CHRTOUT is point data where the feature_id is the unique key, evaluate chunking strategies that consider time, spatial coordinates, and chunks of feature_ids
While LDASOUT and RTOUT are both gridded data, their resolutions are widely different; does the resolution matter in the chunking strategy?
point data will be queried very differently from gridded data by the end user - for point data they may only be interested in fetching variable values for some set of unique keys, for gridded data they may be more interested in spatial statistics - avg, max, min
shorter summary of variables available in the AWS Open Data: https://docs.opendata.aws/nwm-archive/readme.html
complete list of variables: https://ral.ucar.edu/sites/default/files/public/WRFHydroV5_OutputVariableMatrix_V5.pdf