You are on page 1of 12

Crop response to nitrogen fertilizer: The delta yield concept

R. G. Kachanoski
Department of Renewable Resources, University of Alberta, Edmonton, Alberta, Canada T6G 2E3 (e-mail: gary.kachanoski@ualberta.ca). Received 13 January 2009, accepted 2 July 2009.
Kachanoski, R.G. 2009. Crop response to nitrogen fertilizer: The delta yield concept. Can. J. Soil Sci. 89: 543554. Estimation of fertilizer N requirements of crops remains a challenge. Numerous field studies have been carried out to calibrate soil tests against yield response to applied fertilizer N. Synthesis and identification of common crop fertilizer N response across large data sets (years, sites) will allow maximum use of this past work and a framework for comparison of future work. The objective of this paper is to define macro-relationships between the economically optimum fertilizer N rate (EONR) and the yield increase at the EONR defined as the delta yield, DYec, for large data sets of 2nd- and 3rd-order estimates of fertilizer N response functions with both 0th and 1st-order rate relationships between fertilizer nitrogen use efficiency and applied fertilizer N. The derived macro-relationships are curvilinear, depend on the price ratio R0the ratio of the (price per kilogram of fertilizer N)/(price per kilogram grain), and are similar to measurements from data sets of corn fertilizer N response functions spanning decades ('20 yr) and representing areas in both the United States and Canada. The macro-relationships appear to be robust and therefore useful for quantifying (post-harvest analysis) soil fertility, crop fertilizer N requirement, and comparison/classification of N response functions. Key words: Response function, prediction, efciency, economic N rate, corn Kachanoski, R.G. 2009. Reaction des cultures aux engrais azotes : le principe du rendement delta. Can. J. Soil Sci. 89: 543 554. Estimer les besoins en engrais azotes dune culture demeure difficile. De nombreuses etudes sur le terrain ont ete entreprises en vue detalonner lanalyse du sol selon le rendement obtenu apres lapplication dengrais azotes. La synthese ` ` de vastes bases de donnees (annees, sites) et lidentification de la reaction commune des cultures a ces engrais permettront ` dexploiter au maximum les travaux anterieurs tout en offrant un point de comparaison pour les etudes a venir. Cet article ` a pour but de definir les macro-relations liens entre le taux dapplication economiquement optimal pour les engrais N (EONR) et la hausse de rendement obtenue a ce taux, soit le rendement delta ^Yec, pour les grands ensembles de donnees ` de 2e et 3e ordre estimant la reaction aux engrais N compte tenu dun lien dordre 0 ou de 1er ordre entre lutilisation efficace des engrais et la quantite dengrais appliquee. Les macro-relations relevees sont curvilineaires, dependent du ratio des prix R (rapport entre le prix dun kilo dengrais N et dun kilo de grain) et ressemblent aux mesures issues des jeux de donnees sur la reaction du ma s aux engrais N qui couvrent plusieurs decennies (plus de 20 ans), pour les E.-U. et le Canada. Ces macro-relations paraissent robustes et pourraient servir a quantifier (analyse post messianique) la fertilite du ` sol et les besoins dengrais N de la culture, ainsi qua comparer ou a classer les fonctions de reaction aux engrais N. ` ` Mots cles: Fonction de reaction, prevision, efcacite, taux dapplication economique des engrais N, ma s

Can. J. Soil. Sci. Downloaded from pubs.aic.ca by Depository Services Program on 01/10/12 For personal use only.

Accurate prediction of the economically optimum rate of fertilizer N (EONR) for major agricultural crops remains a challenge despite a large number (years, sites) of fertilizer response trials. Most methods for estimating N fertilizer requirements are based on an expected yield goal (or crop yield potential) with subsequent corrections for various N credits based on information about the soil (e.g. soil N test) and previous cropping system. A number of studies examining large data sets, however, have shown that yield (at the EONR) is not an accurate indicator of fertilizer N requirement (Vanotti and Bundy 1994; Kachanoski et al. 1996a; Mulvaney et al. 2005). This is unfortunate given the potential for site-specific fertilizer application and the development of accurate on-the-go combine yield monitors and geographic positioning systems for mapping yield. Given the significant effort and cost of obtaining crop response to applied N fertilizer data, methods of efficiently extracting, synthesizing and summarizing relevant in543

Abbreviations: Y, grain yield; EONR, economically optimum N rate; NUE, fertilizer N use efficiency; Y(N), estimated model for grain yield as a function of applied fertilizer N rate; S, a set of measured grain yield response functions (to applied fertilizer N); Nmax, fertilizer N rate giving maximum grain yield; Ymax, maximum grain yield from application of fertilizer N; Yec, yield at the economically optimum N rate; DYec, yield increase due to application of the economically optimum N rate; DYmax, maximum yield increase due to application of fertilizer N; R, the ratio of the (price per kilogram of fertilizer N)/(price per kilogram grain); f i(N), measured grain yield response to applied fertilizer N; F(n)(a), nth derivative of the i empirical (least-squares fit) fertilizer response curve at fertilizer rate N 0a; B, fitting parameter equal to the initial slope of the empirical fertilizer response function; C, fitting parameter and index of the curvature of the empirical fertilizer response function

