Fig. 4 shows plots of correlation coefficient (r) computed between
state-level yield anomalies and ESI-1 (with additional 4-week averaging
for noise suppression) as a function ofweek of year at the end of the averaging
window (see Section 4.2). Approximate windows of harvesting
date are also indicated for each crop (soybean: Feb.–Apr.; 1st corn crop:
Feb.–Mar.; 2nd corn crop: Jun.–Aug.; cotton: Jun.–Aug.), although harvest
can vary widely with region and year. Each correlation consists of
11 points, one for each year during the baseline period 2003–2013; correlations
of |r| N 0.6 are significant at p b 0.05.