CHIAA RESEARCH REPORT NUMBER 32 RELATIONSHIP OF WEATHER FACTORS AND CROP YIELDS IN ILLINOIS

Transcription

1 CHIAA RESEARCH REPORT NUMBER 32 RELATIONSHIP OF WEATHER FACTORS AND CROP YIELDS IN ILLINOIS by Stanley A. Changnon, Jr. Illinois State Water Survey Urbana, Illinois March 25, 1966 Prepared for Crop-Hail Insurance Actuarial Association 209 West Jackson Boulevard Chicago, Illinois 60606

2 CHIAA RESEARCH REPORT NUMBER 32 RELATIONSHIP OF WEATHER FACTORS AND CROP YIELDS IN ILLINOIS Stanley A. Changnon, Jr. Illinois State Water Survey INTRODUCTION Research on the relationship of crop yields to weather conditions for application to the all-weather peril insurance program was completed in The research was divided into two separate phases. The first and principal phase concerned the determination of regions of homogeneous risk for application in insurance rating within Illinois. The use of the nine crop-reporting districts in Illinois as rate regions does not appear to be satisfactory. Problems can arise largely because several of the crop-reporting districts incorporate areas with quite different soil and weather conditions which produce wide variations in insurance risk within the districts. Thus, a single rate for a crop-reporting district is inequitable. This primary research activity in the crop-yield weather program of 1965 has been based on data concerning technology, soils, weather conditions, and other factors that affect yields. Yieldpredictive equations using 34 years of yield and monthly weather data for each Illinois county were the basic source of data that were used in evaluating and developing the regional patterns of risk. The second phase of the 1965 research has concerned analysis of weatheryield variables using data from the Water Survey's dense raingage networks. The accuracy of corn and bean yields predicted by monthly weather data and technology factors for the 400-square-mile East Central Illinois raingage network was examined in detail in 1964 and was reported in CHIAA Research Report Number

3 -2-22 (1). In 1965, the degree of association between corn yields and individual weekly and monthly weather conditions for a 9-year period from the East Central Illinois network was analyzed. A similar investigation was performed using corn yield and monthly weather data from the Little Egypt raingage network in southern Illinois. Yield data were obtained from 35 farms in the 550-square-mile network area for the years This second phase of the 1965 yieldweather research is primarily basic research performed to ascertain the degree of relationship between corn yields and various weather parameters. Although basic, this information could prove useful during a growing season to validate the predictions of low corn yields, and consequently, the possibility of future loss claims that would occur at the end of a growing season. ACKNOWLEDGMENTS The author is indebted to Philip S. Brown, Secretary of the Crop-Hail Insurance Actuarial Association, and to Harry E. Souza, Director of the Association's computer section, for their assistance in furnishing computer facilities for considerable machine anaylsis related to this research. Special credit is due Dr. J. C. Neill, statistician of the State Water Survey, who has assisted on all phases of this research. Carl Lonnquist of the Survey staff programmed the computer research required at the University of Illinois as well as aiding with the analysis. Dr. Louis M. Thompson, Associate Dean of Agriculture at Iowa State University, made valuable suggestions concerning the analysis and the contents of this report. This entire research was performed under the general direction of Glenn E. Stout, Head of the Atmospheric Sciences Section of the Illinois State Water Survey.

4 -3- REGIONAL RATE EVALUATION Introduction A requirement for determining the regions of all-weather peril insurance rates in Illinois was data from many locations throughout the state that would provide actual or empirical measures of the weather, soil, and technological variables that are the the principal factors affecting yields. and risk variations thereof are based on these variables. Obviously, yields After considering many possible data sources, it was believed that the most meaningful data, albeit empirical, could be derived from the Thompson-type (2) yield-predictive regression equation developed by the Crop-Hail Association in The data input into this equation, which was used in the network predictive studies of 1964, includes corn and soybean yields, preseason precipitation ( September-May ), May temperature, June precipitation and temperature, July precipitation and temperature, and August precipitation and temperature. The input data were for several years of record and could be for a single location or an area of any size. In performing the machine computations to obtain the predictive equations, various mathematical expressions indicating the relationship of the weather variables and technology with yields, were obtained. These expressions are empirical measures of relationships between yields and all related factors at a point or over an area. The plan to determine the regions was: to develop such equations from weather and yield data for the period at several locations, to compare the resulting expressions, and finally to group the similar expressions into regions.

