Glm calibration

Skill cxcscmu/SkillLearnBench/skills/b1-one-shot-claude-sonnet-4-6/temperature-simulation/glm-calibration

Calibrating GLM parameters (Kw, coef_mix_hyp, wind_factor, lw_factor, ch) to minimize RMSE against field observations.From its SKILL.md

Install

npx -y skills add cxcscmu/SkillLearnBench --skill glm-calibration

Assembled from the repository path, not quoted from the project. Check it against their README if it does not work.

SKILL.md

2.6 KB, 740 tokens by cl100k_base, as published. Nobody here has run it

GLM Calibration Skill

Allowed Calibration Parameters

Parameter	Range	Description
`Kw`	[0.1, 0.5]	Light extinction coefficient (m⁻¹) - affects thermocline depth
`coef_mix_hyp`	[0.3, 0.7]	Hypolimnetic mixing coefficient - affects deep mixing
`wind_factor`	[0.7, 1.3]	Wind speed scaling - affects surface mixing
`lw_factor`	[0.7, 1.3]	Longwave radiation scaling - affects surface heat budget
`ch`	[0.0005, 0.002]	Sensible heat transfer coefficient

Physical Intuition

Higher Kw: More light attenuation, shallower thermocline, warmer epilimnion
Higher coef_mix_hyp: More hypolimnetic mixing, warmer deeper waters
Higher wind_factor: More surface mixing, weaker stratification
Higher lw_factor: More downward longwave radiation, warmer surface
Higher ch: More sensible heat exchange with atmosphere

Calibration Strategy

Start with default parameters and compute baseline RMSE
Focus on deep RMSE (depths ≥ 13m): primarily affected by coef_mix_hyp
Focus on overall RMSE: balance all parameters
Use grid search or scipy.optimize for systematic calibration

Modifying glm3.nml Parameters

import re

def update_nml(param, value, nml_path='/root/glm3.nml'):
    with open(nml_path, 'r') as f:
        content = f.read()

    # For Kw in &light section
    if param == 'Kw':
        content = re.sub(r'Kw\s*=\s*[\d.]+', f'Kw = {value}', content)
    elif param == 'coef_mix_hyp':
        content = re.sub(r'coef_mix_hyp\s*=\s*[\d.]+', f'coef_mix_hyp = {value}', content)
    elif param == 'wind_factor':
        content = re.sub(r'wind_factor\s*=\s*[\d.]+', f'wind_factor = {value}', content)
    elif param == 'lw_factor':
        content = re.sub(r'lw_factor\s*=\s*[\d.]+', f'lw_factor = {value}', content)
    elif param == 'ch':
        content = re.sub(r'ch\s*=\s*[\d.]+', f'ch = {value}', content)

    with open(nml_path, 'w') as f:
        f.write(content)

Running GLM and Computing Metrics

import subprocess
result = subprocess.run(['glm'], cwd='/root', capture_output=True, text=True)
if result.returncode != 0:
    print("GLM failed:", result.stderr)
else:
    # Compute metrics from output.nc
    pass

Typical Good Values for Lake Mendota

Kw: ~0.25-0.35 (moderately productive lake)
coef_mix_hyp: ~0.4-0.6
wind_factor: ~0.9-1.1
lw_factor: ~0.95-1.05
ch: ~0.0010-0.0015

What ships with it

Read from the repository

Just SKILL.md. No reference files, no scripts.