Airborne laser scanning (ALS) and area-based modelling enable operational mapping of forest structural attributes, yet cross-border model transfer remains under explored. We assess whether a Swedish-trained “MetaPredictor”based in DeepLab v3 feature extraction from 9×9×60 ALS raster patches fused with tabular metrics via XGBoost, can be transferred to Latvia with minimal local calibration. Swedish training relied on ALS raster stacks aligned to National Forest Inventory (NFI) plots. Targets included mean height, mean diameter, basal area, stem volume, and above-ground biomass (AGB). Latvian evaluation used ALS point clouds processed with lidR to produce equivalent 60-band rasters and plot-level reference attributes. Latvian inputs were standardized using Swedish scalers, passed through DeepLab v3 (ResNet-101) to extract spatial features, concatenated with scaled ALS laser metrics, and evaluated under two regimes: zero-shot transfer (no local labels) and few-shot fine-tuning (1%–20% Latvian plots). Performance was quantified with RMSE, MAE, relative RMSE (rRMSE), and R2. In zero-shot mode, height transferred well (R2=0.91, RMSE 2.60 m, rRMSE 14%), diameter moderately (R2=0.60, RMSE 7.72 cm), while basal area and AGB showed weak to negative generalization (R2=−0.07 and −2.29). Few-shot fine-tuning produced decisive gains: basal area RMSE dropped from 11.9 to 5.9 (R2=0.73), volume from 139.7 to 71.7 (R2=0.83), and AGB from 173.8 to 36.16 (R2=0.86), amounting to 48%–79% error reductions. Height and diameter, already transferable, improved incrementally. Results demonstrate that Swedish ALS-trained models can seed Latvian operational mapping with modest local calibration. The MetaPredictor design, combining deep spatial encodings with tabular metrics, supports rapid deployment of forest attribute mapping in regions with limited labelled data.

Remote sensing techniques are widely used for mapping and monitoring forest attributes, providing valuable information on forest cover, biomass, and overall forest health. In recent years, national airborne laser scanning (ALS) campaigns have been conducted in several countries to map forest resources. When combining ALS with field inventory data, these datasets enable the development of nationwide models for prediction of forest attributes. In this study, we explore the potential of machine learning (ML) to enhance existing modeling approaches for nationwide forest attribute mapping in Sweden. We achieve this by relating ALS data from the most recent ALS campaign of Sweden with field data from the Swedish National Forest Inventory (NFI). By aggregating laser metrics from surveyed areas (NFI plots), as well as over surrounding areas to the plots, we investigate (1) if ML approaches can outperform existing linear regression baseline models and (2) if further enhancements of the predictive capacity can be achieved by including surrounding, spatially correlated ALS data. To this end, we used extreme gradient boosting (XGBoost), as well as a convolutional neural network (CNN), specialized to handle tabular data and spatially correlated data, respectively. The models were evaluated on five forest variables: basal-area weighted mean tree height, basal-area weighted mean stem diameter, basal area, stem volume, and above-ground biomass. All models were evaluated on several nested datasets to assess the robustness, showcasing consistent results across datasets. We achieved significant improvements in prediction accuracy across all investigated forest variables. Furthermore, incorporating surrounding information to the modeling rendered further improvements for diameter, basal area, and biomass predictions. The approaches tested and developed here thus form a promising basis for flexible modeling approaches that can be transferred globally for large-scale forest monitoring and management.

