While the availability of multispectral imagery (MSI) is increasing, there is still much to be learned regarding the spectral sensitivity of deep neural networks (DNNs). This work presents and analyzes DNNs trained to perform object detection in high-resolution multispectral, satellite imagery. These DNNs are first trained and evaluated on variations of the xView dataset with varied bit-depths and spectral band inclusion. Inference is also performed on test imagery with different bit-depths or with missing spectral information in order to analyze their importance for object detection. Initial experiments show an MSI-trained DNN had an average optimal F1-score 3.7 points higher than the best performing RGB network. Additional experiments suggest significant DNN sensitivity to changes in bit-depth and an even greater sensitivity to the loss of any given spectral band. The results given here highlight the need for future research regarding spectral biases of DNNs for object detection in high-resolution MSI.

Here we evaluated the effectiveness of high-resolution multi-spectral (MS) satellite imagery for detection of Camouflage-Covered Military Objects (CCMO) using deep neural networks (DNN). We first divided the eight MS bands from the WorldView 2 & 3 commercial satellites into three separate 3-band MS partitions and then evaluated a variety of DNN models these partitions. The best DNN model using a single 3-band MS partition achieved an F1 score of 84.2% for CCMO detection. This was an 7.7% increase over the best baseline RGB DNN model that had an F1 score of 76.5% for CCMO detection. We then evaluated a variety of techniques to fuse multiple DNN model outputs from the same model architecture to further improve CCMO detection. The best DNN fusion technique improved the F1 score to 91% which is an increase of 14.4% over the best baseline RGB DNN model. Thus, the preliminary results from this study demonstrate significant potential for improving DNN detection for very challenging CCMO objects using the additional information available from high-resolution multi-spectral satellite image bands.

Deep neural network (DNN) algorithms have increased dramatically across a wide variety of remote sensing applications. However, as computer vision (CV) researchers have utilized the power of DNN, additional shortcomings have become more clear, including the reliance of DNNs on texture information rather than shape information to perform feature extraction for CV tasks. This lack of shape can negatively impact many remote sensing applications, where shape information can provide the key discriminatory features in object detection. To improve the utilization of shape information in DNNs, a novel model architecture was developed that incorporated shape information extracted using the Differential Morphological Profile (DMP), which was directly embedded in the DNN. While this network showed the ability to outperform traditional DNNs in CV tasks, there existed several instances in which the tradeoff of spectral and shape information proved ineffective. In this study, we present a Hybrid DMP-enabled network that utilizes both the spectral information of traditional DNNs and the shape information extracted by DMP. Our results show that this network is capable of outperforming competing methods in overhead remote sensing object detection.

Modern satellite sensors can capture reflected optical energy from wavelengths well outside of the range of human perception. Indeed, it is well known that electromagnetic energy from the “non-visible” spectrum can be used to identify materials in geological, oceanographic, and agricultural contexts. What is less understood, however, is if and how this additional spectral information can be leveraged to aid in difficult computer vision tasks, e.g., classification or detection of man-made objects. Thus, we present the results from a series of experiments that evaluate the benefits that Deep Convolutional Neural Networks (DCNN) can garner through the inclusion of more spectral information. For training and testing these networks in image classification, we extract image tiles from the xView multi-spectral dataset. These images contain eight channels of spectral data including five wavelengths bands beyond the three bands traditionally used in computer vision researcher (red, green, and blue). In this work, we report the results of experiments that use an 80–20 train-test split with four DCNN architectures on full 8-band Multispectral Imagery (MSI) and various subsets of this eight banded imagery. The results show that networks trained on MSI have an average testing F1-score around 1.0 point higher than RGB networks trained with the same methods.

We are currently experiencing a deluge of high-resolution electroptical (HR-EO) remote sensing images which can be leveraged for a diverse set of applications, ranging from environmental monitoring to defense and security applications. One of the most challenging application domains is the use of machine learning techniques, such as computer vision, for the enhancement and automation of geospatial big data analytics. In this work, we present techniques for extending deep convolutional neural networks (DCNN) from 3-band color imagery to 4-band and 8-band multispectral remote sensing imagery. Performance comparisons are conducted between DCNN for 3-, 4-, and 8-band imagery data using the Functional Map of the World dataset. In particular, we investigate five distinct DCNN architectures for classification, and show that utilizing more input channels of the imagery typically has a positive impact on classification metrics such as F1-score and raw accuracy. Herein, we evaluated two methods for initializing the DCNN convolutional filters with 4-or 8-band inputs via transfer learning from networks trained on 3-band (RGB) images and show these methods are superior to random initialization. Our findings indicate that additional spectral bands have varied benefits, depending on the image class, where challenging classes saw minimal improvements. We detail these findings along with insights into the implications for multispectral DCNN in particular geospatial big data analytics use-cases.

Here we demonstrate how the results from multiple Deep Neural Networks (DNN) can be fused to improve detections and reduce error when searching for scarce or rare objects in high-resolution satellite imagery. A wide variety of experiments were conducted using the xView dataset to develop and evaluate multiple fusion strategies to improve the detection of Engineering Vehicles (e.g. excavators, cranes, bulldozers, etc.). The results demonstrate that fusion of multiple DNNs can increase the absolute True Positive Rate (TPR) by up to 5% while reducing the total error by ~20-60%. The best results were obtained by partitioning 8-band multi-spectral imagery into three sets of 3-band images to utilize existing RGB models for transfer learning. The multi-DNN results were then fused using fuzzy integrals that also included DNN detections from multiple scales. This multi-DNN fusion approach can be easily extended to a variety of other challenging rare or scarce object detection problems in large remote sensing image datasets.

In the last decade, several high resolution remote sensing benchmark datasets have been developed and publicly released. These datasets, while diverse in design, lack the required intra-class variation for high-performing, machine-assisted visual analytics. More specifically, the disparate datasets are suitable for small, closed system evaluation; however they are not well suited for training of computer vision models that are robust in real-world, non-closed environments encountered in true remote sensing applications. To that end, a benchmark meta-dataset (MDS) was developed to facilitate the training of models for machine-assisted visual analytics. Four existing benchmark datasets were combined to build the original MDS, which excelled for training models for both classification and broad area search applications. In this work, we evaluate an enhanced version of the MDS, MDSv2, by integrating co-occurring classes of two additional recently released, publicly available challenge datasets: xView and Functional Map of the World (FMoW). The MDSv2 has 33 classes with 87,470 total samples. We investigate the utility of three neural architecture search (NAS) deep learning architectures on the MDSv2 for both classification and machine-assisted visual analytics. The NAS models trained with the MDSv2 are able to achieve an average F1 of 98.01% and a powerful 0.934 scanning precision.

