Forest cover typing and species identification are critical to both forest conservation managers and forestry companies interested in their supply inventory. Forest cover typing can consist of reconnaissance mapping over a large area, while species inventories are highly detailed measurements of stand contents and characteristics (tree type, height, density). Timely and accurate information on tree species is crucial for developing strategies for sustainable management and conservation of artificial and natural forests.
In combination with machine learning algorithms, very high-resolution satellite imagery provided an effective solution to reduce the need for labor-intensive and time-consuming field-based surveys.
Tree species spectra, combined with multispectral, radar and DEM data used in machine learning algorithms to produce 10-m resolution tree identification for forestry applications. Axiom’s advanced methods/algorithms/models produced higher classification and mapping accuracies compared to those created with other traditional methods for the Government of Saskatchewan. Machine learning methods including deep learning models were demonstrated to be significant in improving tree species classification accuracy, leading to a 98% accuracy rate. The application was both a cost-effective and scalable solution.