Algorithms for precision forestry inventories based on terrestrial point clouds

Resum

Sustainable forest management helps mitigate climate change and supports a crucial economic sector. Optimizing forest management practice facilitates sustainability, but requires quality information on available resources. Traditional inventories based on periodic field sampling of diameter at breast height and total height and statistical generalization, do not meet current information needs for sustainable multipurpose management. This doctoral thesis focuses on providing new tools for precision forestry inventories through developing algorithms applicable to terrestrial point clouds, which facilitates obtaining accurate geometric representations of trees in a reasonable time. Moreover, such measurements avoid the observer subjectivity of traditional forest inventory methods. In addition, 3D TLS point clouds can be processed automatically, or semi-automatically, by means of mathematical algorithms. Specifically, we developed algorithms which provide automatic solutions for three key issues in forest management: volume equation parametrization, stem shape variables estimation and optimal bucking pattern determination. They constitute a staged workflow in the sense that each algorithm builds on the previous one. The first calculates the basic variables that characterize trees geometrically: (i) diameters of the sections along the stem; (ii) coordinates of the centre of the section (XYZ) and (ii) total height of the tree. Diameters along the stem and height are later used to parametrize wood volume equations. The second algorithm estimates stem shape variables by splitting the stem into evenly spaced sections whose diameters and centres are calculated and used as input variables to automatically calculate the maximum sagitta, sinuosity, and lean of each tree. The third algorithm determines the optimal bucking pattern of each stem, maximizing each tree's economic value in terms of several timber products. It is based on the three-dimensional modelling of stems and includes the diameter and curvature of each log. Algorithm performance was tested in various plots and the advantages and disadvantages of each was compared to traditional techniques. Validation tests for Algorithm 1 (wood volume equation parametrization) were carried out in a Pinus pinaster plot with steep slopes, low branches and dense understory, and 97% of trees were automatically detected and RMSE of the height and diameter estimations was 1.52 m and 1.14 cm, respectively. A volume ratio equation was automatically selected as the best option for the test dataset. Root mean square error (RMSE) in automatic volume estimations was 0.0233 m3 and 0.0149 m3 when diameters were previously reviewed by an operator and anomalous sections redrawn. Algorithm 2 (stem shape variables estimation) was tested in a breeding trial plot of Pinus pinaster, and the results obtained compared with field measurements of straightness and lean based on visual classification. The methodology was robust to errors in estimating section centres, the basis for calculating shape parameters. Besides, its accuracy compared favourably with traditional field techniques, where misclassification is frequent. Testing of Algorithm 3 (optimal bucking of stems) was in a Pinus radiata plot of 120 trees, and the results compared with those obtained with input data that only consider diameters estimated from TLS measurements and taper equations, not take curvature into account. Using TLS and including curvature measurements, provided a more realistic optimal bucking solution and tended to result in lower estimations of commercial value. All three algorithms have a direct application in forest planning, supporting the use of TLS in precision forest inventories. They are completely automatic, aim to be applicable to any species or terrestrial point cloud type and eliminate subjectivity in field measurements and the estimation of variables. Although all show promising results, the systematic investigation of larger test sites with different scanning techniques and forest conditions is desirable.