IN31C-1146
A Matlab Program for Textural Classification Using Neural Networks
A new MATLAB code that provides tools to perform classification of textural images for applications in the Geosciences is presented. The program, here coined TEXTNN, comprises the computation of variogram maps in the frequency domain for specific lag distances in the neighborhood of a pixel. The result is then converted back to spatial domain, where directional or ominidirectional semivariograms are extracted. Feature vectors are built with textural information composed of the semivariance values at these lag distances and, moreover, with histogram measures of mean, standard deviation and weighted fill-ratio. This procedure is applied to a selected group of pixels or to all pixels in an image using a moving window. A feed- forward back-propagation Neural Network can then be designed and trained on feature vectors of predefined classes (training set). The training phase minimizes the mean-squared error on the training set. Additionally, at each iteration, the mean-squared error for every validation is assessed and a test set is evaluated. The program also calculates contingency matrices, global accuracy and kappa coefficient for the three data sets, allowing a quantitative appraisal of the predictive power of the Neural Network models. The interpreter is able to select the best model obtained from a k-fold cross-validation or to use a unique split-sample data set for classification of all pixels in a given textural image. The code is opened to the geoscientific community and is very flexible, allowing the experienced user to modify it as necessary. The performance of the algorithms and the end-user program were tested using synthetic images, orbital SAR (RADARSAT) imagery for oil seepage detection, and airborne, multi-polarimetric SAR imagery for geologic mapping. The overall results proved very promising.
IN31C-1147
AWIPS II+: An Open-Source SOA Solution Enabling Environmental Remote Sensing Integration, Analysis, and Decision Support
Our shared future demands a renewed focus on sound environment stewardship—on the GEOSS socioeconomic imperatives, as well as the interdisciplinary relationships interconnecting our environment, climate, ecosystems, energy, carbon, water—and national security. Data volumes are now measured in the many petabytes. An increasingly urgent and accelerated tempo of changing requirements and responsive solutions demands data exploitation, and transparent, seamless, effortless, bidirectional, and interdisciplinary interoperability across models and observations. There is today a robust working paradigm established with the Advanced Weather Interactive Processing System (AWIPS)—NOAA/NWS's information integration and fusion capability. This process model extends vertically, and seamlessly, from environmental sensing through the direct delivery of societal benefit. NWS, via AWIPS, is the primary source of weather forecast and warning information in the nation. AWIPS is the tested and proven "the nerve center of operations" at all 122 NWS Weather Forecast Offices and 13 River Forecast Centers. Raytheon, in partnership with NOAA, has now evolved AWIPS into an open-source 2nd generation capability to satisfy climate, ecosystems, weather, and water mission goals. Just as AWIPS II supports NOAA decision- making, it is at the same time a platform funded by Raytheon IRAD and Government investment that can be cost-effectively leveraged across all of the GEOSS and IEOS societal benefit areas. The core principles in the AWIPS II evolution to a service-oriented architecture (SOA) were to minimize coupling, increase cohesion, minimize size of code base, maximize simplicity, and incorporate a pull-style data flow. We focused on "ilities" to drive the new AWIPS architecture—our shared architecture framework vision included six elements: • Create a new, low-cost framework for hosting a full range of environmental services, including thick-client visualization via virtual Earth's and GIS • Scale down framework to a small laptop and through workstations to clusters of enterprise servers without software change • "Plug-n-play"— plug-ins can be hot deployable, or system cycled to pick up new plug-ins • Base the framework on highly reusable design patterns that maximize reuse and have datatype independence and fast adaptability • Open Source leveraged to maximize reuse • "Gaming-style" interaction with the data This talk addresses the challenges that we meet to realize benefits in applications that couple environmental data from many disparate remote sensing and ancillary sources and disciplines. By leveraging the existing AWIPS II weather, water, ecosystems, and climate functionality and these six elements, along with well- thought-out displays with the end user's specific needs in mind, we demonstrate an easily adapted, extremely powerful, open-source remote sensing software tool that will help non-geospatial-experts make better use of these remote sensing resources to enhance environmental mapping and analysis and help guide environmental decision making at the national, regional, local and citizen levels.
