Reliability of tomographic image to represent variation in stem wood properties assessed using confusion matrix metrics

The intrinsic benefits of the urban greenery are undeniable, but the lack of planning of the urban forestry has created a great liability for its population. To guarantee environmental preservation with risk reduction, one must invest in research that evaluates the accuracy of technological tools that have been used in tree inspection, the results of which are used for the decision-making on urban tree management. One of these tools is ultrasonic tomography, frequently used to infer the internal condition of the stem. This research aimed to evaluate, using metrics (accuracy, precision, and sensitivity) of the confusion matrix, the quality of the image generated by tomography using conventional ultrasound equipment. For the study, disks of the species Cenostigma pluviosum (Sibipiruna) were used. They were analyzed considering different regions (heartwood and sapwood) and conditions (deterioration and cavities). Masks representative of these different regions and conditions of the wood were compared to tomographic images. In addition, average values ​​of density and stiffness obtained in each region and condition were also used to validate the representation indicated in the tomographic image. Although ultrasonic tomography does not represent exactly the position and size of the internal regions of the trunk (cavities, deteriorations, heartwood, and sapwood), the accuracy (greater than 85%) and coherence with density and stiffness were sufficient to indicate its suitability as support for tree risk assessment.