Heads from a previous run

[DStrom1987]

Hi,

I was wondering if theres a pestpp_option or some other smooth ways of making my pest-ies cycle use the initial heads from a previous run?
When i set it up through model muse it gives me an error and will not proceed. However, the model im running should take around 1 minute to finish with the heads from a successful run as starting heads. With model top it takes roughly 6 minutes which increase when I run 25 models in paralell.

Any advice?

David

[jtwhite79]

Hmm. In an ensemble setting, you'd need initial heads that are coherent with each realization, which is tricky. I guess you could use the state estimation machinery in pestpp-da to do this within the par-obs processes. You'd have to read the simulated heads from a model run for each model cell as "obs" and set a par for each model cells initial head. Then you could tell pestpp-da to transfer those simulated head obs to the initial head pars between cycles and just use a single iter for each cycle. But that's not smooth or easy...

[DStrom1987]

I see... I was thinking more of a static initial head from the first run of the model. Even though it wouldn't progressively update through the iterations, it would still provide a baseline closer to my results rather than just using the model's top or something similar.

Currently it seems to dislike the initial heads entered in the GUI.

[wkitlasten]

I always update my MODFLOW initial heads before starting the PEST process. If you use flopy and external files, something like this (using your own dir/naming conventions of course): sim = flopy.mf6.MFSimulation.load(sim_ws=sim_ws, sim_name=sim_name, load_only=['ic']) for m in sim.model_dict: gwf = sim.get_model(m) hdfile = f"{m}.hds" hdobj = bf.HeadFile(os.path.join(sim_ws, hdfile), precision='double') hdobj.list_records() rec = hdobj.get_data(kstpkper=(0, 0)) for k in range(0, rec.shape[0]): outfile = '{}.ic_head_layer{}.arr'.format(gwf.name, k + 1) # could get slick and read this from the IC package np.savetxt(os.path.join(sim_ws, outfile), rec[k]) gwf.ic.strt.set_data(outfile, layer=k) gwf.ic.write() hdobj.close() I didn't have much luck trying to get simulated head "observations" from previous runs into initial heads for the next run, but my attempts were pre-pestpp-da. This one time, at modelling camp, I used all my initial heads as parameters and "informed them" with simulation performance metrics as "observations" like number of inner and outer iterations, percent discrepancy, simulation time.

…

On Sat, Oct 12, 2024 at 10:28 AM DStrom1987 ***@***.***> wrote: I see... I was thinking more of a static initial head from the first run of the model. Even though it wouldn't progressively update through the iterations, it would still provide a baseline closer to my results rather than just using the model's top or something similar. Currently it seems to dislike the initial heads entered in the GUI. — Reply to this email directly, view it on GitHub <#543 (reply in thread)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/ADSJXRATAE6SFSN77UMKSADZ3A7GVAVCNFSM6AAAAABPZSBD7WVHI2DSMVQWIX3LMV43URDJONRXK43TNFXW4Q3PNVWWK3TUHMYTAOJRHE2DEMA> . You are receiving this because you are subscribed to this thread.Message ID: ***@***.***>

-- "Perfect spheres are pointless."

[DStrom1987]

Interesting. Perhaps it works better to pass it though flopy. I'll definately look into it. Thanks for the ideas.