kelleythompson
commented
2 years ago A new matlab script named StateTLgains.m was developed and added to the github matlab folder. This script runs the StateTL exe using a "StateTL -e" command line argument for every year in the multiyrlist in the control file to calculate and store hourly gains & losses for every year. The gain/loss amounts for each year are saved into a file named StateTL_data_gainsyr.mat (this includes both the subreach gagediffportion variable and the reach based gagedifflast variable). The StateTLgains reads these data and calculates the median hourly gain/loss/error for all the years and saves that as the binary file StateTL_data_gains.mat. For calibration, the StateTL script can then use the fixed/constant hourly gains/losses from this binary file within the simulation. It will average (or median) the hourly values based on the calibavggainloss method and gainsavgwindowdays parameters set in the control file (these are set to 'movingmedian' with a 30 day window which seemed to work well in initial testing. )
The use of the fixed/constant gains & losses can be managed with the control file in the following ways:
- runriverloop=0 and calibconstantgains=1: - this will use the fixed gains&losses for the entire calendar year and the river loop will not be run during execution thereby significantly speeding up each simulation run.
- runriverloop=1 and calibconstantgains=1: - this will use the fixed gains&losses only for the calibration period specified in the control file; this will isolate the simulation to a smaller period (less than a calendar year) but the river loop will be run which will increase the simulation run time.
- runriverloop=1 and calibconstantgains=0: - simulation will be run (like in the past) with gains&losses that are averaged(median) over a specified window but will a function of the calibration parameters used for that run
- calibavggainloss='movingmedian' (current): - method to average (or median) the gain/loss/error values
- gainsavgwindowdays=30 (current): - days over which to average (or median) the gain/loss/error values
- outputgain=0 or 1 (0 currently): trigger to output both actual and
Currently, just one fixed/constant gains/loss/error (gagediffportion and gagedifflast) timeseries is established for use in the simulation runs. However, a future improvement might be to establish several timeseries based on if the year is wet, dry, or average. This could currently be implemented easily similar to the wet/dry/average that is used to fill Qnode; but that wet,dry,avg can currently vary through the year based on the gage flow and it might be best to fix one single year type for use for the whole year.
The gains datafile will need to be rebuilt if there are changes to the river network etc. Although the files can be rebuilt using the StateTLgains script - the rebuilding was also added to the StateTL -r command that is typically used to rebuild all datafiles. With the StateTL -r command, StateTL actually calls the StateTLgains compiled exe which in turn calls new instances of the StateTL script to rebuild the datafiles.
Previously, during calibration the riverloop was rerun and gains/losses recalculated given the changed calibration parameters; then the gains/losses were averaged over some moving window for use in the actual simulation (had been using moving centered median over 14 days). However, it was decided that perhaps the gains and losses should be made entirely independent of the calibration. Therefore, an option was added to fix the gains/losses so they don't change between calibration runs.