cassini-team-todo/eea-waterbase

EEA Waterbase (WISE-4) — v2018.1

Mirror of the European Environment Agency's Waterbase – Water Quality ICM (WISE-4) tabular release, version 2018.1. Contains station-level and water-body-level measurements of chemical and biological determinands in European surface and ground waters, plus the monitoring-site registry with coordinates.

Uploaded here for convenient team access during the 11th CASSINI Hackathon – EU Space for Water. This is a redistribution of the original EEA CSVs, converted to Parquet (snappy compression) for faster loading and type preservation. See Source & Licensing below.

Files

File	Size	Rows	What it is
Waterbase_v2018_1_T_WISE4_DisaggregatedData.parquet	331 MB	33,848,578	Per-sample measurements (one row per sampling date)
Waterbase_v2018_1_T_WISE4_AggregatedData.parquet	82 MB	3,211,183	Per-site yearly aggregates (min/mean/max/median/stddev, LOQ flags)
Waterbase_v2018_1_T_WISE4_AggregatedDataByWaterBody.parquet	507 KB	20,251	Same aggregates rolled up per water body, with per-class site counts
Waterbase_v2018_1_T_WISE4_BiologyEQRData.parquet	587 KB	29,741	Biological Ecological Quality Ratio results per monitoring site
Waterbase_v2018_1_T_WISE4_BiologyEQRClassificationProcedure.parquet	63 KB	2,553	EQR classification boundary values (country × water-body type × determinand)
Waterbase_v2018_1_WISE4_MonitoringSite_DerivedData.parquet	1.56 MB	56,464	Monitoring-site registry: IDs, water-body link, lon/lat, confidentiality flag

Total: ~416 MB across all six files (vs. ~10 GB as CSV).

Schema highlights

Measurement tables (Disaggregated, Aggregated, AggregatedDataByWaterBody)

Shared keys:

• monitoringSiteIdentifier / waterBodyIdentifier + *IdentifierScheme — join keys
• parameterWaterBodyCategory — RW (river), LW (lake), GW (groundwater), TW/CW (transitional/coastal)
• observedPropertyDeterminandCode — typically CAS codes, e.g. CAS_7440-38-2 (arsenic), CAS_14797-55-8 (nitrate)
• procedureAnalysedFraction, procedureAnalysedMedia, resultUom (e.g. ug/L, mg{NO3}/L)
• procedureLOQValue — limit of quantification; paired resultQuality*BelowLOQ flags

Disaggregated adds phenomenonTimeSamplingDate + resultObservedValue. Aggregated / AggregatedByWaterBody add phenomenonTimeReferenceYear, parameterSamplingPeriod, resultNumberOfSamples, and min/mean/max/median/stddev columns. AggregatedByWaterBody additionally provides resultNumberOfSitesClass1..5.

Biology EQR tables

• observedPropertyDeterminandBiologyEQRCode — EEA_* codes (instead of CAS)
• resultEcologicalStatusClassValue, resultEQRValue, resultNormalisedEQRValue
• Classification procedure table gives boundary values for classes 1/2, 2/3, 3/4, 4/5 per country and water-body type

Monitoring sites

monitoringSiteIdentifier, waterBodyIdentifier, confidentialityStatus, lon, lat. The join key for putting any measurement on a map.

Usage

With datasets

from datasets import load_dataset

# Small tables — fine to load fully
sites = load_dataset("cassini-team-todo/eea-waterbase", "monitoring_sites", split="train")
eqr   = load_dataset("cassini-team-todo/eea-waterbase", "biology_eqr",      split="train")

# Large table — stream to avoid materialising all ~34M rows in memory
disagg = load_dataset(
    "cassini-team-todo/eea-waterbase",
    "disaggregated",
    split="train",
    streaming=True,
)
for row in disagg.take(5):
    print(row)

With pandas / pyarrow (direct file access)

import pandas as pd

sites = pd.read_parquet("Waterbase_v2018_1_WISE4_MonitoringSite_DerivedData.parquet")
agg   = pd.read_parquet("Waterbase_v2018_1_T_WISE4_AggregatedData.parquet")

# Read only the columns you need from the 33M-row disaggregated file
cols = ["monitoringSiteIdentifier", "observedPropertyDeterminandCode",
        "phenomenonTimeSamplingDate", "resultObservedValue", "resultUom"]
disagg = pd.read_parquet(
    "Waterbase_v2018_1_T_WISE4_DisaggregatedData.parquet",
    columns=cols,
)

Joining measurements to coordinates

import pandas as pd

sites = pd.read_parquet("Waterbase_v2018_1_WISE4_MonitoringSite_DerivedData.parquet")
agg   = pd.read_parquet("Waterbase_v2018_1_T_WISE4_AggregatedData.parquet")

geo = agg.merge(sites[["monitoringSiteIdentifier", "lon", "lat"]],
                on="monitoringSiteIdentifier", how="left")

Known quirks

• Numeric columns use . as decimal separator; missing values are null.
• parameterSamplingPeriod is an ISO-interval-ish string (2012-01--2012-12), not a proper date.
• phenomenonTimeSamplingDate (Disaggregated) and metadata_beginLifeSpanVersion are proper timestamps in Parquet.
• LOQ handling is explicit: a resultQualityMeanBelowLOQ = 1 flag means the reported mean is a substitution, not a direct measurement.
• Some metadata_observationStatus = U rows carry QC_LEGACY_* remarks — filter if you want only A (accepted) records.

Source & Licensing

• Publisher: European Environment Agency (EEA)
• Original URL: https://www.eea.europa.eu/en/datahub (search "Waterbase – Water Quality ICM") — product discovery page: https://discomap.eea.europa.eu/
• Version: v2018.1, published 2018-04-05
• Format: Original CSVs were converted to Parquet (snappy compression) with pyarrow. No rows were filtered or modified; the UTF-8 BOM on the first column header was stripped. Schema matches the original 1:1.
• Reuse: Governed by the EEA legal notice — https://www.eea.europa.eu/en/legal-notice — which authorises reuse with attribution. Users of this mirror must comply with the EEA's terms. We are only redistributing for hackathon convenience and claim no additional rights.

Attribution

Source: European Environment Agency, Waterbase – Water Quality ICM (WISE-4), version 2018.1.

Citation

European Environment Agency (2018). Waterbase – Water Quality ICM (WISE-4), v2018.1.
https://www.eea.europa.eu/en/datahub

Project context

Part of the 11th CASSINI Hackathon – EU Space for Water. Combined with Copernicus Sentinel-2/3 observations, this in-situ record serves as ground truth for satellite-derived water-quality products.