Use of censor_thresh or censor_val to modify fcst fields

[jwraby]

I have a situation where the post-processed WRF output values for roughness length are all zeros. I believe that I can obtain point observations of roughness length to use for verification. Is there a way to use the functionality of censoring to convert the values in the forecast output from zeros to simulated values obtained from the land surface model used in the WRF? Thanks.

[jprestop]

Hi @jwraby. I needed to request some help from a colleague with this one. We are uncertain about exactly what you want to do. Is it correct that you have point observations of roughness length and want to somehow use them to change the gridded values of roughness length in the model? We are assuming that you have multiple point values that correspond to different locations in your grid. In this case, censor_thresh and censor_val cannot do what you're asking. Censor_thresh and censor_val don't have any spatial information. Since the roughness length is all zeroes, all you can do with censor_thresh and censor_val is reset every point in the grid to a different number (the same number at every point). Point2Grid might be the better tool. It can take point data and put it on a grid (the model grid). It would be helpful to know what you plan to do with the roughness length values after they are on a grid to be able to point you to the right tool.

[jwraby]

Hi Julie. Thanks for your response. I currently do not have the point observations for roughness length. I am planning to work with my group here at ARL at White Sands Missile Range to see how I can obtain them at some point in the future. Our group is working on a project which focuses on computing the surface energy budget (SEB) at high resolution using measurements from a dense array of meteorological towers located near Las Cruces, NM and operated by ARL. Thus, our interest in working with roughness length. My question to the METplus developers involves using those settings in Point-Stat to modify the values of roughness length in the post-processed WRF-LES output we are running in support of the SEB project to effectively change the values used by Point-Stat from zero to values we hope to obtain from the Noah Land surface model (LSM) which the WRF-LES uses for computing surface-based variables like roughness length. Unfortunately, the WRF-LES simulations output we have did not contain the roughness length field and so the post-processed output produced by the UPP contained values for roughness length which were all zeros. My understanding is that Point-Stat can convert the values of roughness length used for computation of the error statistics on the fly using the censoring settings. The intent of all of this is to verify the roughness length values obtained from the LSM against the point observations of roughness length we plan to obtain from the tower measurements. Please ask for further clarification if there is something which is still not clear. I will be happy to answer any questions about this as needed.,

[JohnHalleyGotway]

@jwraby I suspect you're asking more of the censor_thresh and censor_val configuration options than it can handle. Listed below is a selection from the MET User's Guide about those options.

The “censor_thresh” entry is an array of thresholds to be applied to the input data. The “censor_val” entry is an array of numbers and must be the same length as “censor_thresh”. These arguments must appear together in the correct format (threshold and number). For each censor threshold, any input values meeting the threshold criteria will be reset to the corresponding censor value. An empty list indicates that no censoring should be performed. The censoring logic is applied prior to any regridding but after the convert function. All statistics are computed on the censored data. These entries may be used to apply quality control logic by resetting data outside of an expected range to the bad data value of -9999. These entries are not indicated in the metadata of any output files, but the user can set the “desc” entry accordingly.

They really are pretty simple. For example, if you have forecast precip accumulations that are < 0 due to some numerical estimation process, you could use censor_thresh = lt0; censor_val = 0; to reset all negative values to 0. Of if you have observation data that uses a bad data value of -99, you could reset it to MET's internal bad data value with censor_thresh = ==-99; censor_val = -9999;. Or if you have data that must be between 0 and 10, you could use censor_thresh = [ <0, >10 ]; censor_val = [ 0, 10 ]; to reset anything outside that range to the extreme values... or use censor_thresh = [ <0, >10 ]; censor_val = [ -9999, -9999 ]; to reset anything outside the range to bad data.

Another option to provide more sophisticated data processing is the use of Python embedding, as described in Appendix F. Or perhaps you're referring to the incorporation of observation uncertainty as described in this obs_error configuration section for Ensemble-Stat?

[jwraby]

Hi John. Thanks for the examples showing the use of the censor_thresh and censor_val configuration options. I understand how they work better than before. Good suggestion on the possibility of using Python embedding which might enable processing of the input forecast field in a way which would convert the 0 values in the post-processed forecast to an array of values of my choosing. I've used Python embedding for a different type of task, so that is not unfamiliar to me. Very helpful information. This answered my original question.