Land cover (LC) classification remains challenging, even with advanced machine learning (ML) techniques. Pixel-wise convolutional neural networks (CNNs) are currently widely used for image classification. However, CNNs outputs often suffer from blurry boundaries and poorly segmented shapes. Hence, ad-hoc postprocessing may be required to reduce oversmoothing. Conversely, it adds complexity to the classification process. Fortunately, generic foundation models like the segment anything model (SAM), trained on large datasets, have become available. In this study, we propose a novel approach that combines outputs from locally trained CNN models and from the SAM model. We tested the approach on a mosaic of Sentinel-2 (S-2) images with a 10 m resolution, covering four neighboring first-level administrative units in Czechia. S-2 mosaic was classified according to the land use, land-use change, and forestry (LULUCF) nomenclature using CNNs. Separately, this mosaic was segmented by SAM. The SAM output segments were assigned the dominant LC class from the CNN classification. The annotated segments were then overlaid on the original CNN classification to address oversmoothing. This approach resulted in an approximate 2% improvement in overall accuracy (OA) and significantly enhanced the boundary delineation of LC objects in classification outputs, in terms of F1, precision, and recall. The reported 2% improvement is meaningful because it was achieved on top of an already strong baseline (86.87% overall accuracy), making further gains inherently difficult. LULUCF information is critical for monitoring compliance with the Paris Agreement. So, the ability to segment S-2 images into objects that align with LULUCF nomenclature could support future annual LULUCF reporting to the United Nations Framework Convention on Climate Change.

Bark beetle outbreaks pose a severe threat to spruce forests, with the European spruce bark beetle (Ips typographus L.) being the dominant pest in the Czech Republic. Although multispectral imagery using visible (VIS) and near-infrared (NIR) wavelengths has been employed for early detection, it primarily captures visible infestation symptoms rather than the underlying physiological stress, which can, however, be detected using thermal measurements, highlighting the benefit of integrating or complementing multispectral with thermal imagery. We compared a time series of Unmanned Aerial Vehicle (UAV) based thermal and multispectral imagery over a 650—ha coniferous stand in Central Bohemia, acquired at key phases of infestation (a) a year before infestation (August 2020); (b) closely before bark beetle infestation (April 2021); (c) in the initial phase of the green-attack (May 2021); and (d) in green-attack stage, early detection (June 2021), based on the species phenological model and field survey. Comparisons of canopy temperature and Normalised Difference Vegetation Index (NDVI) showed that thermal imagery successfully discriminated between healthy and infested trees seven weeks after beetle attack (green-attack), whereas NDVI differences remained negligible. These results confirm that UAV thermal imaging outperforms multispectral data for the early, individual-tree detection of bark beetle infestation.

Understanding how changes in vegetation cover affect nutrient dynamics is essential for predicting future ecosystem responses to environmental change. In this study, we integrated ground-based observations, remote sensing data, and dynamic process-based modelling to investigate vegetation dynamics under changing environmental conditions and disturbances driven by both natural processes and human activities. Our primary objective was to assess how land-use change, disturbances, and management practices influence nutrient cycling in plant ecosystems.To address this objective, we compiled detailed land-use and land-cover data for 270 currently forested sites across five European countries (Croatia, Hungary, Slovakia, Czech Republic, and Poland). Data sources spanned from the second half of the 18th century to the present and included historical military maps, aerial photographs, satellite imagery, and forest management plans. The analysis revealed that 84% of sites experienced a vegetation change, while one third of them underwent multiple types of change. Management interventions were the most common driver occurring on more than a half of sites, followed by shifts in tree species composition at over one third of sites and deforestation observed at one quarter of sites. Natural disturbances were identified only at one fifth of sites.Subsequently, we simulated vegetation dynamics using the process-based model Biome-BGCMuSo. Each site was modelled under two simulation set-ups: one considering only the current ecosystem state, and another incorporating the documented vegetation changes over the past 200 years. This design enabled us to isolate the effects of historical vegetation dynamics on ecosystem stocks and fluxes. In total, 40 variables related to carbon, nitrogen, and water cycling were analysed. The magnitudes of differences between the two set-ups varied among ecosystem components, sites, and species, and were strongly linked to the type and frequency of vegetation changes. The most pronounced negative effects, reaching up to 50% difference, were observed in soil, litter, and coarse woody debris carbon and nitrogen stocks, as well as in net ecosystem exchange and heterotrophic respiration following deforestation.Our results highlight the critical importance of accounting for historical vegetation changes in ecosystem modelling. By demonstrating how legacy effects shape present-day nutrient dynamics, this study underscores that ecosystem functioning reflects not only current conditions but also the cumulative influence of past land use, management, and disturbance history.The study was funded by the EU NextGenerationEU through the Recovery and Resilience Plan for Slovakia under the project No. 09I03-03-V04- 00130.

