©2023 International Society of Arboriculture 42 classification threshold determined prior, trees were classified as either infested or noninfested. The class predictions were tallied against actual classification of trees in the test data set. A confusion matrix was constructed, and classification accuracy was deter- mined based on the number of correct classifications. Ranking Accuracy for Infestation Level of Surveyed Areas Based on feedback from managers of the surveyed areas, the levels of infestation were considered low for GEC and PR, moderate for SL, and high for ECP and BSA. Using the developed BIS formula, average BIS values for each area were obtained, ranked, and compared to the general level of infestation based on feedback from area managers. Sensitivity to Temporal Progression of Infestation BIS values for the selected 48 infested trees along the ECP area were calculated for July and October. Vio- lin plots for BIS values in July 2020 and October 2020 were generated using the ggplot function from the ggbiplot package. Temporal changes in BIS mean value between July and October were tested using the built-in aov function. RESULTS Index Construction Linearity, Normality, and Outliers As seen from 15 scatter plots between variable pairs (Figure 2), there was no nonlinear correlation detect- able between any variable pair. The Shapiro-Wilk multivariate normality test had a P-value = 3.752 × 10−12 leading to rejection of null hypothesis. The data set did not have normal distribu- tion. Consequently, ICA was performed and the result was compared with PCA. As seen from Figure 3a, the 3D plot of PCA based on the top 3 principal compo- nents indicated that infested and noninfested rain trees were clearly separated along PC1 with most of the data variation also explained along this compo- nent. Meanwhile, the 3D plot of ICA (Figure 3b) also found similar distribution of points since infested and noninfested trees clearly separated along ICA1. Since ICA was designed to separate independent compo- nents from multivariate non-Gaussian data, more clusters detected in the 3D ICA plot were expected. For both 3D plots obtained from PCA and ICA, PC2, PC3, ICA2, and ICA3 did not give clear separation between infested and noninfested trees. Thus, with (b) 25% of original data via random selection. Two subsets (a) were combined to form the training data set, while two subsets (b) were combined to form the testing data set. For the training data set, PCA with scaling was done on recorded variables except for infestation status using the in-built prcomp function. Location of trees was not included in PCA. Visualiza- tion of dimensions and calculation of their corre- sponding eigenvalues and variances were done using the fviz_eig function from the factoextra package. Using the pca3d function from the pca3d package, 3D visualization of PCA score plots for the first 3 components was performed. Dimension Selection and Stopping Rules To determine the number of statistically significant principal components, we employed functions within the PCDimension package to perform a Pseudo-F ratio test (Ter Braak 1990), an eigenvalues P-value test (Ter Braak 1988), a broken stick statistical test (Barton and David 1956), and an Auer-Gervini method (Auer and Gervini 2008). Pseudo-F ratio and eigenvalues P-value tests were applied using the rndLambdaF function for 1,000 iterations at 0.05 significance level. The bsDimension function was used to perform the broken stick test. The AuerGervini function was used to perform the Auer-Gervini method with twicemean, kmean, spectral clustering, and changepoint criteria. Borer Infestation Score (BIS) Statistically significant principal components were identified. Linear combinations of variables along these principal components were determined to derive the formula to calculate the scores for each tree, which are defined as their Borer Infestation Score (BIS). Distribution of BIS values for all trees in the training data set were calculated and plotted using the ggden- sity function from the ggpubr package for infested and noninfested trees. The intersection between BIS density plots of infested and noninfested trees was calculated using the in-built intersect function. The value of this intersection was used as the classifica- tion threshold to distinguish between infested and noninfested trees. Index Validation Classification Accuracy Between Infested and Noninfested Trees The developed BIS formula was applied to calculate BIS values for trees in the test data set. Using the Nguyen Hoang Danh et al: Visual Assessment Method for Lebbek Borer
January 2023
| Title Name |
Pages |
Delete |
Url |
| Empty |
Search Text Block
Page #page_num
#doc_title
Hi $receivername|$receiveremail,
$sendername|$senderemail wrote these comments for you:
$message
$sendername|$senderemail would like for you to view the following digital edition.
Please click on the page below to be directed to the digital edition:
$thumbnail$pagenum
$link$pagenum
Your form submission was a success. You will be contacted by Washington Gas with follow-up information regarding your request.
This process might take longer please wait