5 -4- Data and Analytical Procecures To perform this study using this approach, the necessary seasonal and monthly weather data for the period were determined for each of the 102 counties in Illinois. Each seasonal and monthly weather value used for a county was an average of the weather data from all the weather stations within each county. When no data were available within a county, estimations were made using data from all weather stations surrounding the county. Once this quite laborious process had been completed, the climatological values were entered in punch cards along with the yearly bean and corn yields for each county. These cards were supplied to the Association which in turn calculated both corn yield predictive equations and soybean yield predictive equations for each of the 102 counties. Technology was treated as a curvilinear variable in the equation. Various mathematical expressions of the relationship between corn yield and the various weather factors were made available in the calculations to obtain the county predictive equations. Among these were the simple correlation coefficients, the standard errors of estimate, and multiple correlation coefficients. State maps based on the 102 county values for all these various factors were prepared and analyzed, and four of them are shown in Figure 1. These patterns are based on the simple correlation coefficients between county corn yields and county technology and selected precipitation data. When analyzed, these maps indicated various patterns of risk variance. Subjective grouping of the many patterns was not a feasible means for determining the final risk regions. Therefore, the correlation coefficients for all the variables were analyzed on the University of Illinois 7094 computer using a multivariate program. This

6 -5- statistical procedure provided one basis for the regional grouping of the county data. Other groupings were obtained using the July and August rainfall standard errors, regression coefficients, and the multiple correlation coefficients. The resulting final regional pattern was based on the variance of yields explained by weather which allows definition of areas of different risk. Additional county yield-weather factors were computer-derived on the IBM All the weather conditions were regressed with yields, and various expressions including the multiple correlations and standard errors were derived for each county. The final risk map was also used to determine which one or two readily available data patterns correlated well with it. To this end the final regional risk map was compared with : (1) the individual weather correlation coefficient maps, and (2) other maps such as those for soil types, mean rainfall, and the variability of yields. Such findings would certainly suggest the best data to use in establishing regional rate patterns in other areas (states) with similar climate and soil types so that this lengthy analysis would not have to be repeated in other comparable areas. Results Corn. The patterns based on the correlation coefficients between county corn yields and selected variables are shown in Figure 1. Technology shows an increase in correlation northward across the state. The increase in corn yields because of technological factors (which include all farm practices) in the northernmost counties, where the hightest correlations occurred, was approximately 1. 5 bushels per acre annually during the period. In the areas with correlation coefficients of less than +0. 7, the annual increase in

7 -6- corn yields due to technology was only 0.5 bushels per acre during the period. Figure lb shows the correlation pattern for corn yield and preseason (September-May) precipitation. Although the coefficients were not high, and some negative coefficients were found for several central Illinois counties, the pattern is meaningful. It indicates that preseason precipitation has some small relation with corn yields in the southern third of the state, whereas in the remaining twothirds of Illinois, the amount of preseason precipitation has lesser correlation with corn yields. It should be noted that the effect of monthly and seasonal rainfall on corn yields is curvilinear, and thus simple correlation coefficients are less than a true measure of the degree of correlation between rainfall and yields. However, they do provide a ready means by which to compare counties and to study regionality. Figure 1c presents the correlation pattern derived from coefficients for corn yields and July precipitation. July precipitation has a higher correlation with corn yields than precipitation in any other month. The correlation coefficients for August precipitation and corn yields (Fig. 1d) reveal a west-to-east decrease in the degree of correlation. In general, the patterns of Figure 1 indicate the existence of regions of considerably different response to weather as well as to technological factors, and emphasize the need for the determination of regional rate structures for all-weather peril insurance. In order to determine the value for each county that would best measure the weather-yield relationship and risk variability, a basic measure of the corn yield explained by weather variability had to be established and made comparable

