I just finished a 12-hour flood-frequency refresher, and it struck me how our CE still treats the world as stationary when the governing dynamics are shifting under climate, land use, and institutions. For those planning 2026 CE hours, would you trade some tool tutorials for deeper work in Bayesian uncertainty, causal inference, and socio-hydrology frameworks, or am I overvaluing theory at the expense of field pragmatics?
In our last basin study, we swapped a GIS button-click tutorial for a 90‑min prior‑elicitation lab and, after adding an ENSO covariate, our flood quantile HPD widths shrank about 20% — “stationarity is dead” only matters if you can say what replaces it, which felt like updating firmware before a storm. I’d trade one tool class for Bayesian uncertainty plus a quick DAG walkthrough, but I’d still keep a short annual tool refresher for folks who don’t script much.
I’d trade 2 of those 12 hours for a live Bayesian workflow: sketch a causal DAG for climate/land-use/institution drivers, elicit a prior, and fit a time-varying GEV with a covariate like seasonal soil moisture, then poke it with posterior predictive checks. @hwrigh12’s point lands — once we acknowledged “nonstationary,” our urban retrofit sizing stopped seesawing year to year and the credible bands finally made sense to PMs. Small caveat: keep one focused tool block on reproducible data QA, because bad rating curves and gaps will swamp any fancy prior.
We’ve gotten the most mileage from a 60‑min “ppc sprint”: fit a GEV with time‑indexed covariates, then simulate futures to see where it breaks when ENSO, imperviousness, or permit cycles shift. Building on @michael8126’s DAG idea, we sketch a tiny causal graph to pick covariates, but I’d still keep an hour on reproducible tooling (versioned scripts and seed control) so the results don’t turn into a one‑off magic trick.
But borrowing @sandraW66’s idea, after that ‘12-hour’ refresher I carved out 45 min for a quick DAG + prior check and let a time‑varying GEV include a 1998 change‑point with ENSO; the 1% AEP nudged about 0.3 ft and PPCs finally passed. I’m all for the Bayesian/causal shift, but keep a 30‑min bridge to Bulletin 17C so folks can map the nonstationary fit to LP3 reports.