Barnali M. Dixon

Multispectral landuse classification using neural networks and support vector machines: one or the other, or both?

Land use classification is an important part of many remote sensing applications. A lot of research has gone into the application of statistical and neural network classifiers to remote‐sensing images. This research involves the study and implementation of a new pattern recognition technique introduced within the framework of statistical learning theory called Support Vector Machines (SVMs), and its application to remote‐sensing image classification. Standard classifiers such as Artificial Neural Network (ANN) need a number of training samples that exponentially increase with the dimension of the input feature space. With a limited number of training samples, the classification rate thus decreases as the dimensionality increases. SVMs are independent of the dimensionality of feature space as the main idea behind this classification technique is to separate the classes with a surface that maximizes the margin between them, using boundary pixels to create the decision surface. Results from SVMs are compared with traditional Maximum Likelihood Classification (MLC) and an ANN classifier. The findings suggest that the ANN and SVM classifiers perform better than the traditional MLC. The SVM and the ANN show comparable results. However, accuracy is dependent on factors such as the number of hidden nodes (in the case of ANN) and kernel parameters (in the case of SVM). The training time taken by the SVM is several magnitudes less.

Prediction of Aquifer Vulnerability to Pesticides using Fuzzy Rule-Based Models at the Regional Scale

Contamination of ground water has been a major environmental concern in recent years. The potential for ground-water contamination by pesticides depends on porous media, solute, and hydrologic parameters. Although sophisticated deterministic computer models are available for assessing aquifer-contamination potential on a site-by-site basis, most deterministic models are too complex for vulnerability assessment on a regional scale because they require input data that are spatially and temporally variable, and which may not be available at this scale. Therefore, development of an affordable model that is robust under conditions of uncertainty at the watershed scale with minimum input of field data becomes a useful ground-water management tool. The purpose of this study was to examine the usefulness of fuzzy rule-based techniques in predicting aquifer vulnerability to pesticides at the regional scale. The objectives were to (1) develop fuzzy rule-based models using the same input parameters contained in an index-based model (i.e., the modified DRASTIC model), (2) determine the sensitivity of fuzzy rule model predictions, (3) compare the outputs of the fuzzy rule-based models with those of the modified DRASTIC model and with the results of aquifer water-quality analyses, and (4) examine the spatial variability of field parameters around contaminated wells of the Alluvial aquifer in Woodruff County Arkansas. The fuzzy rule-based model for objective (1) was developed using similar parameter weights and ratings as the modified DRASTIC model. For objective (2), fuzzy rule-based models were created using fewer parameters than the modified DRASTIC model. Sensitivity of the fuzzy rule-based models was determined using different combinations of weights of the four input parameters in DRASTIC. It was found that variations in the weights of the input parameters and number of fuzzy sets influenced the location of the aquifer-vulnerability categories as well as the area within each fuzzy category. The fuzzy rule models tended to predict somewhat higher vulnerabilities of the Alluvial aquifer than the modified DRASTIC model. The fuzzy rule base that had the soil-leaching index (S) as the highest weight was chosen as the best fuzzy rule model in predicting potential contamination by pesticides of the aquifer. In general, the fuzzy rule models tended to overestimate the vulnerability of the aquifer in the study area.

Application of Support Vector Machines for Landuse Classification Using High-Resolution RapidEye Images: A Sensitivity Analysis

