Jill S. Heaton

Applying tree-based ensemble algorithms to the classification of ecological zones using multi-temporal multi-source remote-sensing data

The decision tree method has grown fast in the past two decades and its performance in classification is promising. The tree-based ensemble algorithms have been used to improve the performance of an individual tree. In this study, we compared four basic ensemble methods, that is, bagging tree, random forest, AdaBoost tree and AdaBoost random tree in terms of the tree size, ensemble size, band selection (BS), random feature selection, classification accuracy and efficiency in ecological zone classification in Clark County, Nevada, through multi-temporal multi-source remote-sensing data. Furthermore, two BS schemes based on feature importance of the bagging tree and AdaBoost tree were also considered and compared. We conclude that random forest or AdaBoost random tree can achieve accuracies at least as high as bagging tree or AdaBoost tree with higher efficiency; and although bagging tree and random forest can be more efficient, AdaBoost tree and AdaBoost random tree can provide a significantly higher accuracy. All ensemble methods provided significantly higher accuracies than the single decision tree. Finally, our results showed that the classification accuracy could increase dramatically by combining multi-temporal and multi-source data set.

Detection and classification of invasive saltcedar through high spatial resolution airborne hyperspectral imagery

We evaluated the performance of airborne HyperSpecTIR (HST) images for detecting and classifying the invasive riparian vegetation saltcedar along the Muddy River in Clark County, Nevada. HyperSpecTIR image reflectance spectra (227 bands, 450–2450 nm) were acquired for the following four vegetation covers: invasive saltcedar, native honey mesquite, grassland patches and crops. We compared five feature reduction approaches: band selection based on Jeffreys–Matusita distance, principal component analysis (PCA), minimum noise fraction (MNF), segmented principal component transform (SPCT) and segmented minimum noise fraction (SMNF). In addition, maximum likelihood (ML) and two spectral angle mapper (SAM) classifiers were applied to all extracted bands or features. Classification accuracies were compared among all classification approaches. Although the overall accuracy of maximal likelihood classifiers generally surpassed that of SAM classifiers, the highest overall accuracy was achieved by a SMNF-SAM combination with adjusted angular thresholds for classes. We concluded that high spectral and spatial resolution imagery can be used to detect and classify invasive saltcedar in this arid area.

Spatially explicit decision support for selecting translocation areas for Mojave desert tortoises

Spatially explicit decision support systems are assuming an increasing role in natural resource and conservation management. In order for these systems to be successful, however, they must address real-world management problems with input from both the scientific and management communities. The National Training Center at Fort Irwin, California, has expanded its training area, encroaching U.S. Fish and Wildlife Service critical habitat set aside for the Mojave desert tortoise (Gopherus agassizii), a federally threatened species. Of all the mitigation measures proposed to offset expansion, the most challenging to implement was the selection of areas most feasible for tortoise translocation. We developed an objective, open, scientifically defensible spatially explicit decision support system to evaluate translocation potential within the Western Mojave Recovery Unit for tortoise populations under imminent threat from military expansion. Using up to a total of 10 biological, anthropogenic, and/or logistical criteria, seven alternative translocation scenarios were developed. The final translocation model was a consensus model between the seven scenarios. Within the final model, six potential translocation areas were identified.

A comparison of random forest and Adaboost tree in ecosystem classification in east Mojave Desert

We compared two basic ensemble methods, namely random forest and Adaboost tree for the classification of ecosystems in Clark County, Nevada, USA through multitemporal multisource LANDSAT TM/ETM+ images and terrain-related GIS data layers. Random forest generates decision trees by randomly selecting a limited number features from all available features for node splitting, and each tree cast a vote for the final decision. On the other hand, Adaboost tree is an iterative approach to improve the performance of a weak classifier by assigning weights to training samples, and incorrectly classified training samples will gain a larger weight in the process. We discuss the properties of these two tree-based ensemble methods and compare their classification performances in ecosystem classification. The results show that Adaboost tree can provide higher classification accuracy than random forest in multitemporal multisource dataset, while the latter could be more efficient in computation.

COMPARISON OF EFFECTS OF HUMANS VERSUS WILDLIFE-DETECTOR DOGS

Abstract The use of dogs (Canis lupus familiaris) trained to locate wildlife under natural conditions may increase the risk of attracting potential predators or alter behavior of target species. These potentially negative effects become even more problematic when dealing with threatened or endangered species, such as the Mojave Desert tortoise (Gopherus agassizii). We addressed three concerns regarding use of dogs trained to locate desert tortoises in the wild. First, we looked at the potential for dogs to attract native and non-native predators to sites at a greater rate than with human visitation alone by comparing presence of predator sign before and after visitation by dogs and by humans. We found no significant difference in predator sign based upon type of surveyor. Second, we looked at the difference in risk of predation to desert tortoises that were located in the wild by humans versus humans with wildlife-detector dogs. Over a 5-week period, during which tortoises were extensively monitored and a subsequent period of 1 year during which tortoises were monitored monthly, there was no predation on, nor sign of predator-inflicted trauma to tortoises initially encountered either by humans or wildlife-detector dogs. Third, we looked at movement patterns of tortoises after encounter by either humans or wildlife-detector dogs. Movement of desert tortoises was not significantly different after being found by a human versus being found by a wildlife-detector dog. Based upon these initial results we conclude that use of trained wildlife-detector dogs to survey for desert tortoises in the wild does not appear to increase attraction of predators, increase risk of predation, or alter movement patterns of desert tortoises more than surveys conducted by humans alone.