When sharper is not better: super-resolution of Sentinel-2 for multi-index emergent macrophyte mapping

Deep learning super-resolution sharpens Sentinel-2 imagery from 10 m to 2.5 m. Because emergent reed beds are narrow, often filling only one or two 10 m pixels, a sharper image seems likely to help, yet this had not been tested in optically complex lagoon water. We tested it in the Curonian Lagoon, Lithuania. The standard 10 m image and a super-resolved image were passed through the same multi-index classifier and scored against drone surveys of seven shoreline footprints. Super-resolution did not help. The super-resolved map, from the compact SEN2SRLite model, matched the 10 m map both in accuracy (pooled F1 0.739 against 0.713, a difference the seven footprints cannot resolve) and in mapped area (within 1%). We then repeated the test across five methods, from bicubic interpolation to a latent diffusion model. Spectral consistency varied 7.7-fold and processing time 640-fold, yet accuracy varied by only 0.014 in F1, and no method significantly exceeded the 10 m map; bicubic ranked second. A synthetic-degradation test explains why: the detail super-resolution invents carries about ten times more error than a consistency check reveals, but the conjunction of three index thresholds absorbs it. For threshold-based mapping of emergent macrophytes in optically complex water, a sharper image is not a better one.

Authors

Publication Details

Journal
Figshare
Published
2026-11-21
DOI
https://doi.org/10.6084/m9.figshare.35082814
Primary Topic
Remote-Sensing Image Classification
Type
article
Field-Weighted Citation Impact
0.00
Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
article

When sharper is not better: super-resolution of Sentinel-2 for multi-index emergent macrophyte mapping

Jolita Petkuvienė, Jonas Gintauskas, Diana Vaičiūtė, Edvinas Tiškus et al.
Figshare
Remote-Sensing Image Classification
article

When sharper is not better: super-resolution of Sentinel-2 for multi-index emergent macrophyte mapping

Jolita Petkuvienė, Jonas Gintauskas, Diana Vaičiūtė, Edvinas Tiškus, Martynas Bučas, Rūta Tiškuvienė
article en

Abstract

Deep learning super-resolution sharpens Sentinel-2 imagery from 10 m to 2.5 m. Because emergent reed beds are narrow, often filling only one or two 10 m pixels, a sharper image seems likely to help, yet this had not been tested in optically complex lagoon water. We tested it in the Curonian Lagoon, Lithuania. The standard 10 m image and a super-resolved image were passed through the same multi-index classifier and scored against drone surveys of seven shoreline footprints. Super-resolution did not help. The super-resolved map, from the compact SEN2SRLite model, matched the 10 m map both in accuracy (pooled F1 0.739 against 0.713, a difference the seven footprints cannot resolve) and in mapped area (within 1%). We then repeated the test across five methods, from bicubic interpolation to a latent diffusion model. Spectral consistency varied 7.7-fold and processing time 640-fold, yet accuracy varied by only 0.014 in F1, and no method significantly exceeded the 10 m map; bicubic ranked second. A synthetic-degradation test explains why: the detail super-resolution invents carries about ten times more error than a consistency check reveals, but the conjunction of three index thresholds absorbs it. For threshold-based mapping of emergent macrophytes in optically complex water, a sharper image is not a better one.

Figshare
Remote-Sensing Image Classification
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.

Laika access coming soon