ES&T FEATURES
A reassessment of crop loss from ozone Models suggest that ozone substantially reduces the yields of several crops and that the economic effect of these yield reductions may be important
Walter W. Heck USDA/ARS North Carolina State University Raleigh, N.C. 27650 Richard M. Adams Oregon State University Corvallis, Ore. 97331 William W. Cure North Carolina State University Raleigh, N.C. 27650 Allen S. Heagle USDA/ARS North Carolina State University Raleigh, N.C. 27650 Howard E. Heggestad USDA/ARS Beltsville Agricultural Research Center Beltsville, Md. 20705 Robert J. Kohut Boyce Thompson Institute Cornell University Ithaca, N.Y. 14853 Lance W. Kress Argonne National Laboratory Argonne, Ill. 60439 John O. Rawlings North Carolina State University Raleigh, N.C. 27650 O. Clifton Taylor
University of California Riverside, Calif. 92502
Early estimates of direct crop losses using the limited data available suggest that about $3 billion annually, or about 5.6% of the gross value of farm commodities, would be lost if the country experienced a seasonal 7-h/d mean ozone (0 3 ) concentration of 0.06 ppm (1). The current secondary National Ambient Air Quality Standard (NAAQS) for O3 is based on information about vegetation effects presented in the criteria document, "Air Quality Criteria for Ozone and Other Photochemical Oxidants" (2). A review of the document reveals a lack of field studies in which economically important crops are exposed to O3 at concentrations experienced in the U.S. The National Crop Loss Assessment Network (NCLAN) was initiated to address the significance of crop losses caused by air pollution, with an initial emphasis on O3. The NCLAN approach was discussed in the first NCLAN assessment (3). The Research Management Committee (RMC) coordinates the planning, management, and execution 0013-936X/83/0916-0572A$01.50/0
of the NCLAN program. The program has three primary objectives: • to define the relationships between yields of major agricultural crops and doses of O3, SO2, NO2, and their mixtures as required to satisfy the needs of the economic assessment and to support the development of the NAAQS; • to assess the primary effects on the nation's economy that result from exposing major agricultural crops to O3, SO2, NO2, and their mixtures; and • to aid further understanding of the cause and effect relationships that determine crop responses to pollutant exposures. The NCLAN program includes six field sites that produce experimental data needed to develop dose-response functions. Research at each site is the responsibility of the senior scientists located at the site (Table 1). Detailed descriptions of methodology, results, and interpretation for the experiments at each site will be published separately (4, 5). Open-top chambers were used to expose plants to a range of O3 concentrations, with daily variations determined by changes in ambient O3 concentrations at each site (6-9). In addition, each experiment included a plot, without a chamber, exposed to ambient air. All plots (chambered or nonchambered) were 3 m in diameter. The experimental designs, replications, and treatments differed at each site. However, the basic design at each site included the following 0 3 treatments: • AA—plot, without a chamber, exposed to ambient air; • CF—plot, with a chamber, exposed to charcoal-filtered air; this treatment resulted in O3 concentrations ranging from 20 to 50% of ambient levels because O3 entered through the open top; • NF-1—plot, with a chamber, exposed to nonfiltered air to which sufficient O3 was added (about 0.005 to 0.015 ppm) for 7 h/d to compensate for system losses; and • NF-2, NF-3, and NF-4—treatments that were the same as NF-1 except that additional amounts of O3 were added to the ambient air for 7 h/d. Each day ozone additions were made during the same 7-h period (0900-1600 h standard time). At this time, ambient O3 concentrations are normally the highest, and plants are generally more sensitive than they are during the remaining hours of the day. To allow dew formation, the chamber
© 1983 American Chemical Society
fans were not run between 2100 and 0500 h. Also, exposures were not conducted on rainy days. At each site, the plots that received the AA treatment were used to evaluate the effects of the chambers. Ozone-dispensing and time-sharing monitoring systems were similar to those described by Heagle et al. (8). This assessment presents a summary of NCLAN field research and addresses several issues of importance to the program. Specifically, it includes the development of crop loss functions using the two years of NCLAN data and the data that were collected in several previous studies that employed similar techniques and an economic assessment using both linear and Weibull dose-response functions to determine the effect of model specifications on estimates of economic loss. Crop loss functions In the first NCLAN assessment, relationships between the seasonal 7-h/d mean O3 concentration and crop yield were developed for individual crop cultivars at specific sites (3). The assessment used a seasonal 7-h/d mean O3 concentration of 0.025 ppm as the value from which yield losses were calculated. Although linear, quadratic, logistic, and plateau-linear regression models were used to develop yield loss functions for each of 19 available data sets, the linear model was selected for the first NCLAN assessment since only five of the 19 data sets were significantly improved by the three-parameter models (that is, the quadratic, logistic, and plateaulinear). Because the relative response of the cultivars may be homogeneous, it is worthwhile to find a way to consolidate response curves among cultivars to produce a common crop model for each species as the number of cultivars evaluated increases. The use of a common mathematical model would facilitate this consolidation. Such a function should be flexible enough to accommodate the range of observed responses that may be suggested by the biological behavior. For comparing responses among cultivars and species, it would be useful to express the parameters of the function in biologically meaningful units such as concentration or yield. From a set of alternative functions, we have chosen to reevaluate the linear, quadratic, and plateau-linear functions and to seriously consider the Weibull function. The seasonal 7-h/d mean O3 concentration of 0.025 ppm is a reasonable Environ. Sci. Technol., Vol. 17, No. 12, 1983
573A
TABLE 1
a Site locations, regional coordinators, and senior scientists
Regional coordinators
Site locations
Senior scientists
Northeast Ithaca, N.Y.
