Biodiversity and Ecosystem Monitoring using deep learning

Biodiversity and Ecosystem Monitoring using deep learning | Research Square window.SnipcartSettings = { analytics: { enabled: false } }; (function() { var accessVector = localStorage.getItem('access_vector') || ''; window.dataLayer = window.dataLayer || []; if (accessVector) { window.dataLayer.push({ user: { profile: { profileInfo: { snid: accessVector } } } }); } })(); (function(w,d,s,l,i){w[l]=w[l]||[];w[l].push({'gtm.start':new Date().getTime(),event:'gtm.js'});var f=d.getElementsByTagName(s)[0],j=d.createElement(s),dl=l!='dataLayer'?'&l='+l:'';j.async=true;j.src='https://www.googletagmanager.com/gtm.js?id='+i+dl;f.parentNode.insertBefore(j,f);})(window,document,'script','dataLayer','GTM-K279D39R'); Browse Preprints In Review Journals COVID-19 Preprints AJE Video Bytes Research Tools Research Promotion AJE Professional Editing AJE Rubriq About Preprint Platform In Review Editorial Policies Our Team Advisory Board Help Center Sign In Submit a Preprint Cite Share Download PDF Research Article Biodiversity and Ecosystem Monitoring using deep learning Dilbar Hussain, Turkia Almoustafa, Saba Shakoor, Fahiza Fauz, and 2 more This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-7478081/v1 This work is licensed under a CC BY 4.0 License Status: Posted Version 1 posted You are reading this latest preprint version Abstract The growing impacts of climate change, habitat degradation, and human intervention have intensified the urgency for effective biodiversity and ecosystem monitoring. Traditional methods such as field surveys and manual species identification, though valuable, are increasingly challenged by limitations in scale, frequency, and accuracy. This study addresses the need for scalable, automated solutions by leveraging Deep Learning(DL) models to analyze multi-modal ecological data. The research specifically applies Convolutional Neural Networks(CNNs) for spatial analysis of high-resolution satellite imagery (Sentinel-2 and Landsat) and camera trap datasets (Snapshot Serengeti), and Long Short-Term Memory (LSTM) networks for processing temporal bioacoustic recordings (from Xeno-canto and Rainforest Connection). The methodology involves structured data preprocessing, model training, evaluation using standard metrics, and a final integration into a hybrid framework combining CNN and LSTM outputs. The hybrid model outperformed individual networks with 94.5% accuracy. This research confirms that deep learning, hybrid architectures offer a powerful solution for biodiversity monitoring. Biodiversity Monitoring Remote Sensing Ecosystem monitoring Multi-modal Introduction Monitoring biodiversity has become essential for all creatures' survival, in this changing world, where environmental crises rise by the day. Climate change and creatures activity are imposing havoc on ecosystems, pushing countless species toward extinction while destabilizing the delicate balance of nature 1 . Scientists relied on traditional methods like field surveys and manual species identification for the last decades. These approaches have served us well, but they're painfully limited, time-consuming, and require enormous manpower while only providing snapshots of complex ecological systems. These shortcomings leave gaping holes in understanding, making it clear the world needs smarter, more comprehensive solutions 2 . Today, cutting-edge technologies like satellite imaging and AI, particularly DL, are completely transforming how life is monitored on Earth. DL with its smart neural networks capable of processing information in layers, much like the human brain, thrives on the massive, complex datasets that characterize ecological studies. Whether it's analyzing satellite photos of rainforests, interpreting creatures from bioacoustic recorders, or sorting through millions of camera trap images, AI handles these tasks with superhuman efficiency, giving unprecedented insights 3 . These systems excel at making sense of visual and auditory data, distinguishing between species with astonishing accuracy. Picture a CNN scanning thousands of camera trap photos in a tropical forest, instantly recognizing whether an image contains a jaguar, toucan, or something entirely new. At the forefront of this revolution is CNN, a powerful model of Machine Learning (ML) in biodiversity and ecology. Imagine identifying the complex symmetry of a coral reef, identifying individual species by their unique acoustic signatures. This isn't science fiction, it's happening right now, providing real-time windows into ecosystem health and species distribution that were unimaginable just a decade ago 45 . This research leverages state-of-the-art DL methodologies and extensive ecological datasets to significantly enhance biodiversity monitoring. Specifically, this research uses high-resolution satellite imagery from Sentinel-2 and Landsat to analyze land-use changes, habitat fragmentation, and ecological disturbances at various spatial scales. Additionally, bioacoustic datasets from platforms like Xeno-canto and Rainforest Connection enable detailed acoustic analyses of species distributions and ecosystem health, particularly in biodiverse tropical forests. By integrating these sophisticated models with rich ecological datasets, our research aims to provide precise, scalable, and adaptive biodiversity monitoring solutions, significantly contributing to enhanced ecological management and global conservation efforts. This study primarily focuses on two advanced deep learning models: CNN and LSTM networks. CNN architecture good at processing satellite images and camera trap photographs, providing accurate species identification and habitat assessments through robust visual feature extraction. Concurrently, the LSTM model is deployed to effectively capture temporal dynamics inherent in bioacoustic data, enabling precise analysis of species recognition and ecosystem condition monitoring. The rise of satellite technology with DL has been particularly game-changing. AI algorithms can detect illegal logging operations hidden beneath forest canopies, measure the impact of wildfires on wildlife habitats, and even predict where biodiversity hotspots are most vulnerable. For conservationists fighting to protect endangered ecosystems, these capabilities are proving invaluable 6 . Similar to this, fieldwork is being revolutionized by AI-powered bioacoustic monitoring. Tiny sensors in rainforests can now continuously and in real-time offer data about the existence of species, assisting in the protection of some of the planet's most threatened and biodiverse environments. In Southeast Asian jungles, a system that recognizes Amazonian birds with exceptional precision may entirely fail. Background noise, uneven data quality, and regional differences in species characteristics are some of the challenges. As a result, scientists are developing lifetime learning systems and artificial intelligence (AI) that adapts to new surroundings and continuously learns new things, much like a field biologist acquiring decades of expertise 7 , 8 . These days, systems can predict environmental hazards before they become more serious, acting as ecological crisis early warning systems. The potential of AI for conservation goes much beyond straightforward observation. Understanding complex webs of relationships, such as how topography affects species distribution, how predator-prey dynamics impact entire ecosystems, and how climate shifts modify migration patterns, is what biodiversity is all about. Nature's intricacy necessitates equally complicated instruments. DL is being combined by researchers with other potent analytical techniques such as CNN and LSTM. These hybrid methods are providing previously unattainable insights into the interdependence of nature 9 . Modern conservation, powered by AI takes a comprehensive approach. It evaluates entire ecosystems, assesses the services they provide (from clean water to carbon storage), and develops strategies to protect them sustainably 10 . By analyzing patterns in air quality, temperature changes, and pollution levels, these models help policymakers take preventive action rather than scrambling to fix preventable disasters 11 , 12 . AI aids in the tracking of endangered animals, the fight against wildlife trafficking, and even the prediction of animal movements to avoid conflicts between people and wildlife. These applications provide hope for the preservation of both terrestrial and marine species, directly supporting global sustainability goals 13 . Even agriculture is profiting since AI enables farmers to use less chemicals and preserve local ecosystems while preserving production 14 . Through pushing the limits of DL, innovating to overcome existing constraints, and encouraging interdisciplinary collaboration, this work