Search

Satellite image classification is a challenging problem that lies at the crossroads of remote sensing, computer vision, and machine learning. Due to the high variability inherent in satellite data, most of the current object classification approaches are not suitable for handling satellite datasets. The progress of satellite image analytics has also been inhibited by the lack of a single labeled high-resolution dataset with multiple class labels. In a preliminary version of this work, we introduced two new high resolution satellite imagery datasets (SAT-4 and SAT-6) and proposed DeepSat framework for classification based on "handcrafted" features and a deep belief network (DBN). The present paper is an extended version, we present an end-to-end framework leveraging an improved architecture that augments a convolutional neural network (CNN) with handcrafted features (instead of using DBN-based architecture) for classification. Our framework, having access to fused spatial information obtained from handcrafted features as well as CNN feature maps, have achieved accuracies of 99.90% and 99.84% respectively, on SAT-4 and SAT-6, surpassing all the other state-of-the-art results. A statistical analysis based on Distribution Separability Criterion substantiates the robustness of our approach in learning better representations for satellite imagery., Comment: This is an Accepted Manuscript of an article published by Taylor & Francis Group in Remote Sensing Letters. arXiv admin note: text overlap with arXiv:1509.03602

Machine learning has proven to be useful in classification and segmentation of images. In this paper, we evaluate a training methodology for pixel-wise segmentation on high resolution satellite images using progressive growing of generative adversarial networks. We apply our model to segmenting building rooftops and compare these results to conventional methods for rooftop segmentation. We present our findings using the SpaceNet version 2 dataset. Progressive GAN training achieved a test accuracy of 93% compared to 89% for traditional GAN training., Comment: Accepted too and presented at DMESS 2018 as part of IEEE ICDM 2018

Deep Learning (DL) methods have been transforming computer vision with innovative adaptations to other domains including climate change. For DL to pervade Science and Engineering (S&E) applications where risk management is a core component, well-characterized uncertainty estimates must accompany predictions. However, S&E observations and model-simulations often follow heavily skewed distributions and are not well modeled with DL approaches, since they usually optimize a Gaussian, or Euclidean, likelihood loss. Recent developments in Bayesian Deep Learning (BDL), which attempts to capture uncertainties from noisy observations, aleatoric, and from unknown model parameters, epistemic, provide us a foundation. Here we present a discrete-continuous BDL model with Gaussian and lognormal likelihoods for uncertainty quantification (UQ). We demonstrate the approach by developing UQ estimates on `DeepSD', a super-resolution based DL model for Statistical Downscaling (SD) in climate applied to precipitation, which follows an extremely skewed distribution. We find that the discrete-continuous models outperform a basic Gaussian distribution in terms of predictive accuracy and uncertainty calibration. Furthermore, we find that the lognormal distribution, which can handle skewed distributions, produces quality uncertainty estimates at the extremes. Such results may be important across S&E, as well as other domains such as finance and economics, where extremes are often of significant interest. Furthermore, to our knowledge, this is the first UQ model in SD where both aleatoric and epistemic uncertainties are characterized., Comment: 10 Pages

The impacts of climate change are felt by most critical systems, such as infrastructure, ecological systems, and power-plants. However, contemporary Earth System Models (ESM) are run at spatial resolutions too coarse for assessing effects this localized. Local scale projections can be obtained using statistical downscaling, a technique which uses historical climate observations to learn a low-resolution to high-resolution mapping. Depending on statistical modeling choices, downscaled projections have been shown to vary significantly terms of accuracy and reliability. The spatio-temporal nature of the climate system motivates the adaptation of super-resolution image processing techniques to statistical downscaling. In our work, we present DeepSD, a generalized stacked super resolution convolutional neural network (SRCNN) framework for statistical downscaling of climate variables. DeepSD augments SRCNN with multi-scale input channels to maximize predictability in statistical downscaling. We provide a comparison with Bias Correction Spatial Disaggregation as well as three Automated-Statistical Downscaling approaches in downscaling daily precipitation from 1 degree (~100km) to 1/8 degrees (~12.5km) over the Continental United States. Furthermore, a framework using the NASA Earth Exchange (NEX) platform is discussed for downscaling more than 20 ESM models with multiple emission scenarios., Comment: 9 pages, 5 Figures, 2 Tables

