Remote sensing offers many advantages to previous forest measurements, such as limiting costs and time in the field. Light detection and ranging (lidar) has been shown to enable accurate estimates of forest height. Lidar does produce precise measurements for ground elevation and forest height, where and when it is available. However, it is expensive to collect and does not have wall-to-wall coverage in the United States. In this study, we estimated height using the National Agricultural Imagery Program (NAIP) photogrammetric point clouds to create a predicted height map for managed loblolly pine stands in the southeastern United States. Recent studies have investigated the ability of digital aerial photogrammetry (DAP), and more specifically NAIP, as an alternative to lidar as a means of estimating forest height due to its lower costs, frequency of acquisition, and wall-to-wall coverage across the United States. Field-collected canopy height for 534 plots in Virginia and North Carolina were regressed against the 90th percentile derived from NAIP point clouds. The model for predicted pine height using the 90th percentile of height (P90) is predicted pine height = 1.09(P90) – 0.43. The adjusted R^2 is 0.93, and the RMSE is 1.44 m. This model is being used to produce a 5 m x 5 m canopy height model for all pine stands across Virginia, North Carolina, and Tennessee. NAIP-derived point clouds are thus a viable means of predicting canopy height in southern pines. / M.S. / Collecting accurate measurements of pine plantations is essential to managing them to maximize various ecosystem goods and services. However, it can be difficult and time-consuming to collect these measurements by hand. This study demonstrates that point clouds derived from digital stereo aerial photograms enable calculating forest height to an accuracy sufficient for pine plantation management. We developed a simple linear regression model to predict forest canopy height using the 90th percentile of the photo-derived heights above the ground in a given area. With this model, we created a map of pine plantation canopy heights (consisting of 5 m x 5 m grid cells, each containing a canopy height estimate) for pine forests in Virginia, North Carolina, and Tennessee. Digital aerial photography from the National Agricultural Imagery Program (NAIP) is repeated every three years for a given state, allowing growth to be mapped over time. Photography collected by NAIP and similar programs also has uniform acquisition parameters in a given year applicable over large regions. State- and national photography programs like NAIP are also less expensive than other data sets, like airborne laser scanning data, that enable estimation of tree height.

Understanding the dynamic interplay between fire severity, topography, and tree mortality, is crucial for predicting future forest dynamics and enhancing resilience against climate change-induced wildfire regimes. This thesis develops a multi-sensor approach for automated estimation of tree mortality, then applies it to examine trends in tree mortality over a six-year period across a fire affected study site in the Trinity River basin in Northern California. The Random Forest model uses publicly available USGS 3D Elevation Program Lidar (3DEP) and NAIP imagery as inputs and is likely to be easily adaptable to other landscapes. The model had a Receiver Operating Characteristic Area Under the Curve (ROC AUC) score in training of 0.998. In multiple rounds of validation, using geographically distinct sets of holdout data, had mean accuracy of 0.998. The trained model was then used to assess tree mortality across a patchwork of different levels of burn severity at a site in Northern California. When applied to the study site significant variations were found in tree mortality across different fire severity treatments and landforms. This model shows potential for incorporation into predictive tree mortality models based on landform and climate.