is creating monitoring systems that are as flexible and dynamic as the ecosystems they are studying. This research is a paradigm shift in how to safeguard life on Earth, not just a technological advancement. The final objective? to develop conservation instruments that are accurate enough to monitor individual species, resilient enough to function on a global scale, and intelligent enough to enable better decision-making regarding the preservation of our planet's priceless biodiversity. Recent advancements in AI and DL have substantially enhanced our capability to address complex ecological and environmental challenges. These technological advancements are proving indispensable for sustainable ecosystem management, particularly in biodiversity monitoring and climate change mitigation 15 . ML methods, including neural networks, random forests, and particularly DL, are increasingly effective for processing large, multidimensional datasets, critical for accurate weather forecasting and climate change modeling. These improved models significantly advance our understanding of ecological dynamics and the impacts of environmental changes 16 . Comprehensive datasets, such as BIOTROVE, have provided curated, extensive biodiversity imagery, significantly enhancing DL applications for species identification and ecosystem monitoring. These resources are crucial for supporting ecological research and accurate biodiversity assessments 17 . DL has revolutionized ecological and agricultural monitoring, providing scalable, real-time data essential for analyzing and managing spatial variability and environmental impacts by using remote sensing technologies 18 . AI-based ecosystem modeling has also facilitated deeper insights into ecological interactions. Ecosystem modeling simulate complex relationships within ecological networks, enhancing our predictive capabilities for ecosystem responses under various climate scenarios, thus enabling proactive conservation strategies 19 . AI-driven resource management practices contribute significantly towards sustainable use of resources, reducing ecological footprints, and enhancing overall energy efficiency and sustainability 20 . AI models incorporating fuzzy logic and expert systems effectively predict land-use changes, especially in arid ecosystems where conventional methods are inadequate due to extreme variability. Land use and land cover changes(LULCC) are critical in determining ecosystem biodiversity. These models highlight significant reductions in vegetation areas like grasslands and forests under changing climatic conditions 21 . AI models have dramatically enhanced ecological monitoring with the combination of Light Detection and Ranging (LiDAR) remote sensing technologies, enabling precise climate change detection and predictive modeling crucial for effective ecological management 22 . While AI technologies provide considerable benefits, to pose significant ethical and environmental challenges, particularly concerning their energy consumption and carbon footprints. These concerns highlight the importance of balancing AI’s analytical power with sustainable development goals and minimizing ecological impacts, a crucial area requiring ongoing research and responsible implementation 23 . Additionally, ethical considerations, such as equity, fairness, and global justice, are essential for responsible AI deployment in climate and biodiversity conservation efforts 24 . The concept of sustainable AI emphasizes ecological integrity and social justice, helping to ensure responsible and environmentally considerate. It addresses the lifecycle of AI products, from idea generation and training to implementation and governance, aiming to minimize ecological impacts while maximizing social benefits with the help of AI development practices 25 . Land-use changes due to anthropogenic activities have profound environmental impacts, altering ecosystem functions and atmospheric chemistry. Remote sensing and real-time spatial data are critical tools for monitoring these changes and informing sustainable land-use policies 26 . Phenological studies using AI and remote sensing data have provided new insights into ecosystem resilience and recovery from extreme climatic events. These technologies help identify critical factors influencing ecosystem recovery, emphasizing the necessity of incorporating phenological data into climate modeling to enhance ecological resilience 27 . Studies of climate-driven growth declines, such as those observed in European beech forests, underscore the urgency for integrated ecological monitoring systems utilizing AI technologies to predict and mitigate future impacts of climate change 28 . The convergence of AI, the Internet of Things (IoT), and big data technologies is proving transformative for smart city initiatives aimed at environmental sustainability. These integrated technologies provide robust platforms for monitoring urban ecosystems, managing resources efficiently, and reducing environmental footprints, significantly contributing to urban sustainability goals 29 . Together, these studies highlight the transformative potential and critical considerations associated with the application of AI in biodiversity monitoring and ecosystem management. This research demonstrates that, although AI technologies present powerful tools for ecological management and conservation, their implementation must be managed responsibly to balance technological progress with sustainable environmental stewardship. Methodology The methodology adopted in this research integrates advanced deep learning techniques with diverse ecological datasets, structured clearly into sequential steps to significantly enhance biodiversity monitoring capabilities: Step 1: Data Collection For thorough ecological monitoring, Sentinel-2 and Landsat high-resolution satellite imaging files are obtained, documenting ecological disturbances, habitat fragmentation, and changes in land use. Bioacoustic datasets are available for in-depth audio study on platforms like Rainforest Connection and Xeno-canto. These databases focus mostly on species distribution and ecosystem health, particularly in tropical forests. For visual species identification, camera trap datasets are sourced from Wildlife Insights and Snapshot Serengeti. Step 2: Data Preprocessing To guarantee excellent data quality and applicability for deep learning models, the gathered datasets go through preprocessing. This covers satellite images georeferencing, noise reduction, data normalization, and cleaning. Acoustic data is preprocessed using noise filtering, segmentation, and normalization, whilst camera trap images are preprocessed utilizing scaling, cropping, and augmentation techniques. Step 3: Model Selection and Configuration The selection of DL models is based on how well they work with ecological data: Convolutional Neural Networks (CNNs): CNNs are specifically used because of their remarkable capacity to extract visual elements from satellite and camera trap pictures that are necessary for accurate species identification. Long Short-Term Memory (LSTM) Networks: Sequential bioacoustic data is analyzed using LSTM to efficiently capture temporal dynamics related to ecological variables and species presence. Step 4: Model Training The CNN and LSTM models are trained using preprocessed acoustic data sequences and labeled visual data, respectively, utilizing advanced techniques such as data augmentation to manage sparse datasets and data imbalances in order to maximize their temporal analytic capabilities. Step 5: Model Evaluation and Validation Model performance is evaluated using standard ecological and statistical metrics including: Accuracy : The ratio of correctly predicted observations to the total observations, indicating overall model correctness. Precision : The ratio of correctly predicted positive observations to the total predicted positives, assessing the model's ability to avoid false positives. Recall (Sensitivity) : The ratio of correctly predicted positive observations to the actual positives, measuring the model's effectiveness in identifying true positives. F1 Score : The harmonic mean of precision and recall, providing a balanced measure of model performance, especially useful in the presence of class imbalance. Confusion Matrix : A tabular representation used to visualize model performance, showing true positives, false positives, true negatives, and false negatives. Cross-validation techniques are utilized to ensure the robustness and generalizability of the trained models across diverse ecological scenarios. Step 6: Hybrid Model Development Hybrid approaches are developed by integrating CNN-extracted features with LSTM to enhance prediction accuracy and ecological assessment robustness. Step 7: Addressing Ecological Challenges Specialized techniques