TY - JOUR

T1 - Classification of multilayered forest development classes from low-density national airborne lidar datasets

AU - Valbuena, Ruben

AU - Maltamo, Matti

AU - Packalen, Petteri

PY - 2016/8

Y1 - 2016/8

N2 - Compared with traditional inventory methods based on field plot sampling, airborne laser scanning (ALS) has the potential to assess forest structural properties with greater detail over space and time and at lower cost. Many national ALS survey programmes covering entire countries for topographic mapping are currently in progress and some have provided data that are in the public domain. Although the point density of these datasets is relatively low, there is an interest in developing methods that employ these types of data for categorizing different approaches to forest management. Using Finnish national ALS data with a point density of 0.91 pulses per square metre, we carried out a classification of 252 000 ha of boreal forests into silvicultural development classes (DC) used in practical forest management. Taking into account all eight DCs, the overall accuracy was 74.1 per cent and κ = 0.70. We conclude that the dataset is adequate for discriminating multilayered forests from even-aged ones. This result was compared with a method based on mathematical rules, which succeeded in discriminating multilayered stands with regeneration of shade-intolerant species without the need of field data for training. However, the low point density may hamper the detection of shade-tolerant understories in mature high forests with closed canopies. We, therefore, recommend the use of this supervised classification in the presence of shade-tolerant species.

AB - Compared with traditional inventory methods based on field plot sampling, airborne laser scanning (ALS) has the potential to assess forest structural properties with greater detail over space and time and at lower cost. Many national ALS survey programmes covering entire countries for topographic mapping are currently in progress and some have provided data that are in the public domain. Although the point density of these datasets is relatively low, there is an interest in developing methods that employ these types of data for categorizing different approaches to forest management. Using Finnish national ALS data with a point density of 0.91 pulses per square metre, we carried out a classification of 252 000 ha of boreal forests into silvicultural development classes (DC) used in practical forest management. Taking into account all eight DCs, the overall accuracy was 74.1 per cent and κ = 0.70. We conclude that the dataset is adequate for discriminating multilayered forests from even-aged ones. This result was compared with a method based on mathematical rules, which succeeded in discriminating multilayered stands with regeneration of shade-intolerant species without the need of field data for training. However, the low point density may hamper the detection of shade-tolerant understories in mature high forests with closed canopies. We, therefore, recommend the use of this supervised classification in the presence of shade-tolerant species.

KW - airborne laser scanning

KW - forest structure

KW - stand development

KW - supervised classification

KW - support vector machine

U2 - 10.1093/forestry/cpw010

DO - 10.1093/forestry/cpw010

M3 - Erthygl

VL - 89

SP - 392

EP - 401

JO - Forestry

JF - Forestry

SN - 0015-752X

IS - 4

ER -