b
Leonard Weinstein, Boyce Thompson Institute, Ithaca, N.Y.
Southeast Raleigh, N.C. Beltsville, Md.
b
Walter W. Heck, USDA, North Carolina State University, Raleigh, N.C.
Central states Argonne, III.
6
Joseph E. Miller, Argonne National Lab., Argonne, III
Southwest Shatter and Tracey, Calif.
" O. Clifton Taylor, University of California, Riverside, Calif.
b
Northwest Corvallis, Ore.
b
* Richard Adams, Grady Neely
Eric M. Preston, EPA, Corvallis, Ore.
Robert Amundson, Robert Kohut, John A. Laurence c
William W. Cure, Allen S. Heagle, Howard E. Heggested Lance W. Kress Gail Bingham
a
John O. Rawlings, professor of statistics, North Carolina State University, is the NCLAN stat istician. 0 Members of the Research Management Committee. c Coordinated data analysis and presentation with John Rawlings.
for the assessment of crop loss from O3 for several reasons: • It is flexible enough to cover the range of biological responses ob served; • the parameters are easily inter preted; • since effects are expressed as a proportion of maximum yield, data representing the effects on different cultivars can readily be combined to estimate a common proportional re sponse; • the Weibull form permits testing of homogeneity among the individual cultivars in the common proportional response; and • where homogeneity is found, the common proportional response models can be used to represent the response of the crop as a species. The Weibull function The Weibull model is given as
common control concentration from which to calculate yield losses because it is close to the values found in the CF treatments at NCLAN field sites. During three seasons of NCLAN op erations, the seasonal 7-h/d (09001600 h) mean O3 concentrations for the CF treatments have ranged from 0.012 to 0.026 ppm with a median of 0.020 ppm. In our first report, we also suggested that an ozone level of 0.025 ppm is probably a reasonable "natural" background O3 concentration (3). A review of two references (10, 11) and discussions with A. P. Altshuller (12) suggest that no definitive statement can be made about the natural back ground of O3. There are no long-term studies of natural O3; most measure ments of natural O3 were made at an altitude above 5 km. The season of the year, the particular year, the latitude, the altitude, and local meteorological conditions all influence the natural O3 at any location. However, available data suggest that a seasonal 7-h/d mean O3 concentration range of 0.025 ± 0.01 ppm is a reasonable estimate of natural O3. We anticipate that this range will be found throughout the crop-growing regions of the U.S. Un doubtedly, there are certain locations where using a value of 0.025 ppm will inflate the calculated effects and other locations where it will deflate the cal culated effects. But using this value for a nationwide assessment that spans a number of years should give a reason able interpretation of O3 effects on crop production. It should be recognized that neither the experimental design nor the anal ysis of crop loss models requires a 574A
specific knowledge of the natural O3 concentration. This article and our first report (3) use data from NCLAN and similar data sets to illustrate the development of crop loss functions. Those using the model can choose any natural O3 concentration and calculate yield losses. Alternatively, one can use the current ambient O3 concentrations and calculate yield gains as the con centrations decrease or yield losses as the concentrations increase. This section shows how the indi vidual data sets were analyzed, quali tatively compares the four models identified for possible use, gives a brief overview of the Weibull model, shows how the Weibull model was used to combine data sets, and predicts yield losses using the Weibull model.