such as data augmentation, few-shot learning, and modifications of cross-entropy loss functions are applied to manage challenges related to sparse datasets, data imbalance, and the open-world problem, thereby improving model robustness and adaptability. Step 8: Implementation and Deployment The final validated models are implemented within scalable and adaptable biodiversity monitoring frameworks designed for ecological management. Emphasis is placed on responsible AI deployment practices, ensuring alignment with ecological integrity and sustainable development objectives. Results The proposed methodology was implemented and evaluated using high-resolution satellite imagery, bioacoustic recordings, and camera trap dataset. DL models performed remarkably well in biodiversity and ecosystem monitoring tasks. The astounding outcome shows how well the model can extract important details from complex biological data and correctly identify species even in the face of challenging visual conditions like occlusions or shifting lighting. The CNN model's overall classification accuracy was 93.2%. The results showed a precision score of 92.0%, a recall score of 94.0%, and an F1 score of 0.93. Given its capacity to understand temporal connections, the LSTM proved to be a valuable tool for analyzing time-series ecological data, especially when it came to identifying species-specific vocal patterns. The LSTM model also showed strong performance, particularly when processing bioacoustic data for species detection. It obtained an accuracy of 91.0%, precision and recall of 90.0%, and an F1 score of 0.90. By combining CNN and LSTM models, a hybrid model improves prediction performance. This model merged CNN's spatial analysis capabilities with LSTM's ability to identify temporal patterns. The hybrid model beat all other configurations in Table 1 with 94.5% accuracy, 93.5% precision, 95.2% recall, and an F1 score of 0.94 3031 . These results validate the synergistic effect of combining both models and show the robustness and scalability of this integrated approach. Model evaluation metrics are displayed in Table 1: Model Accuracy (%) Precision (%) Recall (%) F1 Score CNN 93.2 92.0 94.0 0.93 LSTM 91.0 90.0 90.0 0.90 Hybrid Model 94.5 93.5 95.2 0.94 The hybrid model emerged as the most effective solution, showcasing superior performance across all metrics. These results highlight how DL, and hybrid designs in particular, have the potential to transform biodiversity monitoring and promote well-informed conservation efforts. In the fields of image and audio data processing, CNN and LSTM models each showed impressive performance. Conclusion Traditional ecological survey techniques are helpful, but they are often labor-intensive, time-consuming, and have a limited spatial and temporal scope. However, a powerful toolkit for modern conservation efforts is provided by the combination of AI-driven analytics, remote sensing data, and bioacoustic signals. This study demonstrates how DL technology, specifically CNN and LSTM networks, has the potential to revolutionize biodiversity and ecological monitoring. As ecosystems worldwide are threatened by anthropogenic pressures, habitat fragmentation, and climate change, scalable, accurate, and real-time monitoring systems are more crucial than ever. Using both spatial and temporal insights, CNN and LSTM models are employed to create a hybrid strategy that enhances performance on all assessed assessment metrics. CNN's efficiency in processing camera trap and satellite photography data is demonstrated by its capacity to extract spatial features, which allows it to classify habitats and identify species with high accuracy. Because LSTM networks can reliably detect species-specific vocalizations in a variety of ecological circumstances, they hold great promise for capturing temporal dynamics from audio recordings. The hybrid model demonstrates a high degree of versatility by combining visual and aural data streams, which makes it appropriate for deployment in a variety of contexts, from semi-urban green belts to thick tropical rainforests. This implies that multi-modal learning approaches are the best option because there isn't a single DL model that is consistently best for all ecological data sources. The results also show how important interdisciplinary cooperation is becoming. Ecologists, data scientists, software developers, and specialists in remote sensing must work together to do DL research in this field. Encouraging this kind of cooperation makes it possible to create monitoring systems that are both realistically feasible and scientifically sound. Declarations Funding : The British Academy/Cara/Leverhulme Researchers at Risk Research Support Grant. The University of Manchester Conflicts of interest/Competing interests : The authors declare no conflicts of interest, real or perceived. Availability of data and material : The datasets used and/or analysed during the current study available from the corresponding author on reasonable request Code availability : Not applicable Authors' contributions Dilbar Hussain: Technical writing and development of scientific data analysis for the manuscript. Turkia Almoustafa (Corresponding Author): Original idea, writing, and editing for scientific purposes. Saba Shakoor: General methodology and mathematical modelling. Fahiza Fauz: General methodology and mathematical modelling. Rizwan Ahmed: Technical writing and overall compilation for the manuscript. Shahzaib Ansari: Technical writing and editing process. References S. Villon, C. Iovan, M. Mangeas, L. Vigliola, Sensors 2022, 22 . J. Hoffmann, J. Muro, O. Dubovyk, Remote Sens (Basel) 2022, 14 . E. Agrillo, F. Filipponi, A. Pezzarossa, L. Casella, D. Smiraglia, A. Orasi, F. Attorre, A. Taramelli, Remote Sens (Basel) 2021, 13 . M. Lin, T. Lin, L. Jones, X. Liu, L. Xing, J. Sui, J. Zhang, H. Ye, Y. Liu, G. Zhang, X. Lu, Remote Sens (Basel) 2021, 13 . K. N. Shivaprakash, N. Swami, S. Mysorekar, R. Arora, A. Gangadharan, K. Vohra, M. Jadeyegowda, J. M. Kiesecker, Potential for Artificial Intelligence (AI) and Machine Learning (ML) Applications in Biodiversity Conservation, Managing Forests, and Related Services in India , MDPI, 2022. J. Müller, O. Mitesser, H. M. Schaefer, S. Seibold, A. Busse, P. Kriegel, D. Rabl, R. Gelis, A. Arteaga, J. Freile, G. A. Leite, T. N. de Melo, J. G. LeBien, M. Campos-Cerqueira, N. Blüthgen, C. J. Tremlett, D. Böttger, H. Feldhaar, N. Grella, A. Falconí-López, D. A. Donoso, J. Moriniere, Z. Buřivalová, Nat Commun 2023, 14 . P. Bodesheim, J. Blunk, M. Körschens, C. A. Brust, C. Käding, J. Denzler, Mammalian Biology 2022, 102 , 853. S. S. Sethi, A. Bick, R. M. Ewers, H. Klinck, V. Ramesh, M. N. Tuanmu, D. A. Coomes, Nat Ecol Evol 2023, 7 , 1373. C. Zhang, H. Zhan, Z. Hao, X. Gao, Forests 2023, 14 . M. Bayat, H. Burkhart, M. Namiranian, S. K. Hamidi, S. Heidari, M. Hassani, Forests 2021, 12 . S. Subramaniam, N. Raju, A. Ganesan, N. Rajavel, M. Chenniappan, C. Prakash, A. Pramanik, A. K. Basak, S. Dixit, Artificial Intelligence Technologies for Forecasting Air Pollution and Human Health: A Narrative Review , MDPI, 2022. S. C. Izah, A. O. Iyiola, B. Yarkwan, G. Richard, in Visualization Techniques for Climate Change with Machine Learning and Artificial Intelligence , Elsevier, 2022, pp. 123–148. D. A. Isabelle, M. Westerlund, Sustainability (Switzerland) 2022, 14 . B. Garske, A. Bau, F. Ekardt, Sustainability (Switzerland) 2021, 13 . East-West University, Institute of Electrical and Electronics Engineers, Institute of Electrical and Electronics Engineers. Bangladesh Section, IEEE Robotics and Automation Society. Bangladesh Chapter, 2019 1st International Conference on Advances in Science, Engineering and Robotics Technology (ICASERT 2019) : May 3-5, 2019, Dhaka, Bangladesh. B. Bochenek, Z. Ustrnul, Atmosphere (Basel) 2022, 13 . C.