|
|
Impact |
© 2008 Plant Management Network. Forecasting Corn Yield for Eleven States in the Corn Belt: Results for 2001-2005 E. C. A. Runge, Professor and Billie Turner Chair in Production Agronomy (Emeritus), Soil & Crop Sciences Department, Texas A&M University, College Station 77843-2474; and John F. Benci, Consultant, 382 Moore Road, West St. Paul, Manitoba, Canada R4A 7A2 Corresponding author: E. C. A. Runge. e-runge@tamu.edu Runge, E. C. A., and Benci, J. F. 2008. Forecasting corn yield for eleven states in the corn belt: Results for 2001-2005. Online. Crop Management doi:10.1094/CM-2008-0609-03-RS. Abstract Using a model to accurately predict corn yield and production under variable soil and climatic conditions as early as possible has been a goal of agronomists, climatologists and economists for years. The task is difficult because data used to drive most models are not generally collected or available. Physiologic models are desirable for assessing crop development but are difficult to apply to large areas because of date base requirements. The objective of this study was to apply a model for predicting yield and production in 11 corn belt states (Illinois, Indiana, Iowa, Kansas, Michigan, Minnesota, Missouri, Nebraska, Ohio, South Dakota, and Wisconsin). The model uses water supply based on plant-available stored soil moisture at planting and rainfall. Water demand is based on maximum air temperature. These model inputs serve as proxies for evapotranspiration. The model successfully predicted corn yield and production for 2002, 2003, and for the record corn yields in 2004. However, model forecasted yields for 2005, a year of considerable drought stress, under estimated yields by approximately 10%. Sources of error contributing to under prediction of yield and production in 2005 are discussed. Improvement in drought tolerance of today’s corn hybrids is not incorporated into the model and may help explain the under-estimation of forecasted yield in 2005. Introduction The model was derived by growing corn under variable soil moisture conditions at four locations along a 200-mile transect in Illinois from 1969-1971 (6,7). Data on which the model is based captured the effect on corn yield of variable amounts of plant-available stored soil moisture at planting under constant and variable rainfall and temperature during the growing season (6,7). The model used in this study explained 81% (R²) of the corn yield variation (7). Further information on model development can be found in (6,7). DeKalb’s XL45 and 45a hybrids were among the first widely adapted high population single cross hybrids grown over wide geographic areas. Physiologic models are desirable for assessing crop development and ultimately yield but are difficult to apply to large areas because of data base requirements (1,2,5). Plant populations have increased since the model was developed and are one of the main contributors to increased corn yields (4). Assessment of Plant-Available Stored Soil Moisture at Planting, and Climate Data on Corn Yield for the Corn Belt The assessment of plant-available stored soil moisture at planting for the Corn Belt was determined by examining soils for each county using county soil survey reports and more generalized state soil association maps. Two problems present themselves in such an analysis. First, one must determine which soils are being used for corn production and second the amount of plant-available stored soil moisture at planting supplied to the corn plant by these soils must also be determined. In the western Corn-Belt winter and early spring precipitation were used to determine the amount of plant available stored soil moisture at planting. The states of Kansas, Nebraska, and South Dakota and parts of western Iowa had some areas during some years where precipitation did not bring the soils to field capacity. The water balance model utilized in this study uses two climatic inputs; weekly average maximum daily temperature and total weekly precipitation. Real time data for these two weather variables are readily available. Model weeks in our study went from Sunday through Saturday. Weather data for the previous week were available by Tuesday of the following week and calculations were completed by Wednesday. Approximately 100 weather stations were available each week for each state with the range being from 75 to 125 stations per state providing a total of about 1100 weather stations each week. Calculation of county data. All data used in the model were county based
using the procedure outlined in (3,8). The challenge is to estimate weather
for each county in each state each week. The geographic center of each county
was determined and given latitude and longitude coordinates. Likewise, all
weather stations were given latitude and longitude coordinates. A major
difficulty in utilizing weather station or point data in predictive models for
large areas is the variability of these parameters, especially precipitation, in
space and time. An objective technique was utilized to remove undesirable high
frequency noise in the data by computing a space average that interpolated data
from an irregular weather station network to a county midpoint. The objective
analysis scheme adopted was previously reported (3,8). In the analysis the
value at the county grid or midpoint is determined by interpolating weather
station data encircled within a searching radius. In application, the value of
maximum temperature or precipitation at a grid point is determined by taking a
distance weighted average of the weather station data encircled with the
searching radius as: State average corn yields were calculated by multiplying model derived yields for each county by the county acreage, adding all county production together and dividing that total by the total acreage of corn harvested for the state. Model Sensitivity, Relating Experimental Technology to Present Day Technology and 2001 Results The model used is based on plot yields from four Illinois farm locations for three years, 1969-1971 (6). Thus, the corn yields predicted by the model are for 1969-1971 experimental technology. Improvement in corn production technology has surpassed the experimental technology used according to county yield data reported by www.nass.usda.gov in 2001 (6,7). State corn yields from 1960-2000 (www.NASS.USDA.gov) were used to establish trend lines for each state. Examples are given in Figure 1 for Illinois, Iowa, Missouri, and Michigan. Linear trend explained from up to 87% (KS) of the annual corn yield variation from 1960 to 2000. Yield increases per year ranged from 1.48 for Wisconsin to 2.55 bu/acre per year for Kansas. The largest yearly yield increases for this 40-year period occurred in those states that increased acreage of corn grown under irrigation from 1960 to 2000 (Kansas and Nebraska).