8 FIG.I ISO-COEFFICIENT PATTERNS BASED ON CORRELATIONS BETWEEN CORN YIELDS AND VARIOUS COUNTY FACTORS

9 -7- between all 102 counties. It can be assumed that: Yield Variability = Weather-Soil Variability + Technological Variability + Random Variability That is, yields are dependent on the interaction of weather and soil conditions, technology factors, and certain other random factors which likely include other weather and technology variables that are not measured. It is assumed that weather conditions and soils interact as one factor:, which should be labeled a natural factor, and they were treated together in the statistical analyses. If the county corn yields are compared with the county weather factors using a regression relationship, the remaining unexplained variance (standard error 2 ) becomes that variance attributable to the technological factors and the random factors. If this unexplained variance is subtracted from the variance of the actual county corn yields (standard deviation 2 ), the remaining variance is that explainable solely by weather factors included in the regression. The square root of the explained variance divided by the county mean yield expresses the corn yield variability attributed to weather variables as a percent of county mean yield. This index of variation was used to compare counties and to derive the regional patterns shown on Figure 2. The county indices of variation were separated into five groups with approximately equal numbers of counties in each group. The degree of risk is considered directly proportional to the percent of yield explanation by weather, and thus counties in group 1 (Fig. 2), which have the highest percentages, are considered counties of maximum yield risk resulting from weather variables. Group 1, found largely in south-central Illinois, comprises 20 counties in which weather

10 -8- factors explained between 29 and 38 percent of the county mean corn yields. Comparison of this group of counties with the Illinois soil association map (3) reveals that the soils of these 20 counties are slowly to very slowly permeable and light colored. This is also the area of the state where 3-month precipitation droughts are most frequent and most severe (4). Group 5 (Fig. 2) is composed of 22 counties where weather conditions explain-) ed the least amount of variability in the corn yields (8 to 14 percent). Most of these counties are located in northwestern Illinois and have soils that are dark and moderately permeable. Most of the counties in groups 4 and 5 are located in northern Illinois which on the average is the coolest part of the state and the area least subject to droughts. Certain regional anomalies appear on Figure 2. One of these is the moderatel high risk (group 2) area composed of Kankakee, Iroquois, and Livingston counties in east-central Illinois. These counties contain large amounts of sandy, very permeable soils that are adversely affected by droughts. Livingston and Iroquois counties also contain large portions of very slowly permeable soils similar to those in south-central Illinois. Another areal anomaly is the moderately high risk value for Mason County, and this is attributed to the sandy soils of that particular county. Inspection of Figure 2 reveals that the counties surrounding these two areas have much lower risk (percentage) values, and since the weather conditions do not vary significantly, the higher risk in these areas is due to the interactions of weather with the peculiar soils of the areas. The mean percentages listed for each group on Figure 2 indicate the degree of risk or rating that could be assigned to each group. Examination of these group

