Main Workflow Case Study

This study runs the standard hydroseason-v0.2.0 workflow on five Australian catchments using committed 30 m whole-catchment monthly surface-water extent series from DEA Water Observations (2005–2025, 252 months each). The reports are rebuilt offline from those committed series.

Method

analyze_catchment decides each route from the water record alone:

• Seasonal (per_year_detection): circular Kuiper tests reject uniform timing for both annual peaks and troughs (p < 0.05), and at least seven peak and trough cycles are resolved. Each hydrological year runs from one dynamically detected trough to the next.
• Aseasonal (event_characterisation): recurrence is not established. No hydrological years are forced; the report gives wet events, low-extent spells, and overall variability.

The build uses quality_policy="flag": finite monthly observations stay available for cycle mapping, while invalid_pct is carried as a quality state that can make a boundary provisional. Months with no observed extent or 100% invalid coverage remain unusable. The CSV columns are documented in Report CSV columns.

Each report also shows peak and trough timing concentration (R, the mean resultant length) with bootstrap confidence intervals and the timing IQR. These are descriptive; they do not set the route.

What extent can and cannot tell you

invalid_pct uses the fixed historical-mask pixel count as its denominator, and extent_pct uses valid pixels inside the same mask. Surface-water extent is not discharge, depth, storage volume, or ecological condition. Whole-catchment percentages dilute narrow channels across broad dry landscapes, and optical detection can miss water under dense vegetation or persistent cloud.

Results

Catchment	Regime	Route	Peak R	Trough R	Trough R CI low	Peak-month IQR (months)	Hydro Years	Events	Longest Low Spell (months)	Peak Month	Trough Month
Daly River (NT)	seasonal	per_year_detection	0.878	0.777	0.684	2.0	21	21	6	Mar	Nov
Fitzroy River (WA)	seasonal	per_year_detection	0.900	0.892	0.866	2.0	21	18	8	Feb	Nov
Gilbert River (QLD)	seasonal	per_year_detection	0.934	0.908	0.892	1.0	21	24	10	Feb	Nov
Lachlan River (NSW)	aseasonal	event_characterisation	0.393	0.607	0.313	4.0	0	5	55	N/A	N/A
Moonie River (QLD/NSW)	aseasonal	event_characterisation	0.444	0.605	0.423	3.0	0	14	22	N/A	N/A

Kuiper p-values (peak / trough): Daly 0.001 / 0.002, Fitzroy 0.001 / 0.001, Gilbert 0.001 / 0.001, Lachlan 0.262 / 0.139, Moonie 0.232 / 0.009.

Findings

1. Daly, Fitzroy, and Gilbert are seasonal. Both peak and trough timing recur in the same calendar months (all p ≤ 0.002), and each record clears the seven-cycle guard, yielding 21 hydrological years. Daly's March 2011 maximum is kept but marked anomalous because 87.2% of its pixels were invalid; that flags the cycle as provisional without changing the regime or route.
2. Lachlan and Moonie are aseasonal. Lachlan fails both tests. Moonie's troughs recur (p = 0.009) but its peaks do not (p = 0.232), and the method requires both. Both records route to event characterisation: Lachlan has 5 wet events and a 55-month longest low spell; Moonie has 14 events and a 22-month longest low spell. No annual boundaries are forced.