Low flow in Germany, 1992–2025
An area-weighted index built from … gauges of the NIWIS network (the low-flow information system run by Germany's Federal Institute of Hydrology) shows: low-flow spells begin earlier, last longer, and reach larger parts of the river network. The typical year of the past decade has shifted fundamentally.
Every day since 1992
Each cell is one day: the colour shows what share of the gauge network was below its local low-flow threshold (Q95 of the reference period) on that day. Rows are water years (November–October), 1992 at the top, 2025 at the bottom.
The national index
“National” here means: the NIWIS network. Not a single station comes from the state networks of Lower Saxony, Bremen and Hamburg; the north-west is represented only by the federal gauges on the Weser, Aller and Ems. On top of that, the catchments of the major rivers reach far beyond Germany's borders. The index describes the network, not the national territory.
Area-weighted mean across the gauge network, per water year. Switchable by network variant and metric. What the current selection means is stated directly below the controls.
The typical year is shifting
Distribution of low-flow days across all station-years, by decade. It is not only the extremes that are rising: the median moves from … days.
Areas: share of station-years with at least one low-flow day, per 10-day bin; the separated bar on the right collects everything above 200 days (overflow). Vertical dashed line: median of the decade.
Where the trends are
Each dot is one gauge; the colour shows the trend in low-flow days (Mann-Kendall, 1992–2025), the dot size the magnitude of change (Sen's slope). Base map: outline of Germany, federal states and major river courses (Natural Earth); the gauges sit on their rivers.
The map is an overview graphic; its dots are deliberately not individual tab stops. Every gauge can be reached by keyboard through the “Search gauges” field above the map. The detail view below the map shows the same information as the map tooltip, plus NM7Q trend, catchment and the year with the highest value.
Method & limits
Data basis
Daily discharge data from the low-flow information system NIWIS (Federal Institute of Hydrology / LAWA), collected by the responsible federal and state authorities. Per station, the data are subject to the licence terms of the collecting authority (mostly Datenlizenz Deutschland or state licences); this page shows only derived, aggregated metrics. Window: water years 1992–2025 (n = 34). Water year 1991 is incomplete and is therefore left out: NIWIS data begin on 1 January 1991, but the water year starts on 1 November 1990. Its first two months are missing, giving it 304 days instead of 365. Base map of the station map: Natural Earth, 1:10m (public domain), simplified to display scale.
Threshold & metrics
Low-flow day = daily discharge below the station's own Q95 of the fixed reference period 1992–2011. Metrics per station and water year: number of days below Q95, deficit volume (normalised to catchment area, mm) and Standardized Streamflow Index (SSI). Missing-value sentinels (−777) in the raw data are treated as gaps.
Area weighting & nesting
The national index weights stations by their incremental catchment area. Of 191 sink stations, 152 are classified as “nested” (area already contained in the basin outlet) and are excluded from the authoritative primary variant to avoid double counting (~9% of the topology area). A sensitivity run with all stations changes neither the direction nor the significance of the trends.
The reach of the term “national”
The weighted area is larger than Germany: depending on the water year, – to – km² of catchment contribute to the weighting (Germany measures roughly 357,600 km²), because the gauge chains on the Rhine, Elbe, Oder and Danube bring the complete headwaters of their rivers with them. For the five basin outlets it is documented that – km² of that (–% of the maximum weighting mass) lie outside Germany: in Switzerland, Austria, France, Luxembourg, Czechia and above all Poland. The Alpine tributaries Inn and Salzach lie outside these five chains and come on top of that. Counting them in, it is at least roughly 229,000 km² and thus about 45% of the maximum weighting mass. Conversely, the NIWIS network is missing the state networks of Lower Saxony, Bremen and Hamburg entirely: the whole north-west is represented only by federal gauges on the Weser, Aller and Ems, its tributaries (Leine, Hunte, Oste, Ilmenau) not at all. “National” here therefore means the measuring network, not the national territory. Derivation and sources of the area shares: docs/recherche-auslandsanteil.md.
Trend test
Trend-free prewhitening Mann-Kendall with Sen's slope, per station and for the national indices. All four aggregation variants (primary/near-natural, with/without nested stations) show the same picture: days and deficit rising significantly, SSI falling significantly (all p < 0.05). Robust against the reference period (WMO 1991–2020) and event pooling (3/5/7 days).
Testing many gauges at once
Anyone testing – gauges individually will find some “striking” gauges purely by chance: at a 5% threshold, one in twenty on average. That is why we additionally compute a correction (Benjamini-Hochberg, false discovery rate 5%) that accounts for all gauges being tested at the same time. Of the – gauges with significantly rising low-flow days, – remain significant after this correction; for the lowest 7-day discharge (NM7Q) the number of significantly falling gauges drops from – to –. The picture therefore holds; the correction only makes it more cautious. In the station detail view, the corrected value is shown for each individual station.
Extreme-value statistics: an honest non-finding
A non-stationary GEV fit to the national deficit finds no statistically significant trend in extreme behaviour (likelihood-ratio test p ≈ 0.23). With n = 34 this is a power problem, not evidence of stationarity. The shape parameter is unstable at this sample size; return-period estimates for 2018 vary between 12 and 35 years depending on the parametrisation and are deliberately not reported here as a single figure. The only robust statement is the distribution-free one: 2018 had the highest deficit of a 35-year series.
Limits
The near-natural subset covers only ~13–16% of the area of the primary variant and is geographically selective. It provides supporting evidence for the direction, not a “truer” trend. Homogeneity flags (reservoirs, transfers, mining) rest on documented but not exhaustive research. The series only begins in 1992; a longer preceding period was not included. The analysis shows robust trends, but no attribution of causes (climate, land use, water use).
Reproducibility
All figures on this page come unchanged from this project's analysis pipeline (out/national_index.csv, out/station_trends.csv, out/sensitivity.csv, data/interim/trend/). Method in detail: methods.md; results documentation: docs/ergebnisse-phase4.md. The references are file paths in the public repository.