The goal of the project was twofold: (1) to introduce a major innovation by developing a robust and accurate methodology for tree species mapping in operational forestry, and (2) to assess whether a more accurate tree species information would improve the yield predictions accuracy.

Project management: The project contains work packages (WP): project management (WP1), data processing (WP2), method development and assessment (WP3) and dissemination (WP4).

Material and Methods. In WP2, we successfully developed an workflow for automatizing the main data processing steps that combines ‘in-situ’ harvester data and GIS products such as multi-temporal Sentinel-2 imagery, forest state estimates (Skogliga grunddata provided by the Swedish Forest Agency) and height vegetation models derived the Swedish government agency for mapping and land registration (Lantmäteriet). The datasets were acquired over a study area of 53,384 ha across Västernorrland and Jämtland regions, and it is managed by Svenska Cellulosa Aktiebolaget (SCA). The main challenge in WP2 was related to dealing with varying data quality due to cloud occlusions in the satellite imagery.

WP3 was focused on two main activities. The first activity was dedicated to method development or predicting the tree species proportions (relative to total standing volume) using the WP2 outputs. The three main groups of species considered were Spruce, Pine and Deciduous, which are of special interest for the wood industry. In addition, a computational-efficient routine was developed for combining multi-temporal datasets for minimizing the data losses due to the cloud occlusions in satellite imagery. The second activity was focused on assessing the value of tree species information on yield predictions. The assessments were run under three scenarios: (A) baseline case with no species information; (B) using tree species predictions, and (C) using ‘perfect’ tree species information (the ‘ground-truth’ data). Species information was incorporated as auxiliary data in the imputation routines Skogforsk, either as proportions of standing volume per hectare, or as indicator data (0-1) for the dominant species. The yield calculations were preformed using stem price lists for the main tree species were compiled by the Skogforsk specialists using industry data from 2020 that are valid for our study area.

Project dissemination (WP4). A work report will be hosted on Skogforsk’s website followed by an article in Skogforsk’s magazine Vision. Two research articles based on the extended summary (attached) are planned for submission, the first submission being expected for 2020. The attendance to international conferences was limited in 2020 due to cancellations related to COVID-19 (such as the biannual ForestSAT conference). Participation to conferences is therefore sought for 2021.

Main results and discussion. The proposed workflow is robust, relies exclusively on open-source software and can be easily scaled-up for dealing with larger study areas. Besides, it allows for a better use of ‘ground-truth’ data in the forest areas covered by clouds. The results show ca. 100% reduction of the absolute yield losses for Spruce and Pine and ca. 20% reduction for Deciduous species when the predicted tree species information was incorporated into the imputations. Overall, the findings demonstrate that reliable information on species proportion are vital for increasing the yield prediction accuracy. With regard to the effects for the forest industry, the results should be seen as a lower bound since the added value of tree species information is expected to compound further down in the wood supply chain

Senaste anslag