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 presents the first results of the ongoing development of new forest mapping methods for the Swedish national forest mapping case using Airborne Laser Scanning (ALS) data, utilizing the recent findings in machine learning (ML) and Artificial Intelligence (AI) techniques. We used Random Forest (RF) and eXtreme Gradient Boosting (XGBoost) as ML models. In addition, Neural networks (NN) based approaches were utilized in this study. ALS derived features were used to estimate the stem volume (V), above-ground biomass (AGB), basal area (B), tree height (H), stem diameter (D), and forest stand age (A). XGBoost ML algorithm outperformed RF 1 % to 3 % in the R² metric. NN model performed similar to ML model, however it is superior in the estimation of V, AGB, and B parameters.

This paper presents the project ForestMap – a project aiming to develop and distribute new methods, which provide the benefits of accurate forest maps to a global audience. Using the recent developments in remote sensing, machine learning, and Artificial Intelligence (AI) the goal is to export the Scandinavian success stories to a wide range of stakeholders in the world.

Pairs of fully polarimetric ALOS-2 PALSAR-2 L-band SAR images were used to model biomass on backscatter change over seven growth seasons in a hemi-boreal forest. The biomass change was related to backscatter change via consecutive field surveys of 263 field plots with a 10 m radius. To correct for differences in backscatter not related to biomass abundance, a HV-VV polarization ratio based correction, previously used on airborne L-band data, was applied to the data. The uncertainty of obtained predictions (lowest model mean RMSE 65.1 t/ha, lowest model mean bias 7.1 t/ha) was almost identical whether model fitting and prediction used data from the same scene pair, or different scene pairs. This could possibly attest to the feasibility of the backscatter correction for PALSAR-2 data, but no large backscatter offsets were observed for uncorrected data, and significant variance in predictions, due to the inherent noise in the data and the comparatively small area of evaluation plots, inhibit the analysis.

This study investigated the potential of utilizing time series of TanDEM-X phase heights, corrected for penetration depth, to detect silvicultural treatments in hemi-boreal forest. In total, 34 field plots with 40 m radius were used in conjunction with detailed forest management records to construct a reliable data set of treatments. The study area is situated in Remningstorp, a forest test site in southern Sweden. In the analysis, the temporal mean corrected phase heights were compared before and after a silvicultural treatment in order to quantify the effects of thinnings and clear-cuts on the phase height. As expected clear-cuts were highly distinguishable, but thinnings, while exhibiting a negative change in phase height on average, were not individually distinct from all untreated plots. Moreover, the results regarding the utility of applying penetration depth correction for the task were inconclusive. Overall, the results look very promising for using time series of phase height from TanDEM-X to map thinnings and clear-cuts, especially when several observations are available before and after the silvicultural treatment.

When modeling forest variables from InSAR phase height data, phase noise is a nuisance that can be mitigated by spatial averaging. In this empirical study, based on data from a hemi-boreal forest, the relation between field measured forest variables and TanDEM-X phase height and its dependence on the size of the field plots was investigated. Also, the impact of extending the region of phase height extraction beyond the actual plot size was investigated in order to determine the usefulness of this approach when small plot sizes cause significant noise in the estimation of phase height. For fixed field plot sizes (7 m and 10 m radii), the maximum correlation between Lorey's height and phase height occurred for phase height extraction regions multiple times larger in area than the field plots, which were located in homogenous stands.

SLU Forest Map are maps of the Swedish forest state, produced by the Swedish University of Agricultural Sciences (SLU) from satellite images using the Swedish National Forest Inventory sample plots as reference data. Until now, four maps have been produced, in raster format (12.5 × 12.5 m2to 25 × 25 m2cell sizes), with estimates of basal area-weighted mean tree height, basal area-weighted mean stem diameter, stand age, total as well as species-specific stem volume, for the years 2000, 2005, 2010, and 2015. These maps provide publicly available data, free of charge, supporting a wide range of applications; scientific research as well as operational uses in forest management planning, biodiversity assessment, and monitoring. This paper presents SLU Forest Map, the data and methods utilised in the production, and a new consistent evaluation of the estimation accuracy for each variable and mapped year.

