The A-Maize-ing world of seasonal forecasting

The A-Maize-ing world of seasonal forecasting


The Bottom-Line Up Front

Atmospheric and Environmental Research is applying our world-class seasonal forecasting to give the agricultural community information to make better-informed decisions with actionable yield projections for the upcoming season.

The Details

Our forecast for corn yield is primarily based upon our assessment of analog years. An analog year is simply a year (or group of years in our case) in the past that we project to have similar characteristics to the coming year.  We rely on our decades long research into seasonal forecasting, statistical pattern matching, and recent advances in machine learning to guide the selection of the analog years. As the growing season gets underway, we augment our initial seasonal outlook with other relevant factors, such as season-to-date growing degree days, soil moisture and evaporative stress anomalies, in addition to the projected temperature and precipitation outlook.

The analogs are based on science and statistics. First, we use a blend of global atmospheric, surface, and ocean-based inputs that are fed into a model. Our seasonal forecast specialist, Dr. Judah Cohen (who is also the author of the weekly Arctic Oscillation Blog), then determines through statistical methods which past years most closely resemble the coming one. This provides us with the analog years for the Crop Forecast Report. From there, we use the national corn yield data from the National Agricultural Statistics Service (NASS) for each of the analog years.

We base our projected corn yields and trends for the coming season on historical NASS data. Using time series trend analysis, we have been able to determine the amount above or below trend for each year since the early 1960’s.  Figure 1 shows that the vast majority of years have fallen within 10 points on either side of the ‘at trend’ line. All things being equal, the chances of an historically poor year on a national basis are relatively low (~ 10% of years). The only years since 1974 where the national corn yield has been more than 10 points below trend were in years with significant flash drought across a wide portion of the central U.S. (or excessive moisture as was the case in 1993).


Figure 1. Histogram of national corn yields with regard to trend from 1974-2017. Years are listed in chronological order by category. 2018 final yields are not available yet, though expectations are to be around 5 percent above trend.

A hindcast analysis helps demonstrate the overall utility of our forecast methodology. Figure 2 presents the results of the hindcast exercise and illustrates that AER’s forecast (delivered to AER clients in February of each year) would have signaled the correct direction of final yield (relative to trend) in 14 out of the 17 hindcast samples.  This compares with the USDA August projection of correctly predicting the direction of yields relative to trend in 13 out of the 17 years in the hindcast sample.  In years where the projected yield deviates significantly from trend we tend toward increased . Given that the markets can move based on the August estimate, our February projection could be very useful to those who want to decide how to market existing and future corn.   As with any projection, there is uncertainty – but we feel that there is sufficient skill in our forecast that can be leveraged by growers to make important decisions regarding upcoming growing seasons with additional confidence.

Figure 2. AER projected corn yields off trend in February of given year based on analogs (black bars), the USDA’s initial projection in August (red bars), and the final observed trend from NASS (gray bars).

Table 1 enumerates the performance of AER’s projections relative to the August USDA projections.  Keep in mind that AER is able to produce these projections in February, a full 6 months before the USDA August estimate.  The statistics show that the AER forecast exhibits a slightly lower absolute bias and a slightly higher standard deviation with respect to the USDA actual yields.  Both the AER February and USDA August estimates get the direction of deviation (above or below trend) correct about ¾ of the time.  We feel that this is a significant achievement and a testament to our forecast approach, and we see the AER projections as valuable information for producers to have in hand prior to the growing season.  This information should allow for better decisions to be made regarding crop insurance and marketing decisions. 


Projection Performance

USDA August

AER February




Standard Deviation



Frequency of Correct Direction




Table 1.  AER forecast performance relative to USDA August projections – both compared to USDA actual yields (which are released in January following the growing season).

And what’s next …

We believe that the AER methodology will consistently deliver useful insights through our national corn forecast. The details of our forecast, to include the analog years used, will provide a useful baseline to reference throughout the season.

If you are interested in our corn forecast for 2019, please check back in early February for our first forecast. I will also be updating this blog periodically throughout the season. In the meantime, please direct any feedback or questions my way ( ).