Environ. Sci. Technol., Vol. 17, No. 12, 1983
Single data set analyses In fitting the four models (linear, quadratic, plateau-linear, Weibull) to each data set, including data obtained in the AA treatment, an analysis of variance (AOV) was used to obtain an estimate of experimental error for hypothesis testing. For each data set the AOV-mean square error (AOVMSE) term and the reduction in sums of squares were calculated for the three-parameter models relative to the linear model. Comparisons among the reduction in sums of squares, for the three-parameter models relative to the linear model, suggest that in many cases one of the three-parameter models (quadratic, plateau-linear, or Weibull) is preferable to the linear model but in terms of the "goodness of fit," the three-parameter models are indistinguishable from each other. The Weibull function was selected
r=aexp[-(x/ff)']+e
(1)
where Y is the yield and χ is the O3 dose (seasonal 7-h/d mean O3 con centration in ppm). The three param eters to be estimated are a, the hypo thetical maximum yield at zero O3 concentration; σ, the O3 concentration when yield is 0.37a; c, a dimensionless shape parameter (c = 1 gives the ex ponential loss function whereas a larger c [e.g., 4.5] gives a region of almost no loss [a threshold] before the curve starts to drop); and e, the ran dom error associated with each ob servation. Curves showing the effects of c illustrate that both nearly linear and threshold responses are covered over the relevant range of O3 expo sures. The α (yield) will vary with in herent differences in crop cultivar yields. The change in response with O3 concentration is shown by the expo nential part of the model, exp[—(x/ σ)°], which gives the yield response at dose JC as a proportion of the yield at zero dose, a. A priori one expects a to differ among studies, species, and cultivars. However, similarities among proportional responses of crop cultivars are of particular interest and can readily be determined by homogeneity tests on σ and c. (If σ and c show ho mogeneity, the data sets show similar proportional responses to O3.) The yields from plants given the AA treatment were compared with those from plants growing in chambers. The Weibull function was extended to in clude a parameter, 0:2, to account for chamber effects. The value of «2 is a measure of the deviation of the nonchambered plot yield from the cham bered plot yield as shown in the dose-
response curve; in effect, ai causes the proportional yield response to be based only on the chambered plots. The significance of «2 was evaluated with a t-test. Results are shown in Table 2 (see box). The estimates of Weibull parameters (shown with the * ) for individual data sets are also shown in Table 2. A more thorough description of the Weibull approach for use with the NCLAN data is found in Rawlings and Cure (13).
FIGURE 1
The relative response of five major crop species to O3 as predicted by the Weibull model"
^**
**"tlTtr"1
^
^•««,^^
Yield losses with the Weibull model The Weibull function parameters (Table 2) were used to calculate predicted yields for seasonal 7-h/d mean O3 concentrations for 17 of the 19 data sets used in the 1982 report (3) plus the nine new data sets (Table 3). The ozone concentrations chosen were 0.025, 0.04, 0.05, 0.06, and 0.10 ppm. The 0.10-ppm concentration was arbitrarily used as a maximum. The 0.04, 0.05, and 0.06 concentrations were chosen because they cover the range of seasonal mean concentrations found in many parts of the U.S. (Concentrations within the eastern U.S. vary from 0.04-0.07 ppm.) Percentage yield losses for these O3 concentrations of 0.04, 0.05, 0.06, and 0.10 ppm were then calculated relative to the predicted yield at a seasonal 7-h/d mean O3 concentration of 0.025 ppm (Table 3). In the economic analyses, comparisons were made to measured ambient O3 levels, not to a level of 0.025 ppm. When the common proportional responses were homogeneous, the Weibull function parameters for the combined models were used to calculate a percentage yield reduction for the combined data sets (Table 3).
"^*»55gj^(>^
^ ^ >
^ * * ^ » ^ ^4, Corn Wheat * N l g ^
" -,^ ^ ^ ^ iXs^W
Φ
ΪΟ
c 0 α. 85
Cotton
2
e
Ό
sa _ « 0.5
^
Soybeans
c
0
rti
Combination of data sets The Weibull model was used as the basis for testing the homogeneity of response over several data sets. If a common response model is shown for a species in Table 2, this means that the data sets show common &'s and c's but that each data set has a different â. (That is, the homogeneity of the proportional yield responses are tested.) The combined parameters were estimated by weighting so that each study contributed nearly equally to the estimates. The data sets obtained from experiments on different cultivars of cotton (irrigated, droughted) gave a positive test for homogeneity. However, the F statistic was large (3.2), and the analysis of variance showed an ozone-soil moisture interaction. Thus, these data sets were not combined.
'"'••i^g^
^^s^_
Peanuts
0
δ. g Q.
—
1
0 0 Jlf:
!
1
1
1
1
1
1
0.05 Ozone concentration (ppm)
I
ι 0.10
ι
a
The O3 concentration is the seasonal 7-h/d mean. To obtain the proportional response, the à of the Weibull model is set at a value of 1.0. Responses for soybean and wheat are based on the combined models; the response for corn is based on the combined model for PAG 397 and Pioneer 3780; the response for cotton is based on the model for irrigated cotton; and the response for peanut is based on the model for the NC-6 cultivar. The ά-and c parameters are found in Table 2.
