My observations are leaf number of wheat which obtained from multiple experiments. The number of observations are depending on experiment which ranged from 1 to 10. In each experiment, leaf number is a time serial values from 1 to 7.
A processed based crop model APSIM is used to predict leaf number. A optimization procedure is used then to optimize two parameters in the model.
Currently, the optimization procedure is based on the minimum mean squared variation (MSV) to select the better parameters.
All the observations are directly used to calculate MSV during optimization.
My question are
- how to treat all experiments with the same weight as curent method biases to experiment with more observations (experiments with smaller number observations have more errors)?
- how to deal with the time serial data as the current method biases to bigger values (See Exp7 below)?
See picture below for current comparison of observed and predicted leaf number.