IN31C-1148
Landscape-Scale Soil Carbon Inventories by Microclimate Decomposition
Estimation of carbon stocks in rangeland and foothill ecosystems is poised to become an important service once legislation regulating greenhouse gas emissions is passed. Trading of carbon credits and greenhouse gas emission/sequestration budgets for vegetated areas is largely dependent on an accurate and scale- dependent inventory of existing conditions. Soil survey presents one possible resource for surface carbon stocks, however these data are usually not mapped at the landscape-scale. Soil-landscape modeling techniques have been successfully used in several instances to predict the spatial variation in soil carbon. Most of these studies have used site exposure (aspect angle) as a categorical proxy for terrain-induced microclimate. Our objective was to model parameters related to soil microclimate (soil temperature and moisture) for the production of detailed maps of soil carbon and organic matter quality (i.e. C:N ratio). We used a solar radiation model and long-term monitoring of soil moisture and temperature to generate several models of soil microclimate. Parameterization of the ESRA (European Solar Radiation Atlas) solar radiation model (clear-sky version) was accomplished with daily estimates of the Linke turbidity factor, using local pyranometer measurements (11 year record). Our estimated daily radiance values correlated well with local weather station data (R2 = 0.965, p < 0.001). This model is included in the popular, open source GRASS GIS. A preliminary study based on 35 sites, spanning two contrasting landform types (and lithology), revealed a statistically significant relationship between annual radiation load and carbon (R2 = 0.75, p < 0.001). A highly significant relationship between C:N ratio and annual radiation load was identified as well (R2 = 0.99, p < 0.001). Solar radiation models are simple to use, and have the potential to refine previous soil-landscape modeling efforts that relied on aspect class or angle. Models linking surface processes with microclimate can be used to directly generate estimates of carbon, or used to down-scale soil survey-based estimates.
IN31C-1149
The Wildland Fire Emissions Information System: Providing information for carbon cycle studies with open source geospatial tools
A major goal of the North American Carbon Program is to resolve uncertainties in understanding and managing the carbon cycle of North America. As carbon modeling tools become more comprehensive and spatially oriented, accurate datasets to spatially quantify carbon emissions from fire are needed, and these data resources need to be accessible to users for decision-making. Under a new NASA Carbon Cycle Science project, Drs. Nancy French and Tyler Erickson, of the Michigan Technological University, Michigan Tech Research Institute (MTRI), are teaming with specialists with the USDA Forest Service Fire and Environmental Research Applications (FERA) team to provide information for mapping fire-derived carbon emissions to users. The project focus includes development of a web-based system to provide spatially resolved fire emissions estimates for North America in a user-friendly environment. The web-based Decision Support System will be based on a variety of open source technologies. The Fuel Characteristic Classification System (FCCS) raster map of fuels and MODIS-derived burned area vector maps will be processed using the Geographic Data Abstraction Library (GDAL) and OGR Simple Features Library. Tabular and spatial project data will be stored in a PostgreSQL/PostGIS, a spatially enabled relational database server. The browser-based user interface will be created using the Django web page framework to allow user input for the decision support system. The OpenLayers mapping framework will be used to provide users with interactive maps within the browser. In addition, the data products will be made available in standard open data formats such as KML, to allow for easy integration into other spatial models and data systems.
IN31C-1150
Hyperspectral mapping and vulnerability modeling of effects of excessive overland flow on riparian arboreal ecosystems
The destruction of suitability of soil substrates to support riparian ecosystems due to periodic flooding, artificial or excessive water diversions, and overirrigation can last for decades and greatly affect biotic communities habiting these environments. Hyperspectral remote sensing technology with close to 1 m by 1 m pixel resolution and geographic information systems (GIS) offer a viable tool in the rapid analysis of the extent of biochemical, geochemical, and mineralogical changes that can occur due to excessive overland drainage within riparian zones. Hyperspectral data approximate continuous reflectance/emittance spectral measurements over a selected interval of the electromagnetic spectrum. With the advent of new and sophisticated digital sensors – with increased sensitivity – it has become possible to sample the reflection spectra of surficial materials. The interaction of low – pH waters, metals, and sulphate – contaminated water from agricultural practices initiates a sequence of pH-buffering reactions often accompanied by the precipitation of metal-bearing hydroxide and hydroxysulfate minerals that remove dissolved metals from moving water. This precipitation can be detected using hyperspectral imaging. Spectra can be examined for individual absorption features caused by specific chemical bonds in any solid, liquid, or gas. Limited geochemical and mineralogical data for some elements exist from other studies, however, there are no comparable libraries associated with biochemical signatures, a distinct indicator of mineralogical changes in soil composition. In this study we offer unique algorithms to identify and categorize biochemical, geochemical, and mineralogical spectra related to excessive overland drainage, a potential source of environmental problems within many agricultural districts. The common thematic map elements derived from the hyperspectral images are then incorporated into a GIS database. The reflection spectra of the soil substrates on the ground-as defined by image pixels-are in turn compared to laboratory and/or field-derived data. Classification is then based on the similarity of each pixel to a particular spectrum. Band ratioing or math may be done to discriminate potential spectral identities associated with commonly observed substrates in homogeneous patch of target vegetation, soil and water bodies. Geochemical and mineralogical spectral signatures are then determined from a statistical comparison of the reference spectra with the spectra of the pixels being compared with it. The resulting map is finally thresholded to achieve an acceptable confidence level. The imagery developed can then be modeled to determine the potential impact of excessive drainage on agricultural districts and/or related secondary effects due to mineral dissolution or precipitation.