Abstract The classification accuracy of remotely sensed data and its sensitivity to classification algorithms have a critical importance for the geospatial community, as classified images provide the base layers for many applications and models. Support Vector Machines (SVMs), a non-parametric statistical learning method that has recently been used in numerous applications in image processing. The SVMs need user-defined parameters and each parameter has different impact on kernels hence the classification accuracy of SVMs is based upon the choice of the parameters and kernels. The objective of this study is to investigate the sensitivity of SVM architecture including internal parameters and kernel types on landuse classification accuracy of RapidEye imagery for the study area in Turkey. Four types of kernels (linear, polynomial, radial basis function, and sigmoid) were used for the SVM classification. A total of 63 different models were developed and implemented for sensitivity analysis of SVM architecture. The traditional Maximum Likelihood Classification (MLC) method was also performed for comparison. The classification accuracies of the best model for each kernel type and MLC are 85.63%, 83.94%, 83.94%, 83.82% and 81.64% for polynomial, linear, radial basis function, sigmoid kernels and MLC, respectively. The results suggest that the choice of model parameters and kernel types play an important role on SVMs classification accuracy. Best model of polynomial kernel outperformed all SVMs models and gave the highest classification accuracy of 85.63% with RapidEye imagery.

Alternative spatially enhanced integrative techniques for mapping seagrass in Florida's marine ecosystem

Seagrass is an important component of coastal marine ecosystems. Seagrass mapping provides a means for assessing seagrass health by monitoring the spatial distribution and density of seagrass habitat in coastal waters. Recent image processing and satellite technologies present the opportunity to leverage quantitative techniques that have the potential to improve upon traditional photo-interpretation techniques in terms of cost, mapping fidelity, and objectivity. Integrated spatial and spectral processing techniques were identified as an alternative method for mapping seagrass extent and density from an IKONOS satellite image of Springs Coast, Florida. These spatially enhanced integrative mapping techniques objectively standardize seagrass-monitoring efforts and enhance mapping capabilities by characterizing spatial seagrass density gradients. A combination of water column correction, pixel classification, and image segmentation techniques provided a seagrass density index map that represented seagrass density and distribution with high spatial detail and overall accuracy (77%) comparable to photo-interpretation techniques. Satellite imagery-based spatially enhanced image processing techniques were found to provide a consistent, quantitative, and cost-effective alternative for seagrass mapping in Springs Coast with the potential to be transferred to other parts of the world. A cost savings analysis concluded that there was a 13% cost saving using satellite photo-interpretation and a 47% cost saving using enhanced satellite classification when compared to aerial photo-interpretation.

Assessing intrinsic and specific vulnerability models ability to indicate groundwater vulnerability to groups of similar pesticides: A comparative study.

Abstract With continued population growth and increasing use of fresh groundwater resources, protection of this valuable resource is critical. A cost effective means to assess risk of groundwater contamination potential will provide a useful tool to protect these resources. Integrating geospatial methods offers a means to quantify the risk of contaminant potential in cost effective and spatially explicit ways. This research was designed to compare the ability of intrinsic (DRASTIC) and specific (Attenuation Factor; AF) vulnerability models to indicate groundwater vulnerability areas by comparing model results to the presence of pesticides from groundwater sample datasets. A logistic regression was used to assess the relationship between the environmental variables and the presence or absence of pesticides within regions of varying vulnerability. According to the DRASTIC model, more than 20% of the study area is very highly vulnerable. Approximately 30% is very highly vulnerable according to the AF model. When groundwater concentrations of individual pesticides were compared to model predictions, the results were mixed. Model predictability improved when concentrations of the group of similar pesticides were compared to model results. Compared to the DRASTIC model, the AF model more accurately predicts the distribution of the number of contaminated wells within each vulnerability class.

Combining Data Collection from Unmanned Surface Vehicles with Geospatial Analysis: Tools for Improving Surface Water Sampling, Monitoring, and Assessment

Tidal rivers and accompanying coastal environments represent critical links between the open estuary and the local tributary and watershed. In Florida, these coastal ecosystems are multi-use systems. They are often a primary source of water for agricultural, industrial, and human consumption all while functioning as commercial and recreational shipping thoroughfares and receiving storm water runoff and NPDES discharges. In addition to their importance for those direct human uses, their quality and characteristics directly affect the spawning, nursery, and juvenile habitats for numerous commercial and sport fisheries. Thus monitoring and assessment, especially identifying spatial patterns or trends in water chemistry (e.g. temperature, conductivity, salinity, turbidity, chlorophyll, dissolved organic matter and dissolved gasses), of these tidal environments can represent a complex sampling and analysis challenge. There is a perception that sampling and analyzing parameters at regularly spaced intervals over the surface area of a river system will be representative of general trends. However, standard sampling strategy assumes both that parameters will change in a consistent longitudinal and downstream manner and that the average of a parameter is the level where negative impacts occur. Using an innovative combination of unmanned surface vehicles (US V) and geospatial analytical techniques, we will show that this perception is not an entirely accurate or complete view.

