Carson River Watershed Atlas
Every public water measurement in a watershed, on one map
October 4, 2026
The problem
The Carson River has been measured for more than a century: by the U.S. Geological Survey, the Nevada Division of Environmental Protection, the Washoe Tribe, California’s water boards, universities, and volunteers. Almost none of that record is something a resident, a teacher, or a county commissioner could look at. It sits in agency databases built for people who already know how to query them.
The data
- Water quality: every sample in the basin from the Water Quality Portal (EPA and USGS, WQX 3.0), archived as dated snapshots so every build is reproducible.
- Water quantity: USGS daily streamflow, NRCS SNOTEL snowpack, and groundwater levels.
- Geography: NHDPlus flowlines and waterbodies, the Watershed Boundary Dataset, Census TIGER places and population blocks.
- Context: NIFC wildfire perimeters, the U.S. Drought Monitor, state impairment assessments, Nevada’s numeric water quality standards, and the USGS Publications Warehouse.
The hard part was not getting the data but making it agree with itself: more than a dozen parameters reported in dozens of unit and naming variants, nondetects reported every way they can be, and stations from 1900 to this month. A single parameter catalogue and unit harmonizer, with unit tests for every conversion, is what every view downstream trusts.
What I built
Two repositories that work as one system:
- The pipeline (
targetsin R) pulls the raw public record and archives it. - The atlas turns that record and federal hydrography into a GeoPackage for QGIS and a static web map, in numbered build steps.
The map is the whole product: one HTML page, vector tiles in a single PMTiles file, and small JSON files fetched only when needed. No server, no database, no API keys. Inside it:
- The river as a line: every station snapped to the river and placed by distance from its mouth, with towns and reservoirs as landmarks.
- The basin breathes: a water year of snowpack and streamflow played in thirty seconds.
- A trend engine: Seasonal Kendall tests on monthly medians with Sen’s slope for every station and parameter with enough record, with censored values handled rather than dropped.
- Stories told with data: a wildfire or a historic mine traced downstream through the river network, with before/after and upstream/downstream comparisons computed from the samples.
- River Right Now: live flow against normal, weather alerts, fire, and drought, embeddable on any site.
- Look up a date: what the river was doing on a birthday or the day you moved here.
Every watershed is a configuration file, not new code. The Truckee and the Walker followed the Carson from the same pipeline.
Tools
- R (
sf,targets, Parquet viaarrow) for acquisition, harmonization, statistics, and spatial work, in a reproducible pinned environment (renv). - tippecanoe and PMTiles to pack every vector layer into one file a static host can serve by byte range.
- MapLibre GL JS with no framework and no build step, so the page outlives any toolchain.
- QGIS as the cartographic workbench and print atlas.
- GitHub Actions running offline checks on every pull request: configs, scripts, templates, and every unit conversion.
Outcome
- Three atlases live from one codebase: the Carson (2,741 sites, 149,025 measurements), the Truckee, Lake Tahoe to Pyramid Lake (1,995 sites, 255,404 measurements), and the Walker (1,083 sites, 89,599 measurements).
- A new watershed is a configuration file and a data pull, not a software project.