In this study, Synthetic Aperture Radar (SAR) backscatter data from the Swedish airborne CARABAS-II and LORA systems were used to estimate stem density. The analysis were performed at the test site Remingstorp, located in the south of Sweden, consisting mainly of coniferous in hemi-boreal forests. In total, ten 80 m × 80 m forested areas, where all trees were measured in situ, with stem densities in the range of 278-552 stems and stem volumes in the range of 70-550 m3ha-1were analysed. SAR data from CARABAS-II and LORA were acquired in 2006 with nine unique flight headings. Local maxima of the backscatter were used to estimate stem density. The results were compared with in situ data and the accuracy in terms of root mean square error (RMSE) for CARABAS-II and LORA were found to be 81 stems (20.9%) and 82 stems (20.2%), respectively, in the best cases. The accuracy assessment of the stem density were performed after averaging images from each SAR system. This was done in order to suppress noise to a greater extent compared to using single images. The results show a potential to estimate stem density with low-freqency SAR data for the benefit of forest management.

In this study, mapping of tree growth was performed using data from the two nation-wide acquisitions of airborne laser scanning in Sweden. Following the successful first national acquisition performed in 2009 - 2015, a new, repeated, scanning is now launched and ongoing. The first scanning provided new, accurate (in accuracy as well as in spatial resolution) data about the forest and quickly found wide-spread use in the forest industry. It outperformed previous methods and provided a new standard of data capture for forest management planning. The addition of a second scanning provide information also about changes, where forest tree growth is of high interest in the industry. This study presents the first results from large-scale assessment of growth for basal area-weighted mean tree height (H) and mean stem volume (V), using the bi-temporal scannings and sample-plot data from the National Forest Inventory. Growth was most accurately assessed by the direct change metrics of the scannings, although the accuracies were moderate. The accuracy of forecasts, i.e. only utilizing the predicted forest state at the first scanning, were similar for H but inferior for V, though.

The development of drone technology has been rapid in the last decade, providing highly competent and economical platforms for applied remote sensing. Mapping forest using drones is not an economical alternative for most operational applications in forestry and environmental monitoring, due to the limited area covered. However, the potential of drone-based remote sensing is expected to be very large in research applications. In this study, a standard, small, four-rotor drone is used as platform for a multispectral camera to accurately collect 3D-data about the forest canopy, data used to produce maps of forest variables. These maps proved to be very valuable input for development of new models for the exchange of green-house gases in boreal and hemi-boreal forests. The task was performed within the ICOS initiative at three sites in Sweden where each has a 150 m high tower monitoring green-house gas fluxes at many levels in the atmosphere.

Three-year forest above-ground biomass change were measured using L- and P-band Synthetic Aperture Radar (SAR) backscatter. The SAR data were collected in the airborne BioSAR 2007 and BioSAR 2010 campaigns over the hemiboreal Remningstorp test site in southern Sweden. Regression models for biomass were developed using reference biomass maps created using airborne laser scanning data and field measurements. The results from regression analysis show that using HV backscatter (or VH) in a model with above-ground biomass and backscatter change on either natural logarithmic or square root, and decibel scale, respectively, explained most of the variation in the biomass change, both for L- and P-band. In the case of L-band, the two best cases showed R2values of 66%, when comparing two SAR images acquired 2007 and 2010. For P-band using the same models, the best cases showed R2values of 62%. In summary, the results look promising using L- and P-band backscattering for mapping biomass change.

A first evaluation of ALOS-2 PALSAR-2 data for forest stem volume estimation has been performed at a coniferous dominated test site in southern Sweden. Both the Fine Beam Dual (FBD) polarization and the Quad-polarimetric mode were investigated. Forest plots with stem volume reaching up to a maximum of about 620 m3ha−1(corresponding to 370 tons ha−1) were analyzed by relating backscatter intensity to field data using an exponential model derived from the Water Cloud Model. The estimation accuracy of stem volume at plot level (0.5 ha) was calculated in terms of Root Mean Square Error (RMSE). For the best case investigated an RMSE of 39.8% was obtained using one of the FBD HV-polarized images. The corresponding RMSE for the FBD HH-polarized images was 43.9%. In the Quad-polarimetric mode the lowest RMSE at HV- and HH-polarization was found to be 43.1% and 66.1%, respectively.

