Very high-resolution mapping of emerging biogenic reefs using airborne optical imagery and neural network: the honeycomb worm ( Sabellaria alveolata ) case study

Antoine Collin (1, 2, 3) , Stanislas Dubois (4) , Camille Ramambason (2) , Samuel Etienne (5, 2, 3)

1 LETG - Nantes - Littoral, Environnement, Télédétection, Géomatique
2 EPHE - École Pratique des Hautes Études
3 PSL - Université Paris Sciences et Lettres
4 DYNECO - Unité Dynamiques des Écosystèmes Côtiers
5 LETG - Littoral, Environnement, Télédétection, Géomatique UMR 6554

Biogenic reefs provide a wide spectrum of ecosystem functions and services, such as biodiversity hotspot, coastal protection, and fishing practices. Honeycomb worm (Sabellaria alveolata) reefs, in the Bay of Mont-Saint-Michel (France), constitute the largest intertidal biocon-struction in Europe but undergo anthropogenic pressures (aquaculture stemmed food/space competition and siltation, fishing-driven trampling). Very high-resolution (VHR) airborne optical data enable cost-efficient biophysical measurements of reef colonies, strongly expected for conservation approaches. A synergy of remotely sensed airborne optical imagery, calibration/validation photoquadrat ground-truth (202/101, respectively), and artificial neural network (ANN) modelling is first used to map S. alveolata relative abundance, over the largest bioconstruction in Europe. The best prediction of S. alveolata abundance was reached with the infrared–red–green (IRRG) spectral combination and ANN model structured with six neurons (R 2 = 0.72, RMSE = 0.08, and r = 0.85). The six hyperbolic tangent formulas were applied to the three input spectral bands (IRRG) in order to build six hidden neuronal images, resulting in VHR digital S. alveolata abundance model (6547 × 6566 pixels with 0.5 m pixel size). The innovative model revealed undescribed spatial patterns, namely a reef polarization (perpendicular to the shoreline) of S. alveolata abundance: high abundance on forereef and low abundance on backreef. ARTICLE HISTORY

Domaines: Milieux et Changements globaux; Environnement et Société; Méthodologie [stat.ME]; Machine Learning [stat.ML]; Ecosystèmes; Ingénierie de l'environnement; Biodiversité et Ecologie; Traitement du signal et de l'image [eess.SP]; Traitement des images [eess.IV]; Modélisation et simulation; Réseau de neurones [cs.NE]; Bio-informatique [q-bio.QM]; Géographie; Optique / photonique; Traitement du signal et de l'image [eess.SP]; Electromagnétisme; Acoustique [physics.class-ph]

Antoine Collin, Stanislas Dubois, Camille Ramambason, Samuel Etienne. Very high-resolution mapping of emerging biogenic reefs using airborne optical imagery and neural network: the honeycomb worm ( Sabellaria alveolata ) case study. International Journal of Remote Sensing, In press, ⟨10.1080/01431161.2018.1484964⟩. ⟨hal-01822434⟩