Pub Date : 2025-02-05DOI: 10.1016/j.ecoinf.2025.103057
Alexander Kmoch, Clay Taylor Harrison, Jeonghwan Choi, Evelyn Uuemaa
Spatial autocorrelation, the relationship between nearby samples of a spatial random variable, is often overlooked in machine learning models, leading to biased results. This study compares various methods to account for spatial autocorrelation when predicting soil organic carbon (SOC) using random forest models. This kind of systematic comparison has not been done previously. Five models incorporating spatial structure were compared against baseline models with no added spatial components. Cross-validation showed slight improvements in accuracy for models considering spatial autocorrelation, while Shapley Additive Explanations confirmed the importance of spatial variables. However, no decrease in spatial autocorrelation of residuals was observed. Random Forest Spatial Interpolation emerged as the top performer in capturing spatial structure and improving model accuracy. Raster-based models exhibited enhanced prediction detail. The findings emphasize the value of incorporating spatial autocorrelation for better prediction of SOC with machine learning. Considerations such as the spatial distribution of predictions and computational complexity should help guide the selection of suitable approaches for specific spatial modelling tasks.
{"title":"Spatial autocorrelation in machine learning for modelling soil organic carbon","authors":"Alexander Kmoch, Clay Taylor Harrison, Jeonghwan Choi, Evelyn Uuemaa","doi":"10.1016/j.ecoinf.2025.103057","DOIUrl":"10.1016/j.ecoinf.2025.103057","url":null,"abstract":"<div><div>Spatial autocorrelation, the relationship between nearby samples of a spatial random variable, is often overlooked in machine learning models, leading to biased results. This study compares various methods to account for spatial autocorrelation when predicting soil organic carbon (SOC) using random forest models. This kind of systematic comparison has not been done previously. Five models incorporating spatial structure were compared against baseline models with no added spatial components. Cross-validation showed slight improvements in accuracy for models considering spatial autocorrelation, while Shapley Additive Explanations confirmed the importance of spatial variables. However, no decrease in spatial autocorrelation of residuals was observed. Random Forest Spatial Interpolation emerged as the top performer in capturing spatial structure and improving model accuracy. Raster-based models exhibited enhanced prediction detail. The findings emphasize the value of incorporating spatial autocorrelation for better prediction of SOC with machine learning. Considerations such as the spatial distribution of predictions and computational complexity should help guide the selection of suitable approaches for specific spatial modelling tasks.</div></div>","PeriodicalId":51024,"journal":{"name":"Ecological Informatics","volume":"86 ","pages":"Article 103057"},"PeriodicalIF":5.8,"publicationDate":"2025-02-05","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143436486","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"环境科学与生态学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2025-02-03DOI: 10.1016/j.ecoinf.2025.103031
Jelyn Gerkema , Daniel E. Bunker , Andrew M. Cunliffe , Erika Bazzato , Michela Marignani , Tommaso Sitzia , Isabelle Aubin , Stefano Chelli , Julieta A. Rosell , Peter Poschlod , Josep Penuelas , Arildo S. Dias , Christian Rossi , Tanvir A. Shovon , Juan A. Campos , Mark C. Vanderwel , Sharif A. Mukul , Bruno E.L. Cerabolini , Thomas Sibret , Bruno Hérault , Edwin T. Pos
An in-depth understanding of local plant community assembly is critical to direct conservation efforts to promising areas and increase the efficiency of management strategies. This, however, remains elusive due to the sheer complexity of ecological processes. The maximum entropy-based Community Assembly via Trait Selection (CATS) model was designed to quantify the relative contributions of trait-based filtering, dispersal mass effects, and stochastic processes on community assembly. As a maximum entropy model, it does so without introducing additional bias or assumptions. Despite its increasing use, questions regarding its robustness and potential limitations remain. Here, we compared model predictions using either local or database-derived trait values, across different levels of species richness and between different taxonomic levels. A total of 19 datasets and 790 plots were analysed, spanning multiple habitat types (n = 18) and biomes (n = 7). Results indicate trait value origin does indeed influence model outcomes, warranting caution in selecting the method for obtaining trait data. We hypothesise that, for example, intraspecific trait variation combined with trait-based filtering or stochastic processes causes local and database trait values to deviate, potentially even further exacerbated by imputing missing trait data. Furthermore, trait-related information obtained from the model decreased with increasing species richness. We further hypothesise this could signal that stochastic processes are more dominant within species-rich systems, for example, due to functional redundancy or the existence of multiple fitness strategies. This general pattern was conserved across biomes, although with varying strength, showing CATS’ robustness despite these challenges.
