Impact of model choice in predicting urban forest storm damage when data is uncertain

• DOI: 10.1016/j.landurbplan.2022.104467
• Publication Date: 2022-05-18T09:09:57Z
• Authors: Casey Lambert; Shawn Landry; Michael G. Andreu; Andrew Koeser; Gregory Starr; Christina Staudha...
• Keywords: Explanatory power; Basal area

Research that illuminates causes of urban forest storm damage is valuable for planning and management. However, logistical safety concerns often delay post-storm surveys in areas; thus, may include observations with unverified sources damage. While this uncertainty ignored, it can make up a high proportion the number damaged trees. The goal research was to improve understanding techniques modeling forests. Using inventories collected Florida, post-Hurricane Irma (2017), we tested how different imputation methods, procedures, frequency levels could impact overall model results. We utilized machine learning algorithms Random Forest (RF) k-Nearest Neighbors (KNN), generalized linear models (GLM). found GLM RF gave unbiased predictions across all methods rarity levels, while KNN consistently under-predicted Damage influenced some measures performance but did not variable significance. Imputation identified consistent variables most significance within each procedure; however, there variation among ranked moderately important. both plot tree basal area as highly significant predictors, they otherwise disagreed on individual importance. These findings suggest explanatory be achieved by examining relationships, more complex relationships such ones have equal power subsets predictor variables.