544 CANADIAN JOURNAL OF SOIL SCIENCE

formation and insights from large data sets (of crop response to fertilizer N applications), especially relevant relationships between EONR and yield indices, are needed. Beauchamp et al. (1987) compiled data from approximately 250 fertilizer N response trials for corn from field experiments run across southern Ontario, Canada (from 1962 to 1992). An analysis of the data by Kachanoski et al. (1996a) indicated that EONR was poorly predicted by either maximum yield, Ymax, or yield at the EONR, Yec. Kachanoski et al. (1996a) also introduced the concept of delta yield, DYec, defined as the increase in grain yield (over zero fertilizer application) obtained from applying fertilizer at the EONR, and showed that it explained over 50% of the variability of EONR. The relationship of DYec and EONR was found to be highly non-linear. Subsequent studies have also concluded that DYec is a much better indicator of EONR than yield goal (Lory and Scharf 2003; Fixen 2006). Lory and Scharf (2003) evaluated the utility of DYec to predict EONR for corn using data from 298 fertilizer N response experiments across five USA states (Illinois, Minnesota, Missouri, Pennsylvania, Wisconsin). They concluded that farmers should be encouraged to monitor DYec as a more effective indicator of EONR than actual yield. They fit a linear relationship between DYec and EONR, but acknowledged that their data also indicate that the true model is likely nonlinear as originally proposed by Kachanoski et al. (1996a). They also concluded that a greater understanding of the delta yield concept is needed before relying on it to predict fertilizer N requirements. The relationship between yield response and applied N fertilizer, Y(N), is highly non-linear (Dahnke and Olsen 1990; Kachanoski and Fairchild 1996), and often follows a continuous diminishing return response, such that the rate of yield response to fertilizer application decreases progressively with an increase in the application rate (Colwell 1994; Schlegal et al. 2005). The relationship between DYec and EONR is not a fertilizer response function. Rather, it describes a macro-relationship that appears to hold (or not) across a set of Y(N), with each member Y(N) of the set representing an individual fertilizer response experiment. There is no independent theory dictating the model form for Y(N), and the choice of an empirical model for the Y(N) for a given data set can influence the inferred value of EONR (Colwell 1994). Thus, empirical macro-relationships between DYec and EONR for a set of Y(N) may not hold for other data sets, and likely depend on the form of the empirical model used to describe individual Y(N) in a set. The identification of the macro-relationship is similar (but not identical) to using functional normalization/ simultude, an empirical (least squares) approach to coalesce a set of experimentally measured relationships between two variables into a single curve that holds for the entire set (Tillotson and Nielsen 1984).

Can. J. Soil. Sci. Downloaded from pubs.aic.ca by Depository Services Program on 01/10/12 For personal use only.

The objective of the paper is to develop a more theoretical basis for the existence of well-defined macrorelationships between DYec and EONR, which hold for sets (across locations, years) of fertilizer N response functions. General equations to describe the macrorelationship between DYec and EONR are derived and compared for different necessary assumptions regarding the nature of the crop response Y(N) to applied fertilizer N (i.e., constraints necessary for the general equations to hold). Testing the veracity and generality of the necessary constraints is presented as a method of testing the generality and veracity of the macro-relationship between DYec and EONR for a given set of Y(N). Finally the general equations are compared with empirical relationships obtained from past studies. MATERIALS AND METHODS Theory and Governing Equations Let S be a set of crop fertilizer N response measurements fi (N)  S i 01 to M (1)

Where N is the rate of applied fertilizer N (kg N ha1), fi(N) is the measured grain yield at different rates of fertilizer N (kg grain ha1) for a single site/year, i is the index number identifying a crop fertilizer response to applied fertilizer N (kg grain ha1) for a single site/year, and M is the number individual measured crop fertilizer N responses (sites/years) in the set S. We assume that each individual measured crop fertilizer response fi(N) is a continuous diminishing return response, such that the rate of yield response to fertilizer application decreases progressively with an increase in the amount of applied N. Thus, we assume we can represent (approximately) the shape of each of the measured crop fertilizer N responses fi(N) by a leastsquares fitted Taylor polynomial (Burden et al. 1981; Colwell 1994), with a general form fi (N)0Fi (a)'Fi (a)(N (a)' '
=

Fi (a) 2!

==

(N (a)2

Fi(3) (a) F (n) (a) (N (a)3 ':::::' i (N (a)n 3! n! (2)

':::

where F /(a) is the (least-squares estimated) slope (1Y/ i 1N) of the ith fertilizer response curve at N 0a, F//(a) is i the (least-squares estimated) curvature (12Y/1N2) of the ith fertilizer response curve at N 0a, and F(n)(a) is the i (least-squares estimated) nth derivative of the fertilizer response curve. Given the assumption of a continuous diminishing return response, the (least-squares estimated) slope for N 0a00, F/(0) 0(1Y/1N) is also i defined as an index of the maximum fertilizer N use efficiency (NUE) of the crop for that particular site/year

KACHANOSKI * CROP RESPONSE TO NITROGEN FERTILIZER

545

Can. J. Soil. Sci. Downloaded from pubs.aic.ca by Depository Services Program on 01/10/12 For personal use only.

(i.e., the maximum increase in yield for the first increment of applied N fertilizer). With some assumptions, fi(N) has been estimated (by least-squares fitting) with reasonable accuracy using either the first three (2nd-order quadratic) or first four (3rd-order) terms of Eq. 2. The 2nd-order quadratic function is the most common empirical model used to estimate the shape of fi(N) and the 3rd-order estimate allows the fitting of a cubic-sill shape. Since the shape of the response functions fi(N) in S are defined entirely by the (least-squares) shape parameters Fi(a), F/(a), F//(a), and F///(a), the inverse i i i question is asked: What constraints or conditions placed on the shape parameters [Fi(a), F/(a), F//(a), i i and F///(a)] would result in a unique macro-relationi ship [for all fi(N) across S] between DYec and EONR. The application and usefulness of any derived macrorelationships (and the existence of unique measured macro-relationships between DYec and EONR) will depend on the generality of the constraints and the likelihood they hold across the implicit variability in S [due to differences in time, space/place and climate influencing the fi(N) in S]. As will subsequently be shown, for both 2nd- and 3rd-order estimates of fi(N), two different constraints/conditions placed on the shape parameters give rise to unique, but different macro-relationships between DYec and EONR [for all fi(N) across S]. The constraints are quite general and based on assuming a similar (across S) 0th or 1storder rate relationship between NUE and applied N. Thus, the constraints describe general soil fertility conditions. In a manner similar to classifying general kinetic relationships, the generalized rate relationship between NUE and N can be classified by: @(NUE)=@N 00: @(NUE)=@N8(NUE) :
X

