Waterway health by zone
Latest usable satellite reading for each zone, with the change against that zone's own 12-month median. Click a card for the full history. Confidence drops in narrow, dark, tree-lined canals where 10–20 m pixels mix water with banks and docks.
Show as table
How this works
-
Imagery. Sentinel-2 Level-2A surface reflectance (ESA Copernicus),
MGRS tile 17RNJ, read directly from the public AWS Open Data bucket
(
s3://sentinel-cogs). Only passes where at least half the study area is cloud-free are used; clouds and cloud shadows are buffered by ~100 m because haze at cloud edges reads as turbid water. - Where the water is. ESA WorldCover 2021 (10 m) permanent-water class, shrunk by one pixel to drop mixed bank/dock pixels. Each pass then rejects boats, bridges and sun glint (bright shortwave-infrared), and pixels where the atmospheric correction clipped red to zero — common in the darkest tannic canals.
- Atmospheric residual. The standard correction (Sen2Cor) is built for land and leaves a small haze offset over water. Each pass subtracts the darkest 1% of deep offshore water in the red, red-edge, NIR and SWIR bands (dark-object subtraction), where true water-leaving signal is essentially zero.
- Zones. Seven waterway zones are scored separately, using the median of every valid water pixel inside each.
False-color composites
Indicators & health score
| Indicator | Formula (Sentinel-2 bands) | Good | Poor | Weight |
|---|---|---|---|---|
| Chlorophyll-a proxy — NDCI Mishra & Mishra 2012 | (B5 − B4) / (B5 + B4) | ≤ −0.10 (≈7 µg/L) | ≥ +0.10 (≈25 µg/L) | 45% |
| Turbidity (FNU) Dogliotti et al. 2015 | 228.1·ρ / (1 − ρ/0.1641), ρ = B4 | ≤ 3 | ≥ 20 | 40% |
| Floating algae / sargassum — FAI Hu 2009 | B8 − [B4 + (B11 − B4)·(833−665)/(1610−665)], share of water > 0.01 | ≤ 1% | ≥ 15% | 15% |
Each indicator scales linearly from 100 (good) to 0 (poor); the weighted sum is the health score. Good ≥ 70 Fair 45–69 Poor < 45. For regulatory context, Florida's Class III surface-water turbidity standard is 29 NTU above natural background (Rule 62-302.530, F.A.C.); NOAA's estuarine eutrophication assessment treats chlorophyll-a above 20 µg/L as high. The approximate chlorophyll-a equivalents use Mishra & Mishra's NDCI curve, which was fitted in productive, turbid waters — treat them as rough.
Limits — read before acting on a number
- These are remote-sensing screening indicators, not lab measurements. They are uncalibrated against local sampling; the index-to-concentration relationships were developed elsewhere. Use them to decide where and when to sample, and compare against Broward County's ambient water-quality monitoring before drawing conclusions.
- Narrow canals are hard. Middle River, New River and many finger canals are 20–60 m wide, i.e. 2–6 pixels. Bank vegetation, docks, boats and shade leak into those pixels, and the red-edge band is 20 m. These zones carry a confidence rating; treat "low" as "look at the imagery, don't trust the score."
- Shallow, clear water (nearshore reef, sand flats) reflects the bottom, which can read as turbidity. Dark tannic water (dissolved organic matter from runoff) absorbs light and can read as low turbidity even when it isn't.
- Bacteria (enterococci), dissolved oxygen, and nutrients are invisible from orbit. Swim advisories and fish-kill risk still need field data.
- Rainy-season cloud cover (June–October) leaves long gaps between usable passes.
Refreshing the data
Everything on this page is produced by scripts/wq_pipeline.py (Python:
rasterio, numpy, pillow, pyproj), which writes wq-data.js and the images in
wq/img/. A scheduled GitHub Action re-runs it weekly. No accounts or API keys
are needed.