This paper focuses on an innovative application of deep learning (DL) techniques, particularly 3D convolutional neural networks (CNNs), for land cover classification using multispectral Sentinel-2 (S-2) data. In this study, we evaluated the performance of window-pixel-wise 2D, 3D, and 3D Multiscale CNN architectures for land cover classification. 3D and 3D multiscale CNNs were using the spectral dimension as the third dimension for convolutions. Methodology was applied to classify large area (23,217 km2) in Czechia according to the Land use, land-use change and forestry (LULUCF) categories, a key sector in greenhouse gas inventories. The input dataset included S-2 data, along with NDVI, NDVI variance, and SRTM elevation data, all resampled to the 10 m S-2 grid and forming multi-dimensional input 5 x 5 pixel patches. The results show that a 3D CNN with 3 x 3 x 3 spatial-spectral filters and classical training achieved the best F1 score of 0.84, outperforming other proposed CNN architectures and a baseline Random Forest classifier. The study highlights the ability of 3D CNNs to integrate spatial-spectral information, making them highly effective for multispectral data analysis, even with limited (small) training ground truth datasets. This approach provides valuable information for researchers seeking to optimize DL methods for land cover classification, particularly for applications aligned with the LULUCF frameworks.

This study assesses the accuracy of ten satellite-based and reanalysis precipitation datasets available in Google Earth Engine (GEE) using in-situ rain gauge measurements across Czechia, Central Europe, from 2001 to 2021. The gauge-adjusted GSMaP dataset (GSMaP(GA)) was the most accurate dataset overall (Pearson's correlation coefficient r = 0.79), followed by ERA5-Land (r = 0.75), with both showing superior performance for rainy days above 1 mm of precipitation. In contrast, CHIRPS, GLDAS, and PERSIANN-CDR showed the weakest performance (r approximate to 0.41-0.42). All datasets overestimated precipitation on days with no or with very light rain (<= 1 mm/day) and underestimated it during heavy rainfall events ( >5 mm/day). ERA5-Land systematically overestimated annual precipitation by 15-35%, while GSMaP(GA) showed slight underestimation by 0.5-9%. Although absolute errors generally increased with elevation, GSMaP(GA) showed the smallest elevation-related biases, highlighting the importance for gauge-adjustment. Part of the observed spatial and seasonal biases may be explained by the combination of coarse spatial resolution and the challenges of capturing short-lived summer convective storms over complex terrain. Overall, GSMaP(GA) is recommended for most applications due to its superior accuracy, while ERA5-Land is suitable for long-term studies because of its long historical record extending back to the 1950s.