Artificial neural networks are a dominant force in our modern era of data-driven artificial intelligence. The adaptive neuro fuzzy inference system (ANFIS) is a neural network based on fuzzy logic versus a more traditional premise like convolution. Advantages of ANFIS include the ability to encode and potentially understand machine learned neural information in the pursuit of explainable, interpretable, and ultimately trustworthy artificial intelligence. However, real-world data is almost always imperfect, e.g., incomplete or noisy, and ANFIS is not naturally robust. Specifically, ANFIS is susceptible to over inflated uncertainty, poor antecedent (fuzzy set) data alignment, degenerate optimization conditions, and hard to interpret logic, to name a few factors. Herein, we explore the use of possibilistic clustering to identify outliers, specifically typicality degrees, to increase the robustness of ANFIS; or any fuzzy logic neuron/network. Experiments are presented that demonstrate the need and quality of the proposed solutions in the pursuit of robust interpretable machine learned neuro fuzzy logic solutions.

Deep convolutional neural networks (DCNN) have been the dominant methodology in the field of computer vision over the last decade, using various architectural organizations of successive convolutional layers to extract and assemble low level image features into visual component detectors. One of the tradeoffs that have been made as the community has migrated to deep neural models is the loss of explainability and understanding of which salient visual components are being recognized by a model for a particular task. However, there exists a significant heritage in the remote sensing community that has developed advanced algorithms to analyze the signal and structural characteristics of anthropogenic features. One such approach is the use of morphological image processing techniques to extract objects from imagery and aid in the structural analysis of shapes. In particular, the differential morphological profile (DMP) has had great success extracting object shapes, while naturally grouping the extracted shapes into scale ranges. In this research, we present a novel architecture that integrates an explicit (definable and explainable) scaled object extraction into the network architecture, allowing shallower convolutional layers and lower complexity neural models. The architecture is evaluated on a challenging remote sensing dataset of object classes, providing insights to this approach and illuminating future directions of integrating morphology into neural architectures for enhanced explainability.

Here we demonstrate how Deep Neural Network (DNN) detections of multiple constitutive or component objects that are part of a larger, more complex, and encompassing feature can be spatially fused to improve the detection performance of a larger complex feature. A wide variety of experiments were conducted using the public domain xView dataset to develop and evaluate multiple fusion strategies to improve the detection of Construction Sites using DNN detections of constitutive/component objects commonly associated with construction activity, e.g. cement mixers, dump trucks, etc. The results demonstrate that spatial fusion of multi-scale component object DNN detections can reduce the total detection error rate of Construction Sites by ~30-40%. The best results were obtained when local spatial clustering was used to reduce noise in component vehicle object detections generated by scanning candidate Construction Site locations. This multi-scale spatial fusion approach can be easily extended to improve detection performance in a wide variety of other challenging feature/object search and detection problems in large-scale remote sensing image datasets.

Object detection is a challenging but important task for computer vision, and this is especially true in the remote sensing domain where data collections may bring billions of pixels in a single image. There are many methods for object detection, but in recent years the You Only Look Once (YOLO) algorithm has become a leading technique, gaining popularity due to its ability to perform real-time object detection. While YOLO and its successors have shown excellent results in realtime detection, there are many object detection tasks that require better precision, and do not require real-time detection. In this paper, YOLOv3 is compared to other deep neural networks (DNN) for detecting Vehicle Groups in very high resolution remote sensing imagery (VHR-RSI). A unique centerpoint-based dataset is developed by leveraging a novel data framework and combining quality assured chips with regions of interest in the XView Challenge Dataset. This dataset is then used to train state of the art models including two Neural Architecture Search (NAS) variant DNN for object detection. Additionally, a blind test set is developed to further compare our methods with the YOLOv3 algorithm. The results shows that our method detects vehicle groups with a lower false positive rate (FPR) and better true positive rate (TPR) than state-of-the-art YOLOv3 models for the blind test set; achieving a reduction in error rate of 26.70% over YOLOv3 in F1 Score on the blind test set.

Here we demonstrate how Deep Neural Network (DNN) detections of multiple constitutive or component objects that are part of a larger, more complex, and encompassing feature can be spatially fused to improve the search, detection, and retrieval (ranking) of the larger complex feature. Scores computed from a spatial clustering algorithm are normalized to a reference space so that they are independent of image resolution and DNN input chip size. DNN detections from multiple component objects can then be fused with or without human-expert provided weights to improve the retrieval (ranking) of DNN detections of a larger complex feature. We demonstrate the utility of this approach for broad area search and detection of Surface-to-Air Missile (SAM) sites that have a very low occurrence rate (only 16 sites) over a ~90,000 km2study area in SE China. Our spatial fusion approach can be easily extended to a wide variety of other challenging object search and detection problems in large-scale remote sensing image datasets.

Recently, many high-resolution remote sensing imagery (HR-RSI) datasets have been released that have diverse characteristics, such as high inter-class and low intra-class variation. Additionally, a benchmark meta-dataset (MDS) was created by agglomerating object classes from multiple HR-RSI datasets. Previous work has shown that deep convolutional neural networks (DCNN) trained on the MDS perform on par with DCNN trained on constituent benchmark datasets in cross-validation experiments. Here we train a model ensemble on four datasets and compare it with a single robust DCNN trained on the MDS. The goal is to better understand under what conditions an ensemble of models, each trained with distinct datasets, is advantaged or disadvantaged compared to a single model trained with the agglomerated MDS for classification performance.

Here we present techniques and algorithms to apply trained deep convolutional neural networks (DCNN) to high-resolution remote sensing imagery datasets covering large areas of the Earth. First, trained DCNN are used to process broad swaths of imagery in an area of interest (AOI) to produce a classification vector response field (CVRF). The CVRF is then aggregated using mode-seeking algorithms to detect potential objects of interest within the AOI. Our research explores the challenges and opportunities of transitioning DCNN out of the training-validation laboratory setting and into the real-world application domain. We show a scalable approach to leverage state-of-the-art DCNN for broad area automated search, detection, and annotation of objects such as tennis courts, storage tanks, runways, and airplanes.