In addition, county yields from 1960-1998 (NASS-USDA) for three counties were selected within each of the 11 states to gain an appreciation of how well the model might be adapted to present day Corn-Belt technology in each of these 11 states in Figure 2 (9). Model yields were regressed the same as the trend line for actual county yields and the results correlated. R² values ranged from 63 to 92% for county actual versus model derived yields. Examples for Black Hawk Co., IA; Huntington Co., IN; Marshall Co., KS; and Williams Co., OH, are given in Figure 2.
Results for 2001. Results for 2001 served as a test to see how well the model might work in forecasting state average corn yields. It was not possible to obtain county planting date data for 2001, so 25% of the corn was assumed to be planted in each of four consecutive weeks. The consequence is that results do not necessarily mimic corn planting dates that actually occurred and reduced the correlation between actual and model derived corn yields. Final forecasted corn yields during the growing season, using procedures outlined in (9), are given in Table 1 and weekly results are given in Figure 3 for Illinois, Iowa, Minnesota, and Missouri as a percentage of the 2001 trend yield. Weather for 2001 was compared to average weather for each state. Results reported show that average weather gives yields that are 7 to 10% higher than trend line yield (NASS-USDA concurs with this assessment; Carol House, personal communication). Consequently, model derived yields for 2001 were low by from 7 to 10% for all states except Missouri.
2001 conclusions. Many compromises were made to generate the model forecasted yields for 2001. The compromises that impact results include: (i) No planting date data were available to center weather data used in the model to 6 weeks before to 4 weeks after tasseling; (ii) Weather was compared to average weather for the state (this alone reduced model calculated yields by 7 to 10%); and (iii) Data for weather station locations outside of but in adjacent states could not be considered because of inadequate computer memory. All problems from 2001 were resolved and did not impact results in future years. Results for 2002 Planting date and tasseling (silking) date data were made available by NASS-USDA. Planting date data were more complete than were tasseling date data. Yield forecasts for 2002 using tasseling date data forecasted from planting date data are compared to tasseling date data provided by NASS-USDA. In general both ways of determining tasseling dates produced similar results except for SD and MO. Yield forecast results for 2002 are reported as percent of trend line yield for each state. Weather data for 2002 were compared to 2001 weather data rather than using the average weather data for each state as was done in 2001 using the excel based spread sheet model (9). Model forecasted yields for planting date and tasseling (silk) date data are given in Table 2. Weekly changes in model yield forecasts for Illinois, Nebraska (irrigated and non-irrigated), Ohio, and Wisconsin are given in Figures 4. Weekly results for other states are available by request from the corresponding author, E. C. A. Runge. NASS-USDA was particularly interested in model forecasted yields for August and September since NASS’s October forecast is the first yield forecast based on actual harvested samples.