-H. Yang, B. Feuer, Z. Jubery, Z. K. Deng, A. Nakkab, M. Z. Hasan, S. Chiranjeevi, K. Marshall, N. Baishnab, A. K. Singh, A. Singh, S. Sarkar, N. Merchant, C. Hegde, B. Ganapathysubramanian, 2024. H. Han, Z. Liu, J. Li, Z. Zeng, Challenges in remote sensing based climate and crop monitoring: navigating the complexities using AI , Springer Science and Business Media Deutschland GmbH, 2024. N. R. Winter, M. Cearns, S. R. Clark, R. Leenings, U. Dannlowski, B. T. Baune, T. Hahn, From multivariate methods to an AI ecosystem , Springer Nature, 2021. L. Chen, Z. Chen, Y. Zhang, Y. Liu, A. I. Osman, M. Farghali, J. Hua, A. Al-Fatesh, I. Ihara, D. W. Rooney, P. S. Yap, Artificial intelligence-based solutions for climate change: a review , Springer Science and Business Media Deutschland GmbH, 2023. Y. Yu, Y. Cao, D. Hou, M. Disse, A. Brieden, H. Zhang, R. Yu, Journal of Geographical Sciences 2022, 32 , 717. C. Sravani, P. Ananth, C. Sasikala, N. S. Kumar, C. D. V. P. Kumari, V. Thrimurthulu, Remote Sens Earth Syst Sci 2025. J. Cowls, A. Tsamados, M. Taddeo, L. Floridi, AI Soc 2023, 38 , 283. M. Coeckelbergh, AI and Ethics 2021, 1 , 67. A. van Wynsberghe, AI and Ethics 2021, 1 , 213. P. S. Roy, R. M. Ramachandran, O. Paul, P. K. Thakur, S. Ravan, M. D. Behera, C. Sarangi, V. P. Kanawade, Anthropogenic Land Use and Land Cover Changes—A Review on Its Environmental Consequences and Climate Change , Springer, 2022. Y. Li, W. Zhang, C. R. Schwalm, P. Gentine, W. K. Smith, P. Ciais, J. S. Kimball, A. Gazol, S. A. Kannenberg, A. Chen, S. Piao, H. Liu, D. Chen, X. Wu, Nat Clim Chang 2023, 13 , 182. E. Martinez del Castillo, C. S. Zang, A. Buras, A. Hacket-Pain, J. Esper, R. Serrano-Notivoli, C. Hartl, R. Weigel, S. Klesse, V. Resco de Dios, T. Scharnweber, I. Dorado-Liñán, M. van der Maaten-Theunissen, E. van der Maaten, A. Jump, S. Mikac, B. E. Banzragch, W. Beck, L. Cavin, H. Claessens, V. Čada, K. Čufar, C. Dulamsuren, J. Gričar, E. Gil-Pelegrín, P. Janda, M. Kazimirovic, J. Kreyling, N. Latte, C. Leuschner, L. A. Longares, A. Menzel, M. Merela, R. Motta, L. Muffler, P. Nola, A. M. Petritan, I. C. Petritan, P. Prislan, Á. Rubio-Cuadrado, M. Rydval, B. Stajić, M. Svoboda, E. Toromani, V. Trotsiuk, M. Wilmking, T. Zlatanov, M. de Luis, Commun Biol 2022, 5 . S. E. Bibri, A. Alexandre, A. Sharifi, J. Krogstie, Environmentally sustainable smart cities and their converging AI, IoT, and big data technologies and solutions: an integrated approach to an extensive literature review , Springer Nature, 2023. A. O. Almagrabi, Computers, Materials and Continua 2023, 76 , 2079. K. O. Adefemi, M. B. Mutanga, Digital 2025, 5 , 16. Additional Declarations No competing interests reported. Cite Share Download PDF Status: Posted Version 1 posted You are reading this latest preprint version Research Square lets you share your work early, gain feedback from the community, and start making changes to your manuscript prior to peer review in a journal. As a division of Research Square Company, we’re committed to making research communication faster, fairer, and more useful. We do this by developing innovative software and high quality services for the global research community. Our growing team is made up of researchers and industry professionals working together to solve the most critical problems facing scientific publishing. Also discoverable on Platform About Our Team In Review Editorial Policies Advisory Board Help Center Resources Author Services Accessibility API Access RSS feed Manage Cookie Preferences © Research Square 2026 | ISSN 2693-5015 (online) Privacy Policy Terms of Service Do Not Sell My Personal Information {"props":{"pageProps":{"initialData":{"identity":"rs-7478081","acceptedTermsAndConditions":true,"allowDirectSubmit":true,"archivedVersions":[],"articleType":"Research Article","associatedPublications":[],"authors":[{"id":511667041,"identity":"08140992-5231-4da1-aeb6-a66cb00b855d","order_by":0,"name":"Dilbar Hussain","email":"","orcid":"","institution":"Iqra University Main Campus Karachi","correspondingAuthor":false,"prefix":"","firstName":"Dilbar","middleName":"","lastName":"Hussain","suffix":""},{"id":511667043,"identity":"2d7c12c8-6874-44d7-803c-a9084d3d6ad0","order_by":1,"name":"Turkia Almoustafa","email":"data:image/png;base64,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","orcid":"","institution":"University of Manchester","correspondingAuthor":true,"prefix":"","firstName":"Turkia","middleName":"","lastName":"Almoustafa","suffix":""},{"id":511667044,"identity":"07fe8415-3208-4cbd-bbb5-c56b70235f13","order_by":2,"name":"Saba Shakoor","email":"","orcid":"","institution":"University of Hertfordshire","correspondingAuthor":false,"prefix":"","firstName":"Saba","middleName":"","lastName":"Shakoor","suffix":""},{"id":511667045,"identity":"c697835f-5590-4868-9492-005f493d7fdf","order_by":3,"name":"Fahiza Fauz","email":"","orcid":"","institution":"National University of Sciences and Technology (NUST)","correspondingAuthor":false,"prefix":"","firstName":"Fahiza","middleName":"","lastName":"Fauz","suffix":""},{"id":511667046,"identity":"fdb51810-5da0-4e0f-a951-ff323a6ddf81","order_by":4,"name":"Rizwan Ahmed","email":"","orcid":"","institution":"Iqra University Main Campus Karachi","correspondingAuthor":false,"prefix":"","firstName":"Rizwan","middleName":"","lastName":"Ahmed","suffix":""},{"id":511667047,"identity":"d5871e89-c969-498a-9b3f-d067ded53ce1","order_by":5,"name":"Shahzaib Ansari","email":"","orcid":"","institution":"Iqra University Main Campus Karachi","correspondingAuthor":false,"prefix":"","firstName":"Shahzaib","middleName":"","lastName":"Ansari","suffix":""}],"badges":[],"createdAt":"2025-08-28 08:38:17","currentVersionCode":1,"declarations":"","doi":"10.21203/rs.3.rs-7478081/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-7478081/v1","draftVersion":[],"editorialEvents":[],"editorialNote":"","failedWorkflow":false,"files":[{"id":105566808,"identity":"72c03b1e-b774-412b-af17-e6421018a196","added_by":"auto","created_at":"2026-03-27 12:57:24","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":415601,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-7478081/v1/1e680548-4391-4640-b852-2983266dc122.pdf"}],"financialInterests":"No competing interests reported.","formattedTitle":"Biodiversity and Ecosystem Monitoring using deep learning","fulltext":[{"header":"Introduction","content":"\u003cp\u003eMonitoring biodiversity has become essential for all creatures' survival, in this changing world, where environmental crises rise by the day. Climate change and creatures activity are imposing havoc on ecosystems, pushing countless species toward extinction while destabilizing the delicate balance of nature\u003csup\u003e\u003cspan citationid=\"CR1\" class=\"CitationRef\"\u003e1\u003c/span\u003e\u003c/sup\u003e. Scientists relied on traditional methods like field surveys and manual species identification for the last decades. These approaches have served us well, but they're painfully limited, time-consuming, and require enormous manpower while only providing snapshots of complex ecological systems. These shortcomings leave gaping holes in understanding, making it clear the world needs smarter, more comprehensive solutions\u003csup\u003e\u003cspan citationid=\"CR2\" class=\"CitationRef\"\u003e2\u003c/span\u003e\u003c/sup\u003e.\u003c/p\u003e\u003cp\u003eToday, cutting-edge technologies like satellite imaging and AI, particularly DL, are completely transforming how life is monitored on Earth. DL with its smart neural networks capable of processing information in layers, much like the human brain, thrives on the massive, complex datasets that characterize ecological studies. Whether it's analyzing satellite photos of rainforests, interpreting creatures from bioacoustic recorders, or sorting through millions of camera trap images, AI handles these tasks with superhuman efficiency, giving unprecedented insights\u003csup\u003e\u003cspan citationid=\"CR3\" class=\"CitationRef\"\u003e3\u003c/span\u003e\u003c/sup\u003e.\u003c/p\u003e\u003cp\u003eThese systems excel at making sense of visual and auditory data, distinguishing between species with astonishing accuracy. Picture a CNN scanning thousands of camera trap photos in a tropical forest, instantly recognizing whether an image contains a jaguar, toucan, or something entirely new. At the forefront of this revolution is CNN, a powerful model of Machine Learning (ML) in biodiversity and ecology. Imagine identifying the complex symmetry of a coral reef, identifying individual species by their unique acoustic signatures. This isn't science fiction, it's happening right now, providing real-time windows into ecosystem health and species distribution that were unimaginable just a decade ago\u003csup\u003e45\u003c/sup\u003e.\u003c/p\u003e\u003cp\u003eThis research leverages state-of-the-art DL methodologies and extensive ecological datasets to significantly enhance biodiversity monitoring. Specifically, this research uses high-resolution satellite imagery from Sentinel-2 and Landsat to analyze land-use changes, habitat fragmentation, and ecological disturbances at various spatial scales. Additionally, bioacoustic datasets from platforms like Xeno-canto and Rainforest Connection enable detailed acoustic analyses of species distributions and ecosystem health, particularly in biodiverse tropical forests. By integrating these sophisticated models with rich ecological datasets, our research aims to provide precise, scalable, and adaptive biodiversity monitoring solutions, significantly contributing to enhanced ecological management and global conservation efforts. This study primarily focuses on two advanced deep learning models: CNN and LSTM networks. CNN architecture good at processing satellite images and camera trap photographs, providing accurate species identification and habitat assessments through robust visual feature extraction. Concurrently, the LSTM model is deployed to effectively capture temporal dynamics inherent in bioacoustic data, enabling precise analysis of species recognition and ecosystem condition monitoring.