Inka Bohlin

In this study we created national vulnerability models for wind and snow damage based on a combination of remote sensing and Swedish National Forest Inventory (NFI) data and compared the performance of airborne laser scanning (ALS) -driven versus field-measured information in the modelling of damage vulnerability. The empirical training data consisted of circa 42 000 field plots, containing information about recent wind and snow damage, monitored between 2010-2022 in the Swedish NFI. Occurrence of damage was predicted using variables calculated from national ALS data (2009-2023), other mapped products, and weather data. In contrast to earlier large-scale damage models, we used direct ALS metrics to fully exploit ALS-derived forest structural information. Logistic regression was used for modelling, and separate models were created for southern and northern Sweden to consider the geographical differences in forest structure and climate conditions. Field data-based models performed slightly better than RS-based models, resulting in an AUC of 0.8 for northern and 0.73 for southern Sweden. Corresponding results for RS-based models were 0.77 and 0.69. Best models included both forest structural variables (ALS or field-based), tree species, terrain, and weather information. We successfully demonstrated the combination of ALS and NFI data to map forest vulnerability to wind and snow damage, which enables evaluation of the abundance and distributions of the damage vulnerability from stand to national level, and can be applied to a new region or damage type. By locating areas with high damage vulnerability, forest owners can more easily adapt forest management to increasing climate impacts.

Understorey vegetation (USV) plays a vital role in forest ecosystems by influencing biodiversity, nutrient cycling, and disturbance dynamics. Accurate mapping of USV is essential for understanding ecosystem functioning and its relationship with environmental variables across landscapes. The use of remote sensing (RS) for its application to USV prediction has been attempted a handful of times. Open-access Sentinel-1 C-band Synthetic Aperture Radar (SAR) and Sentinel-2 multispectral imagery have been used in this study for USV cover modelling, where USV cover is defined as the surface area (in m2) covered by the USV on the forest floor. Two sources of field reference data used here for USV cover measurements were from: (A) the Swedish National Forest Inventory (NFI) (for year 2019-2024) covering 55,000 km2 area in the Västerbotten County, northern Sweden, and (B) detailed field measurements from a smaller test site – the Krycklan catchment research area covering 70 km2 from 2024. This study aimed to 1) Develop two separate regression-based generalized additive models (GAM), Model A and Model B using an area-based approach integrating Senitnel-1 and 2 metrics and trained using the two field reference datasets A and B, respectively, and 2) Further extend Model A to account for interaction of USV with additional environmental covariate rasters of, e.g., soil moisture, elevation, land-use/land-cover classes, bedrock type, soil type, and bioclimatic metrics such as seasonal and annual temperature and precipitation, acquired over the entire Västerbotten county. Both baseline models A and B were developed using explanatory variables from Sentinel-1 namely, difference between the vertical transmit, vertical receive (VV) backscatter and vertical transmit, horizontal receive (VH) backscatter and total backscatter power (∣VV∣2+∣VH∣2), and from Sentinel-2 namely, Normalized difference vegetation index (NDVI), Visible atmospherically resistant index (VARI), and the difference between surface reflectance of the red-edge band in summer and autumn seasons. Both baseline models A and B - demonstrated comparable performance with similar magnitude of root mean square error (RMSE) and coefficient of determination (R²) values when validated against a common test subset derived from the NFI field reference data. With including the environmental covariates in model A, the USV cover showed correlation with soil moisture, elevation, land-use/land-cover classes, and seasonal and annual temperature and precipitation variables. The inclusion of these variables improved the extended model A performance compared to the baseline model A, with 15% increase of R² and 8% decrease of RMSE values. These results highlight the importance of integrating climate and topographic covariates along with RS data for improved USV prediction and mapping. This study demonstrates the feasibility of large-scale USV prediction using open access Sentinel-1 and 2 data combined with field reference data, and environmental covariates. While SAR and multispectral data provide valuable information, incorporating biophysical and climatic variables substantially enhances model performance. This approach offers a cost-effective and scalable workflow for monitoring USV in boreal forests, benefiting sustainable forest management and biodiversity studies.

Accurate and up-to-date data about growing stock volume are essential for forest management planning. Airborne Laser Scanning (ALS) is known for producing accurate wall-to-wall predictions but the data are at present collected at long time intervals. Digital Photogrammetry (DP) is cheaper and often more frequently available but known to be less accurate. This study investigates the potential of using contemporary DP data together with older ALS data and compares this with the case when only old ALS data are trained with recent field data. Combining ALS data from 2010 to 2011 with DP data from 2015, both trained with National Forest Inventory (NFI) field plot data from 2015, improved predictions of growing stock volume. Validation using data from 100 stands inventoried in 2015 gave an RMSE of 24.3% utilizing both old ALS data and recent DP data, 26.0% for old ALS only and 24.9% for recent DP only. If information about management actions were assumed available, combining old ALS and recent DP gave RMSE of 23.0%, only ALS 23.3% and only DP 23.8%.