THULab/olmoearth_potomac_sewage_spill

Potomac Sewage Spill Reference Dataset (2026) (TsFile)

Apache TsFile version of BAIGroup/OlmoEarth-v1-Potomac-Sewage-Spill-2026.

Overview

Validated, time-aligned reference dataset supporting GeoAI tracking of the 2026 Potomac River sewage spill (Glen Echo, MD — 240–300 million gallons released from the 72-inch Potomac Interceptor on January 19, 2026). The original repository provides an event timeline, hydrologic context, monitoring station locations, and AOI corridor polygons for AI2's downstream Sentinel-1/2 plume detection.

This TsFile copy contains the tabular daily streamflow time series only — the part of the repository that is a time series:

• 8 USGS gauges documented along the corridor; 846 daily flow records, December 2025 – March 2026, daily granularity. Seven gauges carry data; site 01651000 (anacostia_lower, anacostia_lower role) had no NWIS data in the event window and therefore contributes no rows.
• Source: USGS NWIS daily values service, parameter 00060 (discharge, cfs), statistic 00003.
• Each record is one gauge-day, phase-tagged for the four-phase spill chronology: pre_spill_baseline (2025-12-01 → 2026-01-18), active_release (2026-01-19 → 2026-01-24), bypass_period (2026-01-25 → 2026-03-14) and post_recovery (2026-03-15 → 2026-03-31).
• This is an event reference / validation dataset, not a training set: 32 station-phase reference points are far below any fine-tuning threshold. BAI prepares time-aligned reference data; the OlmoEarth team performs the Sentinel-1/2 plume detection.

Schema (TsFile structure)

• Time (INT64, milliseconds) — daily timestamp parsed from the source date column (YYYY-MM-DD), i.e. 00:00:00 UTC of each day.
• site_no (TAG) — USGS station number, e.g. query one gauge with WHERE site_no='01646500'.
• discharge_cfs (FIELD, DOUBLE) — daily mean streamflow in cubic feet per second.
• role (FIELD, STRING) — corridor role of the gauge (e.g. near_spill, tidal_dc, lower_potomac).
• phase (FIELD, STRING) — spill phase for that gauge-day.
• name (FIELD, STRING) — human-readable USGS station name.
• latitude (FIELD, DOUBLE), longitude (FIELD, DOUBLE) — gauge coordinates.

role, name, latitude and longitude are constant per station and are therefore repeated on every row; they are kept as FIELD columns so no source column is dropped. They can simply be ignored when only the flow measurements are needed.

Type mapping: the source discharge_cfs, latitude and longitude are double and stay DOUBLE; the remaining non-time columns are strings and stay STRING.

Usage

Install the Apache TsFile Python SDK (pip install tsfile) and read a converted file:

from pathlib import Path
from tsfile import TsFileReader

path = Path("potomac_flow_daily.tsfile")
with TsFileReader(str(path)) as reader:
    schemas = reader.get_all_table_schemas()
    print("tables:", list(schemas))
    table_name = next(iter(schemas))
    table = schemas[table_name]
    columns = [column.get_column_name() for column in table.get_columns()]
    print("columns:", columns)
    field_names = [
        column.get_column_name()
        for column in table.get_columns()
        if column.get_column_name() not in {"Time", "time"}
    ]
    if field_names:
        with reader.query_table(table_name, field_names[:3], batch_size=1024) as result:
            batch = result.read_arrow_batch()
            if batch is not None:
                print(batch.to_pandas().head())

Source & license

• Original dataset: https://huggingface.co/datasets/BAIGroup/OlmoEarth-v1-Potomac-Sewage-Spill-2026
• Author / publisher: Ziming Qi and BAI Group
• Paper: none
• License: Apache-2.0 (USGS source data and agency reports are public domain)

Conversion notes

• One source file only. data.parquet (846 rows) and tabular/potomac_flow_daily.csv are documented mirrors of the same daily flow table. Only data.parquet was converted; converting both would duplicate every row. The parquet was preferred because it preserves site_no as a zero-padded string (01638500), whereas the CSV parses it as an integer and drops the leading zero.
• Not time series / not included. The geospatial assets (.geojson files such as waterbodies.geojson, monitoring_stations.geojson, spill_metadata.geojson, event_log.geojson), timeline.json, event_log.csv, the CBP manifests, reference_event_logs/, docs and studio//rslearn/ assets are not time series and are not part of this TsFile copy. The original dataset remains available for those geospatial and event-narrative assets.
• Missing values. 18 of the 846 discharge_cfs values are null in the source (real gaps in the USGS daily record); they are preserved as nulls and were not imputed or dropped.