{"title":"Robustness and limitations of maximum entropy in plant community assembly","authors":"Jelyn Gerkema , Daniel E. Bunker , Andrew M. Cunliffe , Erika Bazzato , Michela Marignani , Tommaso Sitzia , Isabelle Aubin , Stefano Chelli , Julieta A. Rosell , Peter Poschlod , Josep Penuelas , Arildo S. Dias , Christian Rossi , Tanvir A. Shovon , Juan A. Campos , Mark C. Vanderwel , Sharif A. Mukul , Bruno E.L. Cerabolini , Thomas Sibret , Bruno Hérault , Edwin T. Pos","doi":"10.1016/j.ecoinf.2025.103031","DOIUrl":"10.1016/j.ecoinf.2025.103031","url":null,"abstract":"<div><div>An in-depth understanding of local plant community assembly is critical to direct conservation efforts to promising areas and increase the efficiency of management strategies. This, however, remains elusive due to the sheer complexity of ecological processes. The maximum entropy-based Community Assembly via Trait Selection (CATS) model was designed to quantify the relative contributions of trait-based filtering, dispersal mass effects, and stochastic processes on community assembly. As a maximum entropy model, it does so without introducing additional bias or assumptions. Despite its increasing use, questions regarding its robustness and potential limitations remain. Here, we compared model predictions using either local or database-derived trait values, across different levels of species richness and between different taxonomic levels. A total of 19 datasets and 790 plots were analysed, spanning multiple habitat types (n = 18) and biomes (n = 7). Results indicate trait value origin does indeed influence model outcomes, warranting caution in selecting the method for obtaining trait data. We hypothesise that, for example, intraspecific trait variation combined with trait-based filtering or stochastic processes causes local and database trait values to deviate, potentially even further exacerbated by imputing missing trait data. Furthermore, trait-related information obtained from the model decreased with increasing species richness. We further hypothesise this could signal that stochastic processes are more dominant within species-rich systems, for example, due to functional redundancy or the existence of multiple fitness strategies. This general pattern was conserved across biomes, although with varying strength, showing CATS’ robustness despite these challenges.</div></div>","PeriodicalId":51024,"journal":{"name":"Ecological Informatics","volume":"86 ","pages":"Article 103031"},"PeriodicalIF":5.8,"publicationDate":"2025-02-03","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143227477","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"环境科学与生态学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2025-02-03DOI: 10.1016/j.ecoinf.2025.103055
Béla J. Szekeres , Máté Natabara Gyöngyössy , János Botzheim
Mosquitoes and other insects are vectors of severe diseases, posing significant health risks to millions worldwide yearly. Effective classification of insect species, particularly through their wingbeat sounds, is crucial for disease prevention and control. Despite recent advancements in Deep Learning, Fourier Neural Operators (FNO), efficient for solving Partial Differential Equations due to their global spectral representations, have yet to be thoroughly explored for real-world time series classification or regression tasks. This study explores the application of FNOs in insect wingbeat sound classification, focusing on their potential for improving the accuracy and efficiency of such tasks, particularly in the fight against mosquito-borne diseases. We introduce CF-ResNet-1D, a novel ResNet-inspired model that integrates Convolutional Fourier Layers, combining the strengths of FNOs and 1D-Convolutional processing. The model is designed to analyze raw time-domain signals, leveraging the parallel spectral processing capabilities of FNOs. Our findings demonstrate that CF-ResNet-1D significantly outperforms traditional spectrogram-based models in classifying insect wingbeat sounds, achieving state-of-the-art accuracy.
