We present a fully convolutional network(FCN) based approach for color image restoration. FCNs have recently shown remarkable performance for high-level vision problem like semantic segmentation. In this paper, we investigate if FCN models can show promising performance for low-level problems like image restoration as well. We propose a fully convolutional model, that learns a direct end-to-end mapping between the corrupted images as input and the desired clean images as output. Our proposed method takes inspiration from domain transformation techniques but presents a data-driven task specific approach where filters for novel basis projection, task dependent coefficient alterations, and image reconstruction are represented as convolutional networks. Experimental results show that our FCN model outperforms traditional sparse coding based methods and demonstrates competitive performance compared to the state-of-the-art methods for image denoising. We further show that our proposed model can solve the difficult problem of blind image inpainting and can produce reconstructed images of impressive visual quality.

On October 20, 2016, Daesh (Islamic State) set fire to the sulphur production site Al-Mishraq as the battle of Mosul in northern Iraq became more intense. An extensive plume of toxic sulphur dioxide and hydrogen sulphide caused comprehensive casualties. The intensity of the SO2 release was reaching levels of minor volcanic eruptions and the plume was observed by several satellites. By investigation of the measurement data from instruments on the MetOp-A, MetOp-B, Aura and Soumi satellites we have estimated the time-dependent source term to 161 kilotonnes sulphur dioxide released into the atmosphere during seven days. A long-range dispersion model was utilized to simulate the atmospheric transport over the Middle East. The ground level concentrations predicted by the simulation were compared with observation from the Turkey National Air Quality Monitoring Network. Finally, the simulation data provided, using a probit analysis of the simulated data, an estimate of the health risk area that was compared to reported urgent medical treatments. This article has been published in Atmospheric Environment, see cross link below.

We develop a statistical method for identifying induced seismicity from large datasets and apply the method to decades of wastewater disposal and seismicity data in California and Oklahoma. The method is robust against a variety of potential pitfalls. The study regions are divided into gridblocks. We use a longitudinal study design, seeking associations between seismicity and wastewater injection along time-series within each gridblock. The longitudinal design helps control for non-random application of wastewater injection. We define a statistical model that is flexible enough to describe the seismicity observations, which have temporal correlation and high kurtosis. In each gridblock, we find the maximum likelihood estimate for a model parameter that relates induced seismicity hazard to total volume of wastewater injected each year. To assess significance, we compute likelihood ratio test statistics in each gridblock and each state, California and Oklahoma. Resampling is used to empirically derive reference distributions used to estimate p-values from the likelihood ratio statistics. In Oklahoma, the analysis finds with extremely high confidence that seismicity associated with wastewater disposal (or other related activities, such as reservoir depletion) has occurred. In California, the analysis finds that seismicity associated with wastewater disposal has probably occurred, but the result is not strong enough to be conclusive. We identify areas where temporal association between wastewater disposal and seismicity is apparent. Our method could be applied to other datasets, extended to identify risk factors that increase induced seismic hazard, or modified to test alternative statistical models for natural and induced seismicity.

The Sun is an effective particle accelerator producing solar energetic particle (SEP) events during which particles up to several GeVs can be observed. Those events observed at Earth with the neutron monitor network are called ground level enhancements (GLEs). Although these events with a high energy component have been investigated for several decades, a clear relation between the spectral shape of the SEPs outside the Earth's magnetosphere and the increase in neutron monitor count rate has yet to be established. Hence, an analysis of these events is of interest for the space weather as well as the solar event community. In this work, SEP events with protons accelerated to above 500 MeV have been identified using data from the Electron Proton Helium Instrument (EPHIN) aboard the Solar and Heliospheric Observatory (SOHO) between 1995 and 2015. For a statistical analysis, onset times have been determined for the events and the proton energy spectra were derived and fitted with a power law. As a result, a list of 42 SEP events with protons accelerated to above 500 MeV measured with the EPHIN instrument onboard SOHO is presented. The statistical analysis based on the fitted spectral slopes and absolute intensities is discussed with special emphasis on whether or not an event has been observed as GLE. Furthermore, a correlation between the derived intensity at 500 MeV and the observed increase in neutron monitor count rate has been found for a subset of events.