a 00), is constant for all estimated response functions in the set S. Constraint 2: Fi (0)0a1Fi (0);a0constant (1st order) This states that all of the estimated fertilizer response functions in the set S have the same 1st-order rate relationship between NUE and N. Thus, the curvature (rate of change of NUE) of a fertilizer response function, for the first increment (i.e., at N 0a 00) of applied fertilizer N, is proportional to the initial slope (maximum NUE) of the estimated response function, and the proportionality parameter a is the same for all estimated response functions in the set S. Equations describing unique (across S) macro-relationships between DYec and EONR are now derived for each of the two constraints, for both the 2nd- and 3rdorder Taylor polynomial estimates of fi(N). It is worth noting that other empirical models and constraints (than the two given above) on the fertilizer response shape parameters may also give rise to additional macrorelationships between DYec and EONR, but this is beyond the scope of this paper. Second-order Approximation of fi(N) Choosing (for convenience) a00, the 2nd-order approximation of fi(N), defined by Yi(N), is given by: Yi (N)0Yi (0)'Bi 1N (Ci 1N 2 (3)
== =

no relationship; or X 0(0; 1; 2 . . .)0 order of relationship

Where Yi(0) 0Fi(0) 0grain yield with no N applied (i.e., N00) for the ith response function Bi 0F/(0) 0slope @Y=@N of the ith response function i (Fi== (0) /C 0 0index of curvature @ 2 Y=@N2 of the ith i 2! response function For convenience, and to emphasis the assumption of a continuous diminishing return response function, the third term in Eq.3 is preceded by a minus sign, with increasing (positive) values of Ci indicating increased convex curvature. Equation.3 is the well-recognized quadratic response model. Regardless of the true underlying shape of the response fi(N), the model parameters Bi and Ci can be viewed as 2nd-order (leastsquares) estimates of the shape parameters for the underlying fertilizer N response curve. Given the assumption of a diminishing return response, the value of Bi is the maximum slope of the estimated shape of response function and is equal to the maximum nitrogen use efficiency NUE (i.e., the change in grain yield per change in applied N, for the first increment of applied N). The value of Ci can then be viewed as the rate of change of the NUE as N increases.

then the 0th-order If we define (X 00) and 1st-order (X 01) rate relationships can be given by: @ 2 Y =@N 2 0Fi (N)8NUE 0 0constant
== ==

NUE 0F/(N) 0(1Y/1N), i

:0th order
=

@ 2 Y =@N 2 0Fi (N)8NUE1 0constant1Fi :1st order and two constraints related to these rate relationships across S can be given as: Constraint 1: Fi (0)0constant across S (0th-order) This states that all of the estimated fertilizer response functions in the set S have the same 0th-order rate relationship between NUE and N. Thus, the curvature (rate of change of NUE) of a fertilizer response function, for the first increment of applied N (i.e. N 0
==

546 CANADIAN JOURNAL OF SOIL SCIENCE

Case 1. Second-order fi(N): F//(0) 0constant across S i In this case, we assume that Bi (the maximum NUE) can vary across S [for different Yi(N)], but the curvature Ci (the rate of change of NUE) is constant for all Yi(N) across S. Setting N 0EONR in Eq.3 it follows that
DYeci 0Y (EONR)i (Y (0)i 0Bi 1EONR(C1EONR2 (4)

DYeci 0Bi 1EONRi (aBi 1EONR2 i 0Bi 1EONRi (1(a1EONRi ) where EONRi 0 Bi ( R 2Ci (12) (11)

Can. J. Soil. Sci. Downloaded from pubs.aic.ca by Depository Services Program on 01/10/12 For personal use only.

Where EONR is obtained by setting the first derivative of Eq.3 equal to R 0the ratio of the (price per kilogram of fertilizer N)/(price per kilogram grain) and solving for N giving EONRi 0 Bi ( R 2C (5)

Substitution of Eq. 10 into Eq. 12 and re-arranging gives: Bi 0R=(1(2a1EONRi ) Finally, substitution of Eq. 13 into Eq. 11 gives: DYec 0R1EONR (1 ( a1EONR) (1 ( 2a1EONR) (14) (13)

Rearrangement of Eq.5 gives: Bi 0 2C 1EONRi 'R Substitution of Eq.6 into Eq.4 gives: DYeci 0(2C 1EONRi 'R)1EONRi (C 1EONR2 i Expanding and simplifying Eq.7 gives: DYec 0R1EONR'C 1EONR2 (8) (7) (6)

As before (Case 1.), the subscript (i) has been dropped from Eq. 14 because it has been assumed (for Case 2.) that a is constant for all Yi(N) across S, and R is again an external boundary (arbitrary) constant. Thus Eq. 14 describes a unique macro-relationship between DYec and EONR that holds for all Yi(N) across S. Third-order Approximation of fi(N) Once again choosing (for convenience) a 00, the 3rdorder approximation of fi(N), defined by Yi(N), is given by an expanded Eq. 3 Yi (N)0Yi (0)'Bi 1N (Ci 1N 2 'Di 1N 3 ; N 5Ni;max and Yi (N)0Yi (Ni;max ); N ]Ni;max (16) (15)