We investigate the use of Deep Neural Networks for the classification of image datasets where texture features are important for generating class-conditional discriminative representations. To this end, we first derive the size of the feature space for some standard textural features extracted from the input dataset and then use the theory of Vapnik-Chervonenkis dimension to show that hand-crafted feature extraction creates low-dimensional representations which help in reducing the overall excess error rate. As a corollary to this analysis, we derive for the first time upper bounds on the VC dimension of Convolutional Neural Network as well as Dropout and Dropconnect networks and the relation between excess error rate of Dropout and Dropconnect networks. The concept of intrinsic dimension is used to validate the intuition that texture-based datasets are inherently higher dimensional as compared to handwritten digits or other object recognition datasets and hence more difficult to be shattered by neural networks. We then derive the mean distance from the centroid to the nearest and farthest sampling points in an n-dimensional manifold and show that the Relative Contrast of the sample data vanishes as dimensionality of the underlying vector space tends to infinity., Comment: Accepted in International Joint Conference on Neural Networks, IJCNN 2016

Satellite image classification is a challenging problem that lies at the crossroads of remote sensing, computer vision, and machine learning. Due to the high variability inherent in satellite data, most of the current object classification approaches are not suitable for handling satellite datasets. The progress of satellite image analytics has also been inhibited by the lack of a single labeled high-resolution dataset with multiple class labels. The contributions of this paper are twofold - (1) first, we present two new satellite datasets called SAT-4 and SAT-6, and (2) then, we propose a classification framework that extracts features from an input image, normalizes them and feeds the normalized feature vectors to a Deep Belief Network for classification. On the SAT-4 dataset, our best network produces a classification accuracy of 97.95% and outperforms three state-of-the-art object recognition algorithms, namely - Deep Belief Networks, Convolutional Neural Networks and Stacked Denoising Autoencoders by ~11%. On SAT-6, it produces a classification accuracy of 93.9% and outperforms the other algorithms by ~15%. Comparative studies with a Random Forest classifier show the advantage of an unsupervised learning approach over traditional supervised learning techniques. A statistical analysis based on Distribution Separability Criterion and Intrinsic Dimensionality Estimation substantiates the effectiveness of our approach in learning better representations for satellite imagery., Comment: Paper was accepted at ACM SIGSPATIAL 2015

Learning sparse feature representations is a useful instrument for solving an unsupervised learning problem. In this paper, we present three labeled handwritten digit datasets, collectively called n-MNIST. Then, we propose a novel framework for the classification of handwritten digits that learns sparse representations using probabilistic quadtrees and Deep Belief Nets. On the MNIST and n-MNIST datasets, our framework shows promising results and significantly outperforms traditional Deep Belief Networks., Comment: Published in the European Symposium on Artificial Neural Networks, ESANN 2015

Obtaining accurate segmentation on large scale images is an open problem in deep learning. The main problem is the amount of labeled data that exists for large scale images. Traditionally, the common solution to this problem is to crop the large images into smaller images to increase the amount of available data and train a Conditional Generative Adversarial Network (CGAN). CGANs are currently the state of the art in image to image translation and provide better accuracy than the traditional method of training a encoder based conv-net architecture to minimize the loss at each pixel. This method can produce noisy and discontinuous images with inaccurate results. We seek to solve this problem by utilizing the concepts of transfer learning and progressive training to create a CGAN that can segment large scale images with a limited amount of labeled data. In transfer learning we recognize that many learned features are applicable to many classes from multiple domains. This introduces the concept of feature reusability, which is the basis for finetuning. Progressive training got its start in training models on the same images at different resolutions. In this work we instead train a GAN on increasing image scales by transferring the weights from the smaller scales to the larger scales. The learned features at the smaller scales are continually reused and applied to larger scales to create a CGAN that can perform accurate segmentation on large scale images. We apply this method to detect building footprints on very high-resolution overhead imagery (e.g Digital Globe and high resolution airborne platforms).