Land Surface Temperature (LST) is one of the key parameters in the physics of land-surface processes on regional and global scales, combining the results of all surface-atmosphere interactions and energy fluxes between the surface and the atmosphere. With the advent of the European Space Agency (ESA) Sentinel 3 (S3) satellite, accurate LST retrieval methodologies are being developed by exploiting the synergy between the Ocean and Land Colour Instrument (OLCI) and the Sea and Land Surface Temperature Radiometer (SLSTR). In this paper we explain the implementation in the Basic ENVISAT Toolbox for (A)ATSR and MERIS (BEAM) and the use of one LST algorithm developed in the framework of the Synergistic Use of The Sentinel Missions For Estimating And Monitoring Land Surface Temperature (SEN4LST) project. The LST algorithm is based on the split-window technique with an explicit dependence on the surface emissivity. Performance of the methodology is assessed by using MEdium Resolution Imaging Spectrometer/Advanced Along-Track Scanning Radiometer (MERIS/AATSR) pairs, instruments with similar characteristics than OLCI/ SLSTR, respectively. The LST retrievals were properly validated against in situ data measured along one year (2011) in three test sites, and inter-compared to the standard AATSR level-2 product with satisfactory results. The algorithm is implemented in BEAM using as a basis the MERIS/AATSR Synergy Toolbox. Specific details about the processor validation can be found in the validation report of the SEN4LST project.

We analyze hourly seismic data measured at the Osservatorio Vesuviano Ovest (OVO, 1972-2014) and at the Bunker Est (BKE, 1999-2014) stations on the Mt. Vesuvius. The OVO record is complete for seismic events with magnitude M > 1.9. We demonstrate that before 1996 this record presents a daily oscillation that nearly vanishes afterwards. To determine whether a daily oscillation exists in the seismic activity of the Mt. Vesuvius, we use the higher quality BKE record that is complete for seismic events with magnitude M > 0.2. We demonstrate that BKE confirms that the seismic activity at the Mt. Vesuvius is higher during nighttime than during day-time. The amplitude of the daily oscillation is enhanced during summer and damped during winter. We speculate possible links with the cooling/warming diurnal cycle of the volcanic edifice, with external geomagnetic field and with magnetostriction that should also stress the rocks. We find that the amplitude of the seismic daily cycle changes in time and has been increasing since 2008. Finally, we propose a seismic activity index to monitor the 24-hour oscillation that could be used to complement other methodologies currently adopted to determine the seismic status of the volcano and to prevent the relative hazard.

Autonomous crop monitoring at high spatial and temporal resolution is a critical problem in precision agriculture. While Structure from Motion and Multi-View Stereo algorithms can finely reconstruct the 3D structure of a field with low-cost image sensors, these algorithms fail to capture the dynamic nature of continuously growing crops. In this paper we propose a 4D reconstruction approach to crop monitoring, which employs a spatio-temporal model of dynamic scenes that is useful for precision agriculture applications. Additionally, we provide a robust data association algorithm to address the problem of large appearance changes due to scenes being viewed from different angles at different points in time, which is critical to achieving 4D reconstruction. Finally, we collected a high quality dataset with ground truth statistics to evaluate the performance of our method. We demonstrate that our 4D reconstruction approach provides models that are qualitatively correct with respect to visual appearance and quantitatively accurate when measured against the ground truth geometric properties of the monitored crops.

The equations governing motion of the satellite under the effect of oblateness of Earth and atmospheric drag have been simulated, for a fixed initial position and three different initial velocities, till satellite collapses on Earth. Simulation of motion of artificial Earth satellite subject to the combined effects of oblate Earth and atmospheric drag is presented. The atmospheric model considered here takes in to account of exponential variation of the density with initial distance of Satellite from Earth's surface, scale height and radial distance. The minimum and maximum values of orbital elements and their variation over a time for different initial velocities have been reported.

