Run2 netcdf rmse evaluation

GLM Evaluation (Improved)

This skill describes how to correctly open and extract depth and temperature from GLM output.nc without coordinate conflict errors, and how to compute exact match RMSE against field data.

Loading and Interpolating

Using xarray to open output.nc will likely result in a MissingDimensionsError because GLM defines a z variable with a dimension that shares its name. Instead, rely on the netCDF4 library natively.

import netCDF4 as nc
import pandas as pd
import numpy as np

# Load simulation
ds = nc.Dataset('output.nc')
times = nc.num2date(ds.variables['time'][:], ds.variables['time'].units)
datetimes = pd.to_datetime([t.strftime('%Y-%m-%d %H:%M:%S') for t in times])

z = ds.variables['z'][:, :, 0, 0]
temp = ds.variables['temp'][:, :, 0, 0]
ns = ds.variables['NS'][:] # Number of simulated layers at each time step

Depth Extraction

GLM saves z as distance from the lake bottom. The distance from the surface down (which field data typically uses) is dynamically derived using surface_z - z:

depths = np.array([z[i, ns[i]-1] - z[i, :ns[i]] if ns[i] > 0 else [] for i in range(len(ns))])

Creating Paired Dataset

To perform the exact datetime and rounded-depth merge:

1. Construct a flat sim_df of valid datetime, rounded depth, and simulated temperature pairs.
2. Format obs datetime to match, and generate its depth_rounded.
3. Perform an explicit merge pd.merge(obs, sim_df, on=['datetime', 'depth_rounded']). Avoid "alternative depth binning" (like groupby().mean()) unless explicitly instructed; GLM layers matching the same rounded integer will create multiple paired rows.