2002 conclusions. Several problems were encountered during the 2002 growing season: (i) Prediction of the tasseling date using planting date data did not agree with silking date data reported for South Dakota by NASS-USDA. This discrepancy may be due to replanting of corn. A similar problem was encountered in Missouri. (ii) If the amount of plant-available stored soil moisture at planting assigned for parts of states is too high or too low the final NASS-USDA reported county and/or crop reporting district yields should provide information. If the assigned plant-available stored soil moisture at planting is larger than it should be, the model forecasted yield should be larger than the NASS-USDA reported county or crop reporting district yield and vice versa. Such comparisons provide a check on soil moisture assigned to counties. No adjustments needed to be made. (iii) It was assumed that irrigated corn growers in Kansas and Nebraska applied 1.3 inches of water each week. This may or may not be correct. NASS-USDA did provide growing season yield forecasts for irrigated and non-irrigated corn in Nebraska for the 2002 year. They did not provide such data for Kansas nor for Kansas and Nebraska in any future year in our study. (iv) Results from the model are in general agreement with the final state yields for these 11 Corn-Belt states except for Iowa, Minnesota, and Ohio. Everyone, including NASS-USDA, underestimated yield and production for Iowa and Minnesota, and over estimated yield for Ohio. Iowa and Minnesota had cool temperatures with clear skies and had a much longer than average maturation period. The result was very high test weights and high yields. The final NASS yield for Ohio was very low and may be due to the assumption that corn was produced on acreages that were drowned out or had low plant populations. Both circumstances were not accounted for by the model used in making yield forecasts reported here for Ohio. (v) The model has the advantage of forecasting final state yields 4 weeks after the corn tassels. Results for 2003 In 2003 tasseling dates were estimated from planting date data furnished by NASS-USDA. The weather was cooler than average and tasseling dates were adjusted by adding one week later when tasseling (silking) date data were made available (NASS-USDA). The effect of weather on corn yield for 2003 was compared to weather that occurred in 2001 and 2002 (9). Results are reported separately and as the average for the two years. Yield forecasts for 2003 are reported in bu/acre rather than as percent of trend line yield. Model forecasts and NASS-USDA yield forecasts for all states for 2003 are given for August, September, and October in Table 3. Final NASS yields for 2003 (January 2004) are also given in Table 3. Weekly changes in yield forecasts for Illinois, Iowa, Ohio, and Wisconsin are given in Figure 5. Weekly results for other states are available by request from the corresponding author, E. C. A. Runge.
2003 conclusions. Several problems were encountered during the 2003 growing season: (i) Weather for 2003 was cooler than average; therefore, it was necessary to adjust the tasseling date by adding 1 week to the period from planting to anthesis. (ii) Weather late in the growing season for 2002 for Iowa and Minnesota resulted in higher than expected corn yields and occurred outside our model time interval; therefore, 2003 yields based on NASS-USDA 2002 yields for these states are too high. (iii) Unusual conditions that occurred in Ohio in 2002 led to lower than expected corn yields; consequently, model forecasted yields for 2003 using NASS-USDA 2002 final yields are too low. See Table 3 for a comparison of yield forecasts based on 2001 and 2002 weather comparisons. (iv) Final model yield forecasts (September in Table 3) are in general agreement with NASS-USDA October yield forecast. The advantage of model forecasted yields is that results are available earlier in the growing season than is possible with NASS-USDA sample based yield forecasts. Results for 2004 The 2004 year was an unusually productive one for the entire Corn Belt as each of the 11 states set new yield records. Two unusual events need to be noted: (i) Temperatures were lower than average and precipitation was average or above; and (ii) Very cool temperatures from 10 to 27 August (maximum daily temperatures in the 60°F range) served to increase model forecasted yields in northern states and northern parts of the more central states. These low maximum temperature data are outside the data set used in generating the model. Also there was no moisture shortage so the cool weather only served to slow corn maturation but did not increase yields as the model predicted. Yields were adjusted for these unusual weather conditions and are reported as adjusted yields. Two yields are given for September and October forecasts as model and adjusted yields for all states in Table 4 and Table 6. The adjustment reduced yields from 25 bu/acre for Wisconsin to zero bu/acre for Missouri. Temperature data used to derive the model are given in Table 5 and were much higher than the maximum daily temperatures that occurred for more than two weeks in August in 2004 (in the 60° range) (5). Minimum temperatures were near freezing in the more northern states. The weather for 10 to 27 August was within the 10-week interval used by the model to forecast yields for Michigan, Minnesota, and Wisconsin, was partially included in Illinois, Indiana, Iowa, Kansas, Nebraska, Ohio, and South Dakota, and was not part of the 10-week interval for most of Missouri: no yield adjustment (Table 4). The adjustment in yield ignored increases in model forecasted corn yield for these 2 weeks of very cool weather since there was no moisture deficit and low temperatures did not increase yield. In 2004 tasseling dates were estimated from planting date data furnished by NASS-USDA and were not adjusted when silking date data became available. The effect of weather on corn yield for 2004 was compared to weather that occurred for 2002 and 2003 and the average yield is reported in Table 4 and Figure 6 (9). Comparisons with 2001 weather were not made in 2004. Weekly results for other states are available by request from the corresponding author, E. C. A. Runge.
