Skill Metrics

Skill results across the experiment configurations (Base, Month/Seasonal, SST, SWE, Chain, Local, Anomaly) for each model family (GP = Gaussian Process, RF = Random Forest, NN = Neural Network, XGB = XGBoost). The experiment configurations themselves are defined on the Experiments page.

These are skill-validation metrics (forecast quality), distinct from the lightweight output smoke checks — see the Design Notes page for that distinction.

Overall [NBS] — RMSE

Model	Base	Month	SST	SWE	Chain	Local	Anomaly
GP	86.678	84.972	85.096	86.292	86.506	88.260	81.753
RF	84.653	85.039	84.979	84.729	84.679	84.750	83.484
NN	87.127	87.438	88.166	88.099	87.671	87.083	82.250
XGB	97.004	98.900	97.751	96.549	89.347	90.650	81.896

Overall [NBS] — R²

Model	Base	Month	SST	SWE	Chain	Local	Anomaly
GP	0.563	0.580	0.579	0.567	0.565	0.547	0.612
RF	0.584	0.580	0.580	0.583	0.583	0.583	0.595
NN	0.559	0.556	0.548	0.549	0.553	0.559	0.607
XGB	0.453	0.432	0.445	0.458	0.536	0.523	0.610

Overall [NBS, P, E, R] — RMSE

Model	Base	Month	SST	SWE	Chain	Local	Anomaly
GP	48.657	47.603	47.770	48.427	48.430	49.438	45.923
RF	47.529	47.751	47.723	47.564	47.546	47.574	46.926
NN	48.914	49.077	49.631	49.603	49.173	48.854	46.208
XGB	53.820	54.807	54.202	53.608	50.389	50.849	46.031

Overall [NBS, P, E, R] — R²

Model	Base	Month	SST	SWE	Chain	Local	Anomaly
GP	0.616	0.633	0.630	0.620	0.620	0.604	0.658
RF	0.634	0.631	0.631	0.633	0.634	0.633	0.643
NN	0.612	0.610	0.601	0.601	0.608	0.613	0.654
XGB	0.531	0.513	0.524	0.534	0.589	0.581	0.657

Overall [NBS] — RMSE by lake

Model	Superior (Local)	Superior (Domain)	Superior (Anomaly)	Michigan-Huron (Local)	Michigan-Huron (Domain)	Michigan-Huron (Anomaly)	Erie (Local)	Erie (Domain)	Erie (Anomaly)	Ontario (Local)	Ontario (Domain)	Ontario (Anomaly)
GP	48.175	49.043	45.287	49.858	50.618	43.768	115.176	111.308	107.810	114.399	112.676	105.568
RF	47.171	46.329	46.230	45.661	46.139	45.244	113.067	111.208	109.385	107.871	109.645	108.300
NN	47.419	48.916	45.704	50.832	50.135	44.096	116.351	111.269	108.204	109.379	113.058	106.390
XGB	50.484	49.379	45.426	57.020	54.916	43.983	122.543	122.215	107.553	109.784	131.333	106.121

Overall [NBS] — R² by lake

Model	Superior (Local)	Superior (Domain)	Superior (Anomaly)	Michigan-Huron (Local)	Michigan-Huron (Domain)	Michigan-Huron (Anomaly)	Erie (Local)	Erie (Domain)	Erie (Anomaly)	Ontario (Local)	Ontario (Domain)	Ontario (Anomaly)
GP	0.594	0.579	0.641	0.538	0.523	0.644	0.455	0.491	0.522	0.479	0.495	0.556
RF	0.611	0.625	0.626	0.612	0.604	0.619	0.475	0.492	0.508	0.537	0.521	0.533
NN	0.607	0.582	0.635	0.519	0.532	0.638	0.444	0.491	0.519	0.524	0.491	0.549
XGB	0.554	0.574	0.639	0.395	0.439	0.640	0.383	0.386	0.525	0.520	0.313	0.552

Overall [NBS] — Bias by lake

Model	Superior (Local)	Superior (Domain)	Superior (Anomaly)	Michigan-Huron (Local)	Michigan-Huron (Domain)	Michigan-Huron (Anomaly)	Erie (Local)	Erie (Domain)	Erie (Anomaly)	Ontario (Local)	Ontario (Domain)	Ontario (Anomaly)
GP	-9.418	-7.013	-9.799	-11.918	-6.147	-7.077	14.506	-4.024	3.498	-28.143	-15.843	-5.076
RF	-10.491	-9.587	-9.952	-4.863	-6.736	-6.866	17.672	8.524	2.935	-2.595	-3.486	-5.002
NN	-3.086	-2.186	-9.193	-5.127	-3.202	-5.007	22.798	0.968	4.432	-19.259	-23.634	-6.919
XGB	-11.908	-9.214	-9.930	-10.291	-10.181	-5.241	-9.576	-17.003	5.406	-9.877	-34.537	-5.185

Summary

"Anomaly-based inputs provide the largest performance gain across all models, with Random Forest offering the most consistent overall performance and Gaussian Process achieving the highest peak skill (but less robust)."