2019 Corn Yield Forecasts: Approach and Interpretation of Results

2019 Corn Yield Forecasts: Approach and Interpretation of Results

Patricio Grassini, UNL Associate Professor of Agronomy and Horticulture, Extension Cropping System Specialist and Water for Food Institute Fellow  |  Jose Andrade, UNL Research Assistant Professor of Agronomy and Horticulture  |  Juan Ignacio Rattalino Edreira, UNL Research Assistant Professor of Agronomy and Horticulture  |  Gonzalo Rizzo, UNL Visiting Scholar  |  Haishun Yang, UNL Associate Professor of Agronomy and Horticulture and Water for Food Institute Fellow  |  Keith Glewen, Nebraska Extension Educator  |  Jennifer Rees, Nebraska Extension Educator  |  Jeff Coulter, Professor and Extension Specialist, University of Minnesota  |  Joe Lauer, Professor, University of Wisconsin-Madison  |  Mark Licht, Extension Cropping System Agronomist, and Sotirios Archontoulis, Assistant Professor, both at Iowa State University  |  Ignacio Ciampitti, Crop Production and Cropping System Specialist and Assistant Professor of Agronomy, Kansas State University  |  Ray Massey, Extension Professor, University of Missouri  |  Peter Thomison, Extension Specialist and Professor, Ohio State University

Corn Belt map showing Locations of forecasted sites during the 2019 crop growing season.
Figure 1. Locations of forecasted sites during the 2019 crop growing season.

The Yield Forecasting Center (YFC) will provide real-time information on corn phenology and forecasts of corn yield potential every three weeks, starting in mid-July, to aid growers and ag industry in making management, logistics, and marketing decisions through the 2019 season. The YFC platform consists of a core team at the University of Nebraska-Lincoln in collaboration with agronomists and extension educators from universities throughout the Corn Belt. Forecasts will be provided for 41 locations (Figure 1), including separate forecasts for rainfed and irrigated corn in regions where both irrigated and rainfed production are important (Nebraska and Kansas). This article summarizes the methodologies used in the YFC to forecast corn phenology and yield and provides guidelines for interpreting the results. 

The Approach

The YFC relies on

  1. a network of collaborators providing local management data and verifying forecasted yields,
  2. precise information on dominant soil types in each region,
  3. measured high-quality, real-time weather data, and
  4. a well-validated crop simulation model (UNL Hybrid Maize).

More information about data sources and modeling can be found at the Global Yield Gap Atlas website.

Local agronomists and extension educators provide information on management in each state and help validate and interpret the forecasts. Information provided by collaborators includes site-specific average planting date (i.e., average calendar date at which 50% of the corn area was planted in the current season), plant density, and hybrid maturity (Table 1). When both rainfed and irrigated production occurs at a location, separate sets of management practices are used for simulating rainfed and irrigated crops at those sites because the water regime has a large influence on production practices. Management data have been further validated using information provided by DuPont Pioneer agronomists, especially with regard to hybrid maturity and plant density. Yield forecasts are based on the two to three dominant soil types at each location. Yield is forecasted separately for each soil type. Yield predictions are subsequently aggregated by averaging the forecasts across soil types, after weighting them according to the prevalence of each soil type in each location (Table 2).

Historical (the last 20+ years) and real-time daily weather data are used for phenology and yield forecasting. Daily weather variables required for simulating real-time crop growth and development include solar radiation, maximum and minimum temperature, precipitation, relative humidity, and wind speed. The YFC relies on measured data collected through state weather networks, including the High Plains Regional Climate Center (HPRCC), the National Weather Service station network (NWS), the Illinois Water and Atmospheric Resources Monitoring Program (WARM), the Ohio State University, Ohio Agricultural Research and Development Center Weather Service (OARDC),  the Indiana Purdue Automated Agricultural Weather Station Network (PAAWS), Purdue Automated Agricultural Weather Station Network (PAAWS), the Southern Research and Outreach Center (SROC) and the Southwest Research and Outreach Center (SWROC) form the University of Minnesota, the Missouri Mesonet (AgEBB), the North Dakota Agricultural Weather Network (NDAWN), and the Michigan State University Enviro-Weather (Figure 1). Weather stations selected from these networks are located in agricultural areas, rather than in urban areas, which helps to ensure the representativeness of the weather data for forecasting corn yields and phenology. Details on the data inputs used as the basis for the forecasts are available in Morell et al (2016).

Graph showing validation of Hybrid-Maize simulations of final yield (bushels per acre) against actual yields in optimally managed crops, under irrigated and rainfed conditions.
Figure 2. Validation of Hybrid-Maize simulations of final yield (bushels per acre) against actual yields in optimally managed crops, under irrigated and rainfed conditions. Adapted from Yang et al (2017).

