Create a test notebook comparing alternate methods for querying the NWM data

[jameshalgren]

Pulling further discussion over from here:
CIROH-UA/awi-ciroh-image#16

[necrisci]

Queries to test might include these time-space combinations. This would just be for querying individual forecasts or collections of individual forecast time series. A similar tree could be built for the analysis, the forcing, and lead time - sliced forecast subsets.

****	****	****	****	Space	****	****	****
****		S (1 point)	M (10 - 150 pts)	L (150 - 1000 pts)	XL (1000 - 10K pts)	XXL (10K - 100K pts)	XXXL (100K - 2.7M pts)
****	S (1 hr - 24 hrs)
****	M (2 d - 10 d)
Time	L (10 d - 2 mo)
****	XL (2 mo +)
****	XXL (40 yr +)

[jameshalgren]

Additional variables:
Number of Threads
Chunking method
Forecast vs. Analysis
Gridded vs. Scalar (some of our methods only work for scalar or are at least presumptive somewhat.)

[jameshalgren]

Here we go... Please drop notes in here...
@ZacharyWills @mgdenno
@hellkite500 @karnesh

[jameshalgren]

ping @RohanSunkarapalli

[jameshalgren]

For AMS January 11th
Tests to perform:

For a given NWM stream segment

• [Critical] build the dataframe for the plot from the interactive map with all traces for any arbitrary date in the record.
• [Optional] assembling an arbitrary 14 day sequence from the retrospective simulation.
• [Bonus] query the forecasts and assemble the 14-day series of same-lead-time forecasts from short range (f006) and medium range (f048).

Note: the 14 days is essentially an arbitrary length -- but should be a good starting point for measurement. Right now, we can do that with the Kerchunk methods, but it takes an awfully long time.

Dataset versions against which to perform tests:
(Bold datasets are currently publicly available)

• NC Reformatted via Kerchunk
• NC NWM Native via Kerchunk
• NC Reformatted and subdivided by HUC via Kerchunk
• NC NWM Native and subdivided by HUC via Kerchunk
• Parquet 1:1 with NWM Native
• Parquet 1:1 with Reformatted
• Parquet 1:1 with NWM Native subdivided by HUC
• Parquet 1:1 with Reformatted subdivided by HUC

Bonus

• Big Query
• [stretch goal] External database

Overall Todo:

• Determine test content for forecast datasets
• Determine test content for retrospective datasets
• Script creation of missing datasets (much of this has been preliminarily templated by @karnesh and @arpita0911patel)
• Develop simple test protocol
• Run and record results

[jameshalgren]

See here. E.g.,

[RohanSunkarapalli]

Is there something I need to study about this?

…

On Fri, Dec 30, 2022 at 2:48 PM James H. ***@***.***> wrote: See here <https://water.noaa.gov/map#forecast-chart>. E.g., [image: image] <https://user-images.githubusercontent.com/53343824/210110621-2daad54e-e71b-4e67-9b68-43b3e7b11404.png> — Reply to this email directly, view it on GitHub <#7 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AN4WNOLE7PXAVALNCWXDA73WP5DC3ANCNFSM577MGIHQ> . You are receiving this because you were mentioned.Message ID: ***@***.*** com>

[jameshalgren]

ping @ZacharyWills @mgdenno @arpita0911patel @KMarkert @imanmaghami