{"title":"Applying Fourier Neural Operator to insect wingbeat sound classification: Introducing CF-ResNet-1D","authors":"Béla J. Szekeres , Máté Natabara Gyöngyössy , János Botzheim","doi":"10.1016/j.ecoinf.2025.103055","DOIUrl":"10.1016/j.ecoinf.2025.103055","url":null,"abstract":"<div><div>Mosquitoes and other insects are vectors of severe diseases, posing significant health risks to millions worldwide yearly. Effective classification of insect species, particularly through their wingbeat sounds, is crucial for disease prevention and control. Despite recent advancements in Deep Learning, Fourier Neural Operators (FNO), efficient for solving Partial Differential Equations due to their global spectral representations, have yet to be thoroughly explored for real-world time series classification or regression tasks. This study explores the application of FNOs in insect wingbeat sound classification, focusing on their potential for improving the accuracy and efficiency of such tasks, particularly in the fight against mosquito-borne diseases. We introduce <em>CF-ResNet-1D</em>, a novel ResNet-inspired model that integrates Convolutional Fourier Layers, combining the strengths of FNOs and 1D-Convolutional processing. The model is designed to analyze raw time-domain signals, leveraging the parallel spectral processing capabilities of FNOs. Our findings demonstrate that <em>CF-ResNet-1D</em> significantly outperforms traditional spectrogram-based models in classifying insect wingbeat sounds, achieving state-of-the-art accuracy.</div></div>","PeriodicalId":51024,"journal":{"name":"Ecological Informatics","volume":"86 ","pages":"Article 103055"},"PeriodicalIF":5.8,"publicationDate":"2025-02-03","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143226970","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"环境科学与生态学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2025-02-01DOI: 10.1016/j.ecoinf.2025.103044
Thi-Thu-Hong Phan, Le Huu Bao Nguyen
This study proposes a robust and accurate approach for classifying rice variety purity to meet stringent agricultural standards. To achieve this, we construct a comprehensive dataset by leveraging diverse types of features encompassing morphological properties, overall image structure, texture information, and color distribution from rice seeds. Subsequently, we employ advanced feature selection techniques, including filter methods (Correlation, Chi-square, ANOVA), wrapper methods (Recursive Feature Elimination — RFE), and embedded methods (Random Forest, Decision Trees), to identify the most significant features. Through rigorous experimentation with eight machine learning algorithms, we find that using Random Forest for feature selection, in combination with the SVM classifier, yields the best performance. Specifically, Random Forest reduces the feature set by more than half, from 172 to 80, remarkably enhancing classification accuracy from 94.73% to 96.11%. This paper highlights the potential of the proposed method to offer a robust and efficient solution for rice seed purity identification in agricultural applications, while also opening up new horizons for similar studies.
{"title":"Enhancing rice seed purity recognition accuracy based on optimal feature selection","authors":"Thi-Thu-Hong Phan, Le Huu Bao Nguyen","doi":"10.1016/j.ecoinf.2025.103044","DOIUrl":"10.1016/j.ecoinf.2025.103044","url":null,"abstract":"<div><div>This study proposes a robust and accurate approach for classifying rice variety purity to meet stringent agricultural standards. To achieve this, we construct a comprehensive dataset by leveraging diverse types of features encompassing morphological properties, overall image structure, texture information, and color distribution from rice seeds. Subsequently, we employ advanced feature selection techniques, including filter methods (Correlation, Chi-square, ANOVA), wrapper methods (Recursive Feature Elimination — RFE), and embedded methods (Random Forest, Decision Trees), to identify the most significant features. Through rigorous experimentation with eight machine learning algorithms, we find that using Random Forest for feature selection, in combination with the SVM classifier, yields the best performance. Specifically, Random Forest reduces the feature set by more than half, from 172 to 80, remarkably enhancing classification accuracy from 94.73% to 96.11%. This paper highlights the potential of the proposed method to offer a robust and efficient solution for rice seed purity identification in agricultural applications, while also opening up new horizons for similar studies.</div></div>","PeriodicalId":51024,"journal":{"name":"Ecological Informatics","volume":"86 ","pages":"Article 103044"},"PeriodicalIF":5.8,"publicationDate":"2025-02-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143102222","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"环境科学与生态学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2025-01-31DOI: 10.1016/j.ecoinf.2025.103060
R. Walcker, C. Le Lay, L. Gandois, A. Elger, V.E.J. Jassey
Large-scale mapping of ecosystem CO2 fluxes between land and atmosphere is a challenging task. Typically, it is based on multispectral satellite images with low (∼250 m) to moderate ground pixel resolution (∼20 m). However, in small, fine-scale ecosystems such as peatlands, the representation of CO2 fluxes heterogeneity across microhabitats is very limited by the ground pixel resolution of satellites. In this context, high ground pixel resolution of drone imagery might prove useful, alone or in synergy with large-scale satellite data, to better support field investigations and carry out rapid carbon assessment at a relatively low cost. Here, we carried out a survey of CO2 exchanges over 4 ha of peatland during the growing season using chamber measurements, as well as simultaneous aerial multispectral orthophotographs acquired by drone. To assess the ability of drone multispectral images at providing relevant information for the prediction of CO2 fluxes, we developed robust linear regression models and used drone imagery to map net ecosystem exchange, ecosystem respiration and gross ecosystem productivity at a very fine scale (∼5 cm). Our predictions were in the range of those found in the satellite remote sensing literature with errors lesser than 0.5 gCO2.m−2.h−1. Our study offers new opportunities to refine large scale satellite assessment of CO2 fluxes on small, valuable peatland areas that can be easily flown over by drone. Moreover, we believe that our results will be of interest for the scientific community as well as environmental managers wishing to carry out rapid carbon assessments of peatlands.