\u003c/p\u003e\u003cp\u003eThe rise of satellite technology with DL has been particularly game-changing. AI algorithms can detect illegal logging operations hidden beneath forest canopies, measure the impact of wildfires on wildlife habitats, and even predict where biodiversity hotspots are most vulnerable. For conservationists fighting to protect endangered ecosystems, these capabilities are proving invaluable\u003csup\u003e\u003cspan citationid=\"CR6\" class=\"CitationRef\"\u003e6\u003c/span\u003e\u003c/sup\u003e. Similar to this, fieldwork is being revolutionized by AI-powered bioacoustic monitoring. Tiny sensors in rainforests can now continuously and in real-time offer data about the existence of species, assisting in the protection of some of the planet's most threatened and biodiverse environments.\u003c/p\u003e\u003cp\u003eIn Southeast Asian jungles, a system that recognizes Amazonian birds with exceptional precision may entirely fail. Background noise, uneven data quality, and regional differences in species characteristics are some of the challenges.\u003c/p\u003e\u003cp\u003eAs a result, scientists are developing lifetime learning systems and artificial intelligence (AI) that adapts to new surroundings and continuously learns new things, much like a field biologist acquiring decades of expertise\u003csup\u003e\u003cspan citationid=\"CR7\" class=\"CitationRef\"\u003e7\u003c/span\u003e,\u003cspan citationid=\"CR8\" class=\"CitationRef\"\u003e8\u003c/span\u003e\u003c/sup\u003e. These days, systems can predict environmental hazards before they become more serious, acting as ecological crisis early warning systems. The potential of AI for conservation goes much beyond straightforward observation. Understanding complex webs of relationships, such as how topography affects species distribution, how predator-prey dynamics impact entire ecosystems, and how climate shifts modify migration patterns, is what biodiversity is all about. Nature's intricacy necessitates equally complicated instruments. DL is being combined by researchers with other potent analytical techniques such as CNN and LSTM. These hybrid methods are providing previously unattainable insights into the interdependence of nature\u003csup\u003e\u003cspan citationid=\"CR9\" class=\"CitationRef\"\u003e9\u003c/span\u003e\u003c/sup\u003e. Modern conservation, powered by AI takes a comprehensive approach. It evaluates entire ecosystems, assesses the services they provide (from clean water to carbon storage), and develops strategies to protect them sustainably\u003csup\u003e\u003cspan citationid=\"CR10\" class=\"CitationRef\"\u003e10\u003c/span\u003e\u003c/sup\u003e.\u003c/p\u003e\u003cp\u003eBy analyzing patterns in air quality, temperature changes, and pollution levels, these models help policymakers take preventive action rather than scrambling to fix preventable disasters\u003csup\u003e\u003cspan citationid=\"CR11\" class=\"CitationRef\"\u003e11\u003c/span\u003e,\u003cspan citationid=\"CR12\" class=\"CitationRef\"\u003e12\u003c/span\u003e\u003c/sup\u003e.\u003c/p\u003e\u003cp\u003eAI aids in the tracking of endangered animals, the fight against wildlife trafficking, and even the prediction of animal movements to avoid conflicts between people and wildlife. These applications provide hope for the preservation of both terrestrial and marine species, directly supporting global sustainability goals\u003csup\u003e\u003cspan citationid=\"CR13\" class=\"CitationRef\"\u003e13\u003c/span\u003e\u003c/sup\u003e. Even agriculture is profiting since AI enables farmers to use less chemicals and preserve local ecosystems while preserving production\u003csup\u003e\u003cspan citationid=\"CR14\" class=\"CitationRef\"\u003e14\u003c/span\u003e\u003c/sup\u003e.\u003c/p\u003e\u003cp\u003eThrough pushing the limits of DL, innovating to overcome existing constraints, and encouraging interdisciplinary collaboration, this work is creating monitoring systems that are as flexible and dynamic as the ecosystems they are studying. This research is a paradigm shift in how to safeguard life on Earth, not just a technological advancement. The final objective? to develop conservation instruments that are accurate enough to monitor individual species, resilient enough to function on a global scale, and intelligent enough to enable better decision-making regarding the preservation of our planet's priceless biodiversity.\u003c/p\u003e\u003cp\u003eRecent advancements in AI and DL have substantially enhanced our capability to address complex ecological and environmental challenges. These technological advancements are proving indispensable for sustainable ecosystem management, particularly in biodiversity monitoring and climate change mitigation\u003csup\u003e\u003cspan citationid=\"CR15\" class=\"CitationRef\"\u003e15\u003c/span\u003e\u003c/sup\u003e. ML methods, including neural networks, random forests, and particularly DL, are increasingly effective for processing large, multidimensional datasets, critical for accurate weather forecasting and climate change modeling. These improved models significantly advance our understanding of ecological dynamics and the impacts of environmental changes\u003csup\u003e\u003cspan citationid=\"CR16\" class=\"CitationRef\"\u003e16\u003c/span\u003e\u003c/sup\u003e.\u003c/p\u003e\u003cp\u003eComprehensive datasets, such as BIOTROVE, have provided curated, extensive biodiversity imagery, significantly enhancing DL applications for species identification and ecosystem monitoring. These resources are crucial for supporting ecological research and accurate biodiversity assessments\u003csup\u003e\u003cspan citationid=\"CR17\" class=\"CitationRef\"\u003e17\u003c/span\u003e\u003c/sup\u003e. DL has revolutionized ecological and agricultural monitoring, providing scalable, real-time data essential for analyzing and managing spatial variability and environmental impacts by using remote sensing technologies\u003csup\u003e\u003cspan citationid=\"CR18\" class=\"CitationRef\"\u003e18\u003c/span\u003e\u003c/sup\u003e.\u003c/p\u003e\u003cp\u003eAI-based ecosystem modeling has also facilitated deeper insights into ecological interactions. Ecosystem modeling simulate complex relationships within ecological networks, enhancing our predictive capabilities for ecosystem responses under various climate scenarios, thus enabling proactive conservation strategies\u003csup\u003e\u003cspan citationid=\"CR19\" class=\"CitationRef\"\u003e19\u003c/span\u003e\u003c/sup\u003e. AI-driven resource management practices contribute significantly towards sustainable use of resources, reducing ecological footprints, and enhancing overall energy efficiency and sustainability\u003csup\u003e\u003cspan citationid=\"CR20\" class=\"CitationRef\"\u003e20\u003c/span\u003e\u003c/sup\u003e.\u003c/p\u003e\u003cp\u003eAI models incorporating fuzzy logic and expert systems effectively predict land-use changes, especially in arid ecosystems where conventional methods are inadequate due to extreme variability. Land use and land cover changes(LULCC) are critical in determining ecosystem biodiversity. These models highlight significant reductions in vegetation areas like grasslands and forests under changing climatic conditions\u003csup\u003e\u003cspan citationid=\"CR21\" class=\"CitationRef\"\u003e21\u003c/span\u003e\u003c/sup\u003e. AI models have dramatically enhanced ecological monitoring with the combination of Light Detection and Ranging (LiDAR) remote sensing technologies, enabling precise climate change detection and predictive modeling crucial for effective ecological management\u003csup\u003e\u003cspan citationid=\"CR22\" class=\"CitationRef\"\u003e22\u003c/span\u003e\u003c/sup\u003e. While AI technologies provide considerable benefits, to pose significant ethical and environmental challenges, particularly concerning their energy consumption and carbon footprints. These concerns highlight the importance of balancing AI\u0026rsquo;s analytical power with sustainable development goals and minimizing ecological impacts, a crucial area requiring ongoing research and responsible implementation\u003csup\u003e\u003cspan citationid=\"CR23\" class=\"CitationRef\"\u003e23\u003c/span\u003e\u003c/sup\u003e. Additionally, ethical considerations, such as equity, fairness, and global justice, are essential for responsible AI deployment in climate and biodiversity conservation efforts\u003csup\u003e\u003cspan citationid=\"CR24\" class=\"CitationRef\"\u003e24\u003c/span\u003e\u003c/sup\u003e.