Soybean is one of the ten greatest crops in the world, answering for billion-dollar businesses every year. This crop suffers from insect herbivory that costs millions from producers. Hence, constant monitoring of the crop foliar damage is necessary to guide the application of insecticides. However, current methods to measure foliar damage are expensive and dependent on laboratory facilities, in some cases, depending on complex devices. To cope with these shortcomings, we introduce an image processing methodology to measure the foliar damage in soybean leaves. We developed a non-destructive imaging method based on two techniques, Otsu segmentation and Bezier curves, to estimate the foliar loss in leaves with or without border damage. We instantiate our methodology in a mobile application named BioLeaf, which is freely distributed for smartphone users. We experimented with real-world leaves collected from a soybean crop in Brazil. Our results demonstrated that BioLeaf achieves foliar damage quantification with precision comparable to that of human specialists. With these results, our proposal might assist soybean producers, reducing the time to measure foliar damage, reducing analytical costs, and defining a commodity application that is applicable not only to soy, but also to different crops such as cotton, bean, potato, coffee, and vegetables.

Face detection is challenging as faces in images could be present at arbitrary locations and in different scales. We propose a three-stage cascade structure based on fully convolutional neural networks (FCNs). It first proposes the approximate locations where the faces may be, then aims to find the accurate location by zooming on to the faces. Each level of the FCN cascade is a multi-scale fully-convolutional network, which generates scores at different locations and in different scales. A score map is generated after each FCN stage. Probable regions of face are selected and fed to the next stage. The number of proposals is decreased after each level, and the areas of regions are decreased to more precisely fit the face. Compared to passing proposals directly between stages, passing probable regions can decrease the number of proposals and reduce the cases where first stage doesn't propose good bounding boxes. We show that by using FCN and score map, the FCN cascade face detector can achieve strong performance on public datasets.

A 23 hour tremor burst was recorded on July 8-9th 2011 at the Katla subglacial volcano, one of the most active and hazardous volcanoes in Iceland. This was associated with deepening of cauldrons on the ice cap and a glacial flood that caused damage to infrastructure. Increased earthquake activity within the caldera started a few days before and lasted for months afterwards and new seismic activity started on the south flank. No visible eruption broke the ice and the question arose as to whether this episode relates to a minor subglacial eruption with the tremor being generated by volcanic processes, or by the flood. The tremor signal consisted of bursts with varying amplitude and duration. We have identified and described three different tremor phases, based on amplitude and frequency features. A tremor phase associated with the flood was recorded only at stations closest to the river that flooded, correlating in time with rising water level observed at gauging stations. Using back-projection of double cross-correlations, two other phases have been located near the active ice cauldrons and are interpreted to be caused by volcanic or hydrothermal processes. The greatly increased seismicity and evidence of rapid melting of the glacier may be explained by a minor sub-glacial eruption. It is also plausible that the tremor was generated by hydrothermal boiling and/or explosions with no magma involved. This may have been induced by pressure drop triggered by the release of water when the glacial flood started. All interpretations require an increase of heat released by the volcano.

Environmental fluctuations have important consequences in the organization of ecological communities, and understanding how such a variability influences the biodiversity of an ecosystem is a major question in ecology. In this paper, we analyze the case of two species competing for the resources within the framework of the neutral theory in the presence of environmental noise, devoting special attention on how such a variability modulates species fitness. The environment is dichotomous and stochastically alternates between periods favoring one of the species while disfavoring the other one, preserving neutrality on the long term. We study two different scenarios: in the first one species fitness varies linearly with the environment, and in the second one the effective fitness is re-scaled by the total fitness of the individuals competing for the same resource. We find that, in the former case environmental fluctuations always reduce the time of species coexistence, whereas such a time can be enhanced or reduced in the latter case, depending on the correlation time of the environment. This phenomenon can be understood as a direct consequence of Chesson's storage effect.

A scenario-based Neo-Deterministic approach to Seismic Hazard Assessment (NDSHA) is available nowadays, which permits considering a wide range of possible seismic sources as the starting point for deriving scenarios by means of full waveforms modeling. The method does not make use of attenuation relations and naturally supplies realistic time series of ground shaking, including reliable estimates of ground displacement, readily applicable to complete engineering analysis. Based on the neo-deterministic approach, an operational integrated procedure for seismic hazard assessment has been developed that allows for the definition of time dependent scenarios of ground shaking, through the routine updating of earthquake predictions, performed by means of the algorithms CN and M8S. The integrated NDSHA procedure for seismic input definition, which is currently applied to the Italian territory, combines different pattern recognition techniques, designed for the space-time identification of strong earthquakes, with algorithms for the realistic modeling of ground motion. Accordingly, a set of deterministic scenarios of ground motion at bedrock, which refers to the time interval when a strong event is likely to occur within the alerted area, is defined both at regional and local scale. CN and M8S predictions, as well as the related time-dependent ground motion scenarios associated with the alarmed areas, are routinely updated since 2006. The prospective application of the time-dependent NDSHA approach provides information that can be useful in assigning priorities for timely mitigation actions and, at the same time, allows for a rigorous validation of the proposed methodology. The results from real-time testing of the time-dependent NDSHA scenarios are illustrated with specific reference to the August 24th, 2016 Central Italy earthquake.