{"title":"High-resolution mapping of peatland CO2 fluxes using drone multispectral images","authors":"R. Walcker, C. Le Lay, L. Gandois, A. Elger, V.E.J. Jassey","doi":"10.1016/j.ecoinf.2025.103060","DOIUrl":"10.1016/j.ecoinf.2025.103060","url":null,"abstract":"<div><div>Large-scale mapping of ecosystem CO<sub>2</sub> fluxes between land and atmosphere is a challenging task. Typically, it is based on multispectral satellite images with low (∼250 m) to moderate ground pixel resolution (∼20 m). However, in small, fine-scale ecosystems such as peatlands, the representation of CO<sub>2</sub> fluxes heterogeneity across microhabitats is very limited by the ground pixel resolution of satellites. In this context, high ground pixel resolution of drone imagery might prove useful, alone or in synergy with large-scale satellite data, to better support field investigations and carry out rapid carbon assessment at a relatively low cost. Here, we carried out a survey of CO<sub>2</sub> exchanges over 4 ha of peatland during the growing season using chamber measurements, as well as simultaneous aerial multispectral orthophotographs acquired by drone. To assess the ability of drone multispectral images at providing relevant information for the prediction of CO<sub>2</sub> fluxes, we developed robust linear regression models and used drone imagery to map net ecosystem exchange, ecosystem respiration and gross ecosystem productivity at a very fine scale (∼5 cm). Our predictions were in the range of those found in the satellite remote sensing literature with errors lesser than 0.5 gCO<sub>2</sub>.m<sup>−2</sup>.h<sup>−1</sup>. Our study offers new opportunities to refine large scale satellite assessment of CO<sub>2</sub> fluxes on small, valuable peatland areas that can be easily flown over by drone. Moreover, we believe that our results will be of interest for the scientific community as well as environmental managers wishing to carry out rapid carbon assessments of peatlands.</div></div>","PeriodicalId":51024,"journal":{"name":"Ecological Informatics","volume":"86 ","pages":"Article 103060"},"PeriodicalIF":5.8,"publicationDate":"2025-01-31","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143102223","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"环境科学与生态学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2025-01-31DOI: 10.1016/j.ecoinf.2025.103058
Clément Violet , Aurélien Boyé , Stanislas Dubois , Graham J. Edgar , Elizabeth S. Oh , Rick D. Stuart-Smith , Martin P. Marzloff
Benthic biogenic habitats are crucial for coastal marine ecosystems, supporting food and shelter for a large range of marine species, but they are increasingly threatened by increasing anthropogenic impacts. While large-scale monitoring data are increasingly available, tools to describe benthic habitat changes in standardised and yet finely resolved manner are still needed. The aim of this study was to define reef benthic habitat states and explore their spatial and temporal variability on a global scale using an innovative clustering pipeline. For this purpose, we used substrate cover data collected along 6554 transects worldwide by citizen scientists contributing to the Reef Life Survey program. We applied an innovative clustering pipeline that combines three algorithms — Uniform Manifold Approximation and Projection (UMAP) for dimension reduction; Hierarchical Density-Based Spatial Clustering of Applications with Noise (HDBSCAN) — to identify benthic habitat states and Shapley values to interpret the clusters identified. This unsupervised pipeline identified 17 distinct clusters worldwide, representing typical temperate and tropical benthic habitats such as large canopy forming algae and branching corals, respectively, as well as transitional states between different habitat states. Temporal site-specific analyses further demonstrated the pipeline's effectiveness in capturing fine-scale habitat dynamics. By providing a standardised, scalable approach, this work enables consistent tracking of benthic habitat changes across spatial and temporal scales worldwide. This study also showcases the potential of integrating the UMAP-HDBSCAN pipeline with Shapley values for clustering noisy ecological data from citizen science initiatives.
