Diagnostics and Calibration

[ephilip]

Good Morning,

I plotted the parameter sensitivity and objective function vs trial using RavenR for first five parameters. The y-axis is labelled as Objective function. But the values are labelled (ranging from 0-15) differently while plotting best objective function vs trial. Shouldn't it be the same? Would you please clarify.

Thank you
Elizabeth

[rchlumsk]

Hi Elizabeth

If I understand your question correctly, the range is a little different in the objective function for the sensitivity plot and the overall trial plot? That makes sense as the scale would just be automatically determined internally by the base R plot function when a plot is generated, there is no functionality within the Raven Ostrich functions to get the full range and use that for everything. You may control this and create your own plot after reading in the data from the Ostrich file with a fixed ylim for each plot if you wish.

Hope that helps!
Rob

[ephilip]

Hi Elizabeth

If I understand your question correctly, the range is a little different in the objective function for the sensitivity plot and the overall trial plot? That makes sense as the scale would just be automatically determined internally by the base R plot function when a plot is generated, there is no functionality within the Raven Ostrich functions to get the full range and use that for everything. You may control this and create your own plot after reading in the data from the Ostrich file with a fixed ylim for each plot if you wish.

Hope that helps!
Rob

Good Morning Rob,
I didn't quite get what you wrote. While plotting:

(1) Ost.plot(ost,'boft')

I get the y-axis "Best Objective function" with value going up to -1 (minimization value)

(2) Ost.plot(ost,'oft')

I get the y-axis Objective function with value ranging from 0-15 in my case

Same type of plot I am getting while plotting the first 5 sensitive parameter.

Would you please clarify the difference between the "Best Objective Function" and "Objective Function" ?

Thank you
Elizabeth

[rchlumsk]

Hi Elizabeth

The Best Objective Function evaluates the best objective function achieved by the calibration for each trial, i.e. the value plotted for each trial will stay the same unless the next trial has a better objective function. In other words, the boft plots the best value so far for each trial. The boft plot will typically change quickly in the first bit of the calibration and then level off and approach the final best calibrated objective function value.

The Objective Function just plots the objective function given in each trial, which will of course have more scatter than the boft plot.

Hope that helps!
Rob

[ephilip]

Good Morning Rob,

Would you please explain the functions

plot(best.params.100$X.A0TLZP.)
plot(best.params.100$X.V0FLAS.)

I am not able to plot these functions. Asking for xlim and ylim. How would you do it in this type of multiple graphs . Would you please give an example?

Thank you
Elizabeth

[rchlumsk]

Hi Elizabeth

The documentation for functions in R, including from the RavenR package, can be found by entering ?<function_name>, and are typically pretty useful with documentation and examples. I would suggest that you take a look there. The text you have here look like just variable names so I can't help without more detail.

As for general R help, the programming forums such as stack overflow may be more helpful and appropriate than this one, I suggest you take a look there for help creating specific plot types outside of the functionality in RavenR.

Cheers,
Rob

[ephilip]

Hi Rob,

Just a quick question, In my model I am getting the best 100 params correctly for all the parameters except two. why I am getting NULL value ?

> best.params.100$par_x1
NULL

Thank you
Elizabeth

Hi Elizabeth

The documentation for functions in R, including from the RavenR package, can be found by entering ?<function_name>, and are typically pretty useful with documentation and examples. I would suggest that you take a look there. The text you have here look like just variable names so I can't help without more detail.

As for general R help, the programming forums such as stack overflow may be more helpful and appropriate than this one, I suggest you take a look there for help creating specific plot types outside of the functionality in RavenR.

Cheers,
Rob

[rchlumsk]

Hi Elizabeth

I don't see enough here to help you out, please send me the entire reproducible example if you would like me to take a look.
Cheers,
Rob

[Ameer]

Often in cold region, gauges have seasonal data (data exists from March to October only). What one should use for winter months when running RAVEN? Should one just populate the gauge observation with -1.2345 for winter period to make the gauge data continuous?
Thank you,
Ameer

[rchlumsk]

Often in cold region, gauges have seasonal data (data exists from March to October only). What one should use for winter months when running RAVEN? Should one just populate the gauge observation with -1.2345 for winter period to make the gauge data continuous?
Thank you,
Ameer

Hi Ameer

I assume this is for flow or level gauge data for observation/diagnostics? In this case yes, you can infill with the missing data flag (-1.2345) and Raven will just use the available data from that series in the calculation of diagnostics. This is true for missing observation data in general. The cautionary note is that this will bias the diagnostic, i.e. if you calibrate the model to this diagnostic you will only be matching observation data within the available range, and the model may not actually work well for the missing period. Just something to consider.

Cheers,
Rob