Freshwater longitudinal connectivity and barrier mapping for migratory fish habitat
Dams, weirs, and culverts fragment river networks in ways that compound across the catchment graph. Satellite-derived DEMs and multispectral imagery can map that graph and flag artificial barriers; they cannot, alone, tell you whether a fish can pass.
Sensors
- TanDEM-X: Global DEM at 12 m posting (0.4 arcsec product), vertical accuracy approximately 2 m absolute at 90th percentile over vegetated terrain. The primary source for drainage-network delineation and hydraulic gradient estimation across large catchments.
- FABDEM (Forest-and-Building-Removed DEM): A bias-corrected derivative of Copernicus DEM that strips canopy and built structures using machine-learning correction trained on ICESat-2 ground returns. Reduces vegetation-induced elevation bias from several metres to sub-metre in many forest types, improving channel-bed extraction under riparian canopy.
- Sentinel-2 MSI: 10 m visible and near-infrared bands, 20 m red-edge and SWIR, 5-day revisit at the equator (2-3 days at mid-latitudes with both satellites). Used for surface-water delineation via MNDWI, detection of backwater impoundment signatures, and identification of sediment-plume discontinuities that indicate flow obstruction.
- Sentinel-1 SAR C-band: 6-day repeat in IW mode, 10 m ground-range resolution. SAR backscatter responds to surface roughness; the abrupt transition from rough turbulent flow below a weir to calm specular water above it produces a detectable backscatter discontinuity even when cloud obscures optical imagery.
- ICESat-2 ATL03/ATL08: Photon-counting lidar with approximately 17 m along-track footprint spacing and sub-decimetre vertical precision over water surfaces. Useful for validating channel-bed elevation at specific cross-sections and for calibrating DEM-derived hydraulic gradients, though spatial coverage is track-limited.
Why the graph matters more than any single dam
River networks are directed graphs: water and fish move from tributary to trunk in one dimension, and every barrier on that path multiplies the isolation of habitat upstream. The dendritic connectivity index (DCI), developed by Petrou and others and formalised in peer-reviewed hydrology literature, weights each barrier by the proportion of total catchment area it disconnects. A small weir near the river mouth can score worse than a large dam high in a headwater tributary, because it severs access to everything above it.
This is why a list of structures is not the same as a connectivity model. Governments and conservation agencies increasingly need the full catchment graph, with barriers located within it and ranked by their marginal effect on connectivity. Satellite-derived data can build that graph at national scale; field surveys then focus effort on the structures that actually matter.
Extracting the river network from elevation data
Standard D8 or multiple-flow-direction algorithms applied to a hydrologically conditioned DEM produce a channel network, a flow-accumulation raster, and Strahler stream-order attribution. The conditioning step, filling sinks and enforcing drainage through flat areas, is sensitive to DEM quality. TanDEM-X at 12 m posting is currently the best freely accessible global product for this purpose, but its surface model retains canopy height over forested riparian zones. FABDEM's vegetation-removal correction reduces this bias materially, though it is not perfect: dense, narrow riparian gallery forest still challenges the correction algorithm, and residual errors of 1 to 3 m are plausible in those settings.
Channel width sets a practical floor on what is detectable. Rivers narrower than roughly 30 m are difficult to delineate reliably from 10 m Sentinel-2 imagery using spectral water indices alone; at 5 m width the problem becomes severe. Headwater streams that are ecologically important for salmonids, for instance, may not appear in the satellite-derived network at all. Any connectivity analysis should state its minimum channel-width threshold explicitly and acknowledge that the upstream headwater graph is likely incomplete.
Detecting barriers: what a floating roof gives away, and what a culvert hides
Large impoundments are straightforward. A reservoir behind a dam produces a persistent, specular open-water polygon upstream of a structure that is itself visible as a bright linear feature in both Sentinel-2 true colour and Sentinel-1 VV backscatter. The backscatter contrast between calm impounded water (very low return, dark) and turbulent tailwater (higher return, brighter) is detectable at 10 m resolution and is largely cloud-independent, which matters in tropical river systems where optical coverage can be sparse for months.
Weirs and low-head structures are harder. A weir 0.5 m tall may impound only a few hundred metres of backwater, producing a subtle spectral change that is easily confused with natural pool-riffle transitions or algal variation. Multi-temporal compositing over a dry-season period, when flow is low and backwater extent is largest, improves detection. Culverts are essentially invisible from orbit: they are buried under roads, produce no surface-water signature, and their passability depends on diameter, gradient, and flow velocity that no current satellite sensor can measure. Any barrier inventory derived from satellite data alone will systematically undercount culverts, and this must be stated in the product metadata.
Road-stream intersection analysis offers a partial workaround. Overlaying a road network (from OpenStreetMap or national datasets) on the satellite-derived channel network identifies every location where a road crosses a watercourse. Each intersection is a candidate culvert or bridge. The distinction between a passable bridge and a blocking culvert still requires field or drone inspection, but the candidate list produced from the spatial intersection is far more targeted than a blind field survey.
Calculating and ranking connectivity
Once the network graph is assembled with barriers attributed to nodes, the DCI can be computed for the current state and for hypothetical removal scenarios. The index ranges from 0 (fully fragmented) to 100 (no barriers). Removing a single barrier at the right node can shift the index dramatically; removing one at the wrong node achieves almost nothing. This prioritisation output is the primary decision-support product: a ranked list of structures where investment in fish-passage retrofitting or removal yields the greatest connectivity gain per unit of catchment area restored.
Barrier passability is a separate question that satellite data cannot answer. A notched weir, a nature-like bypass channel, or a fish ladder may render a structure partially or fully passable depending on species, flow regime, and engineering condition. Passability scores used in European barrier databases such as AMBER (Adaptive Management of Barriers in European Rivers) are assigned through field survey, not remote sensing. The satellite-derived connectivity model should be treated as a spatial prioritisation tool that tells field teams where to look, not as a definitive assessment of fish-passage potential.