IN31C-1151
Building Geospatial Web Services for Ecological Monitoring and Forecasting
The Terrestrial Observation and Prediction System (TOPS) at NASA Ames Research Center is a modeling system that generates a suite of gridded data products in near real-time that are designed to enhance management decisions related to floods, droughts, forest fires, human health, as well as crop, range, and forest production. While these data products introduce great possibilities for assisting management decisions and informing further research, realization of their full potential is complicated by their shear volume and by the need for a necessary infrastructure for remotely browsing, visualizing, and analyzing the data. In order to address these difficulties we have built an OGC-compliant WMS and WCS server based on an open source software stack that provides standardized access to our archive of data. This server is built using the open source Java library GeoTools which achieves efficient I/O and image rendering through Java Advanced Imaging. We developed spatio-temporal raster management capabilities using the PostGrid raster indexation engine. We provide visualization and browsing capabilities through a customized Ajax web interface derived from the kaMap project. This interface allows resource managers to quickly assess ecosystem conditions and identify significant trends and anomalies from within their web browser without the need to download source data or install special software. Our standardized web services also expose TOPS data to a range of potential clients, from web mapping applications to virtual globes and desktop GIS packages. However, support for managing the temporal dimension of our data is currently limited in existing software systems. Future work will attempt to overcome this shortcoming by building time-series visualization and analysis tools that can be integrated with existing geospatial software.
IN31C-1152
Fuel loading and consumption models for assessing carbon release from wildland fires
Total carbon released from wildland fires can be estimated from area burned, fuel loading, fuel consumption,
and pollutant specific emission factors. Two factors alone -- fuel loading and fuel consumption -- account for
up to 80 percent of the error associated with estimating total carbon. We show how ground-based estimates
of these factors can be extrapolated to the continental scale by combining satellite-based vegetation layers
with two software tools, the Fuel Characteristic Classification System (FCCS) and Consume 3.0. The FCCS
quantifies fuelbed layers, including canopy, shrubs, nonwoody, woody, litter/lichen/moss, and duff, via an
explicit linkage to vegetation types. FCCS fuelbeds represent the United States, including Alaska and
Hawai'i, Canada, and a portion of Mexico. We present a fuelbed classification for the continental US
(CONUS) at 1-km resolution, and show how this can be extended to cover North America. Fuel loadings are
calculated from this classification and MODIS vegetation layers, including LAI (Leaf-Area Index) and VCF
(Vegetation Continuous Fields). At each cell we calculate fuel consumption and emissions using Consume
3.0, which imports fuelbed data directly from the FCCS. We embed the models in a decision-support system
to estimate both real-time consumption and carbon release from specific fire events and potential values
given specific parameters for fire weather or fire intensity. Ground-based estimates of fuel loading and
consumption complement and potentially validate remotely sensed estimates, thereby enhancing both real-
time decision support and continental-scale modeling.