Three-dimensional (3D) data about forest captured by airborne laser scanning (ALS) have revolutionized forest management planning. Accurate, updated large-scale maps of forest variables produced with low costs today support greatly improved decisions about silvicultural treatments compared to the past practice based on field surveyed data only. These maps usually lack important information about forest age and site productivity, as this cannot be accurately assessed from the available ALS data. In Sweden, ALS has recently been performed nation-wide, except the mountainous area, to produce a new and accurate digital terrain model (DTM). This DTM enables extremely cost-efficient extraction of 3D data about the forest from other sources than ALS, such as automatic stereo-matching of aerial images as well as from single-pass spaceborne interferometric synthetic aperture radar (InSAR). In contrast to ALS, these data sources can provide low-cost time-series of 3D data. Aerial images of Sweden are often available in archives back to approximately 1960, and the TanDEM-X SAR system has the potential to provide new data every second week over large areas. These data have a potentially high value for forest management planning, since they may provide missing and highly important information - forest site productivity, Site Index (SI) and forest age. This pilot study explores a least-squares minimization approach to estimate forest age and SI from time series of 3D data produced by 1) image matching of DMC aerial images, and 2) TanDEM-X SAR data.

Methods to estimate forest biomass change have been investigated using experimental P-band synthetic aperture radar (SAR) data from the recent airborne campaigns BioSAR 2007 and BioSAR 2010 conducted over a hemiboreal test site in southern Sweden. Regression models based on backscatter change were developed using reference biomass change maps derived from high-density laser scanning data. Different regression models were developed for linear, square root, and logarithmic biomass change scales. The models were compared to the change maps based on laser data using twofold cross-validation, and estimation errors were evaluated using six 80 m by 80 m plots with detailed in situ measurements. The results indicate that the root-mean-square error of biomass change estimates based on P-band SAR backscatter data is about 15% or 20 t/ha. This suggests that not only clear-cuts but also growth and thinning can be measured. Simulations were performed in order to evaluate the possibility of using a spaceborne P-band SAR for measurements of forest biomass change. The simulations show that, with 64 equivalent number of looks (ENL) and a 50% change in biomass, it is possible to correctly indicate whether the forest has gained or lost biomass. Similarly, for a biomass loss of more than 75%, a correct indication of the sign of biomass change can be achieved with only 8 ENL.

Synthetic Aperture Radar (SAR) backscatter data from the Swedish airborne CARABAS-II and LORA systems were used to estimate stem volume at stand level. The study was performed in hemi-boreal forests at the Remningstorp test site, located in southern Sweden. In total, ten 80 m × 80 m stands, where all trees were measured in situ, with stem volumes in the range of 70-530 m3ha-1(on average 347 m3ha-1) were analyzed. SAR data from CARABAS-II and LORA were acquired from two different years, with nine unique flight headings that were repeated for each system and year. Regression analysis was used to estimate stem volume and the accuracy was assessed in terms of Root Mean Square Error (RMSE). As a first step, stem volume was estimated for each flight heading separately. The accuracy assessment was then performed by weighting the separate estimates for each system and year inversely proportionally to the variance about the regression function. The best results for CARABAS-II and LORA showed a relative RMSE of 7% and 24% of the mean stem volume, respectively. In a previous study, stem volume was estimated using LiDAR data and the same forest stands, resulting in an RMSE of about 12%. In conclusion, the estimation accuracy of stem volume using combined low-frequency CARABAS SAR images was found to be superior to that from using LiDAR data for the stands investigated.

