Abstract Detail

Biodiversity Informatics & Herbarium Digitization

Kao, Jenni [1], Huang, Yi [1], Morrison, Glen R [2], Litt, Amy [3].

Predicting future Arctostaphylos species distribution using machine learning.

Arctostaphylos species, or manzanitas, are drought-resistant shrubs that are distributed throughout the diverse California Floristic Province (CFP). Although some species are widespread, most of the 60 species have distributions that are narrowly restricted to a specific region. Many of these are considered rare or threatened. Manzanitas typically live in the chaparral, a habitat known to have hot, dry summers and frequent fires. As climate change increases drought and fires, species already threatened, and many that currently are not, may be put at more risk. This project provides an in-depth analysis from a bioinformatics perspective to predict future ranges of Arctostaphylos species using models of climate change and species distributions. These efforts can be used to evaluate future risks to the species and to direct conservation for those endangered. We will concentrate on three narrow endemic species, A. otayensis, A. cruzensis, and A. purissima to represent different regions of California. We will use locality data from the Consortium of California Herbaria (CCH2), a online database of specimen data, including geographic locality, from California herbaria. Since each species occupies different biogeographic regions, we expect they will have various environmental and climate associations. We will use geographic coordinate data from the specimens in the CCH2 database to extract climate data from the WorldClim database to find the temperature and precipitation range of each species. We will model future distributions using the future climate data modeling by the WorldClim database. To predict future distributions and evaluate their likelihood, we will use the Random Forest algorithm. Whereas other methods predict distributions but are not able to evaluate their likelihood, this algorithm is advantageous because it uses part of the data to model the distribution and part of the data to evaluate the probability. It does this by creating and evaluating a series of decision trees that successively improve the probability of correct predictions. Based on how accurate the estimated distribution is, these results can aid in conservation management for manzanitas and other species that may be at risk due to climate change.

1 - University of California, Riverside, Botany and Plant Sciences, Riverside, CA, 92521, USA
2 - University of California, Riverside, 900 University Ave, Riverside, CA, 92521, United States
3 - University Of California Riverside, Botany Dept, 900 University Ave, Riverside, CA, 92521, United States

Keywords:
manzanita
climate change
California Floristic Province
Machine Learning
conservation.

Presentation Type: Oral Paper
Number: BI&HD I007
Abstract ID:1014
Candidate for Awards:None