed yield and the residual error was reduced to 852 kg/ha. The slope of the regression was unchanged at 0.90103 (+/- 0.027) but the intercept was reduced to 228 (+/- 87) kg/ha (Figure 2). This means that with better phenology estimates the model is driving predictions rather than the intercept.

These improvements might seem to be modest but they are significant and I have not seen better against such a comprehensive test.

Time is of the essence as we will not introduce a new version of APSIM past the middle of May into Yield Prophet. It would be a pity to deny these (perhaps modest) improvements to Yield Prophet and other APSIM users for another year.

Zvi

ApsimBot commented 5 years ago

Author: hol353 Date: 2016-03-07 10:14:00 +0000 UTC

NVT APSIM 76 and New phenology grain yield comparison by cultivar.docx Size: 16485