Accurate parcel geometry is important for many applications in agriculture and forestry. The assessments of agricultural parcel geometries are, e.g., crucial for calculating financial subsidies. Today, authorities use aerial images to check parcels for homogeneous cultivation and incorrect parcel geometries (homogeneity-check). Due to the large number of parcels and limited resources of authority offices, there is a need for an automated method of risk-based selection of a manageable subset of parcels for manual review. To this end, this study evaluated four machine learning approaches; Image Classification, Semantic Segmentation, Edge Detection, and Clustering, with respect to (i) their sensitivity to spatial resolution of aerial images, (ii) the impact of the balancing method used in training, and (iii) their overall performance with respect to preselecting parcels for homogeneity-check. In addition, we introduced a modified version of the commonly used Precision@K metric. The proposed Precision@K% replaces the absolute number of top-ranked items with a relative proportion, facilitating a more scalable evaluation of risk-based selection strategies. The results show that all approaches performed similarly in terms of the area under the receiver operating characteristic curve (AUROC) and the Precision@2% metric; Image Classification: AUROC = 77.4%, Precision@2% = 80%, Semantic Segmentation: AUROC = 83.0%, Precision@2% = 75%, Edge Detection: AUROC = 79.2%, Precision@2% = 85%, and Clustering: AUROC = 84.4%, Precision@2% = 100%. Notably, the Clustering approach consistently maintained a stable Precision@2% of 100%.

The potential of dense Sentinel-2 time series to serve as a basis for operational crop monitoring systems is hindered by cloud cover, especially at high latitudes. Sentinel-1 data can overcome this limitation, but similar to optical data, are prone to saturation, i.e. when changes in vegetation biomass are not reflected in the remote sensing signal. Time-integration is a strategy commonly used in optical remote sensing to mitigate saturation effects. In this pilot study, we tested whether this approach can also improve the relationship between Sentinel-1 backscatter and maize traits, using Sentinel-1 A and B backscatter data. Our test site consisted of a forage maize experimental field in Sweden. Evaluated plant traits included the number of leaves, the phenological stage, the leaf area index, the dry matter yield and the dry matter content. Linear and logistic models were adjusted between time-integrated values of the backscattering coefficient ([Formula: see text], [Formula: see text], [Formula: see text] and [Formula: see text]) and field-measured traits. Our results indicate a good agreement between Sentinel-1 time-integrated signal and maize traits, with [Formula: see text] of 0.97, 0.93, 0.94, 0.95 and 0.86 for phenological stage, leaf number, leaf area index, dry matter yield and dry matter content, respectively, and systematically outperformed models built with non-cumulative [Formula: see text] values. Our findings also indicate that the time-integrated models perform equally well with data acquired from a single Sentinel-1 satellite. These results, if confirmed for a wider range of geographical extent and management conditions, could pave the way for a remote sensing-based, weather-independent and saturation-insensitive decision support tool.

Northern mires are significant natural sources of atmospheric methane (CH4), yet estimating CH4 emissions remains challenging due to their complex spatio-temporal dynamics. While eddy covariance (EC) measurements provide valuable insights into ecosystem-scale CH4 fluxes (FCH4) over mire areas typically < 0.05 km(2), the predictability of FCH4 at the mesoscale (similar to 0.5 - 20 km(2)) of a mire complex based on single-site EC measurements has not been explored. In this study, we utilized a network of four EC towers and developed a machine learning approach that integrates these EC data with comprehensive spatial information on drivers to predict FCH4 across a boreal mire complex in Northern Sweden. For this purpose, environmental driver variables were mapped and area-weighted within dynamic EC flux footprints and related to FCH4 in a spatially-explicit random forest model ('footprint-based model'). For comparison, we also considered a standard random forest model used for gapfilling of FCH4 data that is based on environmental measurements from fixed sensor locations ('biomet model'). For both models, variable importance analysis revealed NDVI as the strongest predictor of temporal FCH4 patterns, followed by air pressure, soil temperature and water table. Adjusting for site-specific carbon-to-nitrogen (C:N) ratios substantially improved model performance. Both models significantly improved estimates of the mire complex average FCH4 compared to simple extrapolation of single-site measurements, reducing the uncertainty from similar to 22 % in 2022 and 32 % in 2023 to <10 % and <25 % for the footprint-based model, and to <11 % and <30 % for the biomet model, respectively. Overall, our findings suggest that the comprehensive spatially-resolved driver information resulted in only marginally improved model performance at our study site. In comparison, the biomet model offers practical advantages through simpler implementation and wider applicability. However, we encourage testing the footprint-based model approach at other more heterogenous sites where it might become superior due to its ability to account for complex site conditions.

