Centre county Pennsylvania, USA | Kevin,I looked at many other possible predictor variables, including percent planted by certain date, but found no combinations that gave better fit (higher Rsquared) than the modified Jun-Jul climate temperature anomaly for predicting US geographical area corn yields. When I tried using the modified Jun-Jul climate temperature anomaly to predict corn yields in smaller geographical areas, it often failed. I found that Jun through Jul temperature anomalies were inversely correlated with climate precipitation anomalies in the US corn growing geographical area but were not correlated in many state level geographical areas.
FWIW, below is the Python code snippet I'm using to modify climate temperature anomaly. Basically it modifies the anomaly by raising it to 1.8 power when the anomaly is greater than +1.5. USDA's climate model for corn yield also uses some form of modified climate temperature anomaly, but its a much more complicated modification that the one I'm using, and gives similar Rsquared performance.
######### JUN-JUL CLIMATE TEMPERATURE ANOMALY MODIFICATION ###########
ANOMALY_TMP_MODIFIED = []
modify_threshold = 1.5
power = 1.8 # anomaly power when anomaly is above threshold
for v in ANOMALY_TMP:
if v > modify_threshold: #condition for modifying
ANOMALY_TMP_MODIFIED.append(v**power) # modify by raising to a power
else:
ANOMALY_TMP_MODIFIED.append(v) # not modified
######################################################### |