Evolutionary Architecture Search for Chlorophyll-$a$ Prediction in Lakes using Sentinel-2

Kursat Komurcu; Linas Petkevicius

cs.NEOct 7, 2026

Evolutionary Architecture Search for Chlorophyll- $a$ Prediction in Lakes using Sentinel-2

Authors: Kursat Komurcu, Linas Petkevicius

Organizations: Vilnius University, Institute of Computer Science, Artificial Intelligence Methods Lab · IRISA, Universite Bretagne Sud · European Commission Joint Research Center

Abstract

Small tabular datasets with expert-designed spectral features are the norm in operational Earth observation, and the networks applied to them are typically hand-designed. We revisit one such published model -- a Sentinel-2 algal bloom classifier -- and ask what architecture search adds, holding the task, the features and the lake-level train/test split of the original study fixed. Searching an extended multilayer-perceptron space with regularized evolution, and selecting on inner-cross-validation AUC only, we find networks that improve held-out AUC from 0.790 to 0.820 and accuracy from 0.733 to 0.748 while using 409 trainable parameters, 26 times fewer than the strongest hand-designed reference. The search converges on a consistent recipe -- a single narrow layer, RMS normalisation, $\tanh$ activation, step-decayed RMSprop and weight averaging -- that a practitioner would be unlikely to reach by default. At 1.6,kB the resulting model is small enough to serve as an onboard screening trigger, which is the setting that motivates the work. Code: https://github.com/VU-AIML/automl4eo-bloom-nas.

Figures & tables

Dimension	Values	Dimension	Values
Blocks	1–6	Optimiser	AdamW, Adam, SGD, RMSprop
Base width	16–256 (9 values)	Learning rate	$10^{-4}$ – $3\times 10^{-2}$
Width profile	constant, taper,	Schedule	constant, cosine, step
	expand, hourglass	Warmup fraction	0.0, 0.05, 0.1
Block type	plain, residual,	Weight decay	0– $10^{-2}$
	multi-head, GLU, GRN	Batch size	8–256

Table 1: Search space, approximately

10^{12}

configurations before the parameter-budget constraint.

Model	CV AUC	Test AUC	Test acc.	$F_{1}$	Params
Baseline-MLP	0.8055	$0.7902\pm.006$	$0.7306\pm.009$	0.7241	10 625
Baseline-Res	0.7588	$0.7543\pm.010$	$0.6989\pm.010$	0.6746	32 129
Baseline-MHA	0.7914	$0.7785\pm.008$	$0.7329\pm.011$	0.7266	19 073
NAS-1	0.8240	$0.8181\pm.001$	$0.7460\pm.004$	0.7442	409
NAS-6	0.8238	$\mathbf{0.8203}\pm.001$	$0.7449\pm.003$	0.7449	433
NAS-8	0.8237	$0.8187\pm.001$	$\mathbf{0.7483}\pm.003$	0.7466	409

Table 2: Held-out performance, mean

\pm

standard deviation over 10 weight seeds. Selection used inner-CV AUC only. NAS-1/6/8 are the three highest-ranked distinct architectures; the remaining seven of the top ten fall within 0.002 AUC of these.

Explore similar work

Landsat-Sentinel-2 Algal Bloom Mapping Using Vision Transformers: Model Description, Implementation, and Examples
Jun 15, 2026Thainara Lima, Vitor MartinsSentinel-2 Coral
a Department of Agricultural & Biological Engineering, Mississippi State University (MSU), Starkville, MS 39762, USA.
LESS: Lightweight Evolutionary Supernet Search in Minutes
Oct 1, 2026Aviral Gandhi, Jinglue Xu, Jialong Li +1Neural Architecture Search Search Algorithms
The University of Tokyo · Sakana AI · Waseda University
Random Cloud: Finding Minimal Neural Architectures Without Training
Apr 29, 2026Javier Gil BlázquezNeural Architecture Search Unstructured Pruning

Jun 15, 2026, cs.CVJ/K move · Enter open · S save

Landsat-Sentinel-2 Algal Bloom Mapping Using Vision Transformers: Model Description, Implementation, and Examples

Thainara Lima, Vitor Martins

a Department of Agricultural & Biological Engineering, Mississippi State University (MSU), Starkville, MS 39762, USA.

Sentinel-2 Coral Vision Transformer

Open paper AskPDF arXiv

Coastal algal bloom monitoring requires frequent, spatially detailed, and globally consistent observations, provided by Landsat-8/9 and Sentinel-2 A/B/C. Together, these missions offer over a decade of medium-resolution multispectral imagery with near-global coverage every 2-3 days, enabling the detection of fragmented bloom structures not resolvable by coarse ocean-color sensors. However, their use in aquatic environments remains challenging due to limited spectral coverage and a lack of harmonized reflectance products. As an alternative to traditional bio-optical methods, deep learning-based image classification offers a data-driven approach that can overcome many of these limitations. This study presents the first successful implementation of vision transformer-based coastal algal bloom mapping using 30-m Landsat-Sentinel-2 images. A globally distributed bloom patch dataset was generated across bloom-prone coastal hotspots worldwide. Four transformer-based architectures were compared against a standard convolutional baseline for fine-scale bloom detection, and assessed under different optical water types and atmospheric and surface conditions. All deep learning models showed strong capabilities in detecting floating bloom areas, with omission and commission errors of 8-65%. Under cloud and glint stress in a time series, the Swin Transformer outperformed traditional spectral-index approaches, which produced widespread false positives, effectively avoiding cloud- and glint-affected pixels. Comparisons with MODIS-derived products further highlighted the benefits of higher spatial resolution in detecting fragmented and irregularly affected blooms. Our findings support deep learning as a reliable tool for medium-resolution, consistent monitoring of floating algal blooms in dynamic coastal environments.

Abstract

Explore similar work

Landsat-Sentinel-2 Algal Bloom Mapping Using Vision Transformers: Model Description, Implementation, and Examples

LESS: Lightweight Evolutionary Supernet Search in Minutes

Random Cloud: Finding Minimal Neural Architectures Without Training