Open Collections

Innovative and conventional modelling for avian acoustic and fire severity analyses

University of British Columbia

Data science, the principles that support extracting knowledge from data, has become increasingly important in natural resources management. This thesis applied both machine learning (innovative modelling) and statistics (conventional modelling) to address two ecological questions. In Chapter 2, Automatic bird sound detection: logistic regression based acoustic occupancy model, logistic models and convolutional neural networks were applied to predict bird presence/absence in audio recordings, in order to improve efficiency in analyzing large audio datasets. The acoustic recordings came from a bird sound detection challenge organized by the Institute of Electrical and Electronics Engineers (IEEE) and covered bird songs and calls in a wide range of environments along with the presence of noise. Based on leave-one-out cross-validation, the final logistic model resulted in an overall accuracy of 75% with a false negative rate of 16%. Compared with a convolutional neural network (CNN) model using the same dataset, the logistic model was about seven times faster in terms of processing time. This bird sound detection model using sound frequency percentiles in a logistic model opens up promising approaches to aid in automatic, accurate, and efficient analysis of large audio datasets for monitoring wildlife communities. In Chapter 3, Previous fire severity enhances reburn severity: a case study in interior British Columbia, Canada, an ordinal logistic model was applied to investigate how previous fires influenced the reburn severity in interior British Columbia, Canada, in order to determine the driver of reburn severity. Previous fires affect rates of fuel consumption and accumulation, thus influencing the probability and severity of subsequent fires. In this study, forest stand structural change due to the first fire (in 2009 or 2010), such as changes in basal area and trees per hectare, were used to model the severity of the reburn in 2017. The ordinal model indicated a positive relationship between fire severities in the Riske Creek area. Specifically, fires in the Riske Creek area might not be able to limit the probability or severity of a reburn after seven or eight years.

Text

2019-11-26

Attribution-NonCommercial-NoDerivatives 4.0 International

http://hdl.handle.net/2429/72418

Forestry

University of British Columbia

UBCV

http://creativecommons.org/licenses/by-nc-nd/4.0/