Accurate land cover classification and detection of objects in high-resolution electro-optical remote sensing imagery (RSI) have long been a challenging task. Recently, important new benchmark data sets have been released which are suitable for land cover classification and object detection research. Here, we present state-of-the-art results for four benchmark data sets using a variety of deep convolutional neural networks (DCNN) and multiple network fusion techniques. We achieve 99.70%, 99.66%, 97.74%, and 97.30% classification accuracies on the PatternNet, RSI-CB256, aerial image, and RESISC-45 data sets, respectively, using the Choquet integral with a novel data-driven optimization method presented in this letter. The relative reduction in classification errors achieved by this data driven optimization is 25%-45% compared with the single best DCNN results.

Deep convolutional neural networks (DCNNs) have recently emerged as a dominant paradigm for machine learning in a variety of domains. However, acquiring a suitably large data set for training DCNN is often a significant challenge. This is a major issue in the remote sensing domain, where we have extremely large collections of satellite and aerial imagery, but lack the rich label information that is often readily available for other image modalities. In this letter, we investigate the use of DCNN for land-cover classification in high-resolution remote sensing imagery. To overcome the lack of massive labeled remote-sensing image data sets, we employ two techniques in conjunction with DCNN: transfer learning (TL) with fine-tuning and data augmentation tailored specifically for remote sensing imagery. TL allows one to bootstrap a DCNN while preserving the deep visual feature extraction learned over an image corpus from a different image domain. Data augmentation exploits various aspects of remote sensing imagery to dramatically expand small training image data sets and improve DCNN robustness for remote sensing image data. Here, we apply these techniques to the well-known UC Merced data set to achieve the land-cover classification accuracies of 97.8 ± 2.3%, 97.6 ± 2.6%, and 98.5 ± 1.4% with CaffeNet, GoogLeNet, and ResNet, respectively.

Deep convolutional neural networks (DCNNs) have recently emerged as the highest performing approach for a number of image classification applications, including automated land cover classification of high-resolution remote-sensing imagery. In this letter, we investigate a variety of fusion techniques to blend multiple DCNN land cover classifiers into a single aggregate classifier. While feature-level fusion is widely used with deep neural networks, our approach instead focuses on fusion at the classification/information level. Herein, we train three different DCNNs: CaffeNet, GoogLeNet, and ResNet50. The effectiveness of various information fusion methods, including voting, weighted averages, and fuzzy integrals, is then evaluated. In particular, we used DCNN cross-validation results for the input densities of fuzzy integrals followed by evolutionary optimization. This novel approach produces the state-of-the-art classification results up to 99.3% for the UC Merced data set and the 99.2% for the RSD data set.

We have developed a fully automated system for change detection of high-resolution satellite imagery. Our system, GeoCDX, is sensor-agnostic, resolution-independent and designed to process the very large volumes of data collected by modern high resolution panchromatic and multispectral imaging satellites. GeoCDX performs fully automated coregistration of imagery; extracts high-level features from the satellite imagery; performs change detection processing to pinpoint locations of change; clusters image tiles to group similar regions of change; and presents results in a variety of ways in an easy-to-use web application that facilitates online discovery, analysis, and dissemination of the change detection results. We applied GeoCDX to 4121 image pairs and successfully coregistered over 91% of the pairs covering a total area greater than 370 000 km2; GeoCDX decreased the average coregistration error from 9.6 ± 8.6 m to 1.8 ± 1.2 m. We show that for some pairs, GeoCDX provides up to a 50% increase in users' efficiency compared to manually performing change detection in common GIS software. Moreover, the change detection assessment performed using GeoCDX was on average four times more accurate compared to the manual approach in large part due to the use of our change intensity map that provides visual cues to the user during exploitation. Finally, change detection analysis using GeoCDX resulted in a missed detection rate of less than 2%.

The Ocean Color CAL/VAL team is evaluating the VIIRS bio-optical products for real-time operations. VIIRS ocean data are being processed using standard government algorithms, and channel calibration and product validation evaluation activities are ongoing. A network of 27 global “Golden Regions” has been established to evaluate and validate bio-optical products. Satellite inter-comparison for data consistency with current ocean color products, and real time vicarious adjustment calculation are performed using in situ water leaving radiance propagated to Top of Atmosphere in coastal and open ocean regions. In addition, routine matchups with VIIRS and MODIS-Aqua are done with in situ data collection from ships and real time coastal AERONET-OC sites. The above activities, product evaluation and tracking of channel stability, are being contributed to the JPSS Team to evaluate the overall mission, including calibration and inter-satellite product consistency. Initial NPP VIIRS ocean bio-optical products are demonstrated with other ocean color satellites.

This letter presents a system to solve the vector-to-imagery building conflation problem. To drive the system, structure outlines in high-resolution images are extracted via a shape-driven level set scheme. Shape and relative position features are then computed for the image-extracted buildings and for vector graphics buildings from a geospatial information system (GIS). These two features are used by a graph-matching procedure that finds correspondences between the image-extracted buildings and those from a GIS. Extensions of our system to vector-to-vector building conflation, generic polygonal object conflation, and image-to-image registration are also possible.

The nine papers in this special issue were originally presented at the 2009 International Geoscience and Remote Sensing Symposium (IGARSS '09), held at the University of Cape Town, Rondebosch, South Africa, from July 12 to 17.

In this paper, we present a novel indexing structure that was developed to efficiently and accurately perform content-based shape retrieval of objects from a large-scale satellite imagery database. Our geospatial information retrieval and indexing system, GeoIRIS, contains 45 GB of high-resolution satellite imagery. Objects of multiple scales are automatically extracted from satellite imagery and then encoded into a bitmap shape representation. This shape encoding compresses the total size of the shape descriptors to approximately 0.34% of the imagery database size. We have developed the entropy-balanced bitmap (EBB) tree, which exploits the probabilistic nature of bit values in automatically derived shape classes. The efficiency of the shape representation coupled with the EBB tree allows us to index approximately 1.3 million objects for fast content-based retrieval of objects by shape.