Current vegetation indices and biophysical parameters derived from optical satellite data for forest monitoring are widely used in various applications but can be limited by atmospheric effects like clouds. Synthetic aperture radar (SAR) data can offer insightful and systematic forest monitoring with complete time series due to signal penetration through clouds and day and night image acquisitions. This study explores the use of SAR data, combined with ancillary data and machine learning (ML), to estimate forest parameters typically derived from optical satellites. It investigates whether SAR signals provide sufficient information for the accurate estimation of these parameters, focusing on two spectral vegetation indices (Normalized Difference Vegetation Index - NDVI and Enhanced Vegetation Index - EVI) and two biophysical parameters (Leaf Area Index - LAI and Fraction of Absorbed Photosynthetically Active Radiation - FAPAR) in healthy and disturbed temperate forests in Czechia and Central Europe in 2021. Vegetation metrics derived from Sentinel-2 multispectral data were used to evaluate the results. A paired multi-modal time-series dataset was created using Google Earth Engine (GEE), including temporally and spatially aligned Sentinel-1, Sentinel-2, DEM-based features and meteorological variables, along with a forest type class. The inclusion of DEM-based auxiliary features and additional meteorological information improved the results. In the comparison of ML models, the traditional ML algorithms, Random Forest Regressor and Extreme Gradient Boosting (XGB) slightly outperformed the Automatic Machine Learning (AutoML) approach, auto-sklearn, for all forest parameters, achieving high accuracies (R2 between 70% and 86%) and low errors (0.055-0.29 of mean absolute error). XGB was the most computationally efficient. Moreover, SAR-based estimations over Central Europe achieved comparable results to those obtained in testing within Czechia, demonstrating their transferability for large-scale modeling. A key advantage of the SAR-based vegetation metrics is the ability to detect abrupt forest changes with sub-weekly temporal accuracy, providing up to 240 measurements per year at a 20 m resolution.

Time series analysis of synthetic aperture radar data (SAR) offers a systematic, dynamic and comprehensive way to monitor forests. The main emphasis of this study is on the identification of the most suitable and best performing Sentinel-1 SAR polarimetric parameters for forest monitoring. This is accomplished through: 1) a pairwise correlation analysis of SAR polarimetric parameters, multispectral optical vegetation indices and ancillary data, 2) a univariate binary time series classification for differentiation between forest types and 3) a visual exploration of time series. For this purpose, 600 validated broad-leaved and 600 coniferous forest areas in Czechia were used. Nine different SAR polarimetric parameters were examined, including VH and VV polarizations, VV/VH and VH/VV polarization ratios, the Radar Vegetation Index, Radar Forest Degradation Index, polarimetric radar vegetation index and the original and modified versions of the dual polarimetric SAR vegetation index. The pairwise correlation analysis revealed that most of the derived SAR polarimetric parameters were functions of each other with nearly identical behavior (r > |0.96|). The strongest correlation of r ~0.50 between SAR and optical features was found for broad-leaved forest for VV/VH and VH/VV. The highest overall accuracy in the time series classification of forest types was achieved by VH (76%), while for VV, VV/VH and VH/VV it was higher than 60%. Furthermore, the time series analysis of these parameters showed seasonal behaviors of the SAR features in both forest types. These results demonstrated the high relevance of using VH, VV, VV/VH and VH/VV time series in forest monitoring compared to other SAR polarimetric parameters. This study also introduces a novel pipeline to generate multi-modal time series datasets in Google Earth Engine (MMTS-GEE), used to generate data for the analysis. MMTS-GEE combines spatially and temporally aligned SAR and multispectral data, extended with topographic and weather data, and a land cover class label. Its high versatility enables its use in time series analyses, intercomparisons and in machine learning applications for tabular time series data. The GEE code for the proposed tool and analysis is freely available to the research community.

This study examines Land Use and Land Cover (LULC) changes over a 30-year period (1994-2024) in the Ahafo Ano Southwest District of Ghana using Landsat imagery from 1994, 2004, 2014, and 2024. Supervised classification based on the Maximum Likelihood algorithm was applied to classify the images into five LULC categories: dense vegetation, sparse vegetation, wetland, bare land, and built-up area. The classification results achieved overall accuracies above 95% with Kappa coefficients of approximately 0.92. The results indicate a substantial expansion of built-up areas, sparse vegetation, and bare land, occurring at the expense of dense vegetation and wetlands. These changes are primarily associated with increasing human activities, particularly agricultural expansion and urbanization, which corresponds with population growth in the study area. The findings provide important insights for land-use planning and the formulation of sustainable development policies aimed at enhancing forest cover, conserving ecosystems, and mitigating carbon stock loss.