11 -9- mean risk values shows that the group 1 counties have a risk three times as great as that for the group 5 counties (33 to 11 percent). If group 5 (lowest risk) is assigned a value of 1. 0 for risk or rating (1.0= 11%), then the risk in group 4 is times as great, that for group 3 is 2. 0, that for group 2 is 2. 36, and that for group 1 is 3.0. Indices of the variation of corn yields explained by technology factors also were computed after regressing corn yields with technology. These values (Fig. 3) reveal that yield variability explained by technology is between 18 and 36 percent of the mean county yields. The pattern resembles that obtained for the weather variables. Another of the computer analysis performed at the University of Illinois was a regression of county technology plus weather variables against corn yields. Most of the resulting multiple correlations were above which indicated that the major variables included in the county equations were explaining more than 80 percent of the yield variations. The variation unexplained by the weather and technology variables can be considered as that attributed to random factors, or by weather-soil and technology factors not included in the equation. The map portraying the percent of yield variability explained by these random factors (Fig. 4) reveals that a few counties in extreme southwestern Illinois had percentages greater than 20. In large portions of central and northwestern Illinois the random factors account for less than 10 percent of the corn yield variability. Comparison of the regional corn risk (rate) map (Fig. 2) with various other patterns including those for the correlation coefficients (Fig. 1) and those for individual weather conditions, such as July rainfall variability, suggest that the

12 -10- best pattern relationship was found with the map depicting the coefficient of variation of corn yields (Fig. 5). This is not surprising since the coefficient of variation is that variation in yields caused by all factors including weather-soil, technology, and random factors. The good relationship between the coefficient of variation map and the weather-corn yield variability map indicates that much of the yield variation in most counties is weather produced. The good agreement between yields can be used to map risk regions in other areas having a climate and soils similar with those in Illinois. Soybeans. A lengthy analysis similar to that described for corn was performed for soybeans to ascertain the risk regions. The index of variation, which is the variability of soybean yields explained by weather conditions expressed as a percent of the county mean bean yields, was computed for each county. The resulting map based on these county indices is shown in Figure 6. The 102 indices were divided into five groups with approximately equal numbers of counties in each group. Group 1 had the highest percentages or weather-yield risk values, ranging from 23 to 31 percent with a mean of 25 percent. The mean percentage of the corn group 1 counties was 33, and thus the eight weather variables did not explain soybean yield variations quite as well as they explained corn yield variations. The counties where weather risk to beans is greatest are in south-central and extreme southern Illinois. Low weather risk to beans is found in northwestern and eastcentral Illinois. In both areas weather explains less than 11 percent of the variability in the county mean yields. Regional anomalies similar to those noted for corn risk appear on the soybean

13 FIG. 2 VARIABILITY OF CORN YIELDS EXPLAINED BY WEATHER (SOIL) EXPRESSED AS PERCENT OF COUNTY MEAN YIELDS FIG. 3 VARIABILITY OF CORN YIELDS EXPLAINED BY TECHNOLOGY EXPRESSED AS PERCENT OF THE COUNTY MEAN YIELDS FIG. 4 PERCENT OF CORN YIELD VARIABILITY EXPLAINED BY RANDOM FACTORS (VARIABILITY UNEXPLAINED BY FIG. 5 COEFFICIENT OF VARIATION OF CORN YIELDS WEATHER AND TECHNOLOGY FACTORS)

14 FIG. 6 SOYBEAN RISK AREAS. VARIABILITY OF SOYBEAN YIELDS EXPLAINED BY WEATHER (SOIL) EXPRESSED AS PERCENT OF COUNTY MEAN YIELDS

15 -11- map, but the differences between these regions and adjacent counties do not show up as clearly. These include the relatively high risk areas in Livingston, Kankakee, Iroquois, and Mason counties. Another areal anomaly appears in northeastern Illinois where Kane, Kendall, and DuPage counties all have moderately high risk values for weather effects on their soybean yields. The mean percentage values shown on Figure 6 for the five groups can be used to compare the regional differences in risk due to weather. If the group 5 mean risk value of 5 percent is equated to 1, the risk of counties in group 4 is 2. 4 times as great as that for group 5; the risk of group 3 is 3. 0, that in group 2 is 4. 0, and that in group 1 is Thus, the rate for all-weather peril insurance in the group 1 counties should be five times the rate in the group 5 counties. The soybean risk map (Fig. 6) was compared with several other readily available patterns, including those for several weather parameters, to ascertain which correlated best and thus could be used to help develop rate regions in other areas where the county weather data needed to develop a map such as that depicted in Figure 6 would be difficult and costly to obtain. As with corn, the one available pattern that correlated best with the Illinois bean risk map was the pattern based on the county coefficients of variation of soybean yields (standard deviation/mean yield). CROP YIELD-WEATHER RELATIONSHIPS East Central Illinois Network The relationships between actual yields over a 9-year period from 108 farms and yields predicted by rainfall and temperature values in the East Central Illinois raingage network were investigated in detail in That study indicated

