Buffalo River Watershed Study

This was a nightly, AI-driven study of how rainfall becomes river rise in the Buffalo River watershed. It ran for 122 days — a full month past its planned 90 — kept going through an unusually wet June that delivered its most valuable data. Across 7 USGS gauges and 37 sub-basins it cataloged 22 rainfall-runoff events plus many sub-threshold near-misses.

What it learned:

• The same inch of rain moves a gauge far more when the ground is already wet — St. Joe rises about 0.7 ft per inch when dry vs. ~2.1 ft per inch when saturated, a 3× swing now captured in per-gauge calibration tables.
• Scattered warm-season rain can produce less rise than a dry-season soak when it falls below a mainstem “coherence” threshold — a counterintuitive pattern the study named and confirmed.
• The upstream→downstream cascade is consistently timed: about 10 hours from St. Joe to Harriet, and 33–45 hours for a headwater pulse at Boxley to reach the lower river.
• Each gauge has a signature — Bear Creek barely responds (a structural low), Pruitt rises in two stages (fast surface, then slow karst), Richland flashes hard.
• The June 22 flash flood (study-record Boxley 9.06 ft, Ponca ~8,700 cfs, Pruitt 14.27 ft) anchored the extreme end of every curve — the single most valuable calibration of the run.

These findings feed the live Creek Intelligence forecasts — watershed alerts, rise likelihoods, and downstream propagation. The larger AI model that authored these nightly write-ups has now been retired; a local model keeps a lightweight nightly check, reviewed by hand from time to time. The record below is frozen as of Day 122 — the final hypothesis document and the full daily archive remain browsable.