The Geospatial Change Detection and exploitation (GeoCDX) is a fully automated system for detection and exploitation of change between multitemporal high-resolution satellite and airborne images. Overlapping multitemporal images are first organized into 256 m × 256 m tiles in a global grid reference system. The system quantifies the overall amount of change in a given tile with a tile change score as an aggregation of pixel-level changes. The tiles are initially ranked by these change scores for retrieval, review, and exploitation in a Web-based application. However, the ranking does not account for the wide variety of change types that are typically observed in the top-ranked change tiles. To automatically organize the wide variety of change patterns observed in multitemporal high-resolution imagery, we perform tile clustering using the competitive agglomeration (CA) algorithm stabilized using the fuzzy c-means (FCM) algorithm. Each resulting cluster contains tiles with a visually similar type of change. By visual inspection of these tile clusters, GeoCDX users can quickly find certain types of change without having to sift through a large number of tiles initially organized solely by their tile change score, thereby reducing the time it takes for users to discover and exploit the change pattern(s) of greatest interest to a given application (e.g., urban growth, disaster assessment, facility monitoring, etc.). The tile clusters also provide a high-level overview of the various types of change that occur between the two observations. This overview is compared with a similar yet more limited view offered by a relevance feedback tool that requires a user to select sample tiles for use as samples in the reranking process.

GeoCDX (Geospatial Change Detection and eXploitation) is an integrated system for detecting change between multi-temporal, high-resolution satellite or airborne images. Overlapping images are organized into 256×256 meter tiles in a global grid system. A tile change score measures the amount of change in the tile which is the aggregation of pixel-level change score. The tiles are initially ranked by these change scores. However, this ranking does not account for the wide variety of change types. To learn the change patterns in the data, we apply the fuzzy c-means clustering algorithm to the tiles. Each resulting cluster contains tiles with similar type of change. Users looking for certain types of change can review the tile clusters rather than the more time consuming process of searching through the tile list based on the initial ranking. The clusters also provide users an overview of various types of change found in the scene.

As the availability of various geospatial data increases, there is an urgent need to integrate multiple datasets to improve spatial analysis. However, since these datasets often originate from different sources and vary in spatial accuracy, they often do not match well to each other. In addition, the spatial discrepancy is often nonsystematic such that a simple global transformation will not solve the problem. Manual correction is labor-intensive and time-consuming and often not practical. In this paper, we present an innovative solution for a vector-to-imagery conflation problem by integrating several vector-based and image-based algorithms. We only extract the different types of road intersections and terminations from imagery based on spatial contextual measures. We eliminate the process of line segment detection which is often troublesome. The vector road intersections are matched to these detected points by a relaxation labeling algorithm. The matched point pairs are then used as control points to perform a piecewise rubber-sheeting transformation. With the end points of each road segment in correct positions, a modified snake algorithm maneuvers intermediate vector road vertices toward a candidate road image. Finally a refinement algorithm moves the points to center each road and obtain better cartographic quality. To test the efficacy of the automated conflation algorithm, we used U.S. Census Bureau's TIGER vector road data and U.S. Department of Agriculture's 1-m multi-spectral near infrared aerial photography in our study. Experiments were conducted over a variety of rural, suburban, and urban environments. The results demonstrated excellent performance. The average correctness measure increased from 20.6% to 95.5% and the average root-mean-square error decreased from 51.2 to 3.4 m.

Here we measure fine-scale polar ice sheet elevation change and estimate decadal mass balance for the Greenland and Antarctic ice sheets from ERS-2 and ENVISat satellite radar altimeter data. The Greenland results show that moderate to slightly positive elevation change rates (ECRs) dominate the interior areas while large negative ECRs are observed for many coastal areas. In Antarctica, glaciers in the coastal areas of the West Antarctic Ice Sheet (WAIS) are experiencing rapid thinning. The strong thinning of some of these glaciers extends well into the interior of the continent. Over the East Antarctic Ice Sheet (EAIS), positive ECRs dominate the interior areas while some glaciers in coastal areas are experiencing thinning. The equivalent mass change for the GIS, EAIS, WAIS, and AIS over the period 1995-2006 is -20, 27, -53, and -26 Gt/yr, respectively.

In this paper, we present an unsupervised change detection approach that combines pixel-based local Haar-like features, color information, vegetation index, and man-made structure features using fuzzy logic rules to provide multi-level change detection results. An illumination invariant descriptor for each pixel is introduced to describe a Haar-like feature in a local area. Hue is used as a color feature in our change detection method. For the purpose of enhancing the change area with man-made structures, we de-weight the level of detected change areas where there are no man-made objects. The comparison of all features is done separately and the decision results are then combined under seven fuzzy logic rules to provide multi-level change detection results. The performance of the change detection is evaluated qualitatively by visual inspection and quantitatively using ground truth. The quantitative test results show that more than 90% of changes are correctly detected.

Here we present an automated image-to-image registration method for high-resolution satellite images that combines a global affine transformation with non-linear local warping. We also present a novel feature matching method based on local image feature similarity, the spatial relationship amongst local extrema points (EPs), as well as k-σ editing techniques to select optimum EPs for global affine transformation. The global affine coefficients are determined using the EP solutions that have similar affine transformations. Finally, local image warping is performed to adjust nonlinear distortions in the imagery caused by topography and/or sensor viewing geometry variations between scenes. Experiments and evaluation results from five test sites show that feature points can be automatically extracted and correctly matched between the input and reference images. The registration accuracy is within 2 pixels and is improved for areas with significant topographic variations.

The demand for high-resolution commercial satellite imagery (HR-CSI) has increased significantly over the last 5 years for a wide variety of applications. This demand has driven an increase in volume, frequency of acquisition, and spatial resolution of HR-CSI. In turn, this has spurred the need for more accurate and time-efficient processing tools for analyzing geospatial information to support user-specific applications. One such application is change detection between multi-temporal HR-CSI data. The significant increase in quantity and quality of multi-temporal HR-CSI data makes traditional manual analysis impractical. Thus, there is a need for a fully automated change detection system that not only identifies areas of change, but also allows users to filter, sort through, and analyze areas of change quickly and efficiently. Here we describe a tool - GeoCDX (Geospatial Change Detection and Exploitation) - to meet this need. GeoCDX is an integrated system that performs image ingestion and registration, feature extraction, and change analysis. The change detection results from GeoCDX are web accessible with additional interfaces to Google Earth™ (GE) and Google Maps™ (GM). In the near future GeoCDX will be integrated with its sister system GeoIRIS (Geospatial Information Retrieval and Indexing System). This integration will produce a very powerful HR-CSI analysis package with the change detection capabilities of GeoCDX and the content-based image retrieval system of GeoIRIS.