2004 conclusions. Several problems were encountered during the 2004 growing season. (i) Weather for 2004 was much cooler than average; therefore, model forecasted yields were adjusted downward except for Missouri. (ii) Weather late in the growing season not included in our 10-week model for 2002 for Iowa and Minnesota resulted in higher than expected corn yields (unusually high test weight of corn in 2002). Therefore the 2004 model yield based on NASS-USDA 2002 final yield data for these states was adjusted downward for the 2004/2002 comparison. Unusual conditions occurred in Ohio in 2002 (88 bu/acre average yield) led to a low model forecasted corn yield for the 2004/2002 comparison so the Ohio forecasted yield for 2004 needed to be adjusted upward. The need to adjust the model forecasted yield upward for the 2004/2002 comparison is confirmed by NASS-USDA’s October forecasted corn yield for Ohio. (iii) Model forecasted corn yields for August and/or September in Table 4 are in general agreement with NASS-USDA’s October yield estimates after they were adjusted for cool weather. (iv) August/September model forecasted yields are in general agreement with NASS-USDA’s October forecasted yields for much of the Corn Belt and are available some one to two months earlier in the growing season. Results for 2005 The contrast in weather between the 2004 and 2005 growing season raised many concerns about how low yields were going to be in 2005. As the harvest started farmers found higher yields than expected. Yields were much lower in 2005 than they were for 2004, but model forecasts underestimated yields for every state. Model yields are given in Table 6 along with those for NASS-USDA forecasts for August, September, and October in 2005, and for January 2006 (the NASS-USDA yield for January 2006 is accepted as the final yield for 2005). Weekly changes in model forecasted yields for 2004 and 2005 are given in Figures 7 for Illinois, Iowa, Minnesota, and Ohio. All forecasted yields for 2005, including NASS-USDA, under estimated yield and production for 2005 for most states.
It was hoped that model forecasted yields in drier than average years would also give early estimates of actual yields as occurred in 2004 when weather was favorable and yield records were set for all 11 states. That was not the result and may be due to two factors. First, the model is derived from corn grown during 1969-1971, and subsequent improvement in drought tolerance of corn hybrids would not be incorporated into the model leading to underestimation of yield in drought years. Second, the underestimation of corn yield in 2005 is due to how 2004 final NASS-USDA yields and weather data were used in the model when comparing 2005 with 2004 (9). In the 2005 model comparison, final NASS-USDA yields for 2004 were used along with the actual weather that occurred in 2004. As discussed previously 2004 model yields were adjusted downward by as much 25 bu/acre in Wisconsin (Table 4); however, by using actual weather that occurred in 2004 in the 2005 comparison, yields in 2005 were underestimated. This problem could have been avoided by using actual unadjusted model forecasted yields for 2004 rather than the final NASS-USDA yields for 2004 since the 2004 weather data could not be changed (Table 4). Using unadjusted model yields for each state for 2004 to calculate 2005 forecasted corn yields would have reduced considerably the under prediction of yields in 2005. The forecasted yields for 2005 would compare to NASS-USDA’s final yield for 2005 as follows (all in bu/acre): Illinois, 4 instead of -11; Indiana, -16 instead of -26; Iowa, -11 instead of -30; Michigan, -17 instead of -22; Minnesota, -2 instead of -18; Missouri, no change; Ohio, +1 instead of -16; South Dakota, +4 instead of -17; and Wisconsin, +3 instead of -22. When yields fluctuate due to factors not incorporated in model inputs, yields must be adjusted accordingly. The high yields for Iowa and Minnesota in 2002, the low yield for Ohio in 2002, and the high yields for 2004 were influenced by events not incorporated in the model. When using the excel model (9) to forecast yields in future years, results will be more accurate if factors not used in the model are taken into consideration. Farmers, advisers and scientists using the model for individual fields or studies are not likely to have the same problems encountered here for large areas. 