Greater sage-grouse populations have decreased steadily since European settlement in western North America. Reduced availability of brood-rearing habitat has been identified as a limiting factor for many populations. We used radio-telemetry to acquire locations of sage-grouse broods from 1998 to 2012 in Strawberry Valley, Utah. Using these locations and remotely-sensed imagery, we proceeded to 1) determine which features of brood-rearing habitat could be identified using widely available, fine-scale imagery 2) assess the scale at which sage-grouse selected brood-rearing habitat in our study area, and 3) create a predictive habitat model that could be applied across our large study area to identify areas of preferred brood-rearing habitat. We used AIC model selection to evaluate support for a list of variables derived from remotely-sensed imagery. We examined the relationship of explanatory variables at three scales (45, 200, and 795 meter radii). Our top model included 10 variables (percent shrub, percent grass, percent tree, percent paved road, percent riparian, meters of sage/tree edge, meters of riparian/tree edge, distance to tree, distance to transmission lines, and distance to permanent structures). Variables from each scale were represented in our top model with the majority of scale-sensitive variables suggesting selection at the larger (795 meter) scale. When applied to our study area our top model predicted 75% of naive brood locations suggesting reasonable success using this method and widely available NAIP (National Agricultural Imagery Program) imagery. We encourage application of this method to other sage-grouse populations and species of conservation concern. The northern river otter is a cryptic semi-aquatic predator that establishes and uses latrines. Highly used river otter latrines indicate otter "activity centers" since frequency of scat deposition is thought to be correlated to frequency of habitat use. We compared an indirect method (scat counts) and a direct method (remote cameras) of determining latrine utilization in order to assess the accuracy of the commonly used indirect method. To further compare these methods we used them to examine effects of anthropogenic disturbance on otters of the Provo River in Utah. We found that overall the direct and indirect methods were highly correlated. There was significant seasonal variation in the degree of correlation between the indirect and direct methods with correlation being significantly higher in the summer. We found similar results when using these methods to examine effects of anthropogenic disturbance. For each method the distance of the latrine to trails was significant in one of the top competing models. We suggest that space use of otters in our study area is being affected by anthropogenic disturbance as measured by distance to trails. We also suggest that scat counts should only be conducted during the summer when they correlate best with actual levels of otter activity.

No description available.

No description available.

This research delves into the application of semantic segmentation in precision agriculture, specifically targeting the automated identification and classification of various irrigation system types within agricultural landscapes using high-resolution aerial imagery. With irrigated agriculture occupying a substantial portion of US land and constituting a major freshwater user, the study's background highlights the critical need for precise water-use estimates in the face of evolving environmental challenges, the study utilizes advanced computer vision for optimal system identification. The outcomes contribute to effective water management, sustainable resource utilization, and informed decision-making for farmers and policymakers, with broader implications for environmental monitoring and land-use planning. In this geospatial evaluation research, we tackle the challenge of intraclass variability and a limited dataset. The research problem centers around optimizing the accuracy in geospatial analyses, particularly when confronted with intricate intraclass variations and constraints posed by a limited dataset. Introducing a novel approach termed "dynamic contrastive learning," this research refines the existing contrastive learning framework. Tailored modifications aim to improve the model's accuracy in classifying and segmenting geographic features accurately. Various deep learning models, including EfficientNetV2L, EfficientNetB7, ConvNeXtXLarge, ResNet-50, and ResNet-101, serve as backbones to assess their performance in the geospatial context. The data used for evaluation consists of high-resolution aerial imagery from the National Agriculture Imagery Program (NAIP) captured in 2015. It includes four bands (red, green, blue, and near-infrared) with a 1-meter ground sampling distance. The dataset covers diverse landscapes in Lonoke County, USA, and is annotated for various irrigation system types. The dataset encompasses diverse geographic features, including urban, agricultural, and natural landscapes, providing a representative and challenging scenario for model assessment. The experimental results underscore the efficacy of the modified contrastive learning approach in mitigating intraclass variability and improving performance metrics. The proposed method achieves an average accuracy of 96.7%, a BER of 0.05, and an mIoU of 88.4%, surpassing the capabilities of existing contrastive learning methods. This research contributes a valuable solution to the specific challenges posed by intraclass variability and limited datasets in the realm of geospatial feature classification. Furthermore, the investigation extends to prominent deep learning architectures such as Segformer, Swin Transformer, Convexnext, and Convolution Vision Transformer, shedding light on their impact on geospatial image analysis. ConvNeXtXLarge emerges as a robust backbone, demonstrating remarkable accuracy (96.02%), minimal BER (0.06), and a high MIOU (85.99%).