Datqorin / Canada
Prepare weather data for a seasonal forecast
A useful dataset connects weather observations to the location, crop and forecast date without including information that would only become available later.
Align time and place
Record the source, units and observation interval. Consider whether a station or gridded estimate represents the field and time period relevant to the task.
Handle gaps explicitly
Distinguish missing measurements from genuine zero values. Fit any learned preparation step on training data and document how filled values may affect the result.
Keep future information out
Build features using only observations available at the forecast date. Separate seasons and related fields when evaluating the model so the test resembles a real forecasting task.