The subscript (i) has been dropped from Eq. 8 because it has been assumed (for Case 1) that C is constant for all Yi(N) across S, and R is an external boundary (arbitrary) constant that holds across all Yi(N). Thus, Eq. 8 describes a unique macro-relationship between DYec and EONR that also holds for all Yi(N) across S. Setting R 00 in Eq. 8 gives the maximum point of the yield response function with N 0Nmax and DYec 0 DYmax. Thus,
2 DYmax 0 C1Nmax

(9)

Case 2. Second-order fi(N): F//(0) 0a*F/(0), i i a0constant across S In this case, we assume that both Bi and Ci can vary for different Yi(N), but Ci is proportional to Bi across S. It follows that
Ci 0a1Bi (10)

where Yi(0), Bi, and Ci are defined as before, and === (Fi (0) /D 0 0index of curvature change @ 3 Y=@N3 of i 3! the ith response function at N00. The assumption of a continuous diminishing return response function requires that Ni,max be defined as the fertilizer rate N where: @2Y @N 2 0 @Y @N 0 0; N 0Ni;max (17)

where a 0proportionality coefficient that is constant for all Yi(N) across S. Substituting Eq. 10 into Eq.3 and setting N 0EONR gives:

This ensures the response function asymptotically approaches a flat (zero slope) line (sill) response at Nmax. Solving Eq. 17 gives the following useful identities

KACHANOSKI * CROP RESPONSE TO NITROGEN FERTILIZER

547

Ni;max 0

Ci 3Di

Bi Ci

(18)

Substitution of Eq. 23 with Di 0C/(3*Ni,max) into Eq. 19 and re-arranging gives: DYec 0R1EONR'C 1EONR2   2C 3 ( 1EONR 3Nmax

Case 3. Third-order fi(N): constant across S In this case, we assume that Bi (the maximum NUE) can vary across S [for different Yi(N)], but the initial curvature Ci (the rate of change of NUE) at N 00, is constant for all Yi(N) across S. Setting N 0Ni,max in Eq. 15 and re-arranging gives
DYi;max 0Y (Ni;max )(Yi (0)

F//(0) 0 i

(24)

Can. J. Soil. Sci. Downloaded from pubs.aic.ca by Depository Services Program on 01/10/12 For personal use only.

2 3 0Bi 1Ni;max (C 1Ni;max 'Di 1Ni:max

(19)

Substituting Bi 0C*Ni,max and Di 0C/(3*Ni,max) (see Eq. 18) into Eq. 19 and re-arranging gives: DYmax 0 C 3 1Nmax
2

Since, in this case, C is constant for all S, Eq. 24 is a macro-relationship which holds for all response curves in S. Equation 24 gives the value of DYec for any paired values of EONR and Nmax, with the paired values of EONR and Nmax defined by Eq. 22. Thus, Eq. 22 and Eq. 24 together give a macro-relationship between DYec and either EONR or Nmax that holds for all response functions in S. Note that if R 00 (thus EONR 0Nmax) then Eq. 24 equals Eq. 20 (which it should).

(20)

As in Case 1, the subscript (i) can be dropped from Eq. 20 because the only shape parameter is C, which, for this case, is constant for all Yi(N) across S. Thus, Eq. 20 is a macro-relationship that holds for all response curves in the set S. Setting the first derivative of Eq. 19 equal to R 0the ratio of the (price per kilogram of fertilizer N)/(price per kilogram grain), N 0EONR and again substituting Bi 0 C*Ni,max and Di 0C/(3*Ni,max) and re-arranging gives: (C=Nmax )1EONR2 (2C 1EONR' (C 1Nmax (R)00

Case 4. Third-order fi(N): Fi//(0) 0a* Fi/(0), a0constant across S As in Case 2 (2nd-order), we assume that both Bi and Ci can vary for different Yi(N), but Ci is proportional to Bi across S. It follows again that:
Ci 0a1Bi (25)

(21)

where a is the proportionality coefficient that is constant for all Yi(N) across S. Substituting Eq. 25 into Eq. 18 gives:  3 a (26) Di 0 1Bi 3 Substituting Eq. 25 and Eq. 26 into Eq. 15, and setting N 0EONR and re-arranging, gives DYi;ec 00Bi 1EONRi (a1Bi 1EONR2 i ' a2 3 Bi 1EONR3 i (27)

where again the subscript i has been dropped because C is constant for all Yi(N) across S, so Eq. 21 holds for all S. Recognizing that Eq. 21 has the general form of a quadratic equation (aX2'bX'c 00, with X 0EONR), the solution for EONR is obtained by finding the roots of Eq. 21, which are: s Nmax 1R EONR 0 Nmax9 C Since (by definition) EONR is less than Nmax, the solution to Eq. 21 is: s Nmax 1R (22) EONR 0 Nmax ( C Setting the first derivative of Eq. 19 equal to R 0the ratio of the (price per kilogram of fertilizer N)/(price per kilogram grain), and re-arranging and solving for Bi gives: Bi 0 R'2C1EONRi (3Di EONR2 i (23)

Recognizing that Eq. 22 also holds for C 0Ci, and substitution of Ci 0a*Bi into Eq. 22 and re-arranging gives Bi 0 R (1 ( a1EONR)2 (28)

Finally, substituting of Eq. 28 into Eq. 27 gives DYec 0R1EONR (1 ( a1EONR ' a2 =31EONR2 ) (1 ( 2a1EONR ' a2 EONR2 ) (29)

As before, the subscript i has been dropped from Eq. 29 because it has been assumed (for this case) that a is

548 CANADIAN JOURNAL OF SOIL SCIENCE

constant for all Yi(N) across S, and R is again an external boundary (arbitrary) constant. Thus it follows that Eq. 29 describes a unique macro-relationship between DYec and EONR that holds for all response curves across S. A summary of the derived macro-relationships (across S) for 2nd- and 3rd-order estimates of fi(N) for the two constraints are given below.