Searching for relevant knowledge across heterogeneous geospatial databases requires an extensive knowledge of the semantic meaning of images, a keen eye for visual patterns, and efficient strategies for collecting and analyzing data with minimal human intervention. In this paper, we present our recently developed content-based multimodal Geospatial Information Retrieval and Indexing System (GeoIRIS) which includes automatic feature extraction, visual content mining from large-scale image databases, and high-dimensional database indexing for fast retrieval. Using these underpinnings, we have developed techniques for complex queries that merge information from heterogeneous geospatial databases, retrievals of objects based on shape and visual characteristics, analysis of multiobject relationships for the retrieval of objects in specific spatial configurations, and semantic models to link low-level image features with high-level visual descriptors. GeoIRIS brings this diverse set of technologies together into a coherent system with an aim of allowing image analysts to more rapidly identify relevant imagery. GeoIRIS is able to answer analysts' questions in seconds, such as "given a query image, show me database satellite images that have similar objects and spatial relationship that are within a certain radius of a landmark."

Here we propose a pixel-based change detector utilizing both spectral and spatial features. Traditional pixel- based change detection methods that only utilize spectral data are inherently sensitive to spectral variation. This presents problems in mitigating the impact of spectral changes due to uninteresting types of change without severely limiting the sensitivity of the detector. Here we introduce a change detector that utilizes both spectral and spatial information, including linear features and texture measures, as a method to decrease sensitivity to spectral variation and increase detection rates. For each pixel, a fuzzy value representing the similarity of each pixel feature (both spectral and spatial) is computed. All features are then fused into a single overall similarity score by weighted averaging. This is followed by thresholding and morphological extraction of the detected regions of change. The algorithm was evaluated using panchromatic and pan-sharpened multi-spectral imagery of Springfield, Missouri acquired during different seasons and covering approximately 20 square kilometers of urban, suburban, and rural terrain. Preliminary results show a 70% change detection probability for types of change unrelated to seasonal variation with rates of only 2.4 uninteresting detections per km 2 and 0.09 false alarms per km 2 .

This paper presents a framework for the efficient retrieval of satellite imagery from large-scale databases. With the ever-expanding volume of imagery being acquired from satellite platforms it has become increasingly important to locate specific areas of interest within a large database of images. Identifying relevant areas within image databases can be thought of as finding the needle in the haystack problem; too often for a particular task or application there exist a small number of useful images hidden among millions of images. The motivation behind the work presented here is that through the use of geospatial image retrieval systems, the number of scenes that image analysts must manually examine may be decreased dramatically. By using a geospatial image retrieval system as a tool, analysts no longer must manually examine the entire database of imagery, but instead can limit their search to a subset identified by our retrieval system. The techniques that are introduced in this paper have been developed in our image retrieval system named GeoIRIS: Geospatial Information Retrieval and Indexing System.

In this paper, several techniques to improve the processing and analysis of decadal elevation-change time series (ECTS) constructed using satellite radar altimeter data over the Antarctic ice sheet are described. First, a method that improves both the quality and quantity of ice-sheet ECTS is introduced. By dynamically selecting the reference month used in constructing ECTS for a given local region, the maximum number of elevation-change estimates are used in a matrix processing technique. This can improve ECTS quality while also generating ECTS in new areas, leading to a significant increase in spatial coverage. Next, an improved autoregressive (IAR) approach is presented for modeling nonlinear medium-period variations in decadal Antarctic ECTS. This builds upon previous work where an AR model was used to characterize seasonal and interannual variations in ice-sheet ECTS. The improved approach avoids overdecomposition by adopting an iterative local average filter to estimate nonlinear medium-period trends. Monte Carlo simulations show that the IAR method significantly outperforms the AR method when realistic nonlinear trends are present. Because of these characteristics, the IAR model is able to adequately characterize seasonal, interannual, and medium-period nonlinear signals present in decadal ECTS and extract their long-term trends with very small error. This is important for accurate measurement of decadal elevation change over the Antarctic ice sheet and for assessing the impact of these changes on the global sea level

Previous research examining coverage expansion due to soft handoff in code division multiple access (CDMA) cellular networks did not address the relationship between the coverage expansion and the amount of carried traffic in the cell. In this paper, we first present a method to estimate reverse-link multicell coverage using a duration-outage approach. Using this, we then proceed to quantitatively analyze reverse-link coverage expansion from soft handoff as a function of the carried traffic in the cell under different propagation environments. Our results show that for modest values of carried traffic, the coverage expansion is almost constant. However, when the amount of carried traffic is large, the coverage expansion can increase dramatically. Next, we examine the dependence of the soft-handoff coverage expansion on key characteristics of the large-scale shadowing environment. Results are presented that quantify the coverage-expansion dependence on the large-scale shadowing's correlation distance (d/sub c/), variability (/spl sigma//sub Z/), path-loss exponent (K/sub 2/), and the correlation between the two-cell large-scale shadowing environments (C/sub Z/). The results indicate that realistic variations in these propagation conditions have a significant, and sometimes dramatic, effect on the coverage expansion provided by soft handoff. Finally, we analyze the capacity improvement from soft handoff and the tradeoff between capacity improvement versus coverage expansion. Understanding and managing the capacity improvement versus coverage expansion tradeoff is critical for achieving optimal CDMA cell-network performance.

We analyzed Antarctic ice sheet elevation change (dH/dt) from 1992-2002 using nearly 307 million elevation change measurements from ERS-1 and ERS-2 satellite radar altimeter data covering an area of about 8.5 million km2. Ten-year elevation change time series were constructed for 1,500 local regions, 22 drainage basins, East and West Antarctica, and the continent as a whole (north of 81.6degS). Almost all basins in East Antarctica were found to have long-term dH/dt trends that fell within plusmn3 cm/yr. In West Antarctica the long-term dH/dt trends varied from -9 cm/yr to +22 cm/yr occur over basin scales. Despite these significant spatial variations, both the East and West Antarctic ice sheets, and the continent as a whole are very close to a state of overall mass balance. The average elevation change for the Antarctic continent from 1992-2002 was found to be 1.4plusmn0.2 cm/yr. Considering decadal accumulation variability, this translates to a net sink of fresh water of 42plusmn23 Gt/yr. The Antarctic continent's decadal contribution to global sea level change is estimated to be -0.12plusmn0.07 mm/yr, which is very close to zero. The plusmn0.07 mm/yr error is a three-fold reduction in uncertainty compared to the most recent best estimate for the Antarctic ice sheet