These combination models permit the development of yield loss estimates for some of our major crop species,fiveof which are shown in Figure 1. The data for soybeans are especially meaningful because data from different cultivars, locations, and years were sufficiently homogeneous that they could be combined to give a single proportional response function. This could not be done with corn since the cultivar Coker 16 data were not homogeneous with the data for the other two cultivars. Comparison of economic losses One objective of the NCLAN pro gram is to assess the primary economic consequences of 0 3 on agriculture, using yield response data such as that developed in the preceding section. Since the four response models dis cussed in the section about crop loss functions predict different yield re sponses, the choice of response model affects the economic assessment. The primary purpose of this section is to test the sensitivity of economic esti mates to yield responses predicted by the linear (plateau-linear for corn) and
A bivariate Weibull function for predicting the effects on
crop yield caused by two pollutants By using a second response term, the Weibull function can be extended into a bivariate function that is useful for predicting the effects on yield that would be caused by treating plants with combinations of 0 3 and S 0 2 (4). The bivariate function is given by: Y=
a-exp[-(x%/ai)ci] X exp[-(x2/ff2)c2]
(2)
where the symbols have the same meaning they had in Equation 1 (sub script 1 is for 0 3 , 2 is for S0 2 ). Model fitting with the bivariate function re quires that five as opposed to three parameters be estimated. The function is based on the assumption that the two pollutants act independently, but multiplicatively, to reduce yield.
Environ. Sci. Technol., Vol. 17, No. 12, 1983
575A
TABLE 2
Open-top chamber effects and Weibull parameters for individual and common ozone dose-crop response data sets a
Crop Soybean Corsoy d Davis* Essex * Hodgson ( F ) ' Hodgson (P)' Williams 8 Common response
Weibull parameters for individual models
Chamber * effect α 2 g/plant
g/plant
-0.75 (0.92) - 2 . 2 6 (5.25) 7.04 (2.65) 1.28(1.33) 0.14(1.87) 3.10(2.33)
15.6(1.23) 31.1 (4.63) 17.1 (2.28) 15.2(7.63) 15.5(2.27) 18.7(2.43)
àc
9
bc ppm
0.129(0.01) 0.129(0.02) 0.168(0.11) 0.207(0.14) 0.153(0.03) 0.188(0.10) 0.150(0.006)
à
c
1.70(0.53) 0.91 (0.29) 1.59(1.53) 0.50 (0.54) 1.57 (1.10) 1.16 (0.94)
" " " " Corn Coker 1 6 " PAG 397 Pioneer 3780 Common response β>Λ
18.3 (8.67) 13.0(7.38) 5.9 (6.28)
Wheat Blueboy II Coker 47-27 Holly Oasis Common response g
0.93 0.70 0.75 0.32
Peanut NC-6
Cotton ' Acala SJ-2(I) Kidney bean California light red Lettuce Empire Turnip Just right Purple top white globe Shogoin Tokyo cross Common response 9 Spinach ' America Hybrid 7 Viroflay Winter bloom Common response » a
— (0.27) 1 (0.23) 2 (0.25) 2 (0.25)
—
240 (5.90) 166(3.80) 149 (3.90)
— 5.88 (0.22) 5.19(0.29) 4.95(0.17) 4.48 (0.20)
—
0.221 (0.05) 0.160(0.00) 0.155(0.00) 0.158(0.00)
4.46 4.28 3.11 3.53
(2.83) (0.72) (0.46) (0.57)
0.175(0.02) 0.171 (0.02) 0.156(0.01) 0.186(0.04) 0.174(0.01)
3.22 (1.33) 2.06 (0.68) 4.95 (2.03) 3.20(1.86) 2.90(0.78)
- 4 8 . 1 (5.80) 1
148(4.70)
0.111 (0.00)
2.21 (0.23)
- 3 . 3 0 (3.72)
41.5(4.90)
0.197(0.02)
1.12(0.42)
1.44(1.00)
16.5(1.10)
0.287 (0.09)
1.77(1.06)
0.098 (0.04)
1.22(0.71)
0.090 0.095 0.096 0.094 0.093
(0.003) (0.005) (0.006) (0.006) (0.003)
3.05 (0.65) 2.51 (0.67) 2.12 (0.64) 3.94(2.01) 2.75 (0.57)
0.142(0.021) 0.139(0.017) 0.129(0.017) 0,127(0.017) 0.135(0.008)
1.65(0.98) 2.68(1.70) 1.99(1.06) 2.07(1.17) 2.08 (0.51)
144(181) 5.57 2.93 2.56 8.99
(0.70) 1 (0.45) 1 (0.38) 1 (2.09) 1
1245(530) 10.89 (0.50) 6.22 (0.35) 4.68 (0.33) 15.25(1.30)
—
—
— — — — —
21.2(3.20) 36.6 (4.90) 41.1 (5.80) 20.8(3.10)
—
Detailed dose-response data are found in References 1, 4, 5, and 14-20. The standard error (SE) is shown in parentheses for all data; all values are ±SE. The SE was calculated using the mean square error term from the AOV. * The o