Application Of Neuro-fuzzy Techniques To Predict Ground Water Vulnerability

There is a need to develop new modeling techniques that assess ground water vulnerability with less expensive data and robust when data are uncertain and incomplete. The specific objectives of this research were to (i) loosely couple Neuro-fuzzy techniques and GIS to predict ground water vulnerability in a relatively large watershed, (ii) examine the sensitivity of the Neuro-fuzzy models by changing training parameters, and (3) determine the effects of the size of the training data sets on model predictions, The Neuro-firzzy models were developed in a JAVA platform using four plausible parameters that are critical in transporting contaminants in and through the soil profile. The models were validated using nitrate-N contamination data. Neuro-fkzy approaches were sensitive to training parameters and scale/size of the training data set. The proposed methodology has potential in facilitating ground water vulnerability modeling at a regional scale, but would require incorporation of appropriate input parameters suitable for the region,

Estimating soil loss from two coastal watersheds in Puerto Rico with RUSLE

The loss of topsoil from the landscape increases sediment loading to adjacent aquatic systems, which may be transported further downstream. Increased sediment loading to downstream ecosystems such as estuaries can negatively impact these environments. This research compares the sediment loss from two watersheds [Jobos Bay (JB) and Rio Espiritu Santo (RES) watersheds] on the island of Puerto Rico. The revised universal soil loss equation (RUSLE) was applied to both watersheds. Results from this study suggested that potential soil loss from both watersheds is relatively low. There is some indication that the RES watershed may have a higher soil loss potential when compared to the JB watershed.

Using the fractal dimension to differentiate between natural and artificial wetlands

Artificial wetlands are characterized by straight lines and simple perimeters such as circles or squares, whereas natural wetlands show far more complex shapes. Fractal dimension analysis provides a quantitative measure of the curves for the edge of an object. This study uses fractal theory to analyze the characteristic of an objects shape to differentiate natural wetlands from artificial wetlands. The objectives of this study were to: 1) determine how the shape complexity metrics varies between raster and vector formats and 2) if there is a quantifiable difference between patch metrics of the fractal dimension of natural vs. man–made wetlands.

Examining spatio–temporal relationships of landuse change, population growth and water quality in the SWFWMD

There are many pressures on Floridas water resources. Industrial, agricultural and urban development over the years has impacted water quality adversely. Deterioration of groundwater quality, a major source of fresh water, is a major concern for long–term sustainable growth. The study area, Southwest Florida Water Management District (SWFWMD) of Florida contains one of the nations fastest growing metropolitan areas. Although landuse changes as a result of population growth is inevitable, it is not too late to try to understand the relationship among landuse change, population growth and environmental dynamics. A thorough understanding of the population growth, landuse change and environmental dynamics is necessary for managing the urban sprawl with minimal environmental impact viz. their impact on water quality and quantity. The objectives of this study were: to explore if there is a spatial relationship exists among NO3 and Bromacil contamination and critical physical/environmental variables and 2) to create a dataset that will be the basis for future study. This was accomplished by studying landuse (1988 and 1999), population (1990s and 2000s), soils, groundwater quality data (1990 and 2000), Floridan Aquifer Vulnerability Assessment (FAVA) and Digital Elevation Models (DEMs). Preliminary results show that contaminated wells were associated with urban and agricultural landuse and sandy soils with high permeability. No significant relationship between population and groundwater quality exists for NO3 and Bromacil contaminated wells.