DANG-ai/SKILLER / skilllearnbench/skills/human_authored/temperature-simulation/glm-calibration
glm-calibration
VerifiedCalibration guidance for GLM tasks. Often most effective after glm-basics has clarified the setup; glm-output is the companion skill for exact final metric computation.
GLM Calibration Guide
Overview
GLM calibration involves adjusting physical parameters to minimize the difference between simulated and observed water temperatures. The goal is to satisfy the task's stated evaluation metrics, not just to lower a single aggregate RMSE.
Suggested Skill Flow
This skill works best after glm-basics has confirmed the file layout and editable parameters. The job of this skill is to find a candidate parameter set that passes the task metrics. Once a candidate setting appears to pass, stop calibration and hand off immediately to glm-output for the final verifier-matching metrics and /root/metrics.json.
Task Checklist
For tasks that restrict the editable parameter set and require an exact self-evaluation:
- Only adjust the parameters explicitly allowed by the task
- Leave protected parameters and initialization profiles unchanged
- Stop only when every required task metric passes; do not use global RMSE alone as the stopping condition
- If a configuration already passes every required task metric, stop the calibration search immediately
- Do not launch a broad refinement search after finding a passing configuration
- If the task requires
/root/metrics.json, hand the final reporting step toglm-outputinstead of a separately improvised evaluator
Important Tips
- In many case, the surface fit becomes acceptable before the deep water is fixed.
- Single-knob moves can improve whole-lake RMSE while leaving the task-defining deep subsets too warm.
- Do not optimize only the global RMSE. Recompute the full task metric set after each promising run.
Key Calibration Parameters
| Parameter | Section | Description | Default | Range |
|---|---|---|---|---|
Kw | &light | Light extinction coefficient (m⁻¹) | 0.3 | 0.1 - 0.5 |
coef_mix_hyp | &mixing | Hypolimnetic mixing coefficient | 0.5 | 0.3 - 0.7 |
wind_factor | &meteorology | Wind speed scaling factor | 1.0 | 0.7 - 1.3 |
lw_factor | &meteorology | Longwave radiation scaling | 1.0 | 0.7 - 1.3 |
ch | &meteorology | Sensible heat transfer coefficient | 0.0013 | 0.0005 - 0.002 |
Parameter Effects
| Parameter | Increase Effect | Decrease Effect |
|---|---|---|
Kw | Less light penetration, cooler deep water | More light penetration, warmer deep water |
coef_mix_hyp | More deep mixing, weaker stratification | Less mixing, stronger stratification |
wind_factor | More surface mixing | Less surface mixing |
lw_factor | More heat input | Less heat input |
ch | More sensible heat exchange | Less heat exchange |
Calibration with Optimization
from scipy.optimize import minimize
def objective(x):
Kw, coef_mix_hyp, wind_factor, lw_factor, ch = x
params = {
'Kw': round(Kw, 4),
'coef_mix_hyp': round(coef_mix_hyp, 4),
'wind_factor': round(wind_factor, 4),
'lw_factor': round(lw_factor, 4),
'ch': round(ch, 6)
}
modify_nml('glm3.nml', params)
subprocess.run(['glm'], capture_output=True)
return calculate_task_metrics(sim_df, obs_df)['overall_rmse']
result = minimize(
objective,
[0.3, 0.5, 1.0, 1.0, 0.0013],
method='Nelder-Mead',
options={'maxiter': 150}
)
Manual Calibration Strategy
- Start from the provided seed and compute the exact deep-band metrics, not just overall RMSE.
- First test changes that improve the deep layer over the full record rather than only cooling the whole lake;
Kwandcoef_mix_hypare usually the first levers to inspect. - Once the annual deep metric improves, use smaller adjustments to
wind_factorandlw_factorto reduce whole-lake bias without undoing the deep-water gains. - Leave
chfor final fine-tuning after the main structure is already close. - Prefer a short hypothesis-driven search: keep candidates that improve both deep metrics together, discard one-knob moves that only make the global RMSE look better, and stop as soon as one candidate passes every required task metric.
- After the first passing configuration is found, rerun GLM once at that setting, switch to
glm-output, write/root/metrics.json, and end the task if the exact final metrics also pass.
Common Issues
| Issue | Likely Cause | Solution |
|---|---|---|
| Surface too warm | Low wind mixing | Increase wind_factor |
| Deep water too warm | Too much light penetration | Increase Kw |
| Weak stratification | Too much mixing | Decrease coef_mix_hyp |
| Overall warm bias | Heat budget too high | Decrease lw_factor or ch |
| Annual deep improves but summer deep still fails | Deep water cooled too weakly during stratified months | Keep Kw high and lower lw_factor slightly |
Best Practices
- Keep parameters within the published ranges
- Use the deep-band metrics as the stopping condition, not surface fit alone
- Save time by evaluating a few structured candidates instead of a long blind optimizer
- After you find a passing parameter set, do not keep refining it blindly
- Rerun GLM once at the chosen setting, then use
glm-outputto write the exact metrics file immediately - If
glm-outputconfirms that the exact final metrics pass, stop there instead of reopening calibration
Related skills
| Skill | Stars | Updated | Works with | Status | Install |
|---|---|---|---|---|---|
| reverse-engineer-anythingmorluto/reaReverse engineer native, managed, Electron/JavaScript, packaged, firmware, and browser targets with REA, and build agent environments (AgentEnv bundles) of web applications from a live instance or a capture, and of Linux desktop apps with a task from a request. Use shipped-artifact or requested runtime evidence to explain features, compare versions, decompile code, or guide a reconstruction. Skip REA for ordinary source-repository architecture analysis. | Stars: 471k | Updated | Claude CodeCodex | Verified | Install |
| diagnosing-superpowersobra/superpowersUse when a superpowers session went wrong and your human partner wants to know why — repeated work, ignored plans, stumbles, poor results, a skill that didn't fire, "it took too long", "why is it so expensive", "what is it doing" — or wants to build a bug report for the superpowers maintainers, for the current session or a past one identified by id or path, on any harness. | Stars: 322k | Updated | Claude CodeCodex | Verified | Install |
| dispatching-parallel-agentsobra/superpowersUse when facing 2+ independent tasks that can be worked on without shared state or sequential dependencies | Stars: 322k | Updated | Claude CodeCodex | Verified | Install |
| executing-plansobra/superpowersUse when executing an implementation plan in the current session as the implementer yourself — your human partner chose inline execution, or no subagent tool is available | Stars: 322k | Updated | Claude CodeCodex | Verified | Install |