Honest limits and what validation looks like
Cloud cover is the most persistent operational constraint. In humid tropical catchments, Sentinel-2 optical coverage may be limited to a handful of usable scenes per year. SAR from Sentinel-1 partially compensates, but SAR-based barrier detection is less sensitive to low-head structures than optical methods. Temporal compositing over multiple years can build adequate coverage, at the cost of potentially mixing pre- and post-construction states if the barrier inventory is changing.
Vertical accuracy in the DEM propagates into errors in hydraulic gradient estimation and, consequently, into the flow-routing algorithm. A 2 m vertical error in TanDEM-X is small relative to a 50 m dam but large relative to a 0.5 m weir. In low-gradient lowland rivers, where many ecologically important barriers are also low-head, DEM-derived channel delineation is least reliable. Lidar from airborne surveys or, where track coverage permits, ICESat-2 transects can anchor calibration at specific cross-sections.
Satellize applies these methods through its satellite-data analytics service, building catchment-graph models from open constellation data and structuring the barrier candidate inventory for client field-validation workflows. The Tonga crop-estimation programme demonstrated the organisation's approach to national-scale spatial analysis from open data; the river-connectivity workflow follows the same principle of producing a decision-ready spatial output rather than a raw data dump.
From map to management decision
The practical output is a GIS layer set: a delineated channel network with Strahler order attribution, a barrier candidate point layer with confidence scores and detection method flagged, a catchment-area polygon per barrier, and a DCI ranking table. From that table, a fisheries agency or hydropower regulator can identify the ten structures whose removal or retrofitting would restore the greatest proportion of accessible habitat. That is a concrete number, attached to specific map locations, which is what a budget decision requires.
Revisiting the analysis annually or after major flood events can detect new barriers (illegal weir construction is not uncommon in some regions) or confirm removal. Sentinel-1's 6-day repeat and Sentinel-2's 5-day repeat provide the temporal density needed for change detection at this scale. The archive depth, Sentinel-1 from 2014 and Sentinel-2 from 2015, allows retrospective analysis of how connectivity has changed over the past decade, which is useful for regulatory compliance reporting and for establishing baselines before restoration investment.
Typical figures
| Primary DEM spatial resolution | 12 m (TanDEM-X 0.4 arcsec); FABDEM matches Copernicus DEM at 30 m globally, 10 m over Europe |
| DEM vertical accuracy (TanDEM-X) | ~2 m absolute at 90th percentile; worse under dense canopy without FABDEM correction |
| Optical imagery resolution | 10 m (Sentinel-2 bands 2, 3, 4, 8); 20 m (SWIR and red-edge bands used for MNDWI) |
| SAR resolution | 10 m ground range (Sentinel-1 IW mode, C-band 5.4 GHz) |
| Revisit frequency | Sentinel-2: 5-day equatorial, 2-3 day mid-latitude; Sentinel-1: 6-day per track |
| Minimum detectable channel width | ~30 m reliable from Sentinel-2 MNDWI; narrower channels require higher-resolution commercial imagery |
| Minimum detectable barrier type | Large dams and reservoirs: high confidence. Low-head weirs (<1 m): moderate confidence with multi-temporal compositing. Culverts: not detectable spectrally; detected only via road-stream intersection |
| Satellite archive depth | Sentinel-1 from 2014; Sentinel-2 from 2015; Landsat back to 1972 for coarser historical context |
| Delivery formats | GeoPackage or Shapefile (network and barrier layers), GeoTIFF (DEM derivatives, flow accumulation), CSV (DCI ranking table), PDF summary report |
| Cloud-cover constraint | Optical methods require multi-date compositing in humid tropics; SAR provides cloud-independent backscatter but lower sensitivity to low-head barriers |
Analytics Satellize can run
| Hydrologically conditioned catchment DEM and drainage network | D8 or MFD flow-routing on FABDEM or TanDEM-X after sink-filling and stream-burning; Strahler order attribution | GeoTIFF flow-accumulation raster and GeoPackage polyline network with stream-order attributes |
| Artificial barrier candidate inventory | MNDWI change detection and SAR backscatter discontinuity analysis at channel nodes; road-stream intersection overlay for culvert candidates | Point GIS layer with barrier type classification, detection-method flag, and confidence score per candidate |
| Catchment-area delineation per barrier | Watershed delineation upstream of each barrier node using conditioned DEM; area computed in equal-area projection | Polygon layer of upstream catchment extents, attributed with area in km² and stream-order composition |
| Dendritic connectivity index (DCI) baseline and removal scenarios | DCI calculation on the network graph weighted by upstream catchment area; scenario modelling for single-barrier and multi-barrier removal | CSV ranking table with current DCI, marginal DCI gain per barrier removal, and cumulative gain curves; visualised in summary PDF |
| Multi-temporal barrier change detection | Annual Sentinel-1 and Sentinel-2 composites compared at channel nodes to flag new impoundment signatures or disappearance of existing ones | Change-alert GIS layer with date range and change-type flag; suitable for regulatory monitoring workflows |
| Field-validation priority list | Spatial join of DCI ranking with barrier confidence scores; low-confidence, high-DCI-gain candidates ranked for ground-truthing | Tabular field-survey schedule with GPS coordinates, satellite-derived attributes, and suggested validation protocol per site |
Who does the work
We can get this done for you. Satellize runs its own analyst desk and a strong science team. You do not buy a data feed and work out what it means; our people source the imagery, run the analysis described on this page, and hand you the answer with its confidence limits stated. Discuss this requirement.