16 -12- the ability of the Thompson-type equation (2) and the network weather data to predict the yields at varying distances away from raingages and over areas of various sizes. This research was reported in detail in Research Report Number 22 (1). In 1965 the same rainfall and temperature data and corn yield data for the period were analyzed further to determine the degree of relationship between the individual weather variables and the resulting corn yields. It should be noted that this research was not pursued under the assumption that extremely high correlations or relationship would be established between any one weather variable and yields in the region. The complex interactions of the many factors affecting crop production obviously indicate that no single weather variable will be highly correlated with the yield. However, knowledge of the degree of correlation between weather conditions and yields from an actual group of farmers was desired. It was also hoped that further analysis into such a basic research area might have application in insurance practices, especially in the determination of the likelihood of losses before the end of the period of crop growth. The research in 1965 using the East Central Illinois network data was restricted to corn yield data from the 60 farms that had complete technology and yield records for the 9-year period of weather data. Thus, there were 540 values for each factor such as June rainfall and nitrogen application. Seasonal, monthly, and weekly temperature and precipitation data were correlated with corn yield data. Much of the basic analysis -was performed on the 7094 computer at the University of Illinois. One of the machine analyses produced a graphical

17 -13- output for each weather condition versus yield. In addition, the machine calculated the fit of the data to linear, quadratic, and cubic relationships. Certain of the monthly and weekly weather variables were found to be linearly related to yields, whereas othersshowed various curvilinear relationships. Seasonal and Monthly. Graphs showing the best-fit curves and their equations for the preaseaon precipitation and seven monthly weather variables analyzed in the East Central Illinois network are displayed in Figure 7. Although the 540 data points are not plotted on each graph, smoothed dashed lines incorporating 95 percent of the data points are shown on all eight graphs to portray the scatter of the data. Interestingly, Figure 7a indicates that either relatively low or high precipitation in the 9-month preseason period may be related to higher corn yields. Included in the indicated least desirable amounts of preseason precipitation, ranging from 17 to 25 inches, is the normal preseason precipitation value for this area of 25 inches. This result differs from that obtained by Thompson (5) for the Corn Belt which indicated that normal preseason precipitation was near optimum for maximum corn yields. The index of correlation calculated for the network cubic regression line was 0. 53, and as shown in Table 1, the standard error of estimate was bushels per acre. The data for May temperature indicated a quadratic represented the relationship (Fig. 7b) with a correlation index of If May is either relatively warm or cold, higher corn yields will result. June mean temperatures and June rainfall also fitted quadratic regression curve,s,but the spread of the dashed lines on Figure 7c and 7d, the low correla-

18 -14- tions in Table 1, and the large standard errors all indicate that the area's corn yields are poorly related to June weather conditions. Figure 7c shows that corn yields tend to decrease with increasing June precipitation, although the curve suggests that if June precipitation is anything less than one inch, yields will not vary. Figure 7d indicates the existence of a curvilinear relationship with the higher yields occurring with June mean temperatures in the range from 71 to 74 degrees. July precipitation (Fig. 7e) exhibited a curvilinear fit with corn yields. The optimum July rainfall for maximum corn yields lies between 6 and 7 inches. However, the index of correlation of indicates that July rainfall explained only 14 percent of the corn yield variability. The relationship of corn yields in the network area with the July mean temperatures was linear and the correlation coefficient of indicates that July temperature explained 25 percent of the yield variations. Unfortunately, the 9-year sampling period on the network did not include an extremely wet August for this area since the highest August rainfall measured was 6. 7 inches. The index of correlation for yields and August precipitation of in Table 1 certainly illustrates the lack of strong association between the two variables. Figure 7g indicates that corn yields were maximizing with August rainfall totals of over six inches, and the quadratic regression line suggests that additional August rainfall would not increase yields. August temperatures fit a quadratic distribution somewhat better than a linear, and as in July, area corn yields decreased with increasing August temperatures (Fig. 7h). Table 1 reveals that August temperatures had an index of 'which was the highest for the