In this study, dense hemi-boreal forest biomass and height estimation was investigated based on optical satellite data and a high quality Digital Elevation Model (DEM) from airborne laser scanning. This analysis was carried out on data collected 2008-2010 over the test site Remningstorp in southern Sweden. The optical sensors SPOT-5 HRS and ASTER were tested to process a Digital Surface Model (DSM), i.e. the vegetation height above mean sea level, that is used together with the DEM (derived from laser data) to calculate a Canopy Height Model (CHM) as the difference between the former ones. By modeling biomass and height using regression analysis on spectral data from SPOT-5 HRG and height metrics from the CHM an improved Root Mean Squared Error (RMSE) and adjusted R2is expected, compared to using the single data sources alone. The best results showed a relative RMSE for standwise prediction of mean biomass and height of 30.3% and 23.3%, respectively. Adding CHM data to a spectral based (HRG) prediction model improved the mapping accuracy roughly 3%. In conclusion, the estimation accuracy did not improve significantly by adding height metrics to spectral data.

The increasing availability of accurate Digital Elevation Models (DEMs) of nation-wide cover has opened new possibilities to produce accurate forest variable estimation using 3D data acquired from aerial imagery. Such data can be produced by automatic matching of stereo images and photogrammetric modeling of the forest canopy height. Using existing accurate DEM information, the forest canopy height above ground is then easily assessed. Today, Airborne Laser Scanning (ALS) is frequently used to capture data for accurate estimation of variables to be used in forest management planning. Recent studies in Scandinavia show estimation accuracies almost as accurate as ALS, using 3D data obtained from standard aerial imagery, at least for the most important forest variables. So far mainly area-based estimation methods at field plot or raster cell level have been studied. This paper reports early results from applying a single-tree modeling approach, corresponding to the Semi-ITC (Individual Tree Crown) method, commonly used in ALS-based applications, using 3D data acquired from aerial DMC imagery. Here, a simplified Semi-ITC method was used to estimate tree height at segment level. The Root Mean Square Error of estimating the maximum tree height was 34% (of the true mean maximum tree height). Clearly, the methodology used shows promising results and has potential to be used in forest management planning.

The ESA funded campaign BioSAR 2010 was carried out at the forestry test site Remningstorp in southern Sweden, in support to the BIOMASS satellite mission under study. Fully polarimetric SAR data were successfully acquired at Land P-band using ONERA's multi-frequency system SETHI. In addition with other data types gathered, e.g. LiDAR and in-situ measurements, the compiled data set will be used for analyses and comparisons with biomass estimation results obtained at the same test site in the campaign BioSAR 2007, in which DLR's E-SAR made the SAR imaging. Detection of forest changes, robustness of biomass retrieval algorithms and long-term P-band coherence will be in focus as well as cross-validations between the two SAR sensors.

The nation-wide Airborne Laser Scanning (ALS) currently performed by the Swedish National Land Survey will provide a new and accurate Digital Elevation Model (DEM). These data will enable new and cost-efficient assessments of vegetation height using Canopy Height Models (CHMs) derived as the difference between a Digital Surface Model (DSM) and the DEM. In this context, the High Resolution Stereoscopic (HRS) sensor onboard SPOT-5 and the airborne Z/I Digital Mapping Camera (DMC) used for operational aerial photography by the Swedish National Land Survey are of main interest. Previous research has shown that reliable tree height data are a powerful source of information for forest management planning. This study investigated the possibilities to map forest variables using CHMs derived from either the SPOT-5 HRS or Z/I DMC sensor together with ALS DEM data, in combination with spectral data from the SPOT-5 High Resolution Geometric (HRG) sensor. The results when using the Z/I DMC CHM in combination with SPOT-5 HRG data showed Root Mean Square Errors for standwise prediction of mean tree height, stem diameter, and stem volume of 7.3%, 9.0%, and 19%, respectively. The SPOT-5 HRS CHM in combination with SPOT-5 HRG data improved the SPOT HRG based estimates from 13% to 10%, 15% to 13%, and 31% to 23%, for tree height, stem diameter, and stem volume, respectively. Adding CHM data to a SPOT-5 HRG based prediction model improved the mapping accuracy between 13% to 44%. In conclusion, the obtained accuracies may be sufficient for operational forest management planning.