Spectral vegetation index (VI) time series data from moderate resolution sensors, such as Earth Observing System (EOS) Moderate Resolution Imaging Spectroradiometer (MODIS), have widely been used to identify "hot spot" areas of vegetation changes and to characterize long-term trends of vegetation changes. The Visible Infrared Imaging Radiometer Suite (VIIRS) sensor series of the Joint Polar Satellite System (JPSS) program is slated to continue the highly calibrated data stream initiated with MODIS. The first VIIRS sensor has operationally been acquiring Earth-reflected radiation since March 2012. The second VIIRS sensor (NOAA-20) was launched in November 2017, which is to continue the measurements made by the first VIIRS sensor. In this study, we examined an integrated use of MODIS and VIIRS VI time series data on capturing vegetation dynamics in the Asia-Pacific region using their overlapped period of observations (2013-2017). Three VIs, the "top-of-canopy (TOC)" normalized difference vegetation index (NDVI), TOC enhanced vegetation index (EVI), and TOC two-band enhanced vegetation index (EVI2), were investigated. For all the three VIs, VIIRS VIs were systematically higher than the MODIS counterparts due mainly to their spectral bandpass differences. However, both VIIRS and MODIS VIs showed the comparable spatial patterns in their temporal variations. Empirical spectral corrections allowed to merge the two data streams, which slightly improved the temporal resolution of the VI temporal profiles. These results suggest the suitability of VIIRS data to extend and merge into the MODIS VI record for long-term vegetation dynamics studies.

The new generation geostationary (GEO) remote sensors (GOES-R ABI, Himawari AHI, and FY4 AGRI) provide high frequency (5-15 minute) observations spatially/spectrally similar to MODIS/VIIRS for land monitoring. These new features of GEO satellite sensors make producing MODIS like land products for terrestrial monitoring possible. The NASA Earth Exchange (NEX) team developed the GEONEX pipeline that is containerized, deployable on NASA Pleiades supercomputer as well as public cloud platforms (e.g. AWS). The processing pipeline is designed to take Himawari Standard Data (HSD) and GOES-16 L1b to generate surface reflectance (SR) and other high-level land remote sensing products. In order to produce low-Earth-orbiting (LEO) remote sensing compatible land products, inter-comparison between Himawari AHI and MODIS Terra/Aqua has been conducted in this research work. Comparisons of TOA reflectance and surface reflectance between AHI and Terra/Aqua are presented. Ray-Matching method was used to locate the co-located pixels, where GEO and LEO sensors look at the land target with similar Viewing Zenith Angle (VZA) and Viewing Azimuth Angle (VAA) simultaneously. Here, we address challenges associated with the selection of qualified pixels of similar solar illumination condition and atmosphere path. We used strict criterion to constrain the pixel selection: the time difference between GEO and LEO observations is less than +-2.5 mins, the cosine of VZA difference is less than 1%, and the VAA difference is less than 10 deg. We also discuss the strong radiometric consistency that the new generation GEO sensors along with the popular LEO sensors would benefit the environmental remote sensing community.

GEONEX is a processing pipeline that produces a suite of satellite land surface products using data streams from the latest geostationary (GEO) sensors including the GOES016/ABI and the Himawari-8/AHI. The suite, created collaboratively by scientists from NASA and NOAA, includes top-of-atmosphere (TOA) reflectances, land surface reflectances (LSRs), vegetation indices, LAI/fPAR, and other downstream products. As a key component of the GEONEX product processing, we have adapted the Multi-Angle Implementation of Atmospheric Correction (MAIAC) algorithm to produce LSRs from the TOA data. Because the algorithm depends on building "stacks" of images, we first run internal geo-registration checks to ensure geo-spatial accuracy and consistency of the input (L1B) data before transferring them from the geostationary projection into a tile system in geographic grids. Scan-time is inferred from metadata and applied to calculate the sun-sensor angles for each grid cell. The MAIAC algorithm is run to detect clouds/shadows, estimate aerosol optical thickness (AOT), perform atmospheric corrections, and generate LSRs. We have processed 18-months (from 2016/04 onward) of AHI data over East Asia and Oceania at a 10-minute time step and 10-months (from 2018/01 onward) of ABI data over North and South Americas at a 15-minute time step. As a verification measure, we compare the GEONEX (AHI/ABI) surface reflectances with the standard MODIS products (MOD09GA) and the MODIS MAIAC products over pixels that have similar sun-view geometries. The results indicate general linear relationships between GEONEX and corresponding MODIS LSRs. In particular, the RMSEs between GEONEX and MOD09 data are comparable to those between MOD09 and MODIS MAIAC products, suggesting that the uncertainties of GEONEX LSRs fall into an acceptable range. However, direct comparisons of LSRs over pixels with different sun-view angles are not as straightforward and require more modeling efforts to correct the directional effects. Evaluation of such angular influences on the downstream products (e.g., vegetation indices) is also under investigation.

