@hcorson-dosch pulled Gridmet driver data from 1980-01-01 to 2020-07-31. We plan on using this moving forward since it is updated more frequently than Daymet. But we need to compare to old PRMS-SNTemp model output to see if there are any big differences in model output caused by changing from Daymet (original driver source) to Gridmet.
I compared model output from Gridmet and Daymet from 1980-01-01 to 2016-01-01. Most segments' temperature predictions error (Gridmet compared to Daymet) were between 1-1.5 C RMSE across all years (see below).
typical timeseries output (segment below has a ~1 C RMSE)
Only one segment, model_idx 452, had an RMSE of greater than 2C and that was because of segment drying up (I think). see plot below. This could be filtered out and maybe already is in the PGDL pre-training phase.
Most segments' flow prediction error (Gridmet compared to Daymet) were between 1-10 cms RMSE across all years (see below, notice log10 scale )
Standardized to mean flow below
typical timeseries output (segment below has a ~1.9 cms RMSE)
Highest RMSE for flow below (~112 cms RMSE)
@hcorson-dosch pulled Gridmet driver data from
1980-01-01
to2020-07-31
. We plan on using this moving forward since it is updated more frequently than Daymet. But we need to compare to old PRMS-SNTemp model output to see if there are any big differences in model output caused by changing from Daymet (original driver source) to Gridmet.I compared model output from Gridmet and Daymet from
typical timeseries output (segment below has a ~1 C RMSE)
Only one segment, model_idx 452, had an RMSE of greater than 2C and that was because of segment drying up (I think). see plot below. This could be filtered out and maybe already is in the PGDL pre-training phase.
![image](https://user-images.githubusercontent.com/4825431/94964303-959c8d00-04c7-11eb-9ab7-69a31f0ae15c.png)
1980-01-01
to2016-01-01
. Most segments' temperature predictions error (Gridmet compared to Daymet) were between 1-1.5 C RMSE across all years (see below).Most segments' flow prediction error (Gridmet compared to Daymet) were between 1-10 cms RMSE across all years (see below, notice log10 scale )
Standardized to mean flow below
![image](https://user-images.githubusercontent.com/4825431/94965807-22484a80-04ca-11eb-8839-516a25fa3b1c.png)
typical timeseries output (segment below has a ~1.9 cms RMSE)
Highest RMSE for flow below (~112 cms RMSE)
![image](https://user-images.githubusercontent.com/4825431/94965907-5459ac80-04ca-11eb-8099-b96d5dfa5f43.png)