The focus of this paper is to present an innovative approach to classify land cover, with specific emphasis on Land Use, Land-Use Change and Forestry (LULUCF) monitoring. LULUCF is a sector in greenhouse gas inventory that tracks changes in greenhouse gas levels in the atmosphere due to land use and land-use change. In this study, we employed Deep Learning classifiers and Random Forest to classify land cover/land use in Czechia, adhering to LULUCF regulations. We evaluated the effectiveness of 2D and 3D Convolutions in Convolutional Neural Networks, with varying filter sizes and training methods, alongside the use of Random Forest classifier. We used Sentinel-2 bands with 10 m and 20 m spatial resolution, NDVI, NDVI variance, and SRTM altitude data to create input paths of 5×5 pixels. The results indicate that the 3D model trained with classical training and 3×3-pixel filters achieved the best F1 Score of 0.84. One significant advantage of using convolutional neural networks is their ability to include information from a pixel’s neighbourhood in the classification process, in contrast to solely considering the pixel itself.

One of the most problematic invasive species in Europe are knotweeds from genus Reynoutria (Fallopia) which have significant negative impact on the native communities as well on human activities. Therefore, they are a target of many control programmes. Due to their high regeneration potential, their management is problematic, and only chemical treatment is reported to be sufficiently effective. The aim of this paper was to describe and analyse the patterns of Reynoutria invasion under long-term chemical treatment with glyphosate-based herbicide in The Morávka river floodplain, Czech Republic. The data covers 17 years of management which started with the European project “Preservation of alluvial forest habitats in the Morávka river basin”. We focus on (i) assessment of Reynoutria distribution during long-term management, (ii) analysis of the change of distribution according to the habitat, and (iii) discussion of the optimal management strategy based on the long-term data. Distribution data was obtained using GNSS field mapping. Before the start of the study in 2007, Reynoutria stands covered 29% of the study area (96.9 ha). As a result of systematic whole area chemical management, the extent decreased to 19.6% (65.3 ha) in 2009, and even reached 14.5% (48.2 ha) in 2013, three years after its end. Due to implementation of local chemical management in the following years, the area of Reynoutria was maintained at similar level, with minimum value 41.8 ha in 2018 and a slight increase in recent mapping in 2023. Beside the extent, the structure and coverage of invaded sites was analysed. There was a clear trend of fragmentation of larger polycormons with high coverage into many smaller and less dense ones as a result of chemical spraying. The average size of Reynoutria stand decreased from 0.61 ha in 2007 to half in 2013 (0.32 ha) to 0.15 ha in 2023. Testing of the effects of time, habitat, and biotope did not reveal significant differences of changes of extent and abundance over different environments (forest, open, bare ground), which indicates that there are no differences in reaction to management in the studied habitat and vegetation types. Our study provides a robust and unique overview of the invasion, reinvasion, and suppression dynamics for an important invasive species. If herbicide management is used, chemical treatment must be quite long-term as even three years of intensive glyphosate foliar spray application was not sufficient for the complete eradication of Reynoutria. Therefore, we propose the following procedure for effective chemical management of Reynoutria: 1) In largely infested sites, the first step is to reduce the distribution of Reynoutria stands to isolated polycormons. This phase can last 3–5 years. 2) After reaching the state of sparse distribution of Reynoutria, we recommend herbicide application only in periods of every 3–5 years depending on the local context and rate of regrowth. 3) At sites exposed to soil disturbances, where the soil is contaminated by fragments of Reynoutria rhizomes, there is a need to apply herbicide immediately to target newly resprouting individuals.

Permanent grasslands play a very important role in the landscape. The loss of permanent grasslands and their subsequent conversion into arable land create erosion-prone agricultural areas in the landscape and have a negative impact on the biodiversity. From this point of view, there is a need for the accurate and effective monitoring of changes in the agricultural landscape along with an assessment of the influence of the agricultural policies on the landscape. Sentinel-2 from the Copernicus programme has improved options for the implementation of remote sensing data into the monitoring of agricultural land. The aim of this study was to evaluate the potential of H2O library and within implemented Automachine learning function (AutoML) and its stacked ensembles for mapping changes from grasslands to arable lands. All results show high overall accuracy from 93.5% to 96.6% and high values of area under the ROC curve (0.94–0.98). Stacked ensembles appear to be the most accurate machine learning models for mapping changes from grasslands to arable lands. The importance of several biological predictors has been tested (FAPAR, FCOVER, LAI, NDVI, etc.) with the help of a heatmap that is part of AutoML function of H2O library.