19 -15- monthly weather variables, and was also higher than those for any of the weekly weather variables.(table 2). Weekly. Maximum temperatures and precipitation amounts by weeks were determined from the raingage and temperature stations in the network area to investigate the degree of association between corn yields and weekly weather conditions. Odell (6), using the Urbana experimental farm data, has shown that precipitation values around the time of corn tasseling (July 24) were more highly correlated with corn yields than precipitation in any other summer period. Specifically, he found that the rainfall in the three consecutive 8-day periods beginning on July 3 to be most beneficial for corn. Results of the weekly analysis based on nine years of data from 60 farms in the network, which has soil types and a climate comparable with Urbana, were not in full agreement with those of Odell. Rainfall amounts in three weeks during the June 1 through August 31 period were found to have moderately good relationships with yields, and the curves of best-fit for these three weeks are shown on Figures 8a, 8c, and 8g. Rainfall in the week of June 29 through July 5 has a positive linear relationship with corn yields, and the correlation coefficient is 0.47 (Table 2) indicates that rain in this week explains about 22 percent of the variability in corn yields. The rainfall in the following week, July 6-12 (Fig. 8c) has a curvilinear relationship with corn yields. Corn yields increase as rainfall amounts increase up to 1. 3 inches, but as rainfall amounts increase beyond 1. 3 inches, the corn yields decrease. These weekly rainfall amounts explain 26 percent of the corn yield variations.

20 -16- TABLE 1 CORRELATION INDICES AND COEFFICIENTS AND STANDARD ERRORS OF ESTIMATE OF CORN YIELDS, AND VARIOUS SEASONAL AND MONTHLY WEATHER CONDITIONS IN EAST CENTRAL ILLINOIS NETWORK Coefficient or Standard Index of Error Weather Condition Correlation (Yield) Preseason Precipitation May Mean Temperature June Precipitation June Mean Temperature July Precipitation July Mean Temperature August Precipitation August Mean Temperature TABLE 2 CORRELATION INDICES AND COEFFICIENTS AND STANDARD ERRORS OF ESTIMATE OF CORN YIELDS AND VARIOUS WEEKLY WEATHER CONDITIONS ON EAST CENTRAL CENTRAL ILLINOIS NETWORK Coefficient or Standard Weather Condition Week Index of Correlation Error (Yield) Precipitation 6/29-7/ Precipitation 7/ Precipitation 8/ Max. Temperature 6/29-7/ Max. Temperature 7/ Max. Temperature 7/ Max. Temperature 8/ Max. Temperature 8/

21 -17- These two early weeks of indicated rainfall-yield relationships are 5 to 1 0 days earlier than those indicated by Odell. In the 9-year network period the average date of tasseling was July 24 which was the same date reported by Odell. Network rainfall in the three weeks from July 13 through August 3, which include two of the 8-day periods shown to be important by Odell, had no correlation with corn yields. Another good nonlinear relationship between weekly rainfall and corn yields was found for the week of August 3-9 (Fig. 8e). The correlation index for this rainfall data and corn yields was and the standard error of estimate was bushels per acre. There were five weeks which had particularly good correlations between their mean maximum temperatures and corn yields, and the best-fit regression curves are indicated on Figure 8. These included the maximum temperatures for the three weeks with good precipitation relationships plus those for the weeks of July and August In the two early weeks which covered the June 29-July 12 period, curvilinear relationships existed between weekly maximum temperatures and the corn yields with optimum temperatures occurring at 85 F. For the July week and those in August, corn yields decreased with increasing mean maximum temperatures. The mean maximum temperatures for the five weeks shown in Table 2 produced correlation indices or coefficients ranging from to The temperature correlations for the three weeks with precipitation correlations were slightly higher than those obtained for precipitation.

