Browse Theses

The US forest products industry is concentrated in the southeastern US (SEUS) and primarily based on the production of loblolly pine (Pinus taeda L.). This region produces greater than half of the nation's timber supply and is the single largest producer in the world, on only approximately 14.4 - 15.8 million hectares of pine plantations. This intensity of production is due in large part to technological advancements made in silviculture over the last 70 years, aided by the use of data products from remote sensing systems and rapid increases in computational power, whereby the fusion of enormous spatial datasets able to inform the dynamics of various ecological processes—such as highly detailed soil products which are mapped across the entire nation at very fine spatial resolutions—are only recently able to be utilized to their full capacity in a logistically practical manner. In these studies, we establish some of the links necessary to prepare forest industry to face many challenges looming in the horizon including remotely sensed biophysical indicators, predictions of response to fertilization, and variable rate fertilizer applications. Leaf area index (LAI)—which is linearly related to productivity and a key criterion for potential nutrient management—was modeled at a 10-meter spatial resolution based on the linear relationship between surface reflectance measurements captured from the Sentinel-2 Multispectral Instrument (MSI), a passive multispectral satellite constellation orbiting Earth, and field measurements collected with a LAI-2200C plant canopy analyzer. Support vector machine (SVM)-supervised classification was used to improve model fit. Results from this study indicated that Sentinel-2's improved spatial and temporal resolutions provided new opportunities to detect within-stand variance with improved accuracy for LAI estimation over other industry-standard remote sensing models at the time. In the following study, we fused historical and contemporary field study data collected from 51 study sites and 5 different Regionwide study trials spanning the entire SEUS to build predictive models using a variation of random forest, a common machine learning technique, that implemented conditional inference, or unbiased recursive partitioning. We modeled and quantified the influence of 69 specific soil variables (and their interactions) on height, basal area, and volume in 1) baseline productivity (no fertilizer applied) and 2) cumulative growth response in the eight years following a fertilizer application greater than 112 kg ha−1 elemental nitrogen and 26 kg ha-1 elemental phosphorus; each model consistently performed well with low margins of error. We rendered predictive soil maps from remotely-sensed classifications of intensively managed pine plantations, revealing potential region-wide growth response to industry-standard application rates of nitrogen and phosphorus to estimate potential carbon storage in approximately 13.8 million km of the SEUS. We estimated approximate eight-year totals ranging from 1.88 billion m3 of baseline volume productivity to 2.38 billion m3 with maximum nutrient additions. Total projected amounts of stored carbon dioxide equivalent were 5.24 billion tonnes CO2e at baseline, and 6.63 billion tonnes CO2e when applied to our maximum fertilization model. Finally we applied precision forestry in an operational loblolly pine plantation context, establishing a methodology for transitioning from traditional stand-level forest management to optimized treatment zones in which forest managers may make optimized nutrient management decisions. Through variable-rate application from fixed-wing aircraft, we showed that thresholds could be created based on remotely sensed LAI measurements, and our analyses exhibited strong statistical evidence that increases in LAI can be detected and attributed to increases in as-applied rates as soon as one growing season post-fertilization.