Land of 10,000 pixels: applications of remote sensing & geospatial data to improve forest management in northern Minnesota, USA
Date
2018
Authors
Engelstad, Peder, author
Falkowski, Michael, advisor
Lefsky, Michael, committee member
Paschke, Mark, committee member
Journal Title
Journal ISSN
Volume Title
Abstract
The use of remote sensing and geospatial data has become commonplace in a wide variety of ecological applications. However, the utility of these applications is often limited by field sampling design or the constraints on spatial resolution inherent in remote sensing technology. Because land managers require map products that more accurately reflect habitat composition at local, operational levels there is a need to overcome these limitations and improve upon currently available data products. This study addresses this need through two unique applications demonstrating the ability of remote sensing to enhance operational forest management at local scales. In the first chapter, remote sensing products were evaluated to improve upon regional estimates of the spatial configuration, extent, and distribution of black ash from forest inventory and analysis (FIA) survey data. To do this, spectral and topographic indices, as well as ancillary geospatial data were combined with FIA survey information in a non-parametric modeling framework to predict the presence and absence of black ash dominated stands in northern Minnesota, USA. The final model produced low error rates (Overall: 14.5%, Presence: 14.3%, Absence: 14.6%; AUC: 0.92) and was strongly informed by an optimized set of predictors related to soil saturation and seasonal growth patterns. The model allowed the production of accurate, fine-scale presence/absence maps of black ash stand dominance that can ultimately be used in support of invasive species risk management. In the second chapter, metrics from low-density LiDAR were evaluated for improving upon estimates of forest canopy attributes traditionally accessed through the LANDFIRE program. To do this, LiDAR metrics were combined with a Landsat time-series derived canopy cover layer in random forest k-nearest neighbor imputation approach to estimate canopy bulk density, two measures of canopy base height, and stand age across the Boundary Waters Canoe Area in northern Minnesota, USA. These models produced strong relationships between the estimates of canopy fuel attributes and field-based data for stand age (R2 = 0.82, RMSE = 10.12 years), crown fuel base height (R2 = 0.79, RMSE = 1.10 m.), live crown base height (R2 = 0.71, RMSE 1.60 m.), and canopy bulk density (R2 = 0.58, RMSE 0.09 kg/m3). An additional standard randomForest model of canopy height was less successful (R2 = 0.33, RMSE 2.08 m). The map products generated from these models improve upon the accuracy of national available canopy fuel products and provide local forest managers with cost-efficient and operationally ready data required to simulate fire behavior and support management efforts.
Description
Rights Access
Subject
canopy fuels
Landsat
random forest
emerald ash borer
black ash
LiDAR