We applied several majority kernels to land classification maps of St. Louis, Missouri. For our study site, the Landsat classification images are further processed by four techniques: (1) morphological filtering, (2) majority filtering by uniform square kernel, (3) majority' filtering by uniform circular kernel, (4) majority filtering by weighted circular kernel. The relationship between the kernel size and the overall land classification is investigated. For our study site, the results show that the majority' filtering using weighted circular kernel increases the overall classification accuracy by 18.8% compared to the raw classified image. This paper discusses the techniques and results of our study

Recent availability of high-resolution satellite imagery provides a new data source to extract small-scale objects such as vehicles. Seldom has vehicle detection been applied to highresolution satellite imagery where panchromatic band resolutions are presently in the range of 0.6-1.0 m. With the limited spatial resolution, reliable vehicle detection can only be achieved by incorporating contextual information. Here we use a GIS road vector map to guide vehicle detection by restricting the search to the road network. A morphological shared-weight neural network (MSNN) is used to classify image pixels on roads into targets and non-targets. Then a target aim point selection algorithm is designed to locate vehicle centroids. The MSNN has been successfully used for automatic target recognition (ATR) from multiple sensors. In this paper, we present experimental results of MSNN vehicle detection applied to 1-m IKONOS panchromatic data. A vehicle image base library is built by collecting more than 300 cars manually from 16 road segments in 3 test images. To suppress false alarms, a screening algorithm is developed using morphological filtering to identify vehicle pixels and non-target pixels that are similar to vehicle pixels. Subimages centered at those pixels are used as positive and negative training samples of the MSNN. The MSNN was trained on subimage samples from 11 road segments. The performance results on the training segments and remaining 5 validation segments are reported. The MSNN shows a good generalization behavior. After learning, the detection rate exceeds 85% with very few false alarms.

A fully automated algorithm for the extraction of building footprints from commercial high-resolution satellite imagery is presented. The algorithm is based on identifying buildings and their shadows in the differential morphological profile (DMP) of 1-m resolution panchromatic imagery. The DMP is a multi-scale image analysis technique where an image profile is constructed through the use of morphological opening and closing operations while varying the size of the structuring element. A variety of shape-based features are utilized to extract buildings and their shadows from the DMP. The algorithm has been tested on an IKONOS image, and the results are promising.

Geosat Follow On (GFO) radar altimeter data are processed with previous altimeter datasets to measure long-term elevation change of higher elevation portions of the southern Greenland ice sheet over time periods from 1978-1988, 1985-2002, and 1978-2002. Average elevation change results indicate approximately zero overall elevation change for all time periods. The results also indicate that upper-elevation thinning in southeast Greenland has been widespread and has existed for several decades.

We analyzed Antarctic ice-sheet elevation change (dH/dt) from 1995 to 2000 using 123 million elevation change measurements from European Remote Sensing 2 ice-mode satellite radar altimeter data covering an area of about 7.2 million km/sup 2/. Almost all drainage basins in east Antarctica had average dH/dt values within /spl plusmn/3.0 cm/year, whereas drainage basins in west Antarctica had substantial spatial variability with average dH/dt values ranging between -11 to +12 cm/year. The east Antarctic ice sheet had a five-year trend of 1/spl plusmn/0.6 cm/year, where 13 out of the 14 basins had either a positive trend or a trend that was not significantly different than zero. The west Antarctic ice sheet had a five-year trend of -3.6/spl plusmn/1.0 cm/year due largely to strong negative trends of around 10 cm/year for basins in Marie Byrd Land along the Pacific sector of the Antarctic coast. The continent as a whole had a five-year dH/dt trend of 0.4/spl plusmn/0.4 cm/year. Finally, time series constructed for the Pine Island, Thwaites, De Vicq, and Land glaciers in west Antarctic showed five-year dH/dt trends from -26 to -135 cm/year that were significantly more negative than the average dH/dt trends in their respective basins. The strongly negative dH/dt values for these coastal glacier outlets are consistent with recently reported results indicating increased basal melting at these glaciers' grounding lines caused by ocean thermal forcing.

We present an autoregressive (AR) model that can effectively characterize both seasonal and interannual variations in ice sheet elevation change time series constructed from satellite radar or laser altimeter data. The AR model can be used in conjunction with weighted least squares regression to accurately estimate any longer term linear trend present in the cyclically varying elevation change time series. This approach is robust in that it can account for seasonal and interannual elevation change variations, missing points in the time series, signal aperiodicity, time series heteroscedasticity, and time series with a noninteger number of yearly cycles. In addition, we derive a theoretically valid estimate of the uncertainty (standard error) in the long-term linear trend. Monte Carlo simulations were conducted that closely emulated actual characteristics of five-year elevation change time series from Antarctica. The Monte Carlo results indicate that the autoregressive approach yields long-term linear trends that are less biased than two other approaches that have been recently used for analysis of ice sheet elevation change time series. In addition, the simulation results demonstrate that the variability (uncertainty) of the long-term linear trend estimates from the AR approach is in very good agreement with the derived theoretical standard error estimates.

In this paper, a genetic-based image segmentation method is proposed which optimizes a fuzzy-set-based evaluation function. A K-Means clustering method is used to generate the initial finely segmented image and to reduce the search space of the image segmentation. A genetic algorithm is then employed to control region splitting and merging to optimize the evaluation function. A critical factor affecting the performance of the segmentation is the choice of the evaluation function in the design of genetic algorithm. Here an evaluation function is defined that incorporates both edge and region information. Considering the edge ambiguity in the image, a novel fuzzy-set-based edge-boundary-coincidence measure is defined and combined with a region heterogeneity measure to guide the genetic algorithm to tune the segmentation. Experimental results on test images show that the genetic segmentation algorithm with the fuzzy-set-based evaluation function performs very well.

We present a fully automated technique for road network extraction from high-resolution multispectral satellite imagery of urban areas. Road segments are iteratively identified by examining contextual length-width features extracted from the multispectral imagery in conjunction with a vegetation index. A straight-line road segments are identified, the endpoints of these line segments are grown, allowing the road network extraction algorithm to track roads around curves and through area that are partially occluded. Long line segments are iteratively added to the road network, and a buffer is set up around them to exclude any line segments that are not close to perpendicular to the identified road network segments. This algorithm is fully automated and requires no interaction with the user after initial setting of several parameters controlling the identification pf potential road pixels, growth of the line segments, and the stopping criteria. The proposed approach yields an accurate road network with minimal interaction from the user. Extraction completeness measures of 82-85% and correctness measures of 71-84% are obtained.

