Loading historical forecasts of future covariates for training - forecast changes each time step

[gconnell-hiringa]

I am struggling to identify the correct method for loading historical forecasts of a future covariate, when the historical forecast is updated every time step.

The scenario is:

• Predict temperature 10 time steps into the future
• Actual temperature is recorded
• Actual pressure has been recorded (Past covariate)
• There is an external forecast for pressure for the next 10 time steps, which can be treated as a known future covariate.
  • The external forecast is updated every time step for the next 10 and is available for live prediction.

I have historical data coming through with columns for:
Time | Actual temp (Target) | Actual Pressure | Pressure forecast T+1 | ... | Pressure forecast T+10

Is it possible to load these in a way that reflects the changes in the forecast? Methods that I have attempted so far, end up providing information to the model that is too far ahead of what is known and makes a mess of the live prediction data formatting.