Prediction of coumestrol content in unirrigated lucerne crops using weather variables

Figures & data

Table 1. The timing of lucerne sample collection from stands at Lincoln and Springston.

Locale	Year	Cutting frequency	Sample date	Onset	End
Lincoln	2014	2, 4, 6, 8, 12 week intervals	Sampled at two-week intervals (six dates)	24 Mar. 2014	2 June 2014
Lincoln	2015	4, 6, 12 week intervals	Sampled at two-week intervals from 14 January (10 dates)	14 Jan. 2015	25 May 2015
Lincoln	2015–2016	8 weeks intervals	Sampled at five and eight weeks regrowth duration on eight dates	2 Nov. 2015	25 May 2016
Springston	2015	Grazed in March.	Sampled on three dates from January to March and two dates in May.	15 Jan. 2015	17 May 2015

Table 2. Damage severity score categories for fungal and insect damage (from James 1971; Harvey & Martin 1980).

Damage score	Score description
1	No symptoms
2	Trace: 1–4% of area affected
3	Light to moderate: 5–20% of area affected
4	Heavy: 21–50% of area affected
5	Severe; over 50% of area affected

Table 3. The refined list of variables used to predict coumestrol content of a lucerne crop, following removal of obsolete, collinear parameters.

Term	DF	F-statistic	P-value	VIF	R-Sq
Regression	9	106.49	<0.001		(0.767)
Days above 95% RH	1	89.36	<0.001	5.36	0.736
Sum of rainfall (mm)	1	24.42	<0.001	7.67	0.653
Development stage	1	13.49	<0.001	6.04	0.009
SMD (%)	1	10.99	0.001	6.53	0.115
Average temperature (°C)	1	8.75	0.003	7.16	0.057
Days in growing period	1	8.61	0.004	4.98	0.129
Average daily sunshine hours	1	2.84	0.093	3.17	0.308
Dry matter yield (t DM/ha)	1	1.40	0.238	2.15	0.025
Date of harvest	1	0.08	0.773	3.11	0.085
Error	279
Total	288

Note: The R² of the regression (0.766) is the adjusted R² and includes all terms. The individual R² values are based on the regression between a single predictor and coumestrol content. Variables are ranked by F-value.

Figure 1. The relationship $(R_{\mathrm{a}\mathrm{d}\mathrm{j}.}^2=0.735)$ between coumestrol content (mg/kg DM) of lucerne and days above 95% RH that the lucerne crop had been exposed to with 95% confidence interval (long-dashed lines) and prediction interval (short-dashed lines). Dotted line is the critical level (25 mg/kg DM) above which lucerne is reported to be a risk for ewe reproductive performance. Data from Lincoln, Canterbury, NZ in 2014 (◇), 2015 (○) and 2016 (×), and from Springston, Canterbury, NZ in 2015 (◆).

Table 4. The variables identified via best subsets regression as best relating to coumestrol content of lucerne.

Variables	Variables used as predictors
Days >95% RH	X		X	X	X	X	X	X
Sum of rainfall (mm)		X	X		X	X	X	X
Average daily sunshine hours				X				X
SMD (%)					X		X	X
Average temperature (°C)						X
Days in growing period							X
R-Sq (adj.)	73.5	65.2	75.4	73.8	75.8	75.5	75.9	75.8
R-Sq (pred.)	73	64.6	74.9	73.3	75.2	74.9	75.3	75.1
Mallows CP	41	143	18	38	14	18	13	15
Standard error of regression	23.9	27.4	23.0	23.7	22.8	22.9	22.7	22.8

Note: For clarity, terms that were not identified by best subsets regression are not included in this table.

Figure 2. Coumestrol content (mg/kg DM) against fungal damage score of lucerne at Lincoln, Canterbury, NZ in 2014 (◇), 2015 (○) and 2016 (×), and from Springston, Canterbury, NZ in 2015 (◆). The relationship (R² = 0.578) is described by the equation: y = 10^{0.40x + 0.11} with 95% confidence interval (long-dashed lines) and prediction interval (short-dashed lines). Dotted line is the critical level (25 mg/kg DM) above which lucerne is reported to be a risk for ewe reproductive performance.

Table 5. Equations for the prediction of coumestrol content of lucerne (mg/kg DM) from the days above 95% RH and/or the sum of rainfall (mm) during the growing period.

		SEcoefficient	$R_{\mathrm{a}\mathrm{d}\mathrm{j}.}^2$	$R_{\mathrm{p}\mathrm{r}\mathrm{e}\mathrm{d}.}^2$
Original dataset
Equation (1)	Y = (6.51 × RH) − 13.3	0.227; 2.35	0.735	0.730
Equation (2)	Y = (0.745 × SR) − 2.58	0.0327; 0.251	0.652	0.646
Equation (3)	Y = (4.80 × RH) + (0.252 × SR) − 13.8	0.398; 0.0489; 2.26	0.754	0.749
Original and independent datasets
Equation (4)	Y = (3.75 × RH) + (0.387 × SR) − 10.74	0.310; 0.0354; 2.22	0.716	0.712

Notes: Equations (1)–(3) use the original dataset, Equation (4) uses the original and independent datasets. Y is coumestrol content, RH is days above 95% RH, and SR is sum of rainfall. The standard errors of the coefficient (SEcoefficient) are for the predictor variable(s) and constant, respectively.

Figure 3. Reported (y) versus predicted (x) coumestrol content (mg/kg DM) of independent data using Equation (3) (Graph A; r = 0.813) and all data using Equation (4) (Graph B; $R_{\mathrm{a}\mathrm{d}\mathrm{j}.}^2=0.716$). The solid regression line in Graph A is described by the equation y = 1.44x + 9.27 and the dashed line is the identity line. For Graph B long-dashed lines are the 95% confidence interval and short-dashed lines are the prediction interval. Dotted line is the level (25 mg/kg DM) above which lucerne is a risk for ewe reproductive performance. Independent data in Graphs A and B are sourced from Logan, Utah, USA in 1961 (●); Brookings, South Dakota, USA in 1963 (▪); Lincoln, Canterbury, NZ in 1980 (▴) and 2016 (▾); and Oamaru, North Otago, NZ in 2015 (▪). Original data in Graph B sourced from Lincoln, Canterbury, NZ in 2014 (◇), 2015 (○) and 2016 (×), and from Springston, Canterbury, NZ in 2015 (◆).

Figure 4. A decision tree to estimate mean coumestrol content of lucerne based on rainfall (mm) and the days above 95% RH that the stand was exposed to during its regrowth period. The regrowth period was the interval between the day forage was removed and the time of sampling, which ranged from four to eight weeks later.

Figure 5. Graph A. Coumestrol content of lucerne plotted against the sum of rainfall (mm) during the regrowth period of plant samples. From left: horizontal lines are mean coumestrol contents for rainfall less than 61 mm, between 61 and 131 mm, and above 131 mm. Graph B. Coumestrol content of lucerne against days above 95% RH for samples with rainfall less than 61 mm during the regrowth period. From left: horizontal lines are mean coumestrol contents for fewer than 5.5 days over 95% RH and more than 5.5 days over 95% RH.

James WC. 1971. An illustrated series of assessment keys for plant diseases, their preparation and usage. Canadian Plant Disease Survey. 51(2):39–65.

 Google Scholar

Harvey IC, Martin RJ. 1980. Leaf spot diseases on lucerne cultivars. New Zealand Journal of Experimental Agriculture. 8(3–4):295–296. doi: 10.1080/03015521.1980.10426276

 Web of Science ®Google Scholar