In this paper, we investigate the usefulness of high-resolution multispectral satellite imagery for classification of urban and suburban areas and present a fuzzy logic methodology to improve classification accuracy. Panchromatic and multispectral IKONOS image datasets are analyzed for two urban locations in this study. Both multispectral and pan-sharpened multispectral images are first classified using a traditional maximum-likelihood approach. Maximum-likelihood classification accuracies between 79% to 87% were achieved with significant misclassification error between the spectrally similar Road and Building urban land cover types. A number of different texture measures were investigated, and a length-width contextual measure is developed. These spatial measures were used to increase the discrimination between spectrally similar classes, thereby yielding higher accuracy urban land cover maps. Finally, a hierarchical fuzzy classification approach that makes use of both spectral and spatial information is presented. This technique is shown to increase the discrimination between spectrally similar urban land cover classes and results in classification accuracies that are 8% to 11% larger than those from the traditional maximum-likelihood approach.

In this paper, we present an object-based approach for urban land cover classification from high-resolution multispectral image data that builds upon a pixel-based fuzzy classification approach. This combined pixel/object approach is demonstrated using pan-sharpened multispectral IKONOS imagery from dense urban areas. The fuzzy pixel-based classifier utilizes both spectral and spatial information to discriminate between spectrally similar road and building urban land cover classes. After the pixel-based classification, a technique that utilizes both spectral and spatial heterogeneity is used to segment the image to facilitate further object-based classification. An object-based fuzzy logic classifier is then implemented to improve upon the pixel-based classification by identifying one additional class in dense urban areas: nonroad, nonbuilding impervious surface. With the fuzzy pixel-based classification as input, the object-based classifier then uses shape, spectral, and neighborhood features to determine the final classification of the segmented image. Using these techniques, the object-based classifier is able to identify buildings, impervious surface, and roads in dense urban areas with 76%, 81%, and 99% classification accuracies, respectively.

We analyzed the effectiveness of generating urban land cover maps from IKONOS imagery. 1-m PAN and 4-m MS IKONOS images were combined to produce two pan-sharpened MS images (PS-MS) with 1-m resolution. The fusion was done with the original 11-bit data and also by scaling the original data to only 8-bits. A parallelepiped supervised classification algorithm was used to process the two PS-MS images as well as the original 4-m MS image. Seven urban land cover classes were used in this study: woods, grass, water, bare soil, commercial building, impervious, and shadow. The classification accuracy was assessed using 256 pixels that were randomly distributed throughout the test site and were independent of the training sites used by the supervised classification algorithm. The results show that classification accuracies on the order of 75-80% are obtained. The best results are obtained using the 4-band 11-bit 1-m PS-MS image, and this yielded an overall accuracy of 83%.

Integration of synthetic-aperture radar data with spaceborne optical data offers potential for improving land-use classification for local and state government applications. Remote sensing techniques using optical sensor data are well established for local and state government applications. As commercial SAR products become available at lower costs, local government applications using optical based classification can incorporate SAR sensor data into their processing. For this study, we orthorectified and analyzed Radarsat SAR data and Landsat TM data that covers Boone County, Missouri. Optical and SAR sensor data are co-registered for data fusion and classification process. Training data from sites throughout the study area are used to verify and validate the classification of five classes: crops, water, built-up, forest and grass. We present preliminary results of our study.

The usefulness of high-resolution satellite imagery for classification of urban and suburban scenes is investigated, and a technique to improve the accuracy of the classification is proposed. The Ikonos commercial remote sensing satellite acquired the imagery (panchromatic and multispectral) used for this study. Both multispectral and pan-sharpened multispectral images are classified using maximum likelihood classification. The overall classification accuracy for both data sets is approximately 81%, however there are significant numbers of misclassifications between the spectrally similar road and building classes and the tree and grass classes. The confusion between the building and road classes was about 22%, and the confusion between the grass and tree classes was about 13%. To decrease the number of misclassifications between the previously mentioned classes, a hierarchical fuzzy classification technique is proposed. The fuzzy classifier makes use of spatial features extracted from the panchromatic data, pan-sharpened multispectral data, and a classification image, generated using maximum likelihood classification. A number of different texture features and directional pixel length-width contextual features are extracted from the panchromatic image data. A fuzzy logic rule based classifier is used in conjunction with these spatial features to perform the classification. The proposed approach decreases the number of misclassifications between the road and building classes and the number of misclassifications between the grass and tree classes raising the overall accuracy to 88%.

Seasat and Geosat satellite radar altimeter measurements over the East Antarctic Ice Sheet (EAIS) are analyzed to determine surface elevation change over the period from 1978 to 1988. The results show that the average elevation change over 1.7 /spl times/10/sup 6/ km/sup 2/ of the EAIS is -0.4/spl plusmn/1.8 cm/yr and is therefore not significantly different than zero. This result is in very good agreement with ERS-1/2 radar altimeter estimates of EAIS elevation change from 1992 to 1996. The combined results suggest that the EAIS has been close to a state of balance for two decades. This further reduces the likelihood that recent changes in snow accumulation have masked a long-term mass imbalance in the EAIS.

The spatial variability of elevation errors in high-resolution digital elevation models (DEMs) derived from stereo-image processing is examined. Error models are developed and evaluated by examining the correlation between various DEM parameters and the magnitude of the observed DEM vertical error. DEM vertical errors were estimated using a dataset of more than 51000 points of known elevation obtained from a kinematic Global Positioning Satellite (GPS) ground survey. Elevation variability and the quality of the stereo-correlation match over small spatial scales were the dominant factors that determined the magnitude of the DEM error at any given location. The error models are strongly correlated with the magnitude of the DEM vertical error and are shown to adequately represent the full range of the observed error. The error models are used to estimate the magnitude of the vertical error for every point in the DEMs. The models are then used to predict the overall error in the DEMs. The results demonstrate that the error models can accurately quantify and predict the spatial variability of the DEM error.

A new method to derive the seasonal elevation signal from a continuous time series of altimeter crossover data is presented. The method utilizes the complete set of intrasatellite crossover data to provide fine temporal resolution and to significantly reduce random measurement errors. This is an improvement over the standard time-series method that only uses those data referenced to a single time period at the beginning of the satellite mission.