\u003c/p\u003e\u003cp\u003eThe concept of sustainable AI emphasizes ecological integrity and social justice, helping to ensure responsible and environmentally considerate. It addresses the lifecycle of AI products, from idea generation and training to implementation and governance, aiming to minimize ecological impacts while maximizing social benefits with the help of AI development practices\u003csup\u003e\u003cspan citationid=\"CR25\" class=\"CitationRef\"\u003e25\u003c/span\u003e\u003c/sup\u003e. Land-use changes due to anthropogenic activities have profound environmental impacts, altering ecosystem functions and atmospheric chemistry. Remote sensing and real-time spatial data are critical tools for monitoring these changes and informing sustainable land-use policies\u003csup\u003e\u003cspan citationid=\"CR26\" class=\"CitationRef\"\u003e26\u003c/span\u003e\u003c/sup\u003e. Phenological studies using AI and remote sensing data have provided new insights into ecosystem resilience and recovery from extreme climatic events. These technologies help identify critical factors influencing ecosystem recovery, emphasizing the necessity of incorporating phenological data into climate modeling to enhance ecological resilience\u003csup\u003e\u003cspan citationid=\"CR27\" class=\"CitationRef\"\u003e27\u003c/span\u003e\u003c/sup\u003e. Studies of climate-driven growth declines, such as those observed in European beech forests, underscore the urgency for integrated ecological monitoring systems utilizing AI technologies to predict and mitigate future impacts of climate change\u003csup\u003e\u003cspan citationid=\"CR28\" class=\"CitationRef\"\u003e28\u003c/span\u003e\u003c/sup\u003e.\u003c/p\u003e\u003cp\u003eThe convergence of AI, the Internet of Things (IoT), and big data technologies is proving transformative for smart city initiatives aimed at environmental sustainability. These integrated technologies provide robust platforms for monitoring urban ecosystems, managing resources efficiently, and reducing environmental footprints, significantly contributing to urban sustainability goals\u003csup\u003e\u003cspan citationid=\"CR29\" class=\"CitationRef\"\u003e29\u003c/span\u003e\u003c/sup\u003e. Together, these studies highlight the transformative potential and critical considerations associated with the application of AI in biodiversity monitoring and ecosystem management. This research demonstrates that, although AI technologies present powerful tools for ecological management and conservation, their implementation must be managed responsibly to balance technological progress with sustainable environmental stewardship.\u003c/p\u003e"},{"header":"Methodology","content":"\u003cp\u003eThe methodology adopted in this research integrates advanced deep learning techniques with diverse ecological datasets, structured clearly into sequential steps to significantly enhance biodiversity monitoring capabilities:\u003c/p\u003e\u003cp\u003e\u003cem\u003eStep 1: Data Collection\u003c/em\u003e For thorough ecological monitoring, Sentinel-2 and Landsat high-resolution satellite imaging files are obtained, documenting ecological disturbances, habitat fragmentation, and changes in land use. Bioacoustic datasets are available for in-depth audio study on platforms like Rainforest Connection and Xeno-canto. These databases focus mostly on species distribution and ecosystem health, particularly in tropical forests. For visual species identification, camera trap datasets are sourced from Wildlife Insights and Snapshot Serengeti.\u003c/p\u003e\u003cp\u003e\u003cem\u003eStep 2: Data Preprocessing\u003c/em\u003e To guarantee excellent data quality and applicability for deep learning models, the gathered datasets go through preprocessing. This covers satellite images georeferencing, noise reduction, data normalization, and cleaning. Acoustic data is preprocessed using noise filtering, segmentation, and normalization, whilst camera trap images are preprocessed utilizing scaling, cropping, and augmentation techniques.\u003c/p\u003e\u003cp\u003e\u003cem\u003eStep 3: Model Selection and Configuration\u003c/em\u003e The selection of DL models is based on how well they work with ecological data:\u003c/p\u003e\u003cp\u003e\u003cul\u003e\u003cli\u003e\u003cp\u003eConvolutional Neural Networks (CNNs): CNNs are specifically used because of their remarkable capacity to extract visual elements from satellite and camera trap pictures that are necessary for accurate species identification.\u003c/p\u003e\u003c/li\u003e\u003cli\u003e\u003cp\u003eLong Short-Term Memory (LSTM) Networks: Sequential bioacoustic data is analyzed using LSTM to efficiently capture temporal dynamics related to ecological variables and species presence.\u003c/p\u003e\u003c/li\u003e\u003c/ul\u003e\u003c/p\u003e\u003cp\u003e\u003cem\u003eStep 4: Model Training\u003c/em\u003e The CNN and LSTM models are trained using preprocessed acoustic data sequences and labeled visual data, respectively, utilizing advanced techniques such as data augmentation to manage sparse datasets and data imbalances in order to maximize their temporal analytic capabilities.\u003c/p\u003e\u003cp\u003e\u003cem\u003eStep 5: Model Evaluation and Validation\u003c/em\u003e Model performance is evaluated using standard ecological and statistical metrics including:\u003c/p\u003e\u003cp\u003e\u003cul\u003e\u003cli\u003e\u003cp\u003e\u003cem\u003eAccuracy\u003c/em\u003e: The ratio of correctly predicted observations to the total observations, indicating overall model correctness.\u003c/p\u003e\u003c/li\u003e\u003cli\u003e\u003cp\u003e\u003cem\u003ePrecision\u003c/em\u003e: The ratio of correctly predicted positive observations to the total predicted positives, assessing the model's ability to avoid false positives.\u003c/p\u003e\u003c/li\u003e\u003cli\u003e\u003cp\u003e\u003cem\u003eRecall (Sensitivity)\u003c/em\u003e: The ratio of correctly predicted positive observations to the actual positives, measuring the model's effectiveness in identifying true positives.\u003c/p\u003e\u003c/li\u003e\u003cli\u003e\u003cp\u003e\u003cem\u003eF1 Score\u003c/em\u003e: The harmonic mean of precision and recall, providing a balanced measure of model performance, especially useful in the presence of class imbalance.\u003c/p\u003e\u003c/li\u003e\u003cli\u003e\u003cp\u003e\u003cem\u003eConfusion Matrix\u003c/em\u003e: A tabular representation used to visualize model performance, showing true positives, false positives, true negatives, and false negatives.\u003c/p\u003e\u003c/li\u003e\u003c/ul\u003e\u003c/p\u003e\u003cp\u003eCross-validation techniques are utilized to ensure the robustness and generalizability of the trained models across diverse ecological scenarios.\u003c/p\u003e\u003cp\u003e\u003cem\u003eStep 6: Hybrid Model Development\u003c/em\u003e Hybrid approaches are developed by integrating CNN-extracted features with LSTM to enhance prediction accuracy and ecological assessment robustness.\u003c/p\u003e\u003cp\u003e\u003cem\u003eStep 7: Addressing Ecological Challenges\u003c/em\u003e Specialized techniques such as data augmentation, few-shot learning, and modifications of cross-entropy loss functions are applied to manage challenges related to sparse datasets, data imbalance, and the open-world problem, thereby improving model robustness and adaptability.\u003c/p\u003e\u003cp\u003e\u003cem\u003eStep 8: Implementation and Deployment\u003c/em\u003e The final validated models are implemented within scalable and adaptable biodiversity monitoring frameworks designed for ecological management. Emphasis is placed on responsible AI deployment practices, ensuring alignment with ecological integrity and sustainable development objectives.\u003c/p\u003e"},{"header":"Results","content":"\u003cp\u003eThe proposed methodology was implemented and evaluated using high-resolution satellite imagery, bioacoustic recordings, and camera trap dataset. DL models performed remarkably well in biodiversity and ecosystem monitoring tasks.\u003c/p\u003e\u003cp\u003eThe astounding outcome shows how well the model can extract important details from complex biological data and correctly identify species even in the face of challenging visual conditions like occlusions or shifting lighting. The CNN model's overall classification accuracy was 93.2%. The results showed a precision score of 92.0%, a recall score of 94.0%, and an F1 score of 0.93. Given its capacity to understand temporal connections, the LSTM proved to be a valuable tool for analyzing time-series ecological data, especially when it came to identifying species-specific vocal patterns. The LSTM model also showed strong performance, particularly when processing bioacoustic data for species detection. It obtained an accuracy of 91.0%, precision and recall of 90.0%, and an F1 score of 0.90.