Detection of the need for seedling stand tending using high-resolution remote sensing data

Abstract

Detection of the need for seedling stand tending using high-resolution remote sensing data Korhonen L., Pippuri I., Packalén P., Heikkinen V., Maltamo M., Heikkilä J. (2013). Detection of the need for seedling stand tending using high-resolution remote sensing data. Silva Fennica vol. 47 no. 2 article id 952. 20 p. AbstractSeedling stands are problematic in airborne laser scanning (ALS) based stand level forest management inventories, as the stem density and species proportions are difficult to estimate accurately using only remotely sensed data. Thus the seedling stands must still be checked in the field, which results in an increase in costs. In this study we tested an approach where ALS data and aerial images are used to directly classify the seedling stands into two categories: those that involve tending within the next five years and those which involve no tending. Standard ALS-based height and density features, together with texture and spectral features calculated from aerial images, were used as inputs to two classifiers: logistic regression and the support vector machine (SVM). The classifiers were trained using 208 seedling plots whose tending need was estimated by a local forestry expert. The classification was validated on 68 separate seedling stands. In the training data, the logistic model's kappa coefficient was 0.55 and overall accuracy (OA) 77%. The SVM did slightly better with a kappa = 0.71 and an OA = 86%. In the stand level validation data, the performance decreased for both the logistic model (kappa = 0.38, OA = 71%) and the SVM (kappa = 0.37, OA = 72%). Thus our approach cannot totally replace the field checks. However, in considering the stands where the logistic model predictions had high reliability, the number of misclassifications reduced drastically. The SVM however, was not as good at recognizing reliable cases.

Remote sensing and forest inventories in Nordic countries – roadmap for the future

“…in Finland. However, the topic is under active research (Naesset and Bjerknes 2001;Korpela et al 2008;Korhonen et al 2013;Ørka et al 2016). Diameter distribution information does not have a high value for non-industrial private forest owners in the harvest decisions (Saad et al 2015), but it can have high value in price negotiations.…”

Improving capacity for large-area monitoring of forest disturbance and recovery

“…ALS metrics used in the analysis included the mean and standard deviation of ALS heights, ALS height percentiles (1%, 5%, 10%, 25%, 50%, 75%, 90%, 95%, 99%; referenced as p01, p05, p10, etcetera), and the percentage of ALS returns within specified height intervals relative to the total number of returns (1,2,3,4,5,6,7,8,9,10,15,20, and 25 m; referenced as d00 for 0-1 m, d01 for 1-2 m, etcetera). Following on the recommendations of Ørka et al (2016) and Korhonen et al (2013), who studied the use of ALS data to characterize regenerating forests, no minimum height threshold (e.g. 2 m) was applied, and only first returns were used when calculating ALS metrics.…”

Towards an enhanced understanding of airborne LiDAR measurements of forest vegetation

“…Some studies have been published that utilize DR LiDAR of approximately the same pulse density in a similar environment (Naesset and Bjerknes 2001;Korpela et al 2008;Korhonen et al 2013a), and some of these studies utilized passive optical data in combination with LiDAR. In addition, there are studies with very high density LiDAR (e.g., paper IV in Nyström 2014), but to my knowledge there are very few studies that have used WF LiDAR.…”

“…As previously mentioned, WF data was able to capture the footprint-scale characteristics of vegetation structure (echo width, energy to amplitude ratio) and also reflectance properties (energy). In Finnish silviculture, the characterization of seedling and sapling stand vegetation is important for determining the pre-commercial treatments of young forests Korhonen et al 2013a). The results of study III are also more widely applicable and indicate the general potential of WF data in the classification of low vegetation types (Vierling et al 2012).…”

“…Treatment needs should therefore perhaps be interpreted directly from LiDAR predictor features, without first predicting the species composition or other stand attributes. On the other hand, treatment needs are not always assessed unambiguously even in the field (Korhonen et al 2013a), which poses challenges for developing RS solutions.…”