In the last decade, advances in sensor and computing technology are revolutionary. The latest-generation of hyperspectral and synthetic aperture radar ((SAR) instruments have increased their spectral, spatial, and temporal resolution. Consequently, the data sets collected are increasing rapidly in size and frequency of acquisition. Remote sensing applications are requiring more computing resources for data analysis. High performance computing (HPC) infrastructure such as clusters, distributed networks, grids, clouds and specialized hardware components, have been used to disseminate large volumes of remote sensing data and to accelerate the computational speed in processing raw images and extracting information from remote sensing data. In previous research we have shown that we can improve computational efficiency of a hyperspectral image denoising algorithm by parallelizing the algorithm utilizing a distributed computing grid. In recent years, computational cloud technology is emerging, bringing more flexibility and simplicity for data processing. Hadoop MapReduce is a software framework for distributed commodity computing clusters, allowing parallel processing of massive datasets. In this project, we implement a software application to map forest aboveground biomass (AGB) with normalized difference vegetation indices (NDVI) using Landsat Thematic Mapper’s bands 4 and 5 (ND45). We present observations and experimental results on the performance and the algorithmic complexity of the implementation. There are three research questions answered in this thesis, as follows. 1) How do we implement remote sensing algorithms, such as forest AGB mapping, in a computer cloud environment? 2) What are the requirements to implement distributed processing of remote sensing images using the cloud programming model? 3) What is the performance increase for large area remote sensing image processing in a cloud environment? / Graduate / 0799 / 0984

<p>Land use and land cover changes significantly affect C storage in terrestrial ecosystems. Programs intended to compensate land owners for the maintenance or enhancement to C stocks are promising, but require detailed and spatially explicit C distribution estimates to monitor the effectiveness of management interventions. Savanna ecosystems are significant components of the global C cycle, however, they have not received much attention for the development of C monitoring approaches. In this dissertation I have investigated three of the aspects related to woody plant cover dynamics in semiarid savannas of central Argentina: spatio-temporal dynamics, precise field surveying and scaling from field to region with the use of freely available remotely sense data. </p><p>To examine the long term changes in woody plant cover, I first carefully extracted information from historical maps of the Caldenal savannas of central Argentina (190,000 km2) in the 1880s to generate a woody cover map that was compared to a 2000s dataset. Over the last ~120 years, woody cover increased across ~12,200 km2 (14.2 % of the area). During the same period, ~5,000 km2 of the original woody area was converted to croplands and ~7,000 km2 to pastures, about the same total land area as was affected by woody plant encroachment. A smaller area, fine scale analysis between the 1960s and the 2000s revealed that tree cover increased overall by 27%, shifting from open savannas to a mosaic of dense woodlands along with additional agricultural clearings. Statistical models indicate that woody cover dynamics in this region were affected by a combination of environmental and human factors.</p><p>To assess the consequences of woody cover dynamics on C, we also measured ecosystem C stocks along a gradient of woody plant density. I characterized changes in C stocks in live biomass (woody and herbaceous, above- and belowground), litter, and soil organic carbon (to 1.5 m depth) pools along a woody plant cover gradient (0 to 94 %). I found a significant increase in ecosystem C stocks with increasing woody cover, with mean values of 4.5, 8.4, 12.4, and 16.5 kg C m-2 for grasslands, shrublands, open and closed forests, respectively. Woody plant cover and soil silt content were the two primary factors accounting for the variability of ecosystem C. I developed simple regression models that reliably predict soil, tree and ecosystem C stocks from basic field measurements of woody plant cover and soil silt content. These models are valuable tools for broad scale estimation if linked to regional soil maps and remotely sensed data, allowing for precise and spatially explicit estimation of C stocks and change at regional scales.</p><p>Finally, I used the field survey data and high resolution panchromatic images (2.5 m resolution) to identify tree canopies and train a regional tree percent cover model using the Random Forests (RF) algorithm. I found that a model with summer and winter tasseled cap spectral indices, climate and topography performed best. Sample spatial distribution highly affected the performance of the RF models. The regression model built to predict tree C stocks from percent tree cover explained 83 % of the variability, and the spatially explicit tree C model prediction presented an root mean squared error (RMSE) of 8.2 tC/ha which represented ~30% of the mean C stock for areas with tree cover. Our analysis indicates that regionally over the last ~120 years, increases in woody plant cover have stored significant amounts of C (95.9 TgC), but not enough to compensate for in C generated by the conversions of woodlands and natural grasslands to croplands and pastures (166.7 TgC), generating a regional net loss of 70.9 TgC. C losses could be even larger in the future if, as predicted, energy crops would trigger a new land cover change phase in this region.</p> / Dissertation