\u003c/p\u003e\u003cp\u003eBy combining CNN and LSTM models, a hybrid model improves prediction performance. This model merged CNN's spatial analysis capabilities with LSTM's ability to identify temporal patterns. The hybrid model beat all other configurations in Table\u0026nbsp;1 with 94.5% accuracy, 93.5% precision, 95.2% recall, and an F1 score of 0.94\u003csup\u003e3031\u003c/sup\u003e. These results validate the synergistic effect of combining both models and show the robustness and scalability of this integrated approach.\u003c/p\u003e\n\u003ch3\u003eModel evaluation metrics are displayed in Table 1:\u003c/h3\u003e\n\u003cp\u003e\u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"No\" id=\"Taba\" border=\"1\"\u003e\u003ccolgroup cols=\"5\"\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e\u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e\u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e\u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c4\" colnum=\"4\"\u003e\u003c/div\u003e\u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c5\" colnum=\"5\"\u003e\u003c/div\u003e\u003cthead\u003e\u003ctr\u003e\u003cth align=\"left\" colname=\"c1\"\u003e\u003cp\u003eModel\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c2\"\u003e\u003cp\u003eAccuracy (%)\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c3\"\u003e\u003cp\u003ePrecision (%)\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c4\"\u003e\u003cp\u003eRecall (%)\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c5\"\u003e\u003cp\u003eF1 Score\u003c/p\u003e\u003c/th\u003e\u003c/tr\u003e\u003c/thead\u003e\u003ctbody\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eCNN\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e\u003cp\u003e93.2\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e\u003cp\u003e92.0\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e\u003cp\u003e94.0\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e\u003cp\u003e0.93\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eLSTM\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e\u003cp\u003e91.0\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e\u003cp\u003e90.0\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e\u003cp\u003e90.0\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e\u003cp\u003e0.90\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eHybrid Model\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e\u003cp\u003e94.5\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e\u003cp\u003e93.5\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e\u003cp\u003e95.2\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e\u003cp\u003e0.94\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003c/tbody\u003e\u003c/colgroup\u003e\u003c/table\u003e\u003c/div\u003e\u003c/p\u003e\u003cp\u003eThe hybrid model emerged as the most effective solution, showcasing superior performance across all metrics. These results highlight how DL, and hybrid designs in particular, have the potential to transform biodiversity monitoring and promote well-informed conservation efforts. In the fields of image and audio data processing, CNN and LSTM models each showed impressive performance.\u003c/p\u003e"},{"header":"Conclusion","content":"\u003cp\u003eTraditional ecological survey techniques are helpful, but they are often labor-intensive, time-consuming, and have a limited spatial and temporal scope. However, a powerful toolkit for modern conservation efforts is provided by the combination of AI-driven analytics, remote sensing data, and bioacoustic signals. This study demonstrates how DL technology, specifically CNN and LSTM networks, has the potential to revolutionize biodiversity and ecological monitoring. As ecosystems worldwide are threatened by anthropogenic pressures, habitat fragmentation, and climate change, scalable, accurate, and real-time monitoring systems are more crucial than ever.\u003c/p\u003e\u003cp\u003eUsing both spatial and temporal insights, CNN and LSTM models are employed to create a hybrid strategy that enhances performance on all assessed assessment metrics. CNN's efficiency in processing camera trap and satellite photography data is demonstrated by its capacity to extract spatial features, which allows it to classify habitats and identify species with high accuracy. Because LSTM networks can reliably detect species-specific vocalizations in a variety of ecological circumstances, they hold great promise for capturing temporal dynamics from audio recordings.\u003c/p\u003e\u003cp\u003eThe hybrid model demonstrates a high degree of versatility by combining visual and aural data streams, which makes it appropriate for deployment in a variety of contexts, from semi-urban green belts to thick tropical rainforests. This implies that multi-modal learning approaches are the best option because there isn't a single DL model that is consistently best for all ecological data sources. The results also show how important interdisciplinary cooperation is becoming. Ecologists, data scientists, software developers, and specialists in remote sensing must work together to do DL research in this field. Encouraging this kind of cooperation makes it possible to create monitoring systems that are both realistically feasible and scientifically sound.\u003c/p\u003e"},{"header":"Declarations","content":"\u003cp\u003e\u003cstrong\u003eFunding\u003c/strong\u003e: The British Academy/Cara/Leverhulme Researchers at Risk Research Support Grant. The University of Manchester\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eConflicts of interest/Competing interests\u003c/strong\u003e: The authors declare no conflicts of interest, real or perceived.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eAvailability of data and material\u003c/strong\u003e: The datasets used and/or analysed during the current study available from the corresponding author on reasonable request\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eCode availability\u003c/strong\u003e: Not applicable\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eAuthors\u0026apos; contributions\u003c/strong\u003e\u0026nbsp;\u003c/p\u003e\n\u003col\u003e\n \u003cli\u003eDilbar Hussain: Technical writing and development of scientific data analysis for the manuscript.\u003c/li\u003e\n \u003cli\u003eTurkia Almoustafa (Corresponding Author): Original idea, writing, and editing for scientific purposes.\u0026nbsp;\u003c/li\u003e\n \u003cli\u003eSaba Shakoor: General methodology and mathematical modelling.\u003c/li\u003e\n \u003cli\u003eFahiza Fauz: General methodology and mathematical modelling.\u003c/li\u003e\n \u003cli\u003eRizwan Ahmed: Technical writing and overall compilation for the manuscript.\u003c/li\u003e\n \u003cli\u003eShahzaib Ansari: Technical writing and editing process.\u003c/li\u003e\n\u003c/ol\u003e"},{"header":"References","content":"\u003col\u003e\n\u003cli\u003eS. Villon, C. Iovan, M. Mangeas, L. Vigliola, \u003cem\u003eSensors\u003c/em\u003e 2022, \u003cem\u003e22\u003c/em\u003e.\u003c/li\u003e\n\u003cli\u003eJ. Hoffmann, J. Muro, O. Dubovyk, \u003cem\u003eRemote Sens (Basel)\u003c/em\u003e 2022, \u003cem\u003e14\u003c/em\u003e.\u003c/li\u003e\n\u003cli\u003eE. Agrillo, F. Filipponi, A. Pezzarossa, L. Casella, D. Smiraglia, A. Orasi, F. Attorre, A. Taramelli, \u003cem\u003eRemote Sens (Basel)\u003c/em\u003e 2021, \u003cem\u003e13\u003c/em\u003e.\u003c/li\u003e\n\u003cli\u003eM. Lin, T. Lin, L. Jones, X. Liu, L. Xing, J. Sui, J. Zhang, H. Ye, Y. Liu, G. Zhang, X. Lu, \u003cem\u003eRemote Sens (Basel)\u003c/em\u003e 2021, \u003cem\u003e13\u003c/em\u003e.\u003c/li\u003e\n\u003cli\u003eK. N. Shivaprakash, N. Swami, S. Mysorekar, R. Arora, A. Gangadharan, K. Vohra, M. Jadeyegowda, J. M. Kiesecker, \u003cem\u003ePotential for Artificial Intelligence (AI) and Machine Learning (ML) Applications in Biodiversity Conservation, Managing Forests, and Related Services in India\u003c/em\u003e, MDPI, 2022.\u003c/li\u003e\n\u003cli\u003eJ. M\u0026uuml;ller, O. Mitesser, H. M. Schaefer, S. Seibold, A. Busse, P. Kriegel, D. Rabl, R. Gelis, A. Arteaga, J. Freile, G. A. Leite, T. N. de Melo, J. G. LeBien, M. Campos-Cerqueira, N. Bl\u0026uuml;thgen, C. J. Tremlett, D. B\u0026ouml;ttger, H. Feldhaar, N. Grella, A. Falcon\u0026iacute;-L\u0026oacute;pez, D. A. Donoso, J. Moriniere, Z. Buřivalov\u0026aacute;, \u003cem\u003eNat Commun\u003c/em\u003e 2023, \u003cem\u003e14\u003c/em\u003e.\u003c/li\u003e\n\u003cli\u003eP. Bodesheim, J. Blunk, M. K\u0026ouml;rschens, C. A. Brust, C. K\u0026auml;ding, J. Denzler, \u003cem\u003eMammalian Biology\u003c/em\u003e 2022, \u003cem\u003e102\u003c/em\u003e, 853.\u003c/li\u003e\n\u003cli\u003eS. S. Sethi, A. Bick, R. M. Ewers, H. Klinck, V. Ramesh, M. N. Tuanmu, D. A. Coomes, \u003cem\u003eNat Ecol Evol\u003c/em\u003e 2023, \u003cem\u003e7\u003c/em\u003e, 1373.\u003c/li\u003e\n\u003cli\u003eC. Zhang, H. Zhan, Z. Hao, X. Gao, \u003cem\u003eForests\u003c/em\u003e 2023, \u003cem\u003e14\u003c/em\u003e.\u003c/li\u003e\n\u003cli\u003eM. Bayat, H. Burkhart, M. Namiranian, S. K. Hamidi, S. Heidari, M. Hassani, \u003cem\u003eForests\u003c/em\u003e 2021, \u003cem\u003e12\u003c/em\u003e.\u003c/li\u003e\n\u003cli\u003eS. Subramaniam, N. Raju, A. Ganesan, N. Rajavel, M. Chenniappan, C. Prakash, A. Pramanik, A. K. Basak, S. Dixit, \u003cem\u003eArtificial Intelligence Technologies for Forecasting Air Pollution and Human Health: A Narrative Review\u003c/em\u003e, MDPI, 2022.\u003c/li\u003e\n\u003cli\u003eS. C. Izah, A. O. Iyiola, B. Yarkwan, G. Richard, in \u003cem\u003eVisualization Techniques for Climate Change with Machine Learning and Artificial Intelligence\u003c/em\u003e, Elsevier, 2022, pp. 123\u0026ndash;148.\u003c/li\u003e\n\u003cli\u003eD. A. Isabelle, M. Westerlund, \u003cem\u003eSustainability (Switzerland)\u003c/em\u003e 2022, \u003cem\u003e14\u003c/em\u003e.\u003c/li\u003e\n\u003cli\u003eB. Garske, A. Bau, F. Ekardt, \u003cem\u003eSustainability (Switzerland)\u003c/em\u003e 2021, \u003cem\u003e13\u003c/em\u003e.\u003c/li\u003e\n\u003cli\u003eEast-West University, Institute of Electrical and Electronics Engineers, Institute of Electrical and Electronics Engineers. Bangladesh Section, IEEE Robotics and Automation Society. Bangladesh Chapter, \u003cem\u003e2019 1st International Conference on Advances in Science, Engineering and Robotics Technology (ICASERT 2019) : May 3-5, 2019, Dhaka, Bangladesh.\u003c/em\u003e\u003c/li\u003e\n\u003cli\u003eB. Bochenek, Z. Ustrnul, \u003cem\u003eAtmosphere (Basel)\u003c/em\u003e 2022, \u003cem\u003e13\u003c/em\u003e.\u003c/li\u003e\n\u003cli\u003eC.-H. Yang, B. Feuer, Z. Jubery, Z. K. Deng, A. Nakkab, M. Z. Hasan, S. Chiranjeevi, K. Marshall, N. Baishnab, A. K. Singh, A. Singh, S. Sarkar, N. Merchant, C. Hegde, B. Ganapathysubramanian, 2024.\u003c/li\u003e\n\u003cli\u003eH. Han, Z. Liu, J. Li, Z. Zeng, \u003cem\u003eChallenges in remote sensing based climate and crop monitoring: navigating the complexities using AI\u003c/em\u003e, Springer Science and Business Media Deutschland GmbH, 2024.\u003c/li\u003e\n\u003cli\u003eN. R. Winter, M. Cearns, S. R. Clark, R. Leenings, U. Dannlowski, B. T. Baune, T. Hahn, \u003cem\u003eFrom multivariate methods to an AI ecosystem\u003c/em\u003e, Springer Nature, 2021.\u003c/li\u003e\n\u003cli\u003eL. Chen, Z. Chen, Y. Zhang, Y. Liu, A. I. Osman, M. Farghali, J. Hua, A. Al-Fatesh, I. Ihara, D. W. Rooney, P. S. Yap, \u003cem\u003eArtificial intelligence-based solutions for climate change: a review\u003c/em\u003e, Springer Science and Business Media Deutschland GmbH, 2023.\u003c/li\u003e\n\u003cli\u003eY. Yu, Y. Cao, D. Hou, M. Disse, A. Brieden, H. Zhang, R. Yu, \u003cem\u003eJournal of Geographical Sciences\u003c/em\u003e 2022, \u003cem\u003e32\u003c/em\u003e, 717.\u003c/li\u003e\n\u003cli\u003eC. Sravani, P. Ananth, C. Sasikala, N. S. Kumar, C. D. V. P. Kumari, V. Thrimurthulu, \u003cem\u003eRemote Sens Earth Syst Sci\u003c/em\u003e 2025.\u003c/li\u003e\n\u003cli\u003eJ. Cowls, A. Tsamados, M. Taddeo, L. Floridi, \u003cem\u003eAI Soc\u003c/em\u003e 2023, \u003cem\u003e38\u003c/em\u003e, 283.\u003c/li\u003e\n\u003cli\u003eM. Coeckelbergh, \u003cem\u003eAI and Ethics\u003c/em\u003e 2021, \u003cem\u003e1\u003c/em\u003e, 67.\u003c/li\u003e\n\u003cli\u003eA. van Wynsberghe, \u003cem\u003eAI and Ethics\u003c/em\u003e 2021, \u003cem\u003e1\u003c/em\u003e, 213.\u003c/li\u003e\n\u003cli\u003eP. S. Roy, R. M. Ramachandran, O. Paul, P. K. Thakur, S. Ravan, M. D. Behera, C. Sarangi, V. P. Kanawade, \u003cem\u003eAnthropogenic Land Use and Land Cover Changes\u0026mdash;A Review on Its Environmental Consequences and Climate Change\u003c/em\u003e, Springer, 2022.\u003c/li\u003e\n\u003cli\u003eY. Li, W. Zhang, C. R. Schwalm, P. Gentine, W. K. Smith, P. Ciais, J. S. Kimball, A. Gazol, S. A. Kannenberg, A. Chen, S. Piao, H. Liu, D. Chen, X. Wu, \u003cem\u003eNat Clim Chang\u003c/em\u003e 2023, \u003cem\u003e13\u003c/em\u003e, 182.\u003c/li\u003e\n\u003cli\u003eE. Martinez del Castillo, C. S. Zang, A. Buras, A. Hacket-Pain, J. Esper, R. Serrano-Notivoli, C. Hartl, R. Weigel, S. Klesse, V. Resco de Dios, T. Scharnweber, I. Dorado-Li\u0026ntilde;\u0026aacute;n, M. van der Maaten-Theunissen, E. van der Maaten, A. Jump, S. Mikac, B. E. Banzragch, W. Beck, L. Cavin, H. Claessens, V. Čada, K. Čufar, C. Dulamsuren, J. Gričar, E. Gil-Pelegr\u0026iacute;n, P. Janda, M. Kazimirovic, J. Kreyling, N. Latte, C. Leuschner, L. A. Longares, A. Menzel, M. Merela, R. Motta, L. Muffler, P. Nola, A. M. Petritan, I. C. Petritan, P. Prislan, \u0026Aacute;. Rubio-Cuadrado, M. Rydval, B. Stajić, M. Svoboda, E. Toromani, V. Trotsiuk, M. Wilmking, T. Zlatanov, M. de Luis, \u003cem\u003eCommun Biol\u003c/em\u003e 2022, \u003cem\u003e5\u003c/em\u003e.\u003c/li\u003e\n\u003cli\u003eS. E. Bibri, A. Alexandre, A. Sharifi, J. Krogstie, \u003cem\u003eEnvironmentally sustainable smart cities and their converging AI, IoT, and big data technologies and solutions: an integrated approach to an extensive literature review\u003c/em\u003e, Springer Nature, 2023.\u003c/li\u003e\n\u003cli\u003eA. O. Almagrabi, \u003cem\u003eComputers, Materials and Continua\u003c/em\u003e 2023, \u003cem\u003e76\u003c/em\u003e, 2079.\u003c/li\u003e\n\u003cli\u003eK. O. Adefemi, M. B. Mutanga, \u003cem\u003eDigital\u003c/em\u003e 2025, \u003cem\u003e5\u003c/em\u003e, 16.\u003c/li\u003e\n\u003c/ol\u003e"}],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":true,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":true,"hideJournal":true,"highlight":"","institution":"","isAcceptedByJournal":false,"isAuthorSuppliedPdf":false,"isDeskRejected":"","isHiddenFromSearch":false,"isInQc":false,"isInWorkflow":false,"isPdf":false,"isPdfUpToDate":true,"isWithdrawnOrRetracted":false,"journal":{"display":true,"email":"[email protected]","identity":"researchsquare","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":true,"externalIdentity":"","sideBox":"","snPcode":"","submissionUrl":"/submission","title":"Research Square","twitterHandle":"researchsquare","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"","reportingPortfolio":"","inReviewEnabled":false,"inReviewRevisionsEnabled":true},"keywords":"Biodiversity Monitoring, Remote Sensing, Ecosystem monitoring, Multi-modal","lastPublishedDoi":"10.21203/rs.3.rs-7478081/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-7478081/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003cp\u003eThe growing impacts of climate change, habitat degradation, and human intervention have intensified the urgency for effective biodiversity and ecosystem monitoring. Traditional methods such as field surveys and manual species identification, though valuable, are increasingly challenged by limitations in scale, frequency, and accuracy. This study addresses the need for scalable, automated solutions by leveraging Deep Learning(DL) models to analyze multi-modal ecological data. The research specifically applies Convolutional Neural Networks(CNNs) for spatial analysis of high-resolution satellite imagery (Sentinel-2 and Landsat) and camera trap datasets (Snapshot Serengeti), and Long Short-Term Memory (LSTM) networks for processing temporal bioacoustic recordings (from Xeno-canto and Rainforest Connection). The methodology involves structured data preprocessing, model training, evaluation using standard metrics, and a final integration into a hybrid framework combining CNN and LSTM outputs. The hybrid model outperformed individual networks with 94.5% accuracy. This research confirms that deep learning, hybrid architectures offer a powerful solution for biodiversity monitoring.\u003c/p\u003e","manuscriptTitle":"Biodiversity and Ecosystem Monitoring using deep learning","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2025-09-09 18:00:25","doi":"10.21203/rs.3.rs-7478081/v1","editorialEvents":[{"type":"communityComments","content":0}],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"researchsquare","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":true,"externalIdentity":"","sideBox":"","snPcode":"","submissionUrl":"/submission","title":"Research Square","twitterHandle":"researchsquare","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"","reportingPortfolio":"","inReviewEnabled":false,"inReviewRevisionsEnabled":true}}],"origin":"","ownerIdentity":"65ab7ea4-7994-45f3-b17b-11934c70412d","owner":[],"postedDate":"September 9th, 2025","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"posted","subjectAreas":[],"tags":[],"updatedAt":"2026-03-26T16:36:45+00:00","versionOfRecord":[],"versionCreatedAt":"2025-09-09 18:00:25","video":"","vorDoi":"","vorDoiUrl":"","workflowStages":[]},"version":"v1","identity":"rs-7478081","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-7478081","identity":"rs-7478081","version":["v1"]},"buildId":"8U1c8b4HqxoKbykW_rLl7","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}