The frequency of wildfires is increasing worldwide, contributing to a third of forest loss in the last two decades. Tracking burned area progression using traditional optical remote sensing is hindered by cloud and smoke coverage. Therefore, this research employs multi-temporal Synthetic Aperture Radar (SAR) satellite data, which are not susceptible to atmospheric effects. Focusing on four Greek wildfires in 2021, the research utilizes unsupervised k-means clustering on bitemporal and multi-temporal SAR polarimetric features. The impact of input feature smoothing with varying moving kernel window sizes was assessed on improving accuracy. The use of these smoothed features led to a substantial improvement in accuracy across all four areas examined, while a window size of 19x19 was chosen as the right balance between preserving fine details and minimizing speckle. Furthermore, adding a filter after clustering to remove areas smaller than 2 hectares led to additional improvements in accuracy, especially in commission error. The results using the defined settings revealed F1 scores of 0.75-0.88, overall accuracy of 81-94%, and omission/commission errors of 33-16% and 14-3%, respectively. Challenges were observed in regions characterized by a substantial share of agricultural areas, while terrain effects revealed no substantial effects on the results. The assumption that the SAR will be sensitive mainly on bigger structural changes was proved in the visual validation using high-resolution imagery. Additionally, a Google Earth Engine (GEE) toolbox ‘Sentinel-1 Burned Area Progression’ (S1-BAP) was developed using the presented methodology and is freely available for the scientific community on GitHub.

This article investigates public spaces near urban rivers that contribute to the interaction between natural and urbanized areas and between people from different socio-economic backgrounds. The main goal of this study was to evaluate the environment of the largest urbanized areas of the Czech Republic, through which a large watercourse flows and creates a direct interaction with the city center. To evaluate the state of connectivity and comfort of urban rivers in the Czech Republic, a set of tools was applied to three cities: Prague, České Budějovice, and Hradec Králové. The methodology was created to correspond to the territory of Central Europe and was used for the specific assessment of rivers in four dimensions: (a) the spatial and visual accessibility, (b) the condition of the green corridor, (c) the condition of public space, and (d) the condition of the first built line. The dimensions are expressed using thirteen quantitative indicators of the environmental condition. The methodology uses the Urban River Sustainability Index (URSI), which was necessary to adjust the calculations of the indicators and resources for the Central European area. The best results were found in the central part of Prague and the worst in the peripheral part of Hradec Králové. The results call for the use of connectivity and comfort assessments of urban rivers for comparison, motivation, and future improvement in practice.

Wildfires are one of the most significant threats to ecosystems and are increasing in frequency globally. The aim of this study is to monitor the evolution of selected wildfires in Greece that occurred during August 2021 using Sentinel-1 SAR data and unsupervised k-means clustering in Google Earth Engine. First, changes in time series after the start of the fire and the influence of precipitation were investigated. In this study, the influence of different speckle filters and post-classification filters on clustering results was tested. The difference Normalized Burn Ratio Index (dNBR) derived from Sentinel-2 data was used as a validation dataset to assess accuracy using the F1-score, overall accuracy, omission and commission error. The best achieved F1-scores were higher than 0.70 with omission error lower than 35% in all selected areas, where the Lee speckle filter with an 11x11 kernel window size and a 2 ha post-classification filter performed the best.

The land system faces many pressures from the provision of biomass resources and space to the economy. The need to understand land use and cover changes and its drivers is of high importance. This work presents an innovative approach by applying a transdisciplinary approach combining the methods of spatial analysis Land Cover Flows with the methods from the concept of socio-economic metabolism, Material and Energy Flow Accounting, Human Appropriation of Net Primary Production (HANPP) and Final Energy Return on Investment (FEROI). Our main aim is to identify the main land use changes and land cover flows, link them to the underlying socio-economic processes and interpret them in a historical context. Our results show that the overall land use intensity is growing although the positive trends of growing grasslands and forests started after the collapse of communism. The growing intensity of agricultural production with increasing suburbanisation reversed these trends. Until the 2000s the HANPP decreased but at the end of the period increased from 55 % in 2012 to 70 % in 2018. Volumes of the extraction of agricultural biomass are growing while the area of agricultural land has decreased. FEROI grew and stabilised to around 1.0 in the last period (2012-2018) comparable to the value found in the year 2001. The suburbanisation rates peaked after the year 2000 at 250 m(2)/km(2)/yr.