Data continuity for the Landsat program relies on accurate cross-calibration among sensors. The Landsat 8 OLI has been shown to exhibit superior performance to the sensors on Landsats 4-7 with respect to radiometric calibration, signal to noise, and geolocation. However, improvements to the positioning of the spectral response functions on the OLI have resulted in known biases for commonly used spectral indices because the new band responses integrate absorption features differently from previous Landsat sensors. The objective of this analysis is to quantify the impact of these changes on linear spectral mixture models that use imagery collected by different Landsat sensors. The 2013 underflight of Landsat 7 and 8 provides an opportunity to cross calibrate the spectral mixing spaces of the ETM+ and OLI sensors using near-simultaneous acquisitions from a wide variety of land cover types worldwide. We use 80,910,343 pairs of OLI and ETM+ spectra to characterize the OLI spectral mixing space and perform a cross-calibration with ETM+. This new global collection of Landsat spectra spans a greater spectral diversity than those used in prior studies and the resulting Substrate, Vegetation, and Dark (SVD) spectral endmembers (EMs) supplant prior global Landsat EMs. We find only minor (-0.01 < u < 0.01) differences between SVD fractions unmixed using sensor-specific EMs. RMS misfit fractions are also small (<98% of pixels with <5% RMSE), in accord with past studies. Finally, vegetation is used as an example to illustrate the empirical and theoretical relationship between commonly used spectral indices and subpixel fractions. SVD fractions unmixed using global EMs thus provide easily computable, linearly scalable, physically based measures of subpixel land cover which can be compared accurately across the entire Landsat 4-8 archive without introducing any additional cross-sensor corrections.

Exploration of hydrocarbon resources is a highly complicated and expensive process where various geological, geochemical and geophysical factors are developed then combined together. It is highly significant how to design the seismic data acquisition survey and locate the exploratory wells since incorrect or imprecise locations lead to waste of time and money during the operation. The objective of this study is to locate high-potential oil and gas field in 1: 250,000 sheet of Ahwaz including 20 oil fields to reduce both time and costs in exploration and production processes. In this regard, 17 maps were developed using GIS functions for factors including: minimum and maximum of total organic carbon (TOC), yield potential for hydrocarbons production (PP), Tmax peak, production index (PI), oxygen index (OI), hydrogen index (HI) as well as presence or proximity to high residual Bouguer gravity anomalies, proximity to anticline axis and faults, topography and curvature maps obtained from Asmari Formation subsurface contours. To model and to integrate maps, this study employed artificial neural network and adaptive neuro-fuzzy inference system (ANFIS) methods. The results obtained from model validation demonstrated that the 17x10x5 neural network with R=0.8948, RMS=0.0267, and kappa=0.9079 can be trained better than other models such as ANFIS and predicts the potential areas more accurately. However, this method failed to predict some oil fields and wrongly predict some areas as potential zones.

In this paper, we study the influence of both long and short skip connections on Fully Convolutional Networks (FCN) for biomedical image segmentation. In standard FCNs, only long skip connections are used to skip features from the contracting path to the expanding path in order to recover spatial information lost during downsampling. We extend FCNs by adding short skip connections, that are similar to the ones introduced in residual networks, in order to build very deep FCNs (of hundreds of layers). A review of the gradient flow confirms that for a very deep FCN it is beneficial to have both long and short skip connections. Finally, we show that a very deep FCN can achieve near-to-state-of-the-art results on the EM dataset without any further post-processing.