In this work, we use a Fully Constrained Least Squares Subpixel Learning Algorithm to unmix global WELD (Web Enabled Landsat Data) to obtain fractions or abundances of substrate (S), vegetation (V) and dark objects (D) classes. Because of the sheer nature of data and compute needs, we leveraged the NASA Earth Exchange (NEX) high performance computing architecture to optimize and scale our algorithm for large-scale processing. Subsequently, the S-V-D abundance maps were characterized into 4 classes namely, forest, farmland, water and urban areas (with NPP-VIIRS-national polar orbiting partnership visible infrared imaging radiometer suite nighttime lights data) over California, USA using Random Forest classifier. Validation of these land cover maps with NLCD (National Land Cover Database) 2011 products and NAFD (North American Forest Dynamics) static forest cover maps showed that an overall classification accuracy of over 91 percent was achieved, which is a 6 percent improvement in unmixing based classification relative to per-pixel-based classification. As such, abundance maps continue to offer an useful alternative to high-spatial resolution data derived classification maps for forest inventory analysis, multi-class mapping for eco-climatic models and applications, fast multi-temporal trend analysis and for societal and policy-relevant applications needed at the watershed scale.

Cloud and cloud shadow detection has important applications in weather and climate studies. It is even more crucial when we introduce geostationary satellites into the field of terrestrial remote sensing. With the challenges associated with data acquired in very high frequency (10-15 mins per scan), the ability to derive an accurate cloud shadow mask from geostationary satellite data is critical. The key to the success for most of the existing algorithms depends on spatially and temporally varying thresholds,which better capture local atmospheric and surface effects.However, the selection of proper threshold is difficult and may lead to erroneous results. In this work, we propose a deep neural network based approach called CloudCNN to classify cloudshadow from Himawari-8 AHI and GOES-16 ABI multispectral data. DeepSAT's CloudCNN consists of an encoderdecoder based architecture for binary-class pixel wise segmentation. We train CloudCNN on multi-GPU Nvidia Devbox cluster, and deploy the prediction pipeline on NASA Earth Exchange (NEX) Pleiades supercomputer. We achieved an overall accuracy of 93.29% on test samples. Since, the predictions take only a few seconds to segment a full multispectral GOES-16 or Himawari-8 Full Disk image, the developed framework can be used for real-time cloud detection, cyclone detection, or extreme weather event predictions.

Uncertainties in input land cover estimates contribute to a significant bias in modeled above ground biomass (AGB) and carbon estimates from satellite-derived data. The resolution of most currently used passive remote sensing products is not sufficient to capture tree canopy cover of less than ca. 10-20 percent, limiting their utility to estimate canopy cover and AGB for trees outside of forest land. In our study, we created a first of its kind Continental United States (CONUS) tree cover map at a spatial resolution of 1-m for the 2010-2012 epoch using the USDA NAIP imagery to address the present uncertainties in AGB estimates. The process involves different tasks including data acquisition ingestion to pre-processing and running a state-of-art encoder-decoder based deep convolutional neural network (CNN) algorithm for automatically generating a tree non-tree map for almost a quarter million scenes. The entire processing chain including generation of the largest open source existing aerial satellite image training database was performed at the NEX supercomputing and storage facility. We believe the resulting forest cover product will substantially contribute to filling the gaps in ongoing carbon and ecological monitoring research and help quantifying the errors and uncertainties in derived products.

The latest generation of geostationary satellites carry sensors such as ABI (Advanced Baseline Imager on GOES-16) and the AHI (Advanced Himawari Imager on Himawari) that closely mimic the spatial and spectral characteristics of Earth Observing System flagship MODIS for monitoring land surface conditions. More importantly they provide observations at 5-15 minute intervals. Such high frequency data offer exciting possibilities for producing robust estimates of land surface conditions by overcoming cloud cover, enabling studies of diurnally varying local-to-regional biosphere-atmosphere interactions, and operational decision-making in agriculture, forestry and disaster management. But the data come with challenges that need special attention. For instance, geostationary data feature changing sun angle at constant view for each pixel, which is reciprocal to sun-synchronous observations, and thus require careful adaptation of EOS algorithms. Our goal is to produce a set of land surface products from geostationary sensors by leveraging NASA's investments in EOS algorithms and in the data/compute facility NEX. The land surface variables of interest include atmospherically corrected surface reflectances, snow cover, vegetation indices and leaf area index (LAI)/fraction of photosynthetically absorbed radiation (FPAR), as well as land surface temperature and fires. In order to get ready to produce operational products over the US from GOES-16 starting 2018, we have utilized 18 months of data from Himawari AHI over Australia to test the production pipeline and the performance of various algorithms for our initial tests. The end-to-end processing pipeline consists of a suite of modules to (a) perform calibration and automatic georeference correction of the AHI L1b data, (b) adopt the Multi-Angle Implementation of Atmospheric Correction (MAIAC) algorithm to produce surface spectral reflectances along with compositing schemes and QA, and (c) modify relevant EOS retrieval algorithms (e.g., LAI and FPAR, GPP, etc.) for subsequent science product generation. Initial evaluation of Himawari AHI products against standard MODIS products indicate general agreement, suggesting that data from geostationary sensors can augment low earth orbit (LEO) satellite observations.

