The pale outline is the full ~3,809 km² basin that drains to the catchment outlet used for this study's 36-year discharge record. The deep teal shape is not an administrative boundary — no such file exists for this project — it is the real, data-derived footprint of every grid cell with a modeled 100-year flood depth (the largest of the five modeled storms, and a superset of the 5/10/25/50-year footprints), i.e. the area this system actually exists to warn about. Its winding shape traces the main drainage channel.
| Level | Return period | Discharge |
|---|
| Peak date | Level | Peak discharge | Duration |
|---|
The system has two parts that share one set of thresholds: a statistical baseline built once from 36 years of historical daily discharge, and a live forecast feed checked against that same historical record and classified against the same thresholds. The six steps below cover both.
36 years (1990–2025) of daily discharge at the Barnawa catchment outlet, sourced from GEOGloWS — a global, physically-based hydrological reanalysis that simulates historical streamflow for river reaches worldwide, including ungauged ones like this catchment. This is the same record used throughout the rest of this site.
The single highest daily discharge in each of the 36 years was extracted, ranked, and fit to a Generalised Extreme Value (GEV) distribution using the method of L-moments — linear combinations of the ranked values that are more stable than ordinary moments on a 36-point sample, and the standard approach in hydrological frequency analysis. GEV is the general family; its shape parameter k decides whether the tail is heavy (k<0), bounded (k>0), or exactly Gumbel (k=0).
k came out at essentially zero, meaning this catchment's floods are well described by the simpler Gumbel (EV-I) distribution — a result of the fit, not an assumption made to simplify it.
The fitted distribution gives a discharge for any return period T (the "1-in-T-year event"): Q(T) = ξ + (α/k)·[1−(−ln(1−1/T))k]. Three return periods were chosen as the reporting thresholds — 5-year (Watch), 25-year (Warning), and 100-year (Emergency) — matching common civil-engineering convention for notable, serious, and design-flood-level events respectively. A 2-year discharge is also shown on the charts as an informal "Advisory" band, but it sits below the reporting threshold and is not a fourth alert level.
The full 36-year daily record was scanned day by day. Any run of days at or above the Watch threshold — allowing a single dry day in between without breaking the run — is grouped into one "episode," recording its peak date, peak discharge, duration, and highest level reached. This recovers the flood episodes on record, including a genuine one-day spike in 2008 just above the Watch threshold.
For the 2012 and 2022 flood seasons (the years with hourly rainfall data available), a rolling 24-hour rainfall total was computed and the most recent point before the discharge peak where it crossed 20 mm — a standard heavy-rainfall threshold — was found by searching backward from the peak. The gap between that rainfall trigger and the discharge peak is the case studies' "lead time," roughly 19–32 hours in these two events. That real lead time is why the live system watches rainfall as well as river flow.
The live discharge forecast comes from GEOGloWS reach 140653280, at the catchment's documented outlet. Before trusting it, its retrospective daily series was joined against this study's own streamflow record: 13,149 overlapping days, zero discrepancy — confirming it is the same reach the historical record itself comes from, not a nearby stand-in. Because of that, the live forecast's daily median is classified directly against the Watch/Warning/Emergency thresholds from step 3. The 25th–75th percentile spread is still shown for context, but only the median drives the classification, to avoid over-warning on the upper band.