22 -18- TABLE 3 CORRELATION INDICES AND COEFFICIENTS AND STANDARD ERRORS OF ESTIMATE OF CORN YIELDS AND VARIOUS SEASONAL AND MONTHLY WEATHER CONDITIONS OF LITTLE EGYPT NETWORK Coefficient or Standard Index of {Error Weather Condition Correlation (Yield) Preseason Precipitation May Mean Temperature May Precipitation June Mean Temperature June Precipitation July Mean Temperature July Precipitation August Mean Temperature August Precipitation Little Egypt Raingage Network A similar analysis using corn yield data for the period of from 35 farms in the Little Egypt raingage network was performed in These data, although a smaller sample size than that from the East Central Illinois network, were selected for study because this southern Illinois network is located in an area with a rainfall climate and soil types quite different from those in central Illinois. Thus, the results should provide a measure of the weather-yield relationships for a significantly different agricultural region. The computer analysis of the corn yields and the seasonal and monthly weather data from this network was performed in much the same fashion as that

23 -19- for the East Central Illinois network. Figure 9 presents graphs similar to those in Figure 7 showing the regression curves of best-fit for eight weather variables. The resulting correlations and standard errors are listed in Table 3. Figure 9a reveals that the preseason (September-May) precipitation has a curvilinear relationship with corn yields, but the low correlation index, 0. 16, in Table 3 suggests that only a very minor relationship exists. In general, the curve has a shape similar to that obtained for preseason precipitation in;the East Central network (Fig. 7a). Corn yields in southern Illinois decrease with increasing May mean temperatures, and the correlation coefficient of indicates that May temperatures explain 19 percent of the variability in corn yields. As shown in Table 3, mean temperature and precipitation in June do not have high correlations with corn yields, and these results agree with those found for June weather and corn yields in the East Central Illinois network. Figure 9d reveals that June precipitation values ranging from 2. 5 inches to 4. 5 inches are associated with higher corn yields. Totals higher than 4. 5 inches and lower than 2. 5 inches are associated with lower corn yields, and in general, this tendency for lower yields with heavier June rainfall compares favorably with the result obtained for the central Illinois network (Fig. 7c). The July mean temperature has a poor correlation of with corn yields, but the curve (Fig. 9e) reveals that yields tend to decrease with increasing temperature. It should be noted that the 7-year sampling period did not include a wide range of July temperautre values, which may partially account for the poor correlation. The adequacy of the network sampling periods is treated in detail

24 -20- in the next section of this report. The July precipitation has a good correlation with corn yields (Fig. 9f) and the correlation index of indicates that the precipitation in July explained 53 percent of the variability on local corn yields. The standard error of estimate of 14.6 bu. /acre is also considerably lower than others shown in Table 3. The optimum July rainfall for corn yields is between 6 and 8 inches, and either lower or higher precipitation in July is associated with lower corn yields. The curvilinear fit of the July rainfall data in the southern network was also found in the central Illinois network, but the correlation index in central Illinois (Table is considerably lower. August mean temperature in the Little Egypt network provides the second highest correlation value shown in Table 3, but it is not as high as the correlation between August temperatures and yields found in central Illinois. August precipitation has a curvilinear fit with yields, but the correlation index of is relatively low and the standard error (Table 3) is large. In general, the shape of the quadratic regression for August precipitation in the East Central Illinois network is similar to the cubic regression curve shown for southern Illinois (Fig. 9h). As August precipitation values increase above 8 inches, yields lower rapidly. Representativeness of Network Weather Conditions One of the factors that affect the yield-weather results derived for the two networks is the relatively short sampling periods and the resulting potential inadequate sampling of weather extremes. For this reason, the range of monthly values obtained in the 9-year central Illinois network sample and in the 7-year