The successful launch of the European Space Agency (ESA) Sentinel-2A (S2-A) on 23 June 2015 with its MultiSpectral Instrument (MSI) provides an important means to augment Earth-observation capabilities following the legacy of Landsat. After the three-month satellite commissioning campaign, the MSI onboard S-2A is performing very well (ESA, 2015). By 3 December 2015, the sensor data records have achieved provisional maturity status and have been accessed in level-1C Top-Of-Atmosphere (TOA) reflectance by the remote sensing community worldwide. Near-nadir observations by the MSI onboard S-2A and the Operational Land Imager (OLI) onboard Landsat 8 were collected during Simultaneous Nadir Overpasses as well as nearly coincident overpasses. This paper presents a processing chain using harmonized S-2A MSI and Landsat 8 OLI sensors to obtain increased temporal resolution in Leaf Area Index (LAI) estimates using the red-edge band B8A of MSI to replace the NIR band B08. Results demonstrate that LAI estimates from the MSI and OLI are comparable, and, given sufficient preprocessing for atmospheric correction and geometric rectification, can be used interchangeably to improve the frequency with which low LAI canopies can be monitored.

Amazon droughts in 2005 and 2010 have raised serious concern about the future of the rainforest. Amazon forests are crucial because of their role as the largest carbon sink in the world which would effect the global warming phenomena with decreased photosynthesis activity. Especially, after a decline in plant growth in 1.68 million km2 forest area during the once-in-a-century severe drought in 2010, it is of primary importance to understand the relationship between different climatic variables and vegetation. In an earlier study, we have shown that non-linear models are better at capturing the relation dynamics of vegetation and climate variables such as temperature and precipitation, compared to linear models. In this research, we learn precise models between vegetation and climatic variables (temperature, precipitation) for normal conditions in the Amazon region using genetic programming based symbolic regression. This is done by removing high elevation and drought affected areas and also considering the slope of the region as one of the important factors while building the model. The model learned reveals new and interesting ways historical and current climate variables affect the vegetation at any location. MAIAC data has been used as a vegetation surrogate in our study. For temperature and precipitation, we have used TRMM and MODIS Land Surface Temperature data sets while learning the non-linear regression model. However, to generalize the model to make it independent of the data source, we perform transfer learning where we regress a regularized least squares to learn the parameters of the non-linear model using other data sources such as the precipitation and temperature from the Climatic Research Center (CRU). This new model is very similar in structure and performance compared to the original learned model and verifies the same claims about the nature of dependency between these climate variables and the vegetation in the Amazon region. As a result of this study, we are able to learn, for the very first time how exactly different climate factors influence vegetation at any location in the Amazon rainforests, independent of the specific sources from which the data has been obtained.