3rd-order DYec 0R1EONR (1 ( 2a3 1EONR ' a2 EONR2 ) 3 DYmax 0 undefined Nmax 01=a3 (1 ( a3 1EONR ' a2 =31EONR2 ) 3 (36)

(37)

Constraint 1: Fi//(0) 0 constant; C0constant across S


Can. J. Soil. Sci. Downloaded from pubs.aic.ca by Depository Services Program on 01/10/12 For personal use only.
2nd-order DYec 0R1EONR'C2 1EONR2
2 DYmax 0 C2 1Nmax Nmax 0 undefined (variable)

(30) (31)

Note: Subscripts have been added to a to indicate the order of the response function model, with a2 and a3 being the 2nd- and 3rd-order proportionality constants, respectively, between the response function curvature and slope at N 00. The values of a2 and a3 are not necessarily related to each other. However, if Nmax is forced to be the same value for a 2nd- and 3rd-order fit to a set of response functions, then a3 02*a2. Experimental Data Sets S: Corn Response to N Fertilizer Beauchamp et al. (1987) conducted a comprehensive review of N requirements for corn in Ontario, Canada, and compiled a data set of 202 field N response trials for corn conducted across southern Ontario from 1962 to 1986. The N response trials were grouped into three significantly different classes based on geographic origin: southwestern, central, and eastern Ontario, with the southwestern Ontario data class further sub-divided into pre-plant N applications and sidedress N applications. An earlier preliminary analysis and application of the delta yield concept to the data set has been given by Kachanoski et al. (1996a). In this paper, a sub-set of 72 field response trials (1962 to 1986) from Beauchamp et al. (1987) for southwestern Ontario (side dress N application) are used to illustrate the application of the delta yield concept. Three trials for southwestern Ontario (side dress N application) were not included because of negative or concave N response. As reported by Beauchamp et al. (1987) a 2nd-order quadratic equation (Eq. 3) was fit to each of the field trials generating 72 pairs of Bi and Ci values. The subsequent analysis gave similar results to the other classes of data. A subset from previously published data from 298 fertilizer N response experiments for corn in Illinois (n 0 54), Minnesota (n 054), Missouri (n 032), Pennsylvania (n 057), and Wisconsin (n 0101) was created and analyzed by Lory and Sharf (2003). They omitted 105 of the response experiments in their analysis because at these sites yield was not responsive to applied fertilizer N. For the remaining 193 responsive sites they obtained an empirical linear macro-relationship between DYec and EONR by regression/curve-fitting of the form EONR 066:4'0:01941DYec r2 00:47 (38)

3rd-order DYec 0R1EONR'C3 1EONR2   2C3 3 ( 1EONR 3Nmax where EONR 0 Nmax ( s Nmax 1R C3 (33)

(32)

2 DYmax 0C3 =31Nmax Nmax 0 undefined (variable)

Note: Subscripts have been added to C to indicate the order of the response function model, with C2 and C3 being the 2nd- and 3rd-order indices, respectively, of the response function curvature at N 00. The values of C2 and C3 and are not necessarily related to each other. However, if Nmax is forced to be the same value for a 2nd- and 3rd-order fit to a set of response functions, then C3 03*C2.

Constraint 2: F//(0) 0a* F/(0); Bi 0a*Ci and a0 i i constant across S


2nd-order DYec 0R1EONR DYmax 0undefined Nmax 01=(21a2 ) (35) (1 ( a2 1EONR) (1 ( 2a2 1EONR) (34)

The empirical macro-relationship (Eq. 38) obtained by Lory and Scharf (2003) is compared against the theoretically derived macro-relationships (Eq. 30 to Eq. 37).

KACHANOSKI * CROP RESPONSE TO NITROGEN FERTILIZER

549

The theoretically derived macro-relationships (Eq. 30 to Eq. 37) are also compared with the results of Fixen (2006). He analyzed two sets of fertilizer N response curves for corn and also obtained empirical macrorelationships between DYec and EONR by curve-fitting. The first empirical relationship obtained by Fixen (2006) for 6 yr of N fertilizer responses for corn at two Minnesota sites (from Fenster et al. 1978) was (after conversion to SI units:
2 EONR 0 0:06811DYec (0:00000561DYec r2 0 0:96

(39)

Can. J. Soil. Sci. Downloaded from pubs.aic.ca by Depository Services Program on 01/10/12 For personal use only.

The second set analyzed by Fixen (2006) was obtained from Fox and Piekielek (1983) for 60 fertilizer N response trials for corn across Pennsylvania and the empirical relationship was (after conversion to SI units)
2 EONR 0 23:6'0:05871DYec (0:00000531DYec r2 0 0:87

(40)

For both Eq. 39 and Eq. 40, Fixen (2006) calculated EONR as the N rate giving 95% of maximum yield rather than defining EONR using a price ratio R. RESULTS AND DISCUSSION Regardless of estimation order (2nd or 3rd) of the assumed model [Yi(N)] of the measured fertilizer N response function fi(n), or the assumed constraint, the derived macro-relationships of EONR versus DYec (1) are highly non-linear, (2) start (as expected) at DYec 00 and EONR 00, (3) have an initial slope determined only by the price ratio R [i.e., 1(EONR)/1(DYec)0R1 at the origin], and (4) have a curvilinear shape determined only by the order (x 00, or x 01) of the rate relationship between NUE and applied N. These common features are illustrated in example plots of the derived macrorelationships (Fig. 1 and Fig. 2). The various coefficients (C and a) for the graphs are representative for data sets of N response by corn and were chosen for convenience to illustrate the nature of the derived functions. A single data pair (DYec, EONR) for a macrorelationship is obtained from each fertilizer response function. The data pair is taken from the location on the estimated fertilizer response function where the slope 1(Y)/1(N)0R. Thus, the price ratio R has a very significant influence on the theoretical macro-relationship between DYec and EONR (Fig. 1). This also suggests that DYec and EONR data pairs calculated with different R values across a set S (of response functions) could have significant increased data scatter in the measured relationship between DYec and EONR. The macro-relationship between DYec and EONR (for a 2nd-order response curve) for constraint 1 (0th-order rate relationship between NUE and N) and constraint 2 (1st-order rate relationship between NUE and N) can be very similar at lower values of DYec and higher R values.