Vegetation characteristics providing spatial heterogeneity at the channel reach scale can produce complex flow patterns and the relationship between plant patterns morphology and flow resistance is still an open question (Nepf 2012). Unlike experiments in laboratory, measuring the vegetation characteristics related to flow resistance on open channel in situ is difficult. Thanks to its high resolution and light weight, scanner lasers allow now to collect in situ 3D vegetation characteristics. In this study we used a 1064 nm usual Terrestrial Laser Scanner (TLS) located 5 meters at nadir above a 8 meters long equipped channel in order to both i) characterize the vegetation structure heterogeneity within the channel form a single scan (blockage factor, canopy height) and ii) to measure the 2D water level all over the channel during steady flow within a few seconds scan. This latter measuring system was possible thanks to an additive dispersive product sprinkled at the water surface. Vegetation characteristics and water surfaces during steady flows from 6 different plant spatial design on channel bottom for 4 plant species were thus measured. Vegetation blockage factors at channel scale were estimated from TLS points clouds and analyzed.

A modified version of a published slug test model for unconfined aquifers is applied to cross-hole slug test data collected in field tests conducted at the Widen site in Switzerland. The model accounts for water-table effects using the linearised kinematic condition. The model also accounts for inertial effects in source and observation wells. The primary objective of this work is to demonstrate applicability of this semi-analytical model to multi-well and multi-level pneumatic slug tests. The pneumatic perturbation was applied at discrete intervals in a source well and monitored at discrete vertical intervals in observation wells. The source and observation well pairs were separated by distances of up to 4 m. The analysis yielded vertical profiles of hydraulic conductivity, specific storage, and specific yield at observation well locations. The hydraulic parameter estimates are compared to results from prior pumping and single-well slug tests conducted at the site, as well as to estimates from particle size analyses of sediment collected from boreholes during well installation. The results are in general agreement with results from prior tests and are indicative of a sand and gravel aquifer. Sensitivity analysis show that model identification of specific yield is strongest at late-time. However, the usefulness of late-time data is limited due to the low signal-to-noise ratios.

Within a GIS environment, we combine field measures of mangrove diameter, mangrove species distribution, and mangrove density with remotely sensed measures of mangrove location and mangrove canopy cover to estimate the mangrove carbon holdings of northern Ecuador. We find that the four northern estuaries of Ecuador contain approximately 7,742,999 t (plus or minus 15.47 percent) of standing carbon. Of particular high carbon holdings are the Rhizophora mangle dominated mangrove stands found in-and-around the Cayapas-Mataje Ecological Reserve in northern Esmeraldas Province, Ecuador and certain stands of Rhizophora mangle in-and-around the Isla Corazon y Fragata Wildlife Refuge in central Manabi Province, Ecuador. Our field driven mangrove carbon estimate is higher than all but one of the comparison models evaluated. We find that basic latitudinal mangrove carbon models performed at least as well, if not better, than the more complex species based allometric models in predicting standing carbon levels. In addition, we find that improved results occur when multiple models are combined as opposed to relying any one single model for mangrove carbon estimates. The high level of carbon contained in these mangrove forests, combined with the future atmospheric carbon sequestration potential they offer, makes it a necessity that they are included in any future payment for ecosystem services strategy aimed at utilizing forest systems to reduce CO2 emissions and mitigate predicted CO2 driven temperature increases.

The wide field of view (WFV) imaging system onboard the Chinese GaoFen-1 (GF-1) optical satellite has a 16-m resolution and four-day revisit cycle for large-scale Earth observation. The advantages of the high temporal-spatial resolution and the wide field of view make the GF-1 WFV imagery very popular. However, cloud cover is an inevitable problem in GF-1 WFV imagery, which influences its precise application. Accurate cloud and cloud shadow detection in GF-1 WFV imagery is quite difficult due to the fact that there are only three visible bands and one near-infrared band. In this paper, an automatic multi-feature combined (MFC) method is proposed for cloud and cloud shadow detection in GF-1 WFV imagery. The MFC algorithm first implements threshold segmentation based on the spectral features and mask refinement based on guided filtering to generate a preliminary cloud mask. The geometric features are then used in combination with the texture features to improve the cloud detection results and produce the final cloud mask. Finally, the cloud shadow mask can be acquired by means of the cloud and shadow matching and follow-up correction process. The method was validated using 108 globally distributed scenes. The results indicate that MFC performs well under most conditions, and the average overall accuracy of MFC cloud detection is as high as 96.8%. In the contrastive analysis with the official provided cloud fractions, MFC shows a significant improvement in cloud fraction estimation, and achieves a high accuracy for the cloud and cloud shadow detection in the GF-1 WFV imagery with fewer spectral bands. The proposed method could be used as a preprocessing step in the future to monitor land-cover change, and it could also be easily extended to other optical satellite imagery which has a similar spectral setting.

