Riparian corridors are not semi-natural refuges
An open-data Earth Engine workflow derives catchment-scale pressure variables for a 444.5 km² chalk-stream catchment and finds the 50-metre river corridor is the most modified land in it.
August 2026 · River Wylye catchment, Wiltshire, United Kingdom · Prepared as an open-data demonstration for a Horizon Europe biodiversity call (HORIZON-CL6-2026-01-BIODIV-01)
Why derive pressure variables at all
Analysing insect decline in rivers requires linking what happens on land to what is counted in the water. For riverflies, the Ephemeroptera, Plecoptera and Trichoptera larvae that anchor freshwater food webs, a substantial share of the drivers is terrestrial: land use in the contributing catchment governs sediment delivery, nutrient loading, thermal regime, and the condition of the riparian corridor that supplies shade and organic matter to the stream.
This analysis sets out a reproducible method for deriving harmonised catchment-scale pressure variables that can be joined statistically to in-stream biological monitoring records (EPT indices, riverfly counts), so that terrestrial drivers can be tested against observed population data rather than assumed.
What was run
The workflow was executed end to end on the River Wylye catchment: HydroBASINS level 8, 444.5 km², resolved into four level-12 sub-catchments. Inputs: HydroBASINS sub-catchment boundaries, Copernicus DEM GLO-30 terrain, 72 Sentinel-2 L2A scenes (1 April to 30 September 2023, cloud probability below 40 %), CORINE Land Cover 2018, and the HydroSHEDS river network (40 mapped reaches). The riparian corridor is a 50-metre half-width buffer on mapped channels. Outputs: a per-sub-catchment pressure-variable table and a 20-metre index and pressure raster stack. All inputs are open. No client or restricted data was used.
The analytical extent was verified, not assumed: retained level-12 sub-catchment area sums to 444.5 km² against a level-8 basin area of 444.5 km², a ratio of 1.00.
Catchment composition, area-weighted: arable 0.431, semi-natural 0.336, pasture 0.209, artificial 0.023, water and wetland 0.001. An arable-dominated lowland chalk catchment with a substantial semi-natural component and very little built development. This is the context against which the corridor figures below should be read.
The result
Comparing the 50-metre corridor with its sub-catchment, all four land-use classes move in the same direction in every one of the four sub-catchments. Sixteen comparisons without an exception.
| Sub-catchment | Δ arable | Δ pasture | Δ artificial | Δ semi-natural |
|---|---|---|---|---|
| 2120402590 (19.3 km²) | −0.501 | +0.513 | +0.086 | −0.098 |
| 2121122150 (158.9 km²) | −0.107 | +0.201 | +0.057 | −0.151 |
| 2120401590 (142.7 km²) | −0.244 | +0.238 | +0.038 | −0.049 |
| 2120401340 (123.6 km²) | −0.012 | +0.276 | +0.045 | −0.309 |
| Mean change | −0.216 | +0.307 | +0.057 | −0.152 |
In mean terms across the catchment, moving from the sub-catchment to the corridor: arable falls from 0.494 to 0.278, pasture rises from 0.184 to 0.491, semi-natural cover falls from 0.295 to 0.144, and artificial cover rises from 0.026 to 0.083, a threefold concentration of built development along the watercourse.
Why the pattern matters more than the numbers
Arable cultivation does retreat from the channel, which is the pattern a buffer-strip narrative would predict. But what occupies the vacated ground is grazing and settlement, not natural vegetation: semi-natural cover is roughly halved in the corridor while built cover triples. These corridors are the most modified land in the catchment, not the least.
For riverfly populations the distinction is direct. Pasture grazed to the bank produces poaching, bank erosion and direct faecal input. Concentrated built development brings point-source discharge, culverting and channel modification. Reduced semi-natural corridor cover means less shading, less leaf-litter input and less woody debris, the resource base for Ephemeroptera, Plecoptera and Trichoptera larvae. A catchment-level land-use statistic would surface none of this. The corridor comparison does.
Two mechanisms remain confounded and cannot be separated with these data: deliberate buffer management, or the simple fact that floodplain land is grazed and historically settled because it is flat, watered and floods. Management records or field survey would distinguish them.
Method in brief
The level-8 basin containing the monitoring site defines the study catchment. Nested level-12 polygons become the analytical units, selected by centroid containment so the retained area matches the basin, with HYBAS_ID kept as the join key to biological records. Slope and elevation come from the Copernicus DEM. The river network is buffered to a stated half-width, 50 m by default, adjustable to the ecological question. A cloud-masked Sentinel-2 growing-season median composite yields NDVI, NDWI, NDMI and BSI, computed both catchment-wide and within the corridor. CORINE classes are grouped into pressure classes and expressed as area fractions per sub-catchment. All variables are aggregated by zonal statistics and exported as a single table: one row per sub-catchment per zone.
The analysis is one parameterised script. Applying it to a different pilot site requires changing one coordinate. Every input is an open, versioned, citable dataset and every processing parameter is recorded, so outputs are findable, accessible, interoperable and reusable by construction rather than by retrofit.
Limitations
Stated here rather than left for a reviewer to find. Four sub-catchments: the direction of every effect is consistent, but four cases do not support a significance claim. The value here is that the method produces the variable, not that the effect is established. CORINE at 100 m under-represents narrow riparian features, so national land-parcel data should replace it for a real pilot. HydroSHEDS resolves 40 reaches at roughly 500 m across this catchment, so corridor geometry is coarse and headwater channels are likely under-represented, and national hydrography would refine the delineation materially. A single growing-season composite from one year leaves inter-annual variability uncharacterised. Spectral indices track biomass and moisture, not habitat quality. And catchment-scale association is not causation: these variables are inputs to driver analysis, not a substitute for ecological and ecotoxicological work.
Reproduce it
The full Earth Engine script and the exported pressure-variable table are available for independent reproduction. Open the script in the Earth Engine Code Editor, set one coordinate, and run.
Download the analysis script (.js) →
Download the pressure-variable table (.csv) →
Data: HydroSHEDS and HydroBASINS (WWF), Copernicus DEM GLO-30, Copernicus Sentinel-2 L2A, CORINE Land Cover 2018. Analysis in Google Earth Engine. Processing date: 3 August 2026.