http://www.fs.fed.us/pnw/fera
IN31C-1153
Mapping Deforestation In Parts Of The Amazon Forest Using JERS-1 SAR Images: Case Study For An Area North West Of Rondonia, Brazil
Deforestation in the Amazon rain forests, especially around Rondonia, is well known and documented in literature. Time series analyses of deforestation from 1975 through 2001 using medium resolution images from the Landsat satellite are available. However, as the Brazilian rain forests are often masked with thick cloud cover, the use of satellite images acquired in the visible and infrared region for deforestation studies poses a serious limitation, as the cloud free scenes for a season in a particular year are few or not available at all. Processing images acquired by JERS-1 from 1993 through 1996, for an area just northwest of the town of Rondonia, Brazil, we demonstrated the potential of SAR images to successfully map deforested areas and to serve as a complimentary data source to optical images for monitoring deforestation activities. An important objective of our study was to devise a simple image processing scheme that could be used by novice users without the need for expensive commercially available software packages. Image pre-processing included importing, georeferencing, and subsetting all images to bring them to a common format, projection and coverage. Image processing involved classification using unsupervised maximum likelihood classifier and by density slicing (grey scale thresholding). The classified product, which showed a strong "salt pepper effect" due to high frequency variations, was smoothed out using a Gamma MAP filter of different kernel sizes to generate the final forest cover map. Simpler low pass filters were also tested. We also tested the effect of image filtering prior to image classification. Amount of deforestation was computed by taking the numeric difference in the number of pixels classified as forest and multiplying this difference by the pixel size. Processing results showed that unsupervised classification using maximum likelihood classifier gave results slightly superior to a simple density slicing. Image filtering prior to classification and post classification gave very comparable results. This simple technique was ineffective in areas where the terrain had high topographic variations, but it worked well in flat terrains.
IN31C-1154
Environmental Remote Sensing Analysis Using Open Source Virtual Earths and Public Domain Imagery
Human activities increasingly impact natural environments. Globally, many ecosystems are stressed to
unhealthy limits, leading to loss of valuable ecosystem services- economic, ecologic and intrinsic. Virtual
earths (virtual globes) (e.g., NASA World Wind, ossimPlanet, ArcGIS Explorer, Google Earth, Microsoft Virtual
Earth) are geospatial data integration tools that can aid our efforts to understand and protect the
environment. Virtual earths provide unprecedented desktop views of our planet, not only to professional
scientists, but also to citizen scientists, students, environmental stewards, decision makers, urban developers
and planners. Anyone with a broadband internet connection can explore the planet virtually, due in large
part to freely available open source software and public domain imagery. This has at least two important
potential benefits. One, individuals can study the planet from the visually intuitive perspective of the synoptic
aerial view, promoting environmental awareness and stewardship. Two, it opens up the possibility of
harnessing the in situ knowledge and observations of citizen scientists familiar with landscape conditions in
their locales. Could this collective knowledge be harnessed (crowd sourcing) to validate and quality assure
land cover and other maps? In this presentation we present examples using public domain imagery and two
open source virtual earths to highlight some of the functionalities currently available. OssimPlanet is used to
view aerial data from the USDA Geospatial Data Gateway. NASA World Wind is used to extract
georeferenced high resolution USGS urban area orthoimagery. ArcGIS Explorer is used to demonstrate an
example of image analysis using web processing services. The research presented here was conducted
under the Environmental Feature Finder project of the Environmental Protection Agency's Advanced
Monitoring Initiative.
Although this work was reviewed by EPA and approved for publication, it may not necessarily reflect official
Agency policy. Use of trade names does not imply endorsement by the authors or the EPA.
http://www.foss4g2007.org/presentations/view.php?abstract_id=257
IN31C-1155
MapReady: An Open Source Tool for the Utilization of SAR in Geospatial Applications
Users of remote sensing data can now benefit from the wide availability of Synthetic Aperture Radar (SAR) satellites, including ERS-2, RADARSAT-2, ALOS PALSAR, Envisat, and TerraSAR-X. As an active sensor, SAR can acquire data independently of weather and at any time of day or night. Unfortunately, SAR data has not seen wide-spread usage by those engaged in mapping or Earth studies. The reason for this is two-fold: 1) the data comes in a format that most geospatial tools cannot ingest, and 2) SAR imagery is subject to geometric distortions that keep it from being co-registered with more conventional imagery. The Alaska Satellite Facility (ASF) has developed the free, open source MapReady Remote Sensing Tool Kit to facilitate the use of SAR data for even novice users of geospatial data. Through MapReady's intuitive GUI interface, the user is able to ingest a SAR image in its native format and process it to an orthorectified image in GeoTIFF format; ready to be used as a layer in a geographic information system (GIS). This presentation will outline the challenges facing the user of SAR and show how they are overcome through the use of MapReady. The principal innovation in the newest generation of SAR satellites is the implementation of polarimetric SAR, for which data exists in two or more polarizations. Like the bands in optical data, polarimetric bands reveal a great deal about targets in the imagery. In its most recent version, MapReady includes the ability to perform polarimetric decompositions and classifications. Specific examples will be shown using polarimetric data from the ALOS PALSAR sensor. Derived products will be analyzed and interpreted to show how SAR polarimetry can be used to perform land classification and identify land change.