Fig. 1. Comparison of the theoretical macro-relationship between DYec [grain yield at the economically optimum N rate (EONR) minus grain yield of the control (N 00)] and EONR for representative sets of 2nd-order fertilizer N response functions for corn under (a) constraint 1 (0th-order rate relationship between crop nitrogen use efciency NUE and N), or (b) constraint 2 (1st-order rate relationship between NUE and N).

For example, the macro-relationships between DYec and EONR for constraint 1 (Fig. 1a) and constraint 2 (Fig. 1b) are quite similar for R 015 and R 020, with EONR values within plus or minus 10 kg N ha1 of each other for all DYec. The differences in the macro-relationships for constraint 1 versus constraint 2 increase as the value of R decreases. At lower values of R and higher values of DYec, EONR asymptotically approaches a maximum value equal to Nmax 01/(2*a) under constraint 2, while for constraint 1, the value of EONR continually increases as DYec increases. This suggests that the supposition that EONR should continually increase as DYec increases (Lory and Scharf 2003) will depend on the value of R chosen and whether there is a 1st (constraint 1) or 2nd- (constraint 2) order relationship between NUE and applied N.

550 CANADIAN JOURNAL OF SOIL SCIENCE

Can. J. Soil. Sci. Downloaded from pubs.aic.ca by Depository Services Program on 01/10/12 For personal use only.

Lory and Scharf (2003) (see Eq. 38) for 193 responsive sites across five states (USA). Lory and Scharf (2003) rejected the notion of a curvilinear relationship between DYec and EONR because there was no indication of curvilinearity in their data. They acknowledged, however, that (1) their empirical relationship (Eq. 38) had an intercept that was significantly greater than zero, (2) theoretically [as suggested by Kachanoski et al. (1996a)] the relationship must pass through the origin (DYec 00, EONR 00), and (3) in their regression they eliminated 105 sites with no N response (i.e., sites with DYec 00, EONR 00 data pairs), which, if included in the data-set, would have required a non-linear macro-relationship for adequate fitting. They speculated that the true model is likely non-linear, starting at the intercept and rapidly approaching the linear trend, but that the lack of locations with small delta yields and the data variability made it impossible to empirically derive the curvature at the lower end (near the origin) of their relationship. The linear empirical macro-relationship between DYec and EONR obtained by Lory and Scharf (2003) (see Eq. 38) and the theoretical macro-relationship using Eq. 30 (for a 2nd-order approximation to fi(N); constraint 1 with C 00.15, and R 00 and R 010) are graphed together in Fig. 3. The theoretical macro-relationship (Eq. 30) for a 2nd-order approximation to fi(N) was chosen because Lory and Scharf (2003) mentioned that a quadratic (2nd-order) model was used to fit individual response functions to obtain the DYec and EONR data pairs. Two R values (R 00, R 010) were graphed because Lory and Scharf (2003) were not sure of the
Fig. 2. An example of the inuence of tting a 2nd-order versus a 3rd-order equation (with same maximum N rate, Nmax) to individual fertilizer N response functions in a set, on the macro-relationship between DYec [grain yield at the economically optimum N rate (EONR) minus grain yield of the control (N 00)] and EONR for the same set.

Fitting a 2nd- or 3rd-order equation to a set of N response functions (for the same R and the same Nmax) changes the macro-relationship between DYec and EONR (Fig. 2). The 3rd order response function asymptotically approaches a flat (slope 00) straight line response of Y versus N at Nmax. This reduces the calculated value of EONR (for a given R and DYec) for a 3rd order function compared to a 2nd-order function with the same Nmax. The differences in the macrorelationships, however, are not very large and the two curves converge as expected to a common slope [1(EONR)/1(DYec]0R1) at the origin (Fig. 2). This suggests that a graph of data pairs (DYec and EONR) from a set S containing a mixture of 2nd- and 3rd-order response functions could appear as a single reasonably well-defined macro-relationship. As mentioned earlier, an empirical linear macrorelationship between DYec and EONR was obtained by

Fig. 3. The similarity in shape of a theoretical macro-relationship between DYec [grain yield at the economically optimum N rate (EONR) minus grain yield of the control (N 00)] and EONR and the linear empirical macro-relationship obtained by Lory and Scharf (2003). The theoretical macro-relationship is for a representative set of 2nd-order fertilizer N response functions for corn under constraint 1 (0th-order rate relationship between crop nitrogen use efciency NUE and N).

