Experiments

A summary of the experiment configurations explored during development. Each experiment varies the domain, initial conditions, framework, or input variables to test a specific hypothesis about what improves forecast skill. Results for these experiments are recorded on the Skill Metrics page.

Last updated from source: 03/04/2026.

Experiment summary

Branch Name	Domain	Initial Conditions	Framework	Variables
Base	GL	None	1-step	Precipitation, Calc. Evaporation, Air Temperature
Seasonal	GL	Month	1-step	Precipitation, Calc. Evaporation, Air Temperature
SST	GL	SST	1-step	Precipitation, Calc. Evaporation, Air Temperature
SWE	GL	SWE	1-step	Precipitation, Calc. Evaporation, Air Temperature
Chain	GL	Month	2-step	Precipitation, Calc. Evaporation, Air Temperature
Local	Lake	None	1-step	Precipitation, Calc. Evaporation, Air Temperature

Definitions

Branch name

Experiment scripts and full skill metrics can be found under the listed branch name.

Domain

• GL (Great Lakes): all lake variables are used as inputs for all targets.
• Lake: only the corresponding lake's variables are used for that lake's targets. For example: erie_lake_precipitation, erie_lake_evaporation, erie_lake_air_temperature → erie_precipitation, erie_evaporation, erie_runoff, erie_nbs.

Initial conditions

• Month: forecast month used as a dummy variable, to test improvements to seasonality.
• SST: mean sea surface temperature on the first day of the forecast month, for all lakes — to test improvements to evaporation.
• SWE: mean snow water equivalent on land on the first day of the forecast month.

Framework

• 1-step: precipitation, evaporation, air temperature → precipitation, evaporation, runoff, nbs
• 2-step: precipitation, evaporation, air temperature → precipitation, evaporation, runoff → nbs