Spatial Tools for Managing the Hemlock Woolly Adelgid in the Southern Appalachians

adelgid;HWA;hemlock;decision tree classification;topographic normalization;forest pest;forest health;infestation risk

Koch, Frank Henry Jr. ; Heather Cheshire, Committee Chair,Hugh Devine, Committee Member,Fred Hain, Committee Member,George Hess, Committee Member,Koch, Frank Henry Jr. ; Heather Cheshire ; Committee Chair ; Hugh Devine ; Committee Member ; Fred Hain ; Committee Member ; George Hess ; Committee Member

University:North Carolina State University

关键词: adelgid;    HWA;    hemlock;    decision tree classification;    topographic normalization;    forest pest;    forest health;    infestation risk;   

Others  :  https://repository.lib.ncsu.edu/bitstream/handle/1840.16/5322/etd.pdf?sequence=1&isAllowed=y

美国|英语

来源: null

【 摘 要 】

Native to Asia, the hemlock woolly adelgid (Adelges tsugae) has recently spread into parts of the southern Appalachian region.This insect pest attacks both native hemlock species (Tsuga canadensis and T. caroliniana), has no natural enemies, and can kill hemlock trees within just a few years.While biological control displays promise for combating the pest, such counter-measures are significantly hampered because neither adelgid nor hemlock distribution patterns have been detailed explicitly.We developed a spatial management system to better target control efforts.The system has two components: (1) a protocol for mapping hemlock stands, and (2) a technique to map areas at risk of imminent hemlock woolly adelgid infestation.To map hemlock stands, we utilized topographically normalized satellite imagery from Great Smoky Mountains National Park.Because hemlocks are difficult to distinguish using just satellite data, we constructed a decision tree classifier that supplemented the imagery with a suite of topographic, environmental, and proximity variables.We then implemented the classifier in a geographic information system and generated hemlock distribution maps.Our final decision tree had 27 terminal nodes and nine variables, with elevation, image band ratios, topographic relative moisture index, and distance to the closest stream among the most important variables.Accuracy assessment—based on field data and aerial photos—of the maps resulting from this tree yielded an overall thematic accuracy of 90% for one study area and 75% accuracy in capturing hemlocks in a second study area.To map areas at risk, we combined known first-year infestation locations from Great Smoky Mountains National Park and the Blue Ridge Parkway with points from uninfested hemlock stands, recording a suite of environmental variables for each point.We applied four different techniques (discriminant analysis, k-nearest neighbor, logistic regression, and decision tree) to generate models from these data in order to predict locations at high risk of imminent hemlock woolly adelgid infestation.We then used the resulting models to generate risk maps of the study region.All techniques performed well, accurately capturing 70-90% of training and validation samples.Discriminant analysis was the most accurate technique, but logistic regression yielded a more practical map from a management standpoint, with large, discrete risk zones.In any case, our results suggest that roads, major trails, and riparian corridors provide an important degree of connectivity enabling long-distance dispersal of the hemlock woolly adelgid, probably by humans or birds.Both components of our hemlock woolly adelgid management system are built on readily available or easily calculable spatial data.Furthermore, they are constructed generally enough that they should be applicable throughout the southern Appalachians.Overlay of derived maps will allow forest managers to prioritize hemlock stands and allocate resources more efficiently.