{"title":"Leveraging citizen science to classify and track benthic habitat states: An unsupervised UMAP-HDBSCAN pipeline applied to the global reef life survey dataset","authors":"Clément Violet , Aurélien Boyé , Stanislas Dubois , Graham J. Edgar , Elizabeth S. Oh , Rick D. Stuart-Smith , Martin P. Marzloff","doi":"10.1016/j.ecoinf.2025.103058","DOIUrl":"10.1016/j.ecoinf.2025.103058","url":null,"abstract":"<div><div>Benthic biogenic habitats are crucial for coastal marine ecosystems, supporting food and shelter for a large range of marine species, but they are increasingly threatened by increasing anthropogenic impacts. While large-scale monitoring data are increasingly available, tools to describe benthic habitat changes in standardised and yet finely resolved manner are still needed. The aim of this study was to define reef benthic habitat states and explore their spatial and temporal variability on a global scale using an innovative clustering pipeline. For this purpose, we used substrate cover data collected along 6554 transects worldwide by citizen scientists contributing to the <em>Reef Life Survey</em> program. We applied an innovative clustering pipeline that combines three algorithms — <em>Uniform Manifold Approximation and Projection</em> (<em>UMAP</em>) for dimension reduction; <em>Hierarchical Density-Based Spatial Clustering of Applications with Noise</em> (<em>HDBSCAN</em>) — to identify benthic habitat states and Shapley values to interpret the clusters identified. This unsupervised pipeline identified 17 distinct clusters worldwide, representing typical temperate and tropical benthic habitats such as large canopy forming algae and branching corals, respectively, as well as transitional states between different habitat states. Temporal site-specific analyses further demonstrated the pipeline's effectiveness in capturing fine-scale habitat dynamics. By providing a standardised, scalable approach, this work enables consistent tracking of benthic habitat changes across spatial and temporal scales worldwide. This study also showcases the potential of integrating the <em>UMAP-HDBSCAN</em> pipeline with Shapley values for clustering noisy ecological data from citizen science initiatives.</div></div>","PeriodicalId":51024,"journal":{"name":"Ecological Informatics","volume":"86 ","pages":"Article 103058"},"PeriodicalIF":5.8,"publicationDate":"2025-01-31","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143395917","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"环境科学与生态学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Mapping habitats on coastal dunes, crucial yet highly vulnerable ecosystems, requires objectivity and repeatability, which are still lacking in the implementation of the Habitats Directive. Although remote sensing offers promising solutions, the effectiveness of distinguishing habitats on coastal dunes from satellite imagery remains uncertain. In this study, we compare crisp and fuzzy classification approaches using WorldView-3 imagery to map coastal dune habitats in two Natural Parks of Tuscany (Italy).
Field-collected vegetation data were classified into Annex I habitats of Habitats Directive and EUNIS habitats. Using field data as reference, we performed image classifications with a crisp method (Random Forests) and three fuzzy methods, namely Random Forests, Spectral Angle Mapper and Multiple Endmember Spectral Mixture Analysis. Metrics of overall accuracy and Mantel tests were used to compare the results.
EUNIS habitats exhibited the best performance in terms of classification accuracy, likely due to the simpler classification system. We observed a great disparity among habitats, with coastal dune scrubs and white dunes generally achieving the highest accuracy. Fuzzy classifications, despite yielding lower overall accuracy than the crisp classification, provided a more realistic representation of vegetation patterns, highlighting the inherent fuzziness of vegetation in coastal dunes. Despite challenges related to image resolution and habitat heterogeneity, combining satellite imagery with field surveys proved valuable for mapping coastal dune habitats, contributing essential data to the conservation of these fragile ecosystems. We provide a novel and effective tool, which will reduce the economic and physical efforts needed for habitat search and sampling in the field.