Current optical vegetation indices (VIs) for monitoring forest ecosystems are well established and widely used in various applications, but can be limited by atmospheric effects such as clouds. In contrast, synthetic aperture radar (SAR) data can offer insightful and systematic forest monitoring with complete time series (TS) due to signal penetration through clouds and day and night image acquisitions. This study aims to address the limitations of optical satellite data by using SAR data as an alternative for estimating optical VIs for forests through machine learning (ML). While this approach is less direct and likely only feasible through the power of ML, it raises the scientific question of whether enough relevant information is contained in the SAR signal to accurately estimate VIs. This work covers the estimation of TS of four VIs (LAI, FAPAR, EVI and NDVI) using multitemporal Sentinel-1 SAR and ancillary data. The study focused on both healthy and disturbed temperate forest areas in Czechia for the year 2021, while ground truth labels generated from Sentinel-2 multispectral data. This was enabled by creating a paired multi-modal TS dataset in Google Earth Engine (GEE), including temporally and spatially aligned Sentinel-1, Sentinel-2, DEM, weather and land cover datasets. The inclusion of DEM-derived auxiliary features and additional meteorological information, further improved the results. In the comparison of ML models, the traditional ML algorithms, RFR and XGBoost slightly outperformed the AutoML approach, auto-sklearn, for all VIs, achieving high accuracies ($R^2$ between 70-86%) and low errors (0.055-0.29 of MAE). In general, up to 240 measurements per year and a spatial resolution of 20 m can be achieved using estimated SAR-based VIs with high accuracy. A great advantage of the SAR-based VI is the ability to detect abrupt forest changes with sub-weekly temporal accuracy.

Current optical vegetation indices (VIs) for monitoring forest ecosystems are widely used in various applications. However, continuous monitoring based on optical satellite data can be hampered by atmospheric effects such as clouds. On the contrary, synthetic aperture radar (SAR) data can offer insightful and systematic forest monitoring with complete time series due to signal penetration through clouds and day and night acquisitions. The goal of this work is to overcome the issues affecting optical data with SAR data and serve as a substitute for estimating optical VIs for forests using machine learning. Time series of four VIs (LAI, FAPAR, EVI and NDVI) were estimated using multitemporal Sentinel-1 SAR and ancillary data. This was enabled by creating a paired multi-temporal and multi-modal dataset in Google Earth Engine (GEE), including temporally and spatially aligned Sentinel-1, Sentinel-2, digital elevation model (DEM), weather and land cover datasets (MMT-GEE). The use of ancillary features generated from DEM and weather data improved the results. The open-source Automatic Machine Learning (AutoML) approach, auto-sklearn, outperformed Random Forest Regression for three out of four VIs, while a 1-hour optimization length was enough to achieve sufficient results with an R2 of 69-84% low errors (0.05-0.32 of MAE depending on VI). Great agreement was also found for selected case studies in the time series analysis and in the spatial comparison between the original and estimated SAR-based VIs. In general, compared to VIs from currently freely available optical satellite data and available global VI products, a better temporal resolution (up to 240 measurements/year) and a better spatial resolution (20 m) were achieved using estimated SAR-based VIs. A great advantage of the SAR-based VI is the ability to detect abrupt forest changes with a sub-weekly temporal accuracy.

The paper presents research conducted in the context of the professional education of teachers in kindergartens and primary schools focused on personality development, emotional intelligence and the creation of a favourable classroom climate. The research aims to confirm or refute the usefulness of such education at other than a pedagogical faculty, and to find how specific areas of this new form of education contributed to the participants’ teaching practice. The effect of the training was evaluated using questionnaires collected from 27 participants before and after the programme. The programme's benefits were analysed based on closed questions using statistical methods, while open-ended questions were processed by content analysis. The results suggest that a training programme using elements from different managerial field appears to be highly suitable and beneficial. The paper is an appeal that approaches from various fields can be used in further education, thus expanding the skills portfolio of modern teachers.