2005 conclusions. The model forecasted yields for 2005: (i) Underestimated the actual corn yield for all 11 states by approximately 10%. (ii) The underestimation has two components. First is the improved drought tolerance of corn hybrids since the model was derived. Secondly the unusually cool weather that occurred in August 2004 did not increase yields due to a complete lack of drought stress during the cool weather period. Therefore, the use of NASS-USDA final 2004 yields without being able to adjust 2004 weather lead to underestimation of 2005 yields. The 2004 unadjusted model yields should have been used instead of NASS-USDA final state yields to make the 2005 comparison. If that would have been done, 2005 yield underestimation would have been reduced or even eliminated in some states. (iii) Model forecasted yields will be more accurate if factors influencing corn yield outside model inputs can be quantified. This is due to the fact that current year model forecasted yields are calculated by comparing them to previous year’s yield and weather (9). Final Conclusions The model developed earlier (7,9) captures most yearly variations in corn yield in 11 Corn-Belt states of Illinois, Indiana, Iowa, Kansas, Michigan, Minnesota, Missouri, Nebraska, Ohio, South Dakota, and Wisconsin in which it was applied. The following comments are made: (i) The model forecasts a final yearly corn yield 4 weeks after tasseling (anthesis date). This is considerably earlier than forecasts made from sample data used by NASS-USDA. (ii) Model performance over large areas like the Corn Belt was reassuring. Performance will be more accurate if factors influencing yield not incorporated into model inputs are taken into account. For example, in some years factors impact corn yield that are not repeated in most years and are not a part of model inputs. For example, high test weight corn in Iowa and Minnesota in 2002 was due to a cool and prolonged maturation period for the corn crop. This phenomenon was not repeated in subsequent years of the study. The low corn yield for Ohio in 2002 was due to drowned out and low stand corn areas. Evidently low stand levels occurred in many fields and decreased the overall state yield. The under estimation of yield in 2005 was partially due to the unusually low temperatures from 10 to 27 August. 2004. (iii) Data for tassel/silking dates are not available over large areas like the Corn Belt and it was better to forecast a tassel/silk date from planting date data. Model use requires an accurate forecast of the tasseling date. NASS-USDA data are reliable for planting date but are incomplete and less reliable for silking dates. (iv) Weather station data are used to calculate county weather data. Additional weather stations would improve estimation of county weather data and improve the county yield forecasts. (v) Assessment of plant-available stored soil moisture at planting for wide areas is a challenge but was not a problem in this study. (vi) The model correctly forecasted corn yields in most states in most years with the exceptions noted above. Model forecasts for individual fields should be more accurate than for the entire Corn Belt. Individual farmers, advisers and scientists will find the model useful in predicting yields for their fields or test plots. On-site weather data, at least rainfall, needs to be assessed for each field or location to obtain accurate yield forecasts. Acknowledgment Financial support for the study in 2002, 2003, and 2004 from NASS-USDA, Washington, DC, is gratefully acknowledged. Literature Cited 1. Anapalli, S. S., Ma, L., Nielsen, D. C., Vigil, M. F., and Ahuja, L. R. 2005. Simulating planting date effects on corn production using RZWQM and DERES-maize models. Agron. J. 97:58-71. 2. Baez-Gonzalez, A. D., Kiniry, J. R., Maas, S. J., Tiscareno, M. L., Macias, C. J., Mendoza, J. L., Richardson, C. W., Salinas, G. J., and Manjarrez, J. R. 2005. Large-area maize yield forecasting using leaf area index based yield model. Agron. J. 97:418-425. 3. Benci, J. F., Runge, E. C. A., Dale, R. F., Duncan, W. R., Curry, B., and Schaal, L. A. 1975. Impacts of climate change on the biosphere. Pages 4.3-4.37 in: CIAP Monog. 5, Part 2: Climatic Effects. Dept. of Transportation, Climatic Impact Assessment Prog., Off. of the Secr. of Transportation, Washington, DC. 4. Duvick, D. N., and Cassman, K. G. 1999. Post-green revolution trends in yield potential of temperate maize in the north-central United Statees. Crop Sci. 39:1622-1630. 6. Leeper, R. A., Runge, E. C. A., and Walker, W. M. 1974. Effect of plant-available stored soil moisture on corn yields. I. Constant climatic conditions. Agron. J. 66:723-728. 7. Leeper, R. A., Runge, E. C. A., and Walker, W. M. 1974. Effect of plant-available stored soil moisture on corn yields. II. Variable climatic conditions. Agron. J. 66:728-733. 8. Runge, E. C. A., and Benci, J. F. 1975. Modeling corn production: Estimating production under variable soil and climatic conditions. Pages 194-214 in: Proc. of the 30th Ann. Corn and Sorghum Res. Conf. Am. Seed Trade Assoc. Washington, DC. 9. Runge, E. C. A., and Benci, J. F. 2008. User friendly corn yield prediction model. Online. Crop Management doi:10.1094/CM-2008-0609-04-RS. |