A new analysis of Seasat and Geosat satellite radar altimeter measurements over the Greenland ice sheet was performed to determine surface elevation change. The new analysis includes twice as many measurements and has 50% greater spatial coverage than the authors' previous study. In addition, a precise global ocean reference network created from four years of Topex/Poseidon altimeter data is used to obtain improved estimates of altimeter orbit errors and measurement system biases. The results show that the average elevation change of the southern Greenland ice sheet above 2000 m from 1978 to 1988 is not significantly different than zero. This contradicts earlier and even more recent studies that reported positive ice sheet growth rates and suggested increased precipitation due to a warmer polar climate.

A threshold retracking algorithm for processing ice-sheet altimeter data is presented. The primary purpose for developing this algorithm is detection of ice-sheet elevation change, where it is critical that a retracking algorithm produce repeatable elevations. The more consistent an algorithm is in selecting the retracking point the less likely that errors and/or biases will be introduced by the retracking scheme in the elevation-change measurement. The author performed extensive comparisons between the threshold algorithm and two widely used ice-sheet retracking algorithms on Geosat datasets comprised of over 60000 crossover points. The results show that the threshold retracking algorithm, with a 10% threshold level, produces ice-sheet surface elevations that are more repeatable than the elevations derived from the other retracking algorithms. For this reason, the threshold retracking algorithm has been adopted by NASA/GSFC as an alternative to their existing algorithm for production of ice sheet altimeter datasets under the NASA Pathfinder Program. The threshold algorithm will be used to re-process existing ice-sheet altimeter datasets and to process the datasets from future altimeter missions.

The NASA and ESA retracking algorithms are compared with an algorithm based upon a combined surface and volume (S/V) scattering model. First, the S/V, NASA, and ESA algorithms were used to retrack over 1.3 million altimeter return waveforms from the Greenland and Antarctic ice sheets. The surface elevations from the S/V algorithm were compared with the elevations produced by the NASA and ESA algorithms to determine the relative accuracy of these algorithms when subsurface volume scattering occurs. The results show that the ESA/sub 25%/ algorithm produced slightly higher surface elevations than the S/V algorithm. The NASA retracking algorithm produced lower surface elevations than the SN retracking algorithm, with average differences ranging from -0.3 to -0.9 m. The lower NASA elevations can only account for a portion of previously reported differences between altimeter and geoceiver surface elevations, suggesting that the remainder is probably due to orbital differences. Next, by analyzing several thousand satellite crossover points from the Greenland and Antarctic ice sheets, the author estimated the repeatability of the surface elevations derived from the different retracking algorithms. The elevations derived from the ESA/sub 25%/ and S/V algorithm had the smallest standard deviations for the crossover differences for a time period where no significant change in surface elevation should occur. The NASA standard deviations were approximately 0.2 m larger than those from the ESA/sub 25%/ and S/V algorithm, which represents an average increase in error of approximately 0.5 m in the datasets. Since previous ice-sheet growth estimates have been based upon the elevations produced by the NASA retracking algorithm, further work needs to be conducted to determine if the ESA/sub 25%/ or S/V retracking algorithms produce growth estimates that are significantly different from the previous estimates.

The extinction coefficient of snow k/sub e/ along the central portion of the Greenland ice sheet is mapped using data from the Seasat (1978) and Geosat (1985-1989) altimeters. The extinction coefficient is obtained by fitting altimeter waveforms with a surface/volume scattering model. The authors find that in the lower latitudes the Seasat and Geosat extinction coefficients are very nearly the same, while in a specific higher latitude region of the ice sheet the Seasat k/sub e/ values exceed the Geosat values by over 100%. By analyzing 18 months of the Geosat data, the author quantified the variability inherent in the extinction coefficient measurements. The results show that the observed temporal variation in the extinction coefficient from 1978 to 1985 is three times larger than the measured variability. This indicates that the average grain size of the near surface snow in this region may have decreased during the time span between the two altimeter datasets. The temporal change in extinction coefficient found in this study demonstrates the important contributions that time-series analysis of satellite datasets can make to the study of the polar ice sheets. In addition, these results have important implications for the study of long-term elevation change over the ice sheets using altimeter data. The author's study demonstrates that significant biases could be introduced into ice-sheet elevation change estimates because of temporal variations in the surface conditions of the ice sheet. Future investigations of ice-sheet mass balance using altimetry data should be aware of this possibility.

The authors compare the performance of different altimeter retracking algorithms for measuring ice sheet elevations and growth rates. The results show that the threshold, ESA, and S/V retracking algorithms produce growth rates that are 30-55% smaller than those produced by the NASA algorithm. Based upon a comparison of crossover-point standard deviations, the analysis indicates that the surface elevation estimates produced by these algorithms are more repeatable than the NASA surface elevations. An analysis of the NASA algorithm shows that a mixing of its 5 and 9 parameter functional fits on the crossover-point altimeter waveforms occurs in over 70% of the crossover data. The mixing of the functional fits is shown to reduce the repeatability of the NASA elevations and this may be responsible for the larger estimates of ice sheet growth produced by the NASA retracking algorithm. The extremely close agreement between the standard deviations and the growth-rate estimates from the threshold, ESA, and S/V retracking algorithms: lead the authors to conclude that 0.10 m/yr is a more accurate estimate of the growth of the Greenland ice sheet from 1978-1987 (south of 72/spl deg/N).>

An algorithm based on a combined surface and volume scattering model that is used to retrack individual altimeter waveforms from the ice sheets is developed. Because the combined model is nonlinear, an iterative least-squares procedure is used to fit the combined model to the return waveforms. This retracking algorithm can be used to assess the accuracy of elevations produced by current retracking algorithms, which do not account for subsurface volume scattering. This is extremely important if repeated altimeter elevation measurements are to be used to accurately detect changes in the mass balance of the ice sheets. In addition, by analyzing the distribution of the model parameters over large portions of the ice sheet, quantitative estimates of regional and seasonal variations in the near-surface properties of the ice sheets can be obtained.>

Measurements taken by an X-band pulse radar system in East Antarctica in 1987 were analyzed to determine the depth of penetration of radar energy in polar firn. A minimum estimate of the penetration depth at 10 GHz for the cold and dry polar plateau region is approximately 4.7 m. Spatial variations in the amount of signal penetration were observed and are related to latitude, surface elevation, and mean annual temperature. The amount of signal penetration and its spatial variation are important factors that should be considered when processing datasets from microwave remote sensing systems that operate in the polar regions.>