Hotstart Levels

Hotstart Levels

Building project forcing can take a long time, especially LCS life cycles. When a case is run again (same project, PSE, alternative, workflow and storm database), CHESS compares the new config with the cached *_config_file.mat and reuses as much as it safely can. The comparison is hotstart_level.m / hotstart_plan.m, the action is apply_hotstart_level.m; the library (resolve_hotstart_state.m) and the GUI share both.

Levels

The first matching row wins.

Level	Triggered by a change in	What is rebuilt	What is kept
resample	mcs_nYears, mcs_nLC, lcs_restrict_sampling_rate, the storm database build date; a request the cache does not cover (WLP/WHP, Timeseries switched on, a new storm type such as XC → CC)	Everything: project forcing, and for LCS new life cycles (with the convergence check)	Nothing
reuse_sampling	storm_duration; any forcing adjustment input: apply_depth_limitation, apply_random_tides, tide_file, swl_slr, toe / wall bottom / berm geometry, struc_type	Project forcing, rebuilt from the storm database; uncertainty and adjustments applied again	LCS: the same life cycles (storm IDs, years, order) from *_sampling_seed.mat, the XC copula values and the sampling mode. Tide start hours when the tide settings are unchanged
full	Nothing above	Nothing	The cached project forcing as it is
cold	No cache, or a cache that fails to load	Everything	Nothing

PROS and DET have no random sampling, so reuse_sampling and resample both rebuild their forcing from the storm database.

Only the run's own project forcing and LCS seed file are ever deleted, never folders: other storm databases share the workflow folder. Each level prints a Hotstart: line naming the field that triggered it.

Why these levels

• Responses are always recomputed. Response inputs (crest, slopes, armor, switches) change only step 6, which is cheap.
• reuse_sampling keeps alternatives comparable. Two LCS alternatives that differ only in geometry or adjustments see the same storms in the same years, so differences in the results come from the design, not from sampling noise.
• resample protects against stale statistics. A new life-cycle count, sampling mode or storm database invalidates the sampled population.

Reusing another case (LCS)

With load_project_forcing = 1 and ref_case_name, an LCS case reuses the reference case's life cycles instead of its own: storm identities from the reference's seed file, its sampling mode and its XC copula values. Storm values are rebuilt from the storm database. When the reference has no project forcing or no usable seed file, CHESS warns and samples fresh storms.

GUI

Pressing Run on a case that already has results opens a dialog: Hotstart / Coldstart / Load existing / Cancel. It states the level and the reason. Hotstart is hidden when the level is resample or the needed cache is missing. Every GUI run backs up the case first and restores it if the run fails.