KACHANOSKI * CROP RESPONSE TO NITROGEN FERTILIZER

551

Can. J. Soil. Sci. Downloaded from pubs.aic.ca by Depository Services Program on 01/10/12 For personal use only.

trials in Minnesota and Pennsylvania and the theoretically derived macro-relationships [for a 2nd-order model of Yi(N)] under constraint 1 and constraint 2 is illustrated in Fig. 4a and Fig. 4b, respectively. Fixen (2006) calculated EONR as the N rate giving 95% of maximum yield, which for a 2nd-order approximation to the N response function is equivalent (analysis not shown) to choosing a different R value for each response function based on maximum yield. For yield levels indicated by Fixen (2006) and C value 00.1 (for comparison), the R value associated with the N rate giving 95% of maximum yield would vary systematically (with maximum yield) from approximately 5 to 20. Thus, the analysis (and the empirical relationship) given by Fixen (2006) may be a composite function convoluting (from variable R) an underlying macro-relationship, which is a function of R. The slopes of the empirical relationships at DYec 00 suggest effective R values of approximately R 015 and R 017 for Eq. 39 and Eq. 40, respectively. Without access to the original data it is not possible to determine the underlying macro-relationship or to test the veracity of assuming either constraint 1 or constraint 2. For the set of fertilizer N response trials (1962 to 1986) from Beauchamp et al. (1987) for southwestern Ontario (side-dress N application), the curvatures Ci of individual response functions (from a 2nd-order fit) varied considerably, but were also highly correlated (r2 00.92) to the response function slope Bi at N 00 (i.e., to maximum NUE) (Fig. 5). This suggests that all of the fertilizer response functions in this set S have a similar 1st-order
Fig. 4. Comparison of the empirical macro-relationships between DYec [grain yield at the economically optimum N rate (EONR) minus grain yield of the control (N 00)] and EONR obtained by Fixen (2006) using a date set with variable price ratios R, and theoretical macro-relationships for representative sets of 2nd-order fertilizer N response functions for corn under (a) constraint 1 (0th-order rate relationship between nitrogen use efciency NUE and N), or (b) constraint 2 (1st-order rate relationship between NUE and N).

price ratio R used to calculate EONR in the data for three of the five states (USA). Without access to the original data, it is not possible to determine the veracity of either constraint 1 or constraint 2 across the data set, so constraint 1 was assumed and a representative C value (C 00.15) chosen to illustrate the remarkable similarity in the shape of the theoretical and empirical macrorelationships, including what could easily be misinterpreted as an apparent linear trend for larger values of DYec. The theoretical macro-relationship, however, also includes (correctly) the near origin curvilinearity and starts at the origin (DYec, EONR) 0(0, 0). The similarity of the empirical non-linear macrorelationships between DYec and EONR obtained by Fixen (2006) (Eq. 39 and Eq. 40) for the N response

Fig. 5. A linear relationship between the curvature C and initial slope B of individual fertilizer N response functions suggesting constraint 2 (1st-order rate relationship between nitrogen use efciency NUE and N) holds for this set of response functions. The values of C and B were obtained from tting a 2nd-order grain yield response function to each of the 72 fertilizer N response trials for corn (1962 to 1986) from Beauchamp et al. (1987) for southwestern Ontario (side dress N application).

552 CANADIAN JOURNAL OF SOIL SCIENCE

Can. J. Soil. Sci. Downloaded from pubs.aic.ca by Depository Services Program on 01/10/12 For personal use only.

Fig. 6. A comparison of predicted versus measured macro-relationships between DYec [grain yield at the economically optimum N rate (EONR) minus grain yield of the control (N00)] and EONR for the set of 72 fertilizer N response trials for corn (1962 to 1986) from Beauchamp et al. (1987) for southwestern Ontario (side dress N application). The only parameter used in the prediction is the slope a00.0027 of the linear relationship between the curvature C and initial slope B of individual fertilizer N response functions for this set of response functions (see Fig. 5).

rate relationship between NUE and N (i.e., constraint 2). The proportionality constant for the 1st-order relationship (constraint 2) is obtained from the slope of Ci versus Bi and is a2 00.0027 for this set S (Fig. 5). The appropriate theoretical macro-relationship between DYec and EONR for this set S is given by Eq. 34 (i.e., for a 2nd-order model of fertilizer N response and constraint 2) and fits the measured data across a range of selected price ratios R (Fig. 6). It is worth noting that the theoretical/predicted macro-relationships in Fig. 6 are not obtained by curve-fitting Eq. 34 to the measured data. Rather, the only necessary parameter (i.e., a) used for the predictions from Eq. 34 is the slope of Ci versus Bi (a2 00.0027, Fig. 5) and arbitrary values of the price ratio R. Similar curves of predicted versus measured DYec and EONR are obtained for any value of R, with the

goodness-of-fit directly related to the strength of the correlation (r2 00.92, Fig. 5) between Ci and Bi. The correlation between Ci and Bi is also a measure of the strength of the assumption of constraint 2 and 1st-order rate relationship between NUE and N. If the correlation between Ci and Bi for this set was perfect (r2 01), then the fit between the predicted and measured macro-relationship between DYec and EONR would also be perfect. For any single selected price ratio R, the different derived macro-relationships [for 2nd- and 3rd-order estimates of fi(N) and for the two constraints] can give rise to relatively similar shapes of predicted DYec versus EONR depending on the choice of either C (constraint 1) or a (constraint 2). Thus, curve-fitting a derived macrorelationship equation to measured DYec versus EONR data from a single R value cannot be used to infer the