{"title":"Fuzzy approaches provide improved spatial detection of coastal dune EU habitats","authors":"Emilia Pafumi , Claudia Angiolini , Giovanni Bacaro , Emanuele Fanfarillo , Tiberio Fiaschi , Duccio Rocchini , Simona Sarmati , Michele Torresani , Hannes Feilhauer , Simona Maccherini","doi":"10.1016/j.ecoinf.2025.103059","DOIUrl":"10.1016/j.ecoinf.2025.103059","url":null,"abstract":"<div><div>Mapping habitats on coastal dunes, crucial yet highly vulnerable ecosystems, requires objectivity and repeatability, which are still lacking in the implementation of the Habitats Directive. Although remote sensing offers promising solutions, the effectiveness of distinguishing habitats on coastal dunes from satellite imagery remains uncertain. In this study, we compare crisp and fuzzy classification approaches using WorldView-3 imagery to map coastal dune habitats in two Natural Parks of Tuscany (Italy).</div><div>Field-collected vegetation data were classified into Annex I habitats of Habitats Directive and EUNIS habitats. Using field data as reference, we performed image classifications with a crisp method (Random Forests) and three fuzzy methods, namely Random Forests, Spectral Angle Mapper and Multiple Endmember Spectral Mixture Analysis. Metrics of overall accuracy and Mantel tests were used to compare the results.</div><div>EUNIS habitats exhibited the best performance in terms of classification accuracy, likely due to the simpler classification system. We observed a great disparity among habitats, with coastal dune scrubs and white dunes generally achieving the highest accuracy. Fuzzy classifications, despite yielding lower overall accuracy than the crisp classification, provided a more realistic representation of vegetation patterns, highlighting the inherent fuzziness of vegetation in coastal dunes. Despite challenges related to image resolution and habitat heterogeneity, combining satellite imagery with field surveys proved valuable for mapping coastal dune habitats, contributing essential data to the conservation of these fragile ecosystems. We provide a novel and effective tool, which will reduce the economic and physical efforts needed for habitat search and sampling in the field.</div></div>","PeriodicalId":51024,"journal":{"name":"Ecological Informatics","volume":"86 ","pages":"Article 103059"},"PeriodicalIF":5.8,"publicationDate":"2025-01-30","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143402725","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"环境科学与生态学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2025-01-29DOI: 10.1016/j.ecoinf.2025.103054
Víctor Manuel Romeo Jiménez , Jesús Santiago Notario del Pino , José Manuel Fernández-Guisuraga , Miguel Ángel Mejías Vera
Soil organic carbon (organic C) and pH are key soil properties and valuable indicators of soil quality, whereas phosphate retention capacity (P retention) is a diagnostic property to define andic soil properties and andic soils, with all of them typically interrelated in volcanic ash (i.e., andic) soils. In this paper, we examined the potential of a random forest (RF) regression model to predict field-measured soil pH, organic C and P retention capacity from several biophysical (type and fraction of the plant cover), bioclimatic (maximum temperature of the warmest month, precipitation and temperature seasonality, and precipitation of the driest quarter), and topographic (ruggedness and curvature of the slope) predictors in a protected forest area in Tenerife, Canary Islands. Piecewise structural equation modeling (pSEM) was subsequently used to unravel the complex, direct and indirect relationships between the biophysical, bioclimatic and topographic variables, and the selected soil properties. The RF regression model accounted for the properties of interest with varying degrees of accuracy, from organic C (R2 = 0.67; RMSE = 29.86), to P retention capacity (R2 = 0.44; RMSE = 18.84) and soil pH (R2 = 0.31; RMSE = 0.43). The pSEM model revealed that P retention capacity is strongly linked to organic C in volcanic ash soils, and thus indirectly to the environmental variables shaping organic C variability, namely fractional vegetation cover and precipitation seasonality.
