The Sentinel-2 program provides the opportunity to monitor terrestrial ecosystems with a high temporal and spectral resolution. In this study, a multi-temporal Sentinel-2 data set was used to classify common tree species over a mature forest in central Sweden. The tree species to be classified were Norway spruce (Picea abies), Scots pine (Pinus silvestris), Hybrid larch (Larix × marschlinsii), Birch (Betula sp.) and Pedunculate oak (Quercus robur). Four Sentinel-2 images from spring (7 April and 27 May), summer (9 July) and fall (19 October) of 2017 were used along with the Random Forest (RF) classifier. A variable selection approach was implemented to find fewer and uncorrelated bands resulting in the best model for tree species identification. The final model resulting in the highest overall accuracy (88.2%) came from using all bands from the four image dates. The single image that gave the most accurate classification result (80.5%) was the late spring image (27 May); the 27 May image was always included in subsequent image combinations that gave the highest overall accuracy. The five tree species were classified with a user’s accuracy ranging from 70.9% to 95.6%. Thirteen of the 40 bands were selected in a variable selection procedure and resulted in a model with only slightly lower accuracy (86.3%) than that using all bands. Among the highest ranked bands were the red edge bands 2 and 3 as well as the narrow NIR (near-infrared) band 8a, all from the 27 May image, and SWIR (short-wave infrared) bands from all four image dates. This study shows that the red-edge bands and SWIR bands from Sentinel-2 are of importance, and confirms that spring and/or fall images capturing phenological differences between the species are most useful to tree species classification.

Precision forestry allows decision-making on tree level or pixel level, as compared to stand-level data. However, little is known about the importance of precision in thinning decisions and its long-term effects on within-stand variation, stand economy and growth. In this study, silviculture was optimized for Net Present Value (NPV) in 20 conifer-dominated forest stands in hemi-boreal southern Sweden. The precision-thinning approach, Precision Thinning (PT), is compared with a stand-level approach, Stand Level Thinning (SLT) that is optimized for the same criteria but based on stand-level data. The results suggest no substantial long-term benefit or drawback in implementing thinning decisions based on pixel-level data as compared to stand-level data when optimizing stand economy. The result variables NPV and Mean annual increment of living stem volume (MAI_net) were not higher for PT than for SLT. The within-stand variation in basal area (m²/ha⁻¹) was lower at the end of the rotation compared to the start of the simulation for both SLT and PT. At the end of the rotation, SLT had higher variation in basal area compared to PT. However, pixel-level information enables adapting the silviculture to the within-stand variation which may favour other forest management goals than strictly financial goals.

Most of our knowledge about wood production of Scots pine (Pinus sylvestris L.) and Norway spruce (Picea abies (L.) Karst.) and associated silvicultural guidelines are based on field experiments. These have been established in rather small, homogenous stands. In practical forestry there is probably a comparatively larger gradient in within-stand variation due to varying site conditions and less controlled silviculture than in experiments. The extent of the within-stand variation in coniferous production stands and how thinning guides are used in relation to the within-stand variation, is not well understood. Also, the freely available Forest resource maps (sv. Skogliga grunddata) and satellite data offer the possibility to accounts for the within-stand variation in forest management, but this is also poorly researched. This thesis evaluates within-stand variation at first thinning: its extent, its effect on silviculture and its importance for future stand development. Additionally, optical satellite data from Sentinel-2 is used to detect thinning operations, estimate growth after thinning and classify tree species. The thesis is mainly based on a survey carried out in the fall of 2018 in planted conifer-dominated production stands planned for first commercial thinning in which the thinning method of the forest workers was observed. The survey was inventoried directly after thinning and three growing seasons later. The survey showed an unprecedented within-stand variation before thinning in stem volume, stem density, dominant height, mean height quadratic mean diameter and basal area. The thinning operations did not reduce the within-stand variation in any of the attributes measured with the relative standard deviation. The stands were thinned heavily, and the harvested basal area increased with basal area before thinning at sample plot level, which suggest an ambition to reduce the variation. The stands were also monitored using Sentinel-2 satellite data. The thinning detection model separated unthinned, lightly thinned and heavily thinned sample plots with a moderate overall accuracy of 62% (Kappa of 0.34). A set of satellite images over the whole observation period was used estimate the periodical annual volume increment after thinning and did so with a root mean squared error (RMSE) of 1.8 m3 ha-1 y-1 (relative RMSE: 24%). The long-term effects of optimizing the thinning regime on pixel level versus conventional stand-level thinning was evaluated using the Heureka system. No benefits in terms of stand economy or production was found, but the within-stand variation in basal area decreased over the rotation. Tree-species classification, rendering maps with the dominant tree species at pixel level over a forest holding, were made using multi-temporal Sentinel-2 satellite data and the Random Forest classifier. The major tree species in the forest holding were Scots pine, Norway spruce, Pedunculate oak (Quercus robur), Birch (Betula spp.) and Hybrid larch (Larix × marschlinsii). These species were classified with a high overall accuracy of 88.2% (Kappa of 0.82). This thesis illustrates that considerable within-stand variation could be expected before and after first thinning for coniferous dominated stands in southern Sweden. The average stand basal area after thinning was consistently lower than the required basal area in the thinning guides from the Swedish Forest Agency, which means that reduced total production over the rotation may be a result. The increasing harvested basal area with basal area before thinning, suggests an ambition to reduce the within-stand variation in basal area. Thinning at the pixel level by adapting the thinning regime to the within-stand variation did not have any long-term effects on stand economy or volume production compared to conventional stand-level thinning. Despite the non-significant results, high-resolution maps are probably needed anyway to support forest workers in thinning operations to avoid heavy thinning. The Sentinel-2 satellite data proved its relevance for practical forestry for thinning detection, assessing growth after thinning, and classifying tree species. These methods can be used in combination the already existing Forest resource maps to reduce uncertainties for the management of planted forest.