Machine learning for non-experts: A more accessible and simpler approach to automatic benthic habitat classification DOI Creative Commons
Chloe A. Game, Michael B. Thompson, Graham D. Finlayson

et al.

Ecological Informatics, Journal Year: 2024, Volume and Issue: 81, P. 102619 - 102619

Published: May 10, 2024

Automating identification of benthic habitats from imagery, with Machine Learning (ML), is necessary to contribute efficiently and effectively marine spatial planning. A promising method adapt pre-trained general convolutional neural networks (CNNs) a new classification task (transfer learning). However, this often inaccessible non-specialist, requiring large investments in computational resources time (for user comprehension model training). In paper, we demonstrate simpler transfer learning framework for classifying broad deep-sea habitats. Specifically, take an 'off-the-shelf' CNN (VGG16) use it extract features (pixel patterns) images (without further The default outputs VGG16 are then fed Support Vector (SVM), classical than deep networks. For comparison, also train the remaining layers using stochastic gradient descent. discriminative power these approaches demonstrated on three datasets (574–8353 images) Norwegian waters; each unique imaging platform. Benthic broadly classified as Soft Substrate (sands, muds), Hard (gravels, cobbles boulders) Reef (Desmophyllum pertusum). We found that relatively simplicity SVM classifier did not compromise performance. Results were competitive consistently high, test accuracy ranging 0.87 0.95 (average = 0.9 (±0.04)) across datasets, somewhat increasing dataset size. Impressively, results achieved 2.4–5× faster training had significantly less dependency high-specification hardware. Our suggested approach maximises conceptual practical simplicity, representing realistic baseline novice users when approaching habitat classification. This has wide potential. It allows automated image grouping aid annotation or selection, well screening old-datasets. especially suited offshore scenarios can provide quick, albeit crude, insights into presence, allowing adaptation sampling protocols near real-time.

Language: Английский

Improving National Forest Mapping in Romania Using Machine Learning and Sentinel-2 Multispectral Imagery DOI Creative Commons
Mohamed Islam Keskes,

Aya Hamed Mohamed,

Stelian Alexandru Borz

et al.

Remote Sensing, Journal Year: 2025, Volume and Issue: 17(4), P. 715 - 715

Published: Feb. 19, 2025

Forest attributes, such as standing stock, diameter at breast height (DBH), tree height, and basal area, are critical for effective forest management; yet, traditional estimation methods remain labor-intensive often lack the spatial detail required contemporary decision-making. This study addresses these challenges by integrating machine learning algorithms with high-resolution remotely sensed data rigorously collected ground truth measurements to produce accurate, national-scale maps of attributes in Romania. To ensure reliability model predictions, extensive field campaigns were conducted across representative Romanian forests. During campaigns, detailed recorded every within selected plots. For each tree, DBH was measured directly, heights obtained either direct measurement—using hypsometers or clinometers—or, when not feasible, applying well-established DBH—height allometric relationships that have been calibrated local types. comprehensive approach collection, supplemented an independent dataset from Brasov County using same protocols, allowed robust training validation models. evaluates performance three algorithms—Random (RF), Classification Regression Trees (CART), Gradient Boosting Tree Algorithm (GBTA)—in predicting Sentinel-2 satellite imagery. While Random consistently delivered high R2 values low root mean square errors (RMSE) all GBTA showed particular strength CART excelled area but less reliable other attributes. A sensitivity analysis multiple resolutions revealed varied significantly changes resolution, emphasizing importance selecting appropriate scale accurate mapping. By focusing on both methodological advancements applications rigorous, empirical this provides a clear solution problem obtaining reliable, spatially attribute maps.

Language: Английский

Citations

2
Making Ecosystem Modeling Operational–A Novel Distributed Execution Framework to Systematically Explore Ecological Responses to Divergent Climate Trajectories DOI Creative Commons
Jeroen Steenbeek, Pablo Ortega, Raffaele Bernardello

et al.

Earth s Future, Journal Year: 2024, Volume and Issue: 12(3)

Published: March 1, 2024

Abstract Marine Ecosystem Models (MEMs) are increasingly driven by Earth System (ESMs) to better understand marine ecosystem dynamics, and analyze the effects of alternative management efforts for ecosystems under potential scenarios climate change. However, policy commercial activities typically occur on seasonal‐to‐decadal time scales, a span widely used in global modeling community but where skill level assessments MEMs their infancy. This is mostly due technical hurdles that prevent MEM from performing large ensemble simulations with which undergo systematic assessments. Here, we developed novel distributed execution framework constructed low‐tech freely available technologies enable analysis linked ESM/MEM prediction ensembles. We apply this scale, assess how retrospective forecast uncertainty an initialized decadal ESM predictions affects mechanistic spatiotemporal explicit trophodynamic MEM. Our results indicate internal variability has relatively low impact comparison broad assumptions related reconstructed fisheries. also observe sensitive specificities. case study warrants further explorations disentangle impacts change, fisheries scenarios, ecological hypotheses, variability. Most importantly, our demonstrates simple free empower any group fundamental capabilities operationalize modeling.

Language: Английский

Citations

9
Avoiding confusion: Modelling image identification surveys with classification errors DOI Creative Commons
Michael A. Spence, Jon Barry,

Thomas Bartos

et al.

Methods in Ecology and Evolution, Journal Year: 2025, Volume and Issue: unknown

Published: Jan. 21, 2025

Abstract Automated systems driven by machine learning is becoming increasingly used as an environmental monitoring tool. A common approach to use classification algorithms identify counts of categories (e.g. species) from images. However, the can be biased in presence error. To draw valid conclusions, it crucial incorporate these errors into analysis and interpretation algorithm results. We introduce a general framework for describing with classifiers, including data both classifier confusion matrix. The incorporates uncertainty matrix well generating process. By treating latent variables, our allows wide range processes. illustrate methods three case studies based on simulated different processes, zooplankton Celtic Seas English Channel. widely applicable many subject areas where occur.

Language: Английский

Citations

1
On the use of hydrodynamic modelling and random forest classifiers for the prediction of hypoxia in coastal lagoons DOI Creative Commons
Irene Simonetti,

Claudio Lubello,

Lorenzo Cappietti

et al.

The Science of The Total Environment, Journal Year: 2024, Volume and Issue: 951, P. 175424 - 175424

Published: Aug. 12, 2024

Hypoxia is one of the fundamental threats to water quality globally, particularly for partially enclosed basins with limited renewal, such as coastal lagoons. This work proposes combined use a machine learning technique, field observations, and data derived from hydrodynamic heat exchange numerical model predict, forecast up 10 days in advance, occurrence hypoxia eutrophic lagoon. The random forest algorithm used, training validating set models classify dissolved oxygen levels Orbetello lagoon, central Mediterranean Sea (Italy), has provided test case assessing reliability proposed methodology. Results proved that methodology effective providing reliable short-term evaluation DO levels, high resolution both time space throughout an entire An overall classification accuracy 91 % was found models, score identifying severe - i.e. hourly lower than 2 mg/l 86 %. predictors extracted allows us overcome intrinsic limitation modelling approaches which rely on input relatively few, local measurements, inability capture spatial heterogeneity distributions, unless several measuring points are available. methodological approach application similar environments.

Language: Английский

Citations

6
Machine-learning aiding sustainable Indian Ocean tuna purse seine fishery DOI Creative Commons
Nerea Goikoetxea, Izaro Goienetxea, José A. Fernandes

et al.

Ecological Informatics, Journal Year: 2024, Volume and Issue: 81, P. 102577 - 102577

Published: March 26, 2024

Among the various challenges facing tropical tuna purse seine fleet are need to reduce fuel consumption and carbon footprint, as well minimising bycatch of vulnerable species. Tools designed for forecasting optimum fishing grounds can contribute adapting changes in fish distribution due climate change, by identifying location new suitable grounds, thus reducing search time. While information about high probability find species could result a reduction. The present study aims at contributing more sustainable cleaner fishing, i.e. catching same amount target with less consumption/emissions lower bycatch. To achieve this, catches species, silky shark accidental have been modelled machine learning models Indian Ocean using inputs historical catch data these fleets environmental data. resulting show an accuracy 0.718 0.728 SKJ YFT, being absences (TPR = 0.996 0.993 respectively) better predicted than or low catches. In case BET, which is not main this fleet, that previous Regarding shark, presence/absence model provides 0.842. Even though model's performance has room improvement, work lays foundations process avoiding only input forecast provided near real time earth observation programs. future be improved knowledge conditions influencing becomes available.

Language: Английский

Citations

5
Exploring coral reef communities in Puerto Rico using Bayesian networks DOI Creative Commons
John F. Carriger, William S. Fisher

Ecological Informatics, Journal Year: 2024, Volume and Issue: 82, P. 102665 - 102665

Published: June 5, 2024

Most coral reef studies focus on scleractinian (stony) corals to indicate condition, but there are other prominent assemblages that play a role in ecosystem structure and function. In Puerto Rico these include fish, gorgonians, sponges. The U.S. Environmental Protection Agency conducted unique surveys of communities across the southern coast included simultaneous measurement all four assemblages. Evaluating results from community perspective demands endpoints for assemblages, so patterns were explored by probabilistic clustering measured variables with Bayesian networks. found have stronger associations within than between taxa, unsupervised learning identified three cross-taxa relationships potential ecological significance. Clusters each assemblage constructed using an expectation-maximization algorithm created factor node jointly characterizing density, size, diversity individuals taxon. clusters characterized variables, taxa examined, such as stony fish variables. Each nodes then used create set meta-factor further summarized aggregate monitoring taxa. Once identified, taxon-specific meta-clusters represent can be examined regional or site-specific basis better understand risk assessment, management delivery services.

Language: Английский

Citations

5
Automated species classification and counting by deep-sea mobile crawler platforms using YOLO DOI Creative Commons
Luciano Ortenzi, Jacopo Aguzzi, Corrado Costa

et al.

Ecological Informatics, Journal Year: 2024, Volume and Issue: 82, P. 102788 - 102788

Published: Aug. 20, 2024

Edge computing on mobile marine platform is paramount for automated ecological monitoring. The goal of demonstrating the computational feasibility an Artificial Intelligence (AI)-powered camera fully real-time species-classification deep-sea crawler platforms was searched by running You-Only-Look-Once (YOLO) model edge device (NVIDIA Jetson Nano), to evaluate achievable animal detection performances, execution time and power consumption, using all available cores. We processed a total 337 rotating video scans (∼180°), taken during approximately 4 months in 2022 at methane hydrates site Barkley Canyon (Vancouver Island; BC; Canada), focusing three abundant species (i.e., Sablefish Anoplopoma fimbria , Hagfish Eptatretus stoutii Rockfish Sebastes spp.). trained 1926 manually annotated frames showed high test performances terms accuracy (0.98), precision recall (0.99). then applied videos. In 288 videos we detected 133 Sablefish, 31 Hagfish, 321 nearly (about 0.31 s/image) with very low consumption (0.34 J/image). Our results have broad implications intelligent Indeed, YOLO can meet operational-autonomy criteria fast image processing limited energy loads. • Edge-computing allows robots detect, classify count animals situ. An routine tuned operate Wally deep-sea. were Nano, seeking load. Processing sustain autonomy

Language: Английский

Citations

4
Random forest regression models in ecology: Accounting for messy biological data and producing predictions with uncertainty DOI

Caitlin I. Allen Akselrud

Fisheries Research, Journal Year: 2024, Volume and Issue: 280, P. 107161 - 107161

Published: Sept. 6, 2024

Language: Английский

Citations

4
Machine learning for ecological analysis DOI

Zhengyang Yu,

Chongfeng Bu, Yanjie Li

et al.

Chemical Engineering Journal, Journal Year: 2025, Volume and Issue: unknown, P. 160780 - 160780

Published: Feb. 1, 2025

Language: Английский

Citations

0
An integrative machine learning approach to understanding South Pacific Ocean albacore tuna habitat features DOI Creative Commons
Lei Liu, Rong Wan, Feng Wu

et al.

ICES Journal of Marine Science, Journal Year: 2025, Volume and Issue: 82(1)

Published: Jan. 1, 2025

Abstract This study employs a random forest model combined with interpretable machine learning techniques to analyze the habitat preferences of South Pacific albacore tuna, incorporating broad range marine environmental variables. Among these, several factors derived from mesoscale eddy structures, including polarity, radius, and kinetic energy, are integrated further enhance characterization features. Interpretable methods were applied provide intuitive visualizations tuna preferences, focus on most influential factors, seawater temperature, dissolved oxygen concentration, normalized radius. Seawater temperature concentration directly linked physiological needs while characteristics influence foraging behavior by altering water column properties. provides comprehensive perspective mechanisms driving its oceanographic variables, providing valuable insights for developing location-based, practical science-based management strategies fishery resources.

Language: Английский

Citations

0