Santiago Beguería

(full size)

I’ve been playing around for a while with USDA NASS crop statistics, and have built a simple yet powerful Bayesian hierarchical network for predicting crop yields early in the season. Since the first USDA winter wheat yield forecast is due in less than a week (the next crop production report is programmed on May 10^th), I thought it might be fun to present my own forecast.

The above map depicts the expected yields at the state level, in bushels per acre. By placing the pointer over a given state it is possible to get the numeric values of the expected yield with its prediction interval, and the anomaly in percentage over the last 5‑years average. It is possible to switch the representation between yield and anomaly. This is the complete dataset shown on the map:

The expected yield informs about the development of the crop, especially if it is compared with the average of the previous years, but the yield alone does not inform about the expected winter wheat production. The final production is the result of the yield and the acreage (the area planted), which varies from year to year. By combining the expected yield with the estimated crop acreage, it is possible to estimate the production. A production map can be found below:

(full size)

The total US winter wheat production is estimated at 1,575,383 (x 1,000) bushels, or 34,299 (x 1,000) metric tons. This is a variation of +1% with respect to last week’s estimate. As a reference, the total US winter wheat production in the period 2014–2016 was: 1,377,216, 1,374,690, 1,671,532 (x 1,000) bushels, or 29,985, 29,930, 36,393 (x 1,000) metric tons, respectively.

Production forecasts by state can be found below: