Savvas Karatsiolis
Research Subjects
Savvas Karatsiolis has a total of 20 co-authors. Here are the top co-authors by number of shared publications:
- Andreas Kamilaris
Cyens Centre of Excellence, Dimarchou Lellou Dimitriadi 23, Nicosia 1016, Cyprus
11 shared publications - Chirag Padubidri
CYENS Center of Excellence Nicosia Cyprus
7 shared publications - Christos N. Schizas
University of Cyprus, Nicosia, Cyprus
4 shared publications Andreas Kamilaris
3 shared publicationsNicolai Petkov
Department of Intelligent Systems Group, Bernoulli Institute for Mathematics, Computer Science and Artificial Intelligence, University of Groningen, Groningen, The Netherlands
2 shared publications- Panos Hadjinicolaou
Climate and Atmosphere Research Center (CARE-C), The Cyprus Institute, Nicosia, Cyprus
2 shared publications
Latest Publications
- 15 September 2026
- Published by Springer Nature
- p. 321-332
Spectral Feature Refinement: Enhancing Robustness to Image Corruptions
- 15 September 2026
- Published by Springer Nature
- p. 321-332
Spectral Feature Refinement: Enhancing Robustness to Image Corruptions
- Open Access
- 6 September 2026
- Published by Elsevier in Urban Climate
- Vol. 69
Decomposing urban heat islands at the micro-scale: A hybrid physics-inspired and data-driven modelling framework
- 6 September 2026
- Published by Elsevier in Urban Climate
- Vol. 69
Open AccessDecomposing urban heat islands at the micro-scale: A hybrid physics-inspired and data-driven modelling framework
- Open Access
- 20 May 2026
- Published by Elsevier in SSRN Electronic Journal
Decomposing Urban Heat Islands at Micro-Level: A Hybrid Physics-Inspired and Data-Driven Modelling Framework
We propose a hybrid physics-inspired, data-driven framework that decomposes the canopy-layer urban-rural air temperature difference into a location-specific Urban Heat Island potential (UHIp) and a weather-driven modulation (UHIm), with a learnable vehicle-induced convective bias correction. The framework is applied to Nicosia, Cyprus - a compact, hot semi-arid Mediterranean city - using 700,000 measurements acquired during August - September 2025 from custom sensors on the Municipality of Nicosia's operational fleet, referenced to the Athalassa meteorological station. Learned embeddings over 863 grid cells of 200×200 m implicitly encode local morphology, while a 92-hour antecedent meteorological record drives UHIm estimation. Validation against six independent stations yields a MAE of 0.25_1.09 ◦C. The model recovers a physically coherent diurnal reversal: a morning cool island regime (mean ΔT of −0.7 to −1.3◦C) transitions to a city-wide heat island in the afternoon (+2.4 to +3.1◦C) that persists nocturnally (+1.4 to +1.6◦C). UHIp correlates with surface and terrain characteristics, confirming physical interpretability. Local Climate Zone stratification reveals the highest UHIp in compact low-rise zones and the lowest in open low-rise, with realised urban-rural differences reaching 3-4◦C mid-afternoon. Crucially, during the 42◦C+ heatwave of 9-11 August, UHIm approaches zero, indicating that extreme atmospheric overheating suppresses the urban-rural differential precisely when thermal stress peaks. Trained on opportunistic municipal fleet data, this physics-inspired decomposition yields interpretable, high-resolution micro-scale UHI characterisation, directly relevant to urban planning and climate-resilience strategies in Mediterranean cities.
- 20 May 2026
- Published by Elsevier in SSRN Electronic Journal
Open AccessDecomposing Urban Heat Islands at Micro-Level: A Hybrid Physics-Inspired and Data-Driven Modelling Framework
We propose a hybrid physics-inspired, data-driven framework that decomposes the canopy-layer urban-rural air temperature difference into a location-specific Urban Heat Island potential (UHIp) and a weather-driven modulation (UHIm), with a learnable vehicle-induced convective bias correction. The framework is applied to Nicosia, Cyprus - a compact, hot semi-arid Mediterranean city - using 700,000 measurements acquired during August - September 2025 from custom sensors on the Municipality of Nicosia's operational fleet, referenced to the Athalassa meteorological station. Learned embeddings over 863 grid cells of 200×200 m implicitly encode local morphology, while a 92-hour antecedent meteorological record drives UHIm estimation. Validation against six independent stations yields a MAE of 0.25_1.09 ◦C. The model recovers a physically coherent diurnal reversal: a morning cool island regime (mean ΔT of −0.7 to −1.3◦C) transitions to a city-wide heat island in the afternoon (+2.4 to +3.1◦C) that persists nocturnally (+1.4 to +1.6◦C). UHIp correlates with surface and terrain characteristics, confirming physical interpretability. Local Climate Zone stratification reveals the highest UHIp in compact low-rise zones and the lowest in open low-rise, with realised urban-rural differences reaching 3-4◦C mid-afternoon. Crucially, during the 42◦C+ heatwave of 9-11 August, UHIm approaches zero, indicating that extreme atmospheric overheating suppresses the urban-rural differential precisely when thermal stress peaks. Trained on opportunistic municipal fleet data, this physics-inspired decomposition yields interpretable, high-resolution micro-scale UHI characterisation, directly relevant to urban planning and climate-resilience strategies in Mediterranean cities.
- Open Access
Poultry Farm Intelligence: An Integrated Multi-Sensor AI Platform for Enhanced Welfare and Productivity
Poultry farming faces increasing pressure to meet productivity targets while ensuring animal welfare and environmental compliance. Yet many small and medium-sized farms lack affordable, integrated tools for continuous monitoring and decision-making, relying instead on manual, reactive inspections. This paper presents Poultry Farm Intelligence (PoultryFI) - a modular, cost-effective platform that integrates six AI-powered modules: Camera Placement Optimizer, Audio-Visual Monitoring, Analytics & Alerting, Real-Time Egg Counting, Production & Profitability Forecasting, and a Recommendation Module. Camera layouts are first optimized offline using evolutionary algorithms for full poultry house coverage with minimal hardware. The Audio-Visual Monitoring module extracts welfare indicators from synchronized video, audio, and feeding data. Analytics & Alerting produces daily summaries and real-time notifications, while Real-Time Egg Counting uses an edge vision model to automate production tracking. Forecasting models predict egg yield and feed consumption up to 10 days in advance, and the Recommendation Module integrates forecasts with weather data to guide environmental and operational adjustments. This is among the first systems to combine low-cost sensing, edge analytics, and prescriptive AI to continuously monitor flocks, predict production, and optimize performance. Field trials demonstrate 100% egg-count accuracy on Raspberry Pi 5, robust anomaly detection, and reliable short-term forecasting. PoultryFI bridges the gap between isolated pilot tools and scalable, farm-wide intelligence, empowering producers to proactively safeguard welfare and profitability.
Open AccessPoultry Farm Intelligence: An Integrated Multi-Sensor AI Platform for Enhanced Welfare and Productivity
Poultry farming faces increasing pressure to meet productivity targets while ensuring animal welfare and environmental compliance. Yet many small and medium-sized farms lack affordable, integrated tools for continuous monitoring and decision-making, relying instead on manual, reactive inspections. This paper presents Poultry Farm Intelligence (PoultryFI) - a modular, cost-effective platform that integrates six AI-powered modules: Camera Placement Optimizer, Audio-Visual Monitoring, Analytics & Alerting, Real-Time Egg Counting, Production & Profitability Forecasting, and a Recommendation Module. Camera layouts are first optimized offline using evolutionary algorithms for full poultry house coverage with minimal hardware. The Audio-Visual Monitoring module extracts welfare indicators from synchronized video, audio, and feeding data. Analytics & Alerting produces daily summaries and real-time notifications, while Real-Time Egg Counting uses an edge vision model to automate production tracking. Forecasting models predict egg yield and feed consumption up to 10 days in advance, and the Recommendation Module integrates forecasts with weather data to guide environmental and operational adjustments. This is among the first systems to combine low-cost sensing, edge analytics, and prescriptive AI to continuously monitor flocks, predict production, and optimize performance. Field trials demonstrate 100% egg-count accuracy on Raspberry Pi 5, robust anomaly detection, and reliable short-term forecasting. PoultryFI bridges the gap between isolated pilot tools and scalable, farm-wide intelligence, empowering producers to proactively safeguard welfare and profitability.
- Open Access
Domain Adaptation via Feature Refinement
We propose Domain Adaptation via Feature Refinement (DAFR2), a simple yet effective framework for unsupervised domain adaptation under distribution shift. The proposed method synergistically combines three key components: adaptation of Batch Normalization statistics using unlabeled target data, feature distillation from a source-trained model and hypothesis transfer. By aligning feature distributions at the statistical and representational levels, DAFR2 produces robust and domain-invariant feature spaces that generalize across similar domains without requiring target labels, complex architectures or sophisticated training objectives. Extensive experiments on benchmark datasets, including CIFAR10-C, CIFAR100-C, MNIST-C and PatchCamelyon-C, demonstrate that the proposed algorithm outperforms prior methods in robustness to corruption. Theoretical and empirical analyses further reveal that our method achieves improved feature alignment, increased mutual information between the domains and reduced sensitivity to input perturbations.
Open AccessDomain Adaptation via Feature Refinement
We propose Domain Adaptation via Feature Refinement (DAFR2), a simple yet effective framework for unsupervised domain adaptation under distribution shift. The proposed method synergistically combines three key components: adaptation of Batch Normalization statistics using unlabeled target data, feature distillation from a source-trained model and hypothesis transfer. By aligning feature distributions at the statistical and representational levels, DAFR2 produces robust and domain-invariant feature spaces that generalize across similar domains without requiring target labels, complex architectures or sophisticated training objectives. Extensive experiments on benchmark datasets, including CIFAR10-C, CIFAR100-C, MNIST-C and PatchCamelyon-C, demonstrate that the proposed algorithm outperforms prior methods in robustness to corruption. Theoretical and empirical analyses further reveal that our method achieves improved feature alignment, increased mutual information between the domains and reduced sensitivity to input perturbations.
- Open Access
- 9 December 2024
- Published by MDPI AG in Remote Sensing
- Vol. 16 (23) , 4611
A Weakly Supervised Multimodal Deep Learning Approach for Large-Scale Tree Classification: A Case Study in Cyprus
Forest ecosystems play an essential role in ecological balance, supporting biodiversity and climate change mitigation. These ecosystems are crucial not only for ecological stability but also for the local economy. Performing a tree census at a country scale via traditional methods is resource-demanding, error-prone, and requires significant effort by a large number of experts. While emerging technologies such as satellite imagery and AI provide the means for achieving promising results in this task with less effort, considerable effort is still required by experts to annotate hundreds or thousands of images. This study introduces a novel methodology for a tree census classification system which leverages historical and partially labeled data, employs probabilistic data imputation and a weakly supervised training technique, and thus achieves state-of-the-art precision in classifying the dominant tree species of Cyprus. A crucial component of our methodology is a ResNet50 model which takes as input high spatial resolution satellite imagery in the visible band and near-infrared band, as well as topographical features. By applying a multimodal training approach, a classification accuracy of 90% among nine targeted tree species is achieved.
- 9 December 2024
- Published by MDPI AG in Remote Sensing
- Vol. 16 (23) , 4611
Open AccessA Weakly Supervised Multimodal Deep Learning Approach for Large-Scale Tree Classification: A Case Study in Cyprus
Forest ecosystems play an essential role in ecological balance, supporting biodiversity and climate change mitigation. These ecosystems are crucial not only for ecological stability but also for the local economy. Performing a tree census at a country scale via traditional methods is resource-demanding, error-prone, and requires significant effort by a large number of experts. While emerging technologies such as satellite imagery and AI provide the means for achieving promising results in this task with less effort, considerable effort is still required by experts to annotate hundreds or thousands of images. This study introduces a novel methodology for a tree census classification system which leverages historical and partially labeled data, employs probabilistic data imputation and a weakly supervised training technique, and thus achieves state-of-the-art precision in classifying the dominant tree species of Cyprus. A crucial component of our methodology is a ResNet50 model which takes as input high spatial resolution satellite imagery in the visible band and near-infrared band, as well as topographical features. By applying a multimodal training approach, a classification accuracy of 90% among nine targeted tree species is achieved.
- Open Access
- 28 December 2023
- Published by MDPI AG in Remote Sensing
- Vol. 16 (1) , 142
Scalable Retrieval of Similar Landscapes in Optical Satellite Imagery Using Unsupervised Representation Learning
Global Earth observation is becoming increasingly important in understanding and addressing critical aspects of life on our planet, including environmental issues, natural disasters, sustainable development, and others. Finding similarities in landscapes may provide useful information regarding applying contiguous policies, by making similar decisions or learning from best practices for events and occurrences that previously occurred in similar landscapes in the past. However, current applications of similar landscape retrieval are limited by a moderate performance and the need for time-consuming and costly annotations. We propose splitting the similar landscape retrieval task into a set of smaller tasks that aim to identify individual concepts inherent to satellite images. Our approach relies on several models trained using unsupervised representation learning on Google Earth images to identify these concepts. We show the efficacy of matching individual concepts for retrieving landscape(s) similar to a user-selected satellite image of the geographical territory of the Republic of Cyprus. Our results demonstrate the benefits of breaking up the landscape similarity task into individual concepts closely related to remote sensing, instead of applying a single model targeting all underlying concepts.
- 28 December 2023
- Published by MDPI AG in Remote Sensing
- Vol. 16 (1) , 142
Open AccessScalable Retrieval of Similar Landscapes in Optical Satellite Imagery Using Unsupervised Representation Learning
Global Earth observation is becoming increasingly important in understanding and addressing critical aspects of life on our planet, including environmental issues, natural disasters, sustainable development, and others. Finding similarities in landscapes may provide useful information regarding applying contiguous policies, by making similar decisions or learning from best practices for events and occurrences that previously occurred in similar landscapes in the past. However, current applications of similar landscape retrieval are limited by a moderate performance and the need for time-consuming and costly annotations. We propose splitting the similar landscape retrieval task into a set of smaller tasks that aim to identify individual concepts inherent to satellite images. Our approach relies on several models trained using unsupervised representation learning on Google Earth images to identify these concepts. We show the efficacy of matching individual concepts for retrieving landscape(s) similar to a user-selected satellite image of the geographical territory of the Republic of Cyprus. Our results demonstrate the benefits of breaking up the landscape similarity task into individual concepts closely related to remote sensing, instead of applying a single model targeting all underlying concepts.
- Open Access
- 25 October 2023
- Published by MDPI AG in Preprints.org
Scalable Retrieval of Similar Landscapes in Optical Satellite Imagery Using Unsupervised Representation Learning
Global Earth Observation (EO) is becoming increasingly important in understanding and addressing critical aspects of life on our planet about environmental issues, natural disasters, sustainable development and others. EO plays a key role in making informed decisions on applying or reforming land use, responding to disasters, shaping climate adaptation policies etc. EO is also becoming a useful tool for helping professionals make the most profitable decisions, e.g., in real estate or the investment sector. Finding similarities in landscapes may provide useful information regarding applying contiguous policies, taking alike decisions or learning from best practices on events and happenings that have already occurred in similar landscapes in the past. However, current applications of similar landscape retrieval are limited by moderate performance and the need for time-consuming and costly annotations. We propose splitting the similar landscape retrieval task into a set of smaller tasks that aim at identifying individual concepts inherent to satellite images. Our approach relies on several models trained with Unsupervised Representation Learning (URL) on Google Earth images to identify these concepts. We show the efficacy of matching individual concepts for tackling the task of retrieving similar landscape(s) to a user-selected satellite image with a proof-of-concept application of the proposed approach on the geographical territory of the Republic of Cyprus. Our results demonstrate the efficacy of breaking up the landscape similarity task into individual concepts closely related to remote sensing instead of trying to capture all concepts and image semantics with a single model like a single RGB semantics model.
- 25 October 2023
- Published by MDPI AG in Preprints.org
Open AccessScalable Retrieval of Similar Landscapes in Optical Satellite Imagery Using Unsupervised Representation Learning
Global Earth Observation (EO) is becoming increasingly important in understanding and addressing critical aspects of life on our planet about environmental issues, natural disasters, sustainable development and others. EO plays a key role in making informed decisions on applying or reforming land use, responding to disasters, shaping climate adaptation policies etc. EO is also becoming a useful tool for helping professionals make the most profitable decisions, e.g., in real estate or the investment sector. Finding similarities in landscapes may provide useful information regarding applying contiguous policies, taking alike decisions or learning from best practices on events and happenings that have already occurred in similar landscapes in the past. However, current applications of similar landscape retrieval are limited by moderate performance and the need for time-consuming and costly annotations. We propose splitting the similar landscape retrieval task into a set of smaller tasks that aim at identifying individual concepts inherent to satellite images. Our approach relies on several models trained with Unsupervised Representation Learning (URL) on Google Earth images to identify these concepts. We show the efficacy of matching individual concepts for tackling the task of retrieving similar landscape(s) to a user-selected satellite image with a proof-of-concept application of the proposed approach on the geographical territory of the Republic of Cyprus. Our results demonstrate the efficacy of breaking up the landscape similarity task into individual concepts closely related to remote sensing instead of trying to capture all concepts and image semantics with a single model like a single RGB semantics model.
- Open Access
- 18 January 2023
- Published by Elsevier in Fire Safety Journal
- Vol. 136
Examining the potential of mobile applications to assist people to escape wildfires in real-time
- 18 January 2023
- Published by Elsevier in Fire Safety Journal
- Vol. 136
Open AccessExamining the potential of mobile applications to assist people to escape wildfires in real-time
- Open Access
- 26 September 2022
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Geoscience and Remote Sensing
- Vol. 60 (01962892) , 1-13
Exploiting Digital Surface Models for Inferring Super-Resolution for Remotely Sensed Images
Despite the plethora of successful super-resolution (SR) reconstruction (SRR) models applied to natural images, their application to remote sensing imagery tends to produce poor results. Remote sensing imagery is often more complicated than natural images, has its peculiarities such as being of lower resolution, contains noise, and often depicts large textured surfaces. As a result, applying nonspecialized SRR models like the enhanced SR generative adversarial network (ESRGAN) on remote sensing imagery results in artifacts and poor reconstructions. To address these problems, we propose a novel strategy for enabling an SRR model to output realistic remote sensing images: instead of relying on feature-space similarities as a perceptual loss, the model considers pixel-level information inferred from the normalized digital surface model (nDSM) of the image. This allows the application of better-informed updates during the training of the model which sources from a task (elevation map inference) that is closely related to remote sensing. Nonetheless, the nDSM auxiliary information is not required during production, i.e., the model infers an SR image without additional data. We assess our model on two remotely sensed datasets of different spatial resolutions that also contain the DSMs of the images: the Data Fusion 2018 Contest (DFC2018) dataset and the dataset containing the national LiDAR flyby of Luxembourg. We compare our model with ESRGAN, and we show that it achieves better performance and does not introduce any artifacts in the results. In particular, the results for the high-resolution DFC2018 dataset are realistic and almost indistinguishable from the ground-truth images.
- 26 September 2022
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Geoscience and Remote Sensing
- Vol. 60 (01962892) , 1-13
Open AccessExploiting Digital Surface Models for Inferring Super-Resolution for Remotely Sensed Images
Despite the plethora of successful super-resolution (SR) reconstruction (SRR) models applied to natural images, their application to remote sensing imagery tends to produce poor results. Remote sensing imagery is often more complicated than natural images, has its peculiarities such as being of lower resolution, contains noise, and often depicts large textured surfaces. As a result, applying nonspecialized SRR models like the enhanced SR generative adversarial network (ESRGAN) on remote sensing imagery results in artifacts and poor reconstructions. To address these problems, we propose a novel strategy for enabling an SRR model to output realistic remote sensing images: instead of relying on feature-space similarities as a perceptual loss, the model considers pixel-level information inferred from the normalized digital surface model (nDSM) of the image. This allows the application of better-informed updates during the training of the model which sources from a task (elevation map inference) that is closely related to remote sensing. Nonetheless, the nDSM auxiliary information is not required during production, i.e., the model infers an SR image without additional data. We assess our model on two remotely sensed datasets of different spatial resolutions that also contain the DSMs of the images: the Data Fusion 2018 Contest (DFC2018) dataset and the dataset containing the national LiDAR flyby of Luxembourg. We compare our model with ESRGAN, and we show that it achieves better performance and does not introduce any artifacts in the results. In particular, the results for the high-resolution DFC2018 dataset are realistic and almost indistinguishable from the ground-truth images.