High resolution tree cover classification maps are needed to increase the accuracy of current land ecosystem and climate model outputs. Limited studies are in place that demonstrates the state-of-the-art in deriving very high resolution (VHR) tree cover products. In addition, most methods heavily rely on commercial softwares that are difficult to scale given the region of study (e.g. continents to globe). Complexities in present approaches relate to (a) scalability of the algorithm, (b) large image data processing (compute and memory intensive), (c) computational cost, (d) massively parallel architecture, and (e) machine learning automation. In addition, VHR satellite datasets are of the order of terabytes and features extracted from these datasets are of the order of petabytes. In our present study, we have acquired the National Agriculture Imagery Program (NAIP) dataset for the Continental United States at a spatial resolution of 1-m. This data comes as image tiles (a total of quarter million image scenes with ~60 million pixels) and has a total size of ~65 terabytes for a single acquisition. Features extracted from the entire dataset would amount to ~8-10 petabytes. In our proposed approach, we have implemented a novel semi-automated machine learning algorithm rooted on the principles of "deep learning" to delineate the percentage of tree cover. Using the NASA Earth Exchange (NEX) initiative, we have developed an end-to-end architecture by integrating a segmentation module based on Statistical Region Merging, a classification algorithm using Deep Belief Network and a structured prediction algorithm using Conditional Random Fields to integrate the results from the segmentation and classification modules to create per-pixel class labels. The training process is scaled up using the power of GPUs and the prediction is scaled to quarter million NAIP tiles spanning the whole of Continental United States using the NEX HPC supercomputing cluster. An initial pilot over the state of California spanning a total of 11,095 NAIP tiles covering a total geographical area of 163,696 sq. miles has produced true positive rates of around 88 percent for fragmented forests and 74 percent for urban tree cover areas, with false positive rates lower than 2 percent for both landscapes.

The NASA Earth Exchange (NEX) is a collaborative computing platform that has been developed with the objective of bringing scientists together with the software tools, massive global datasets, and supercomputing resources necessary to accelerate research in Earth systems science and global change. NEX is funded as an enabling tool for sustaining the national climate assessment. Over the past five years, researchers have used the NEX platform and produced a number of data sets highly relevant to the National Climate Assessment. These include high-resolution climate projections using different downscaling techniques and trends in historical climate from satellite data. To enable a broader community in exploiting the above datasets, the NEX team partnered with public cloud providers to create the OpenNEX platform. OpenNEX provides ready access to NEX data holdings on a number of public cloud platforms along with pertinent analysis tools and workflows in the form of Machine Images and Docker Containers, lectures and tutorials by experts. We will showcase some of the applications of OpenNEX data and tools by the community on Amazon Web Services, Google Cloud and the NEX Sandbox.

Plant phenology and maximum photosynthetic state determine spatiotemporal variability of gross primary productivity (GPP) of vegetation. Recent warming induced impacts accelerate shifts of phenology and physiological status over Northern vegetated land. Thus, understanding and quantifying these changes are very important. Here, we investigate 1) how vegetation phenology and physiological status (maximum photosynthesis) are evolved over last three decades and 2) how such components (phenology and physiological status) contribute on inter-annual variation of the GPP during the last three decades. We utilized both long-term remotely sensed (GIMMS (Global Inventory Modeling and Mapping Studies), NDVI3g (Normalized Difference Vegetation Index 3rd generation) and MODIS (Moderate Resolution Imaging Spectroradiometer)) to extract larger scale phenology metrics (growing season start, end and duration); and productivity (i.e., growing season integrated vegetation index, GSIVI) to answer these questions. For evaluation purpose, we also introduced field-measured phenology and productivity datasets (e.g., FLUXNET) and possible remotely-sensed and modeled metrics at continental and regional scales. From this investigation, we found that onset of the growing season has advanced by 1.61 days per decade and the growing season end has delayed by 0.67 days per decade over the circumpolar region. This asymmetric extension of growing season results in a longer growing-season trend (2.96 days per decade) and widespread increasing vegetation-productivity trend (2.96 GSIVI per decade) over Northern land. However, the regionally-diverged phenology shift and maximum photosynthetic state contribute differently characterized productivity, inter-annual variability and trend. We quantified that about 50 percent, 13 percent and 6.5 percent of Northern land's inter-annual variability are dominantly controlled by the onset of the growing season, the end of the growing season and the maximum photosynthetic state, respectively. Productivity characterization over the other approximately 30 percent region has been driven by these co-dominant drivers. Our study clearly shows that regionally different contribution of phenological and physiological components on characterizing vegetation production over the last three decades.