KACHANOSKI * CROP RESPONSE TO NITROGEN FERTILIZER

553

validity of an assumed constraint or derived fitting parameter (C or a). Changing the value of R with fixed values of C (constraint 1) or a (constraint 2), however, results in significantly different shapes of predicted DYec versus EONR (Fig. 1). Thus, examining the goodness of fit between measured and predicted macro-relationships (DYec versus EONR) across a range of R values (e.g., Fig. 6) and testing the relationship between response curvature C and slope B (e.g., Fig. 5) across the set of response functions is recommended as a way of testing the veracity of a constraint/macro-relationship between DYec and EONR. Unfortunately, the variability in measured yield response to N and the well-demonstrated ability for different yield response models (2nd- or 3rdorder under different constraints) to equally fit measured data suggest that goodness of fit, even for multiple R values, may not be sufficient to test the validity of assumptions for a particular macro-relationship (response function order or constraint). By corollary, the similarity of derived macro-relationships between DYec versus EONR across significantly different assumptions (response function order or constraint), suggests the function shapes are robust and likely the reason they seem to hold across data sets spanning decades and large areas. It also means they are useful as yield indices of N response (post-analysis), soil fertility assessment, and for comparison of N response across years/sites. It is interesting to speculate whether direct measures of the order of relationship between NUE and applied N (e.g., isotope studies) would help define and evaluate criteria for soil test calibration, or provide a way for determining meaningful inclusion (or not) of a particular response function into a particular set S. The influence of site, soil, and climate characteristics on set (S) characteristics could then be explored. Kachanoski et al. (1996b) suggested that estimates of the spatial pattern of DY, using yield measurements from side-by side field strips (with and without fertilizer N) and a suitable macro-relationship between DY versus EONR, could be used to predict the spatial patterns of EONR and/or to rank different fields for response to applied fertilizer N. If a macro-relationship between DY versus EONR holds for a set of fertilizer response functions (crop and region), then the usefulness of this approach would depend primarily on the temporal stability of the ranking (field, location) and spatial patterns of DY. Murrell (2007) indicated that application of this approach would require fundamentally new types of information to be collected on farms, including geographically referenced (location, field) measurements and records of crop yield with and without fertilizer N. CONCLUSIONS Unique and useful macro-relationships between DYec and EONR should be expected across sets (S) of 2nd- or 3rd-order estimates of the shape of measured fertilizer

N response functions fi(N) provided either of the constraints examined hold across S. Since the two constraints chosen represent a range of 0th to 1st-order rate relationships between NUE and applied N, the common features of the macro-relationships for both constraints (Eq. 34 to Eq. 37) are robust and likely the reason that macro-relationships between DYec and EONR seem to hold across data sets spanning decades and large areas.

Can. J. Soil. Sci. Downloaded from pubs.aic.ca by Depository Services Program on 01/10/12 For personal use only.

ACKNOWLEDGEMENTS Financial support for this study was provided by the Natural Sciences and Engineering Research Council of Canada (NSERC).
Beauchamp, E. G., Newdick, P. G. and Sheard, R. W. 1987. Nitrogen requirements for corn in southern Ontario. Department of Land Resource Science Publication, University of Guelph, Guelph, ON. Burden, R. L., Faires, J. D. and Reynolds, A. C. 1981. Numerical analysis. 2nd ed. Prindle, Weber, and Schmidt, Boston, MA. 591 pp. Colwell, J. D. 1994. Estimating fertilizer requirements: A quantitative approach. CAB International, Wallingford, UK. Dahnke, W. C. and Olsen, R. A. 1990. Soil test correlation, calibration, and recommendation. Pages 4571 in R. L Westerman, ed. Soil testing and plant analysis. SSSA, Madison, WI. Fenster, W. E., Overdahl, C. J., Randall, G. W. and Schoper, R. P. 1978. Effect of nitrogen fertilizer on corn yield and soil nitrates. Misc. Rept 1531978. Agric. Exp. Station, University of Minnesota, St. Paul, MN. Fixen, P. E. 2006. Use of yield goals for providing N rate suggestions: General concept. Proceedings of the North Central Extension-Industry Soil Fertility Conference. 22: 57 66. Fox, R. H. and Piekielek, W. P. 1983. Response of corn to nitrogen fertilizer and the prediction of soil nitrogen availability with chemical tests in Pennsylvania. Bulletin 843. The Pennsylvania State University, Agric. Exp. Station, University Park, PA. Kachanoski, R. G. and Fairchild, G. L. 1996. Field scale fertilizer recommendations: The spatial scaling problem. Can. J. Soil Sci. 76: 16. Kachanoski, R. G., Fairchild, G. L. and Beauchamp, E. G. 1996a. Yield indices for corn response to applied fertilizer: Application in site specic crop management. Pages 425432 in P. C. Robert, R. H. Rust, and W. E. Larson, eds. Proc. Third International Conference on Precision Agriculture, Minneapolis, Minnesota. 1996 Jun. 2326. ASA, CSSA, and SSSA, Madison, WI. Kachanoski, R. G., OHalloran, I. P., Aspinall, D. and von Bertoldi, P. 1996b. Delta yield: mapping fertilizer nitrogen requirements for crops. Better Crops 80: 2023. Lory, J. A. and Scharf, P. C. 2003. Yield goal versus delta yield for predicting fertilizer nitrogen need in corn. Agron. J. 95: 994999. Mulvaney, R. L., Khan, S. A. and Ellsworth, T. R. 2005. Need for a soil-based approach in managing nitrogen fertilizers for protable corn production. Soil Sci. Soc. Am. J. 70: 172182.

554 CANADIAN JOURNAL OF SOIL SCIENCE Murrel, T. S. 2007. The delta yield concept: an update. Better Crops 91: 2021. Schlegel, A. J., Grant, C. A. and Havlin, J. L. 2005. Challenging approaches to nitrogen fertilizer recommendations in continuous cropping systems in the great plains. Agron. J. 97: 391398. Tillotson, P. M. and Nielsen, D. R. 1984. Scale factors in soil science. Soil Sci. Soc. Am. J. 48: 953959. Vanotti, M. B. and Bundy, L. G. 1994. An alternative rationale for corn N recommendations. J. Prod. Agric. 7: 249256.

Can. J. Soil. Sci. Downloaded from pubs.aic.ca by Depository Services Program on 01/10/12 For personal use only.

You might also like