25 -21- southern Illinois sample were compared with long-term point weather records in the network areas. Thompson (5) has indicated that the period in Illinois was a uniquely good corn-weather period. The data sampled from the East Central Illinois network provided a range of values for preseason precipitation, May temperatures, June precipitation, and July precipitation that are representative of the extremes that have occurred in this area over a 60-year period. However, 10 percent of the time the mean temperature in June, July, and in August have been higher than the highest sampled in the 9-year period, and the results obtained for these monthly mean temperatures are probably somewhat affected by this lack of extremes. Similarly, August rainfall totals higher than the maximum value obtained (6. 7 inches) have occurred 5 percent of the time in past years. The 7-year period of data from the Little Egypt network provided values of May temperatures, July precipitation, and August precipitation that incorporated the extremes of record for the area. However, the sample of preseason precipitation, June temperature, July temperature, and August temperature were quite unrepresentative, and data for June precipitation also was somewhat nonrepresentative because they failed to sample 5 percent of the very high and 5 percent of the very low June amounts that have occurred in the area. Preseason precipitation in the southern area has ranged from a low of 19 inches to a high of 62 inches, and yet the sampling period gave a range from 23 to 36 inches. The June mean temperatures in the area have been higher than the highest sampled (77 degrees) 25 percent of the time and lower than the lowest sampled 10 percent of the time. Mean temperatures in the seven Julys in the

26 period were notably cool and quite similar. Fifty percent of the time July temperatures have been higher than any sampled in the 7-year period, and 10 percent of the time they have been lower. August mean temperatures in the period also represented inadequate sampling. The long-term area records reveal August temperatures have been higher than the maximum value sampled (79.2 degrees) 25 percent of the time and went lower than the lowest sampled (74. 5 degrees) in 10 percent of the years. Thus, the correlation results for the June, July, and August mean temperatures obtained for the southern Illinois network cannot be considered representative, and those obtained for temperatures in the central Illinois network are also somewhat less than representative.

27 FIG. 7 RELATION OF CORN YIELDS WITH SEASONAL AND MONTHLY WEATHER CONDITIONS IN EAST CENTRAL ILLINOIS NETWORK,

28 FIG. 8 RELATION OF CORN YIELDS WITH WEEKLY WEATHER CONDITIONS IN EAST CENTRAL ILLINOIS NETWORK,

29 FI6.9 RELATION OF CORN YIELDS WITH SEASONAL AND MONTHLY WEATHER CONDITIONS IN THE LITTLE EGYPT NETWORK,

30 -23- REFERENCES 1. Changnon, S. A., Prediction of Corn and Soybean Yields Using Weather Data. CHIAA Research Report 22, Chicago, Illinois. 2. Thompson, L. M., Weather and Technology in the Production of Corn and Soybeans. CAED Report 17, Iowa State University, Ames. 3. Wascher, H. L., J. B. Fehrenbacher, R. T. Odell, and P. T. Veale, Illinois Soil Type Description. AG-1443, Agricultural Experiment Station, University of Illinois, Urbana. 4. Huff, F. A., and S. A. Changnon, Drought Climatology of Illinois. Bulletin 50, Illinois State Water Survey, Urbana. 5. Thompson, L. M., Weather Variability and the Need for A Food Reserve. CAED Report 26, Iowa State University, Ames. 6. Odell, R. T., "Effects of Weather on Corn and Soybean Yields, " Illinois Research, University of Illinois Agricultural Experiment Station, Urbana.