{"title":"Prediction of some soil properties in volcanic soils using random forest modeling: A case study at chinyero special nature reserve (Tenerife, canary islands)","authors":"Víctor Manuel Romeo Jiménez , Jesús Santiago Notario del Pino , José Manuel Fernández-Guisuraga , Miguel Ángel Mejías Vera","doi":"10.1016/j.ecoinf.2025.103054","DOIUrl":"10.1016/j.ecoinf.2025.103054","url":null,"abstract":"<div><div>Soil organic carbon (organic C) and pH are key soil properties and valuable indicators of soil quality, whereas phosphate retention capacity (P retention) is a diagnostic property to define andic soil properties and andic soils, with all of them typically interrelated in volcanic ash (i.e., andic) soils. In this paper, we examined the potential of a random forest (RF) regression model to predict field-measured soil pH, organic C and P retention capacity from several biophysical (type and fraction of the plant cover), bioclimatic (maximum temperature of the warmest month, precipitation and temperature seasonality, and precipitation of the driest quarter), and topographic (ruggedness and curvature of the slope) predictors in a protected forest area in Tenerife, Canary Islands. Piecewise structural equation modeling (pSEM) was subsequently used to unravel the complex, direct and indirect relationships between the biophysical, bioclimatic and topographic variables, and the selected soil properties. The RF regression model accounted for the properties of interest with varying degrees of accuracy, from organic C (R<sup>2</sup> = 0.67; RMSE = 29.86), to P retention capacity (R<sup>2</sup> = 0.44; RMSE = 18.84) and soil pH (R<sup>2</sup> = 0.31; RMSE = 0.43). The pSEM model revealed that P retention capacity is strongly linked to organic C in volcanic ash soils, and thus indirectly to the environmental variables shaping organic C variability, namely fractional vegetation cover and precipitation seasonality.</div></div>","PeriodicalId":51024,"journal":{"name":"Ecological Informatics","volume":"86 ","pages":"Article 103054"},"PeriodicalIF":5.8,"publicationDate":"2025-01-29","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143102127","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"环境科学与生态学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pastures are a primary food source for livestock in Australia, with various pasture species grown in rotations. Accurate prediction of pasture availability is critical for effective farm management, livestock growth, and maintaining the supply chain. Environmental factors, particularly climate, heavily influence pasture yield. However, different pasture species respond to climate attributes with varying time lags; for example, one species might be more influenced by last week’s weather while another by the previous month’s highlighting the nuanced temporal dependencies. This time-lagging effect complicates the development of machine-learning models that can learn the temporal dependencies to predict pasture yield. To address this, our study proposes an averaging-based feature engineering approach, effectively capturing the varying temporal dependencies across pasture species and also allowing interpretation of the dependencies. Utilizing remote sensing and climate data, covering 196 farms (and 6885 paddocks) across Australia, we applied several machine learning techniques, including XGBoost, random forest, linear regression, deep neural networks, stacking, and bootstrapping. Our results show that incorporating averaging-based feature-engineered climate attributes significantly improves pasture yield predictions, with enhancements of up to 20.28%, 31.81%, and 31.11% across the three evaluation measures, RMSE, MAE, and R, respectively. This approach also enhances interpretability, revealing diverse time-lagging effects on different pasture species. XGBoost-based feature importance analysis further unveils insights into the influence of each climate attribute and its temporal dependencies on pasture yield.
{"title":"Feature engineering on climate data with machine learning to understand time-lagging effects in pasture yield prediction","authors":"Thirunavukarasu Balasubramaniam , Wathsala Anupama Mohotti , Kenneth Sabir , Richi Nayak","doi":"10.1016/j.ecoinf.2025.103011","DOIUrl":"10.1016/j.ecoinf.2025.103011","url":null,"abstract":"<div><div>Pastures are a primary food source for livestock in Australia, with various pasture species grown in rotations. Accurate prediction of pasture availability is critical for effective farm management, livestock growth, and maintaining the supply chain. Environmental factors, particularly climate, heavily influence pasture yield. However, different pasture species respond to climate attributes with varying time lags; for example, one species might be more influenced by last week’s weather while another by the previous month’s highlighting the nuanced temporal dependencies. This time-lagging effect complicates the development of machine-learning models that can learn the temporal dependencies to predict pasture yield. To address this, our study proposes an averaging-based feature engineering approach, effectively capturing the varying temporal dependencies across pasture species and also allowing interpretation of the dependencies. Utilizing remote sensing and climate data, covering 196 farms (and 6885 paddocks) across Australia, we applied several machine learning techniques, including XGBoost, random forest, linear regression, deep neural networks, stacking, and bootstrapping. Our results show that incorporating averaging-based feature-engineered climate attributes significantly improves pasture yield predictions, with enhancements of up to 20.28%, 31.81%, and 31.11% across the three evaluation measures, RMSE, MAE, and R<span><math><msup><mrow></mrow><mrow><mn>2</mn></mrow></msup></math></span>, respectively. This approach also enhances interpretability, revealing diverse time-lagging effects on different pasture species. XGBoost-based feature importance analysis further unveils insights into the influence of each climate attribute and its temporal dependencies on pasture yield.