SWOT obs coverage gap for type=3 (lake_on_river) — need LakeSP integration

[jameshgrn]

Problem

aggregate_swot_observations reads from SWOT RiverSP parquet files only. Type=3 (lake_on_river) reaches have 13-54% observation coverage vs 97-99% for type=1 rivers, because lake reach measurements in RiverSP are mostly sentinel values (-1e12) that get filtered out by quality gates.

Coverage by type (current)

Type	Description	Coverage
1	River	97-99%
3	Lake on river	13-54%
4	Dam	80-90%
5	Unreliable	95-99%
6	Ghost	0% (correct)

Type=3 detail by region

Region	Total type=3	With obs	%
AF	1,551	318	20.5%
AS	7,631	1,896	24.8%
EU	4,159	888	21.4%
NA	9,194	1,248	13.6%
OC	808	436	54.0%
SA	3,485	521	14.9%

Root cause

• SWOT publishes lake observations in the LakeSP product, not RiverSP
• Type=3 reach_ids DO appear in RiverSP parquet files (6,551 of 9,194 NA type=3), but river-specific fields (slope, width) are mostly sentinels
• Our quality filters correctly reject these sentinel values → NULL obs stats for most lake reaches

Proposed fix

Integrate SWOT LakeSP parquet data for type=3 reaches:

1. Acquire LakeSP parquet files (same source as RiverSP)
2. Map LakeSP observations to SWORD reach_ids (lake_id → reach_id mapping needed)
3. Aggregate WSE stats from LakeSP (slope/width may not apply the same way for lakes)
4. Mark data source in a new column or in slope_obs_quality so downstream users know which product the stats came from

Notes

• swot_obs (max passes per 21-day cycle) is >0 for most type=3 reaches — SWOT CAN observe them, we just aren't reading the right product
• This does NOT affect type=1 river coverage (97-99%) or any shipped v17c columns
• Lake WSE from LakeSP may be more reliable than RiverSP anyway (purpose-built for lake surfaces)