Hybrid-Maize is a corn simulation model developed by UNL researchers. It simulates daily corn growth and development and final grain yield under irrigated and rainfed conditions. The model estimates "yield potential," which is the yield obtained when the crop is not limited by nutrient supply, diseases, insect pressure, or weed competition — conditions that represent an "optimal management" scenario. It also assumes a uniform plant stand at the specified plant population and no problems of flooding or hail. Although the model can account for drought stress and high temperatures during vegetative growth and grain filling, it will be unlikely to portray well crop yield in two situations:

  1. a season when a crop suffers very severe heat and drought stress during the silking and pollination (i.e., flowering) window of about seven days, and
  2. when an early frost kills the crop well before grain filling is completed.

However, even with these relatively uncommon provisos, the model has been successfully evaluated across a wide range of environments where corn produced yields varying from near zero to 300 bu/ac (Figure 2). More details on the Hybrid-Maize model are available in this article by Yang et al (2017).

Previous evaluations indicated that our forecasts captured the spatial pattern in rainfed and irrigated corn yields across the US Corn Belt. However, these previous evaluations also indicated that rainfed yields were underestimated at sites with a shallow groundwater table during the cropping season, and especially during the critical silking and pollination period, as is the case in many regions in the central and eastern Corn Belt. During the 2019 crop season, we will use a new approach to account for the positive impact of groundwater table on corn yield and stability. For a given location, the fraction of fields with a shallow groundwater table will be derived from maps on tile drainage because the latter is a robust indicator of the presence of the groundwater table. For fields with a shallow groundwater table, we will assume that there is no water limitation, and subsequently, the average forecasted yield for the given location will be estimated as the average of the forecasted yields for fields with and without a groundwater table, weighted by their respective shares of corn harvested area at that location.

Table 3. Environmental factors accounted for by the Yield Forecast Center forecasts.

Accounted factors

Non-accounted factors

Solar radiation
Temperature
Relative humidity
Wind speed
Reference evapotranspiration
Precipitation
Soil type
Soil water at planting
Irrigation
Plant population density
Hybrid maturity
Planting date
Shallow groundwater table

Nutrient supply
Incidence of biotic stresses
   (weeds, insect pests, pathogens)
Flooding
Hail
Non-uniform stand
Soil crusting
Severe heat/drought stress around silking
Early killing frost
Green snap/lodging

How Real-time Corn Yield Potential is Forecast

Diagram showing how the corn yield forecasts are performed with Hybrid-Maize
Figure 3. a) Diagram showing how the yield forecasts are performed with Hybrid-Maize. b) The distribution of possible end-of-season yields. c) Comparison of the distribution of forecasted yields against the long-term (20+ year) mean yield (black arrow) allows to estimate the probability (P) of above-average end-of-season yield (dashed area).

Hybrid-Maize uses measured weather data to simulate crop growth until the forecast date (solid black line in Figure 3a). It will then use historical (20+ years) weather data to predict all possible weather scenarios for the rest of the season (blue dotted lines in Figure 3a). This results in a range of possible final yields (blue distribution in Figure 3b). By comparing the distribution of forecasted yields (blue-shaded area in Figure 3c) against the average yield simulated for the same location using only the historical weather data (vertical arrow in Figure 3c), it is possible to determine the likelihood (probability) for the current season’s yields to be below, near, or above long-term average yield (dashed area in Figure 3c). Yields reported by the YFC more closely track yields achieved in fields that did not suffer severe yield losses due to pests, hail, waterlogging, poor establishment, or which were substantially affected by a stress not accounted for in the Hybrid-Maize model. Finally, it is important to keep in mind that the yield forecast is not field-specific. Instead, it represents an estimate of the average corn yield potential for a given location in the absence of the yield-reducing factors.

What to Expect During the Growing Season

Early in the season (June), yield forecasts for the current season mainly rely on historical weather and the range of forecasted yields will be wide and almost identical to the range of yields based on the historical weather data alone. As the season progresses (early-July to August), and more of the current season’s weather data are used in the simulations, the range of forecast yields will start to converge and may deviate from the long-term average yield, depending upon the similarity of the current season’s weather relative to the long-term average weather patterns. As the crop approaches maturity (end of August-September), the range of forecasted yields will further converge and the final yield can be forecasted with greater confidence.

How to Use These Forecasts to Inform Farm Decisions

When complemented with other sources of information, personal experience, and best judgment, the information provided by the YFC allows adjusting yield goals for the current growing season in comparison to normal years, and in some cases this insight can support management decisions such as timing and amounts of fertilization and irrigation, or application of pesticides during the grain-filling period. Information from simulations during grain filling provides additional information to guide marketing decisions for all participants in the corn supply chain at regional and national levels (producers, livestock sector, the ethanol industry, insurance companies, companies involved in commercialization and transportation, and policymakers).

Yield forecasts will be released as CropWatch articles every two or three weeks during the 2019 crop season, starting from mid-July. We will also provide information on the range of forecasted potential yields, probabilities for above-, near-, or below-normal yields, and also information about crop phenology and the probability of an early killing frost. Forecasts will be discussed in relation to current season’s weather versus the normal weather for each location to identify regions where yields are predicted to be well above, below, or near the long-term average.