</div></div>","PeriodicalId":51024,"journal":{"name":"Ecological Informatics","volume":"86 ","pages":"Article 103011"},"PeriodicalIF":5.8,"publicationDate":"2025-01-28","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143102487","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"环境科学与生态学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2025-01-27DOI: 10.1016/j.ecoinf.2025.103047
Haibin Han , Bohui Jiang , Hongliang Huang , Yang Li , Jianghua Sui , Guoqing Zhao , Yuhan Wang , Heng Zhang , Shenglong Yang , Yongchuang Shi
Achieving energy-efficient, precise, and overall efficient production of Antarctic krill (Euphausia superba) is critical for realizing sustainable and ecological fisheries in the context of ongoing natural and anthropogenic climate change. In this study, we comprehensively analyzed commercial E. superba statistics and multivariate marine environmental data from 2010 to 2022 using the gravity center of the fishing ground method, dynamic sliding window, 3DCNN, and 3DCNN-ConvLSTM models. Results: 1) Inter-annual and inter-weekly catch varied significantly, with the total weekly catch evenly distributed between 0 and 2600 tons. The annual gravity center of the fishing grounds varied considerably between years and was mainly concentrated around the islands and in the strait. 2) Neither long- nor short-time-series historical data led to the best prediction. The optimal sliding window size for the 3DCNN was 4, whereas it was 11 for the 3DCNN-ConvLSTM model. 3) Climate change must be considered when selecting data, and the addition of biased data may negatively affect the model's predictive performance. 4) When using an optimal sliding window, the 3DCNN model outperformed the 3DCNN-ConvLSTM model. 5) The 3DCNN model tends to learn information about the environmental variables with the most significant differences in different categories of fishing grounds. This study aids in efficient selection of the most relevant historical data and an optimal model for developing a prediction model for high-catch fishing grounds, thereby providing a scientific foundation for clean production, sustainable development, and effective management of the E. superba fishery.
{"title":"Clean fishing: Construction of prediction model for high-catch Antarctic krill (Euphausia superba) fishing grounds based on deep learning and dynamic sliding window methods","authors":"Haibin Han , Bohui Jiang , Hongliang Huang , Yang Li , Jianghua Sui , Guoqing Zhao , Yuhan Wang , Heng Zhang , Shenglong Yang , Yongchuang Shi","doi":"10.1016/j.ecoinf.2025.103047","DOIUrl":"10.1016/j.ecoinf.2025.103047","url":null,"abstract":"<div><div>Achieving energy-efficient, precise, and overall efficient production of Antarctic krill (<em>Euphausia superba</em>) is critical for realizing sustainable and ecological fisheries in the context of ongoing natural and anthropogenic climate change. In this study, we comprehensively analyzed commercial <em>E. superba</em> statistics and multivariate marine environmental data from 2010 to 2022 using the gravity center of the fishing ground method, dynamic sliding window, 3DCNN, and 3DCNN-ConvLSTM models. Results: 1) Inter-annual and inter-weekly catch varied significantly, with the total weekly catch evenly distributed between 0 and 2600 tons. The annual gravity center of the fishing grounds varied considerably between years and was mainly concentrated around the islands and in the strait. 2) Neither long- nor short-time-series historical data led to the best prediction. The optimal sliding window size for the 3DCNN was 4, whereas it was 11 for the 3DCNN-ConvLSTM model. 3) Climate change must be considered when selecting data, and the addition of biased data may negatively affect the model's predictive performance. 4) When using an optimal sliding window, the 3DCNN model outperformed the 3DCNN-ConvLSTM model. 5) The 3DCNN model tends to learn information about the environmental variables with the most significant differences in different categories of fishing grounds. This study aids in efficient selection of the most relevant historical data and an optimal model for developing a prediction model for high-catch fishing grounds, thereby providing a scientific foundation for clean production, sustainable development, and effective management of the <em>E. superba</em> fishery.</div></div>","PeriodicalId":51024,"journal":{"name":"Ecological Informatics","volume":"86 ","pages":"Article 103047"},"PeriodicalIF":5.8,"publicationDate":"2025-01-27","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143102122","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"环境科学与生态学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
扫码关注我们
求助内容：
应助结果提醒方式：
京公网安备 11010802042870号