Natural and planted forests, covering approximately 31% of the Earth’s land area, are crucial for global ecosystems, providing essential services such as regulating the water cycle, soil conservation, carbon storage, and biodiversity preservation. However, traditional forest mapping and monitoring methods are often costly and limited in scale, highlighting the need to develop innovative approaches for tree detection that can enhance forest management. In this study, we present a new dataset for tree detection, VHRTrees, derived from very high-resolution RGB satellite images. This dataset includes approximately 26,000 tree boundaries derived from 1,496 image patches of different geographical regions, representing various topographic and climatic conditions. We implemented various object detection algorithms to evaluate the performance of different methods, propose the best experimental configurations, and generate a benchmark analysis for further studies. We conducted our experiments with different variants and hyperparameter settings of the YOLOv5, YOLOv7, YOLOv8, and YOLOv9 models. Results from extensive experiments indicate that, increasing network resolution and batch size led to higher precision and recall in tree detection. YOLOv8m, optimized with Auto, achieved the highest F1-score (0.932) and mean Average Precision (mAP)@0.50 Intersection over Union threshold (0.934), although some other configurations showed higher mAP@0.50:0.95. These findings underscore the effectiveness of You Only Look Once (YOLO)-based object detection algorithms for real-time forest monitoring applications, offering a cost-effective and accurate solution for tree detection using RGB satellite imagery. The VHRTrees dataset, related source codes, and pretrained models are available at https://github.com/RSandAI/VHRTrees.

High latitude mires are key ecosystems in the context of climate change since they store large amounts of carbon while constituting an important natural source of methane (CH4). However, while a growing number of studies have investigated methane fluxes (FCH4) at the plot- (~1 m2) and ecosystem-scale (~0.1-0.5 km2) across the boreal biome, variations of FCH4 magnitudes and drivers at the mesoscale (i.e., 0.5-20 km2) of a mire complex are poorly understood. This study leveraged a network of four eddy-covariance flux towers to explore the spatio-temporal variations in ecosystem-scale FCH4 across a boreal mire complex in northern Sweden over 3 years (2020-2022). We found a consistent hierarchy of drivers for the temporal variability in FCH4 across the mire complex, with gross primary production and soil temperature jointly emerging as primary controls, whereas water table depth had no independent effect. In contrast, peat physical and chemical properties, particularly bulk density and C:N ratio, were identified as significant baseline constraints for the spatial variations in FCH4 across the mire complex. Our observations further revealed that the 3-year mean annual FCH4 across the mire complex ranged from 7 g C m-2 y-1 to 11 g C m-2 y-1, with a coefficient of variation of 16% that is similar to the variation observed among geographically distant mire systems and peatland types across the boreal biome. Thus, our findings highlight an additional source of uncertainty when scaling information from single-site studies to the mire complex scale and beyond. Furthermore, they suggest an urgent need for peatland ecosystem models to resolve the mesoscale variations in FCH4 at the mire complex level to reduce uncertainties in the predictions of peatland carbon cycle-climate feedbacks.

Radar remote sensing observations are predominantly affected by the concentration and spatial distribution of water in natural scenes. This motivates the utilization of high-resolution spaceborne radar observations for monitoring the water status of vegetation and the impacts of climate change on forests globally. While current satellite-based synthetic aperture radar observations are limited to temporal resolutions of days, tower-based radar observations of forests are capable of capturing detailed sub-daily physiological responses to variations in soil water availability and meteorological conditions. Such experiments demonstrate the scientific value of prospective sub-daily space-borne observations in the future. The BorealScat tower-based radar experiment conducted in southern Sweden from 2017 to 2021 has captured various ecophysiological phenomena in a boreo-nemoral forest, including water stress and degradation induced by spruce bark beetles (Ips typographus). To gain a deeper insight into the sub-daily impacts of forest water dynamics on radar observations, the BorealScat-2 tower-based radar experiment was initiated in a boreal forest, located in northern Sweden in 2022. Along with in-situ sensors characterizing the water status on the tree level and an eddy-covariance flux tower, this initiative aims to compile a comprehensive and open dataset. The goal is to enhance our understanding and modelling of the relationship between traditional ground-based forest information, eddy-covariance flux measurements and radar remote sensing observables. The data gathered by BorealScat-2 stands out as the most radiometrically precise high-resolution time series ever recorded in forest environments, resolving the subtle water content-induced signatures in radar measurements. Preliminary findings from the 2022 growing season, highlight the detectability of a diurnal radar signature across all conventional radar remote sensing bands (i.e. C-, L- and P-band). Moreover, metrics akin to tree water deficit, as measured by high-resolution point dendrometers, can be derived from interferometric radar observations. The fine temporal resolution of the data also unveils distinct signatures corresponding to intercepted precipitation in time series measurements. These findings underscore the need for sub-daily observations from space-borne satellites to monitor vegetation water status.

The early detection of forest damage is critical for minimizing economic losses and improving climate-smart forestry practices. This study applies time series analysis to monitor tree health and detect early signs of damage in the research park Svartberget in northern Sweden. High-resolution data in red (R), green (G), and blue (B) bands (RGB), and multispectral data in G, R, red edge (RE), and near-infrared (NIR) bands were collected at varying intervals from 31 May to 9 November 2024, aligned with the phenological stages of the boreal forest vegetation season, using a unmanned aerial vehicle (UAV) at a flight altitude of 80 m above ground level. A total of 756 trees were manually digitized and classified into three categories: damaged (30 trees), birch (33 trees), and coniferous (693 trees). This classification enabled spectral trajectory analysis to evaluate changes in tree vitality. By identifying trends in single bands and spectral indices sensitive to tree health, this study aims to develop a replicable methodology for monitoring forest vitality. The results will contribute to broader efforts in mapping early forest damage and advancing climate-smart forestry practices. Future work will also explore data acquisition at higher flight altitudes (100 m and 120 m), incorporate LiDAR and hyperspectral data, and apply additional vegetation indices and spectral ratios to enhance the detectability of forest health.

In this study, for the first time, sub-daily radar reflectivity variations are compared with a hydraulic forest model. Multi-polarimetric time series at P- to L-band (435, 600 and 1270 MHz) over a coniferous boreal forest stand, acquired by the BorealScat-2 tomographic tower radar, are analyzed. Attenuation was measured by observing a large trihedral reflector in the forest. The hydraulic forest model was parameterized based on tree sensor data, and driven by evapotranspiration measurements. The forest canopy reflectivity is found to co-vary mainly with the crown water potential and content at P-band, and with the attenuation at L-band. A frequency dependence is apparent, with a positive linear relation at P-band transitioning to a negative nonlinear relation at L-band. The attenuation co-varies mainly with the stem water potential and content at P-band, while it varies in between that of the crown and stem at L-band. In conclusion, it is shown that radar can be used to sense forest hydraulic properties, with prospects of possibly driving forest hydraulic models.

In this study the performance of 2D and 3D segmentation approaches for single-tree canopy delineation in a mixed boreal forest was investigated. Using high-resolution UAV LiDAR data collected from Svartberget research park in northern Sweden. The accuracy of delineations was compared for three tree classes: coniferous, birch, and damaged trees. The 2D approach relied on traditional local maxima segmentation techniques, while the 3D approach utilized the point cloud data to incorporate structural information. Preliminary results indicate that the 3D segmentation method provides more precise canopy delineation, particularly for complex and overlapping canopies. The enhanced accuracy of the 3D approach (ForAINet) is expected to contribute significantly to forest management and ecological monitoring including biodiversity studies. At the conference, we will present detailed results highlighting the advantages of 3D segmentation over 2D methods in terms of accuracy and reliability across different tree classes.

Effective forest monitoring requires the accurate detection of trees, but traditional methods usually fail due to cost, scale, and precision limitations. This study applies transfer learning using the state-of-the-art YOLOv8 model to improve tree detection from very high-resolution drone imagery. A pre-trained YOLOv8 network was initially trained on the high-resolution VHRTrees satellite imagery dataset and fine-tuned on a smaller very high-resolution drone image dataset. It will resolve issues related to both limited data and domain differences. Experiments explored the impact of spatial resolution at 15 cm and 30 cm, data augmentation, and layer freezing. The results showed that the best accuracy is achieved by fine-tuning using 15 cm resolution while freezing 10 layers; this outperforms models without any transfer learning and those trained from scratch. Data augmentation further reduced false positives, enhancing detection reliability. These findings highlight transfer learning as a cost-effective means to achieve both scalable and precise ecological monitoring-feasible for several applications in forest management and environmental care.

The future role of boreal forests in the global carbon cycle is uncertain given the rapid climate change in high latitudes. At the landscape scale, heterogeneity in stand age and land cover, contributions from terrestrial and aquatic fluxes, and harvest export may create complex carbon cycle-climate interactions. However, the integrated response of the net landscape carbon balance (NLCB) to inter-annual variations (IAVs) in environmental conditions is poorly understood. Here, we used tall-tower eddy covariance and stream monitoring to integrate terrestrial and aquatic carbon fluxes with harvest export for a 68 km2 boreal catchment in Sweden during 2016-2020. This actively managed forest landscape acted as a net carbon sink with a 5-year mean (+ standard deviation) NLCB of 128+55 g C m-2 yr-1. The NLCB IAV included a reduced sink (36 g C m-2 yr-1) during the cool/ cloudy year 2017. In the other four years, featuring a drought summer (2018) and an exceptionally warm/wet winter (2020), the landscape acted as a significant sink (127-180 g C m-2 yr-1). The NLCB IAV corresponded primarily to variations in landscape respiration, followed by GPP and harvest export, with negligible contributions from landscape CH4 and aquatic carbon fluxes. The NLCB IAV was not correlated to any single environmental factor. However, daily NLCB contrastingly responded to key environmental factors as a function of forest aboveground biomass and mire contributions. Overall, our study indicates that the annual carbon sink-strength of the managed boreal forest landscape may be resilient to a wide range of IAVs in environmental conditions.

Building on the positive experiences with open forest map data in Scandinavia, it is evident that extending a similar solution globally has the potential to revolutionize forest management and business on a worldwide scale. While forest management in the Nordic countries can certainly be enhanced, the most rapid solution for climate change mitigation involves providing other nations with opportunities akin to those that have benefited the forestry sector in Sweden during the initial stages of digitalization. In the proposed project, we aim to create a novel hierarchical decision-making system for efficient forest mapping, leveraging a diverse range of remote sensing data sources with varying resolutions. This hierarchical system will be developed using state-of-the-art AI methods, complemented by results from traditional computer vision techniques such as texture analysis, saliency, and probabilistic object representation. A significant strength of the project lies in using the forest data and maps of Sweden and Finland as test beds to benchmark the methodology developed. We are confident that this project will make substantial contributions to climate change mitigation, biodiversity enhancement, and other societal values. Moreover, it aims to foster the creation of new business models by developing an innovative methodology for the next generation of forest maps. Our vision is to adapt the success story of open forest map data from the Nordic region globally, harnessing the power of advanced AI technology and integrated use of remote sensing and field data.

The boreal forest is an important global carbon sink, but its response to drought remains uncertain. Here, we compiled biometric- and chamber-based flux data from 50 boreal forest stands to assess the impact of the 2018 European summer drought on net ecosystem production (NEP) across a 68 km 2 managed landscape in northern Sweden. Our results reveal a non-uniform reduction in NEP (on average by 80 ± 16 g C m −2 yr − 1 or 57 ± 13%) across the landscape, which was greatest in young stands of 20–50 years (95 ± 39 g C m − 2 yr − 1 ), but gradually decreased towards older stands (54 ± 57 g C m − 2 yr − 1 ). This pattern was attributed to the higher sensitivity of forest-floor understorey to drought and its decreasing contribution to production relative to trees during stand development. This suggests that an age-dependent shift in understorey–tree composition with increasing stand age drives the drought response of the boreal forest NEP. Thus, our study advocates the need for partitioning ecosystem responses to improve empirical and modelling assessments of carbon cycle–climate feedbacks in boreal forests. It further implies that the forest age structure may strongly determine the carbon sink response to the projected increase in drought events across the managed boreal landscape.

In this study, a novel approach to map forest attributes has been investigated for boreal forests in Sweden. The methodology relies on machine learning, utilizing a combination of remote sensing data and field data for both training and evaluating the proposed models. To ensure the accuracy in estimating forest attributes at any given time, the approach incorporates a broad range of available remote sensing data including airborne laser scanning (ALS) data, weekly satellite data from Sentinel-1 and Sentinel-2, and global forest map data. However, in this study focus has been on utilizing ALS data. The field data utilized in the study are derived from the Swedish National Forest Inventory and encompass measurements of key forest variables such as above-ground biomass, stem volume, basal area-weighted mean tree height, basal area-weighted mean diameter at breast height, and basal area. The potential of exporting knowledge gained from mapping Sweden to other forested landscapes such as in Latvia, using model updating with limited reference data from the new targeted area will be the next step to investigate. Here, data from Sweden were used to take the first steps towards developing a mapping methodology. The results demonstrate a promising potential of the proposed approach that will showcase new possibilities to share knowledge of updated forest mapping using the increasing flow of high-precision remote sensing data.

This paper introduces an AI-based approach to detect human-made objects and changes in these on land parcels. To this end, we used binary image classification performed by a convolutional neural network. Binary classification requires the selection of a decision boundary, and we provided a deterministic method for this selection. Furthermore, we varied different parameters to improve the performance of our approach, leading to a true positive rate of 91.3% and a true negative rate of 63.0%. A specific application of our work supports the administration of agricultural land parcels eligible for subsidiaries. As a result of our findings, authorities could reduce the effort involved in the detection of human made changes by approximately 50%.

Unravelling slow ecosystem migration patterns requires a fundamental understanding of the broad-scale climatic drivers, which are further modulated by fine-scale heterogeneities just outside established ecosystem boundaries. While modern Unoccupied Aerial Vehicle (UAV) remote sensing approaches enable us to monitor local scale ecotone dynamics in unprecedented detail, they are often underutilised as a temporal snapshot of the conditions on site. In this study in the Southern Alps of New Zealand, we demonstrate how the combination of multispectral and thermal data, as well as LiDAR data (2019), supplemented by three decades (1991–2021) of treeline transect data can add great value to field monitoring campaigns by putting seedling regeneration patterns at treeline into a spatially explicit context. Orthorectification and mosaicking of RGB and multispectral imagery produced spatially extensive maps of the subalpine area (~4 ha) with low spatial offset (Craigieburn: 6.14 ± 4.03 cm; Mt Faust: 5.11 ± 2.88 cm, mean ± standard error). The seven multispectral bands enabled a highly detailed delineation of six ground cover classes at treeline. Subalpine shrubs were detected with high accuracy (up to 90%), and a clear identification of the closed forest canopy (Fuscospora cliffortioides, >95%) was achieved. Two thermal imaging flights revealed the effect of existing vegetation classes on ground-level thermal conditions. UAV LiDAR data acquisition at the Craigieburn site allowed us to model vegetation height profiles for ~6000 previously classified objects and calculate annual fine-scale variation in the local solar radiation budget (20 cm resolution). At the heart of the proposed framework, an easy-to-use extrapolation procedure was used for the vegetation monitoring datasets with minimal georeferencing effort. The proposed method can satisfy the rapidly increasing demand for high spatiotemporal resolution mapping and shed further light on current treeline recruitment bottlenecks. This low-budget framework can readily be expanded to other ecotones, allowing us to gain further insights into slow ecotone dynamics in a drastically changing climate.