This work demonstrates the development and implementation of a Fully Constrained Least Squares (FCLS) unmixing model developed in C++ programming language with OpenCV package and boost C++ libraries in the NASA Earth Exchange (NEX). Visualization of the results is supported by GRASS GIS and statistical analysis is carried in R in a Linux system environment. FCLS was first tested on computer simulated data with Gaussian noise of various signal-to-noise ratio, and Landsat data of an agricultural scenario and an urban environment using a set of global end members of substrate (soils, sediments, rocks, and non-photosynthetic vegetation), vegetation that includes green photosynthetic plants and dark objects which encompasses absorptive substrate materials, clear water, deep shadows, etc. For the agricultural scenario, a spectrally diverse collection of 11 scenes of Level 1 terrain corrected, cloud free Landsat-5 TM data of Fresno, California, USA were unmixed and the results were validated with the corresponding ground data. To study an urbanized landscape, a clear sky Landsat-5 TM data were unmixed and validated with coincident World View-2 abundance maps (of 2 m spatial resolution) for an area of San Francisco, California, USA. The results were evaluated using descriptive statistics, correlation coefficient, RMSE, probability of success, boxplot and bivariate distribution function. Finally, FCLS was used for sub-pixel land cover analysis of the monthly WELD (Wen-enabled Landsat data) repository from 2008 to 2011 of North America. The abundance maps in conjunction with DMSP-OLS nighttime lights data were used to extract the urban land cover features and analyze their spatial-temporal growth.

Tree cover delineation is a useful instrument in deriving Above Ground Biomass (AGB) density estimates from Very High Resolution (VHR) airborne imagery data. Numerous algorithms have been designed to address this problem, but most of them do not scale to these datasets, which are of the order of terabytes. In this paper, we present a semi-automated probabilistic framework for the segmentation and classification of 1-m National Agriculture Imagery Program (NAIP) for tree-cover delineation for the whole of Continental United States, using a High Performance Computing Architecture. Classification is performed using a multi-layer Feedforward Backpropagation Neural Network and segmentation is performed using a Statistical Region Merging algorithm. The results from the classification and segmentation algorithms are then consolidated into a structured prediction framework using a discriminative undirected probabilistic graphical model based on Conditional Random Field, which helps in capturing the higher order contextual dependencies between neighboring pixels. Once the final probability maps are generated, the framework is updated and re-trained by relabeling misclassified image patches. This leads to a significant improvement in the true positive rates and reduction in false positive rates. The tree cover maps were generated for the whole state of California, spanning a total of 11,095 NAIP tiles covering a total geographical area of 163,696 sq. miles. The framework produced true positive rates of around 88% for fragmented forests and 74% for urban tree cover areas, with false positive rates lower than 2% for both landscapes. Comparative studies with the National Land Cover Data (NLCD) algorithm and the LiDAR canopy height model (CHM) showed the effectiveness of our framework for generating accurate high-resolution tree-cover maps.

Urban expansion and the associated changes in land cover have important climatic, hydrologic, biophysical and ecologic and socio-economic impacts on the environment. Yet, despite todays abundance of remote sensing data, an automated characterization of large-scale historical changes in urban spatial extent remains a challenge due to the inherent complexity and variability of the urban environment, the lack of a spectral signature unique to urban land cover, and the absence of an unambiguous definition of what is urban versus non-urban.Here we present a consistent, robust, scalable, physically- based methodology for characterization of urban expansion using Landsat observations. We use atmospherically corrected Landsat Global Land Survey time series, Web-enabled Landsat data time series, DMSP-OLS and NPP-VIIRS nighttime lights, for mapping the built-up and vegetated components of urban settlements at 30m resolution through multi- temporal standardized spectral mixture analysis. The methodology is tested and validated over the North American continent where it provides a first quantification of urban expansion and vegetation abundance changes from 1990 to 2010.

This work investigates variations in satellite-measured greenness of Amazon forests using ten years of NASA Moderate Resolution Imaging Spectroradiometer (MODIS) enhanced vegetation index (EVI) data. Corruption of optical remote sensing data with clouds and aerosols is prevalent in this region; filtering corrupted data causes spatial sampling constraints, as well as reducing the record length, which introduces large biases in estimates of greenness anomalies. The EVI data, analyzed in multiple ways and taking into account EVI accuracy, consistently show a pattern of negligible changes in the greenness levels of forests both in the area affected by drought in 2005 and outside it. Small random patches of anomalous greening and browning-especially prominent in 2009-appear in all ten years, irrespective of contemporaneous variations in precipitation, but with no persistence over time. The fact that over 90% of the EVI anomalies are insignificantly small-within the envelope of error (95% confidence interval) in EVI-warrants cautious interpretation of these results: there were no changes in the greenness of these forests, or if there were changes, the EVI data failed to capture these either because the constituent reflectances were saturated or the moderate resolution precluded viewing small-scale variations. This suggests a need for more accurate and spatially resolved synoptic views from satellite data and corroborating comprehensive ground sampling to understand the greenness dynamics of these forests.