Nonlinear plant-scale interactions controlling the soil-water balance are generally not valid at larger spatial scales due to spatial heterogeneity in rainfall and vegetation type. The relationships between spatially averaged variables are hysteretic even when unique relationships are imposed at the plant scale. The characteristics of these hysteretic relationships depend on the size of the averaging area and the spatial properties of the soil, vegetation, and rainfall. We upscale the plant-scale relationships to the scale of a regional land-surface model based on simulation data obtained through explicit representation of spatial heterogeneity in rainfall and vegetation type. The proposed upscaled function improves predictions of spatially averaged soil moisture and evapotranspiration relative to the effective-parameter approach for a water-limited Texas shrubland. The degree of improvement is a function of the scales of heterogeneity and the size of the averaging area. We also find that single-valued functions fail to predict spatially averaged leakage accurately. Furthermore, the spatial heterogeneity results in scale-dependent hysteretic relationships for the statistical-dynamic and Montaldo & Albertson approaches.

This paper considers the quantification of the prediction performance in Gaussian process regression. The standard approach is to base the prediction error bars on the theoretical predictive variance, which is a lower bound on the mean square-error (MSE). This approach, however, does not take into account that the statistical model is learned from the data. We show that this omission leads to a systematic underestimation of the prediction errors. Starting from a generalization of the Cram\'er-Rao bound, we derive a more accurate MSE bound which provides a measure of uncertainty for prediction of Gaussian processes. The improved bound is easily computed and we illustrate it using synthetic and real data examples. of uncertainty for prediction of Gaussian processes and illustrate it using synthetic and real data examples.

In this work we address the task of semantic image segmentation with Deep Learning and make three main contributions that are experimentally shown to have substantial practical merit. First, we highlight convolution with upsampled filters, or 'atrous convolution', as a powerful tool in dense prediction tasks. Atrous convolution allows us to explicitly control the resolution at which feature responses are computed within Deep Convolutional Neural Networks. It also allows us to effectively enlarge the field of view of filters to incorporate larger context without increasing the number of parameters or the amount of computation. Second, we propose atrous spatial pyramid pooling (ASPP) to robustly segment objects at multiple scales. ASPP probes an incoming convolutional feature layer with filters at multiple sampling rates and effective fields-of-views, thus capturing objects as well as image context at multiple scales. Third, we improve the localization of object boundaries by combining methods from DCNNs and probabilistic graphical models. The commonly deployed combination of max-pooling and downsampling in DCNNs achieves invariance but has a toll on localization accuracy. We overcome this by combining the responses at the final DCNN layer with a fully connected Conditional Random Field (CRF), which is shown both qualitatively and quantitatively to improve localization performance. Our proposed "DeepLab" system sets the new state-of-art at the PASCAL VOC-2012 semantic image segmentation task, reaching 79.7% mIOU in the test set, and advances the results on three other datasets: PASCAL-Context, PASCAL-Person-Part, and Cityscapes. All of our code is made publicly available online.

We propose a data-driven, coarse-graining formulation in the context of equilibrium statistical mechanics. In contrast to existing techniques which are based on a fine-to-coarse map, we adopt the opposite strategy by prescribing a probabilistic coarse-to-fine map. This corresponds to a directed probabilistic model where the coarse variables play the role of latent generators of the fine scale (all-atom) data. From an information-theoretic perspective, the framework proposed provides an improvement upon the relative entropy method and is capable of quantifying the uncertainty due to the information loss that unavoidably takes place during the CG process. Furthermore, it can be readily extended to a fully Bayesian model where various sources of uncertainties are reflected in the posterior of the model parameters. The latter can be used to produce not only point estimates of fine-scale reconstructions or macroscopic observables, but more importantly, predictive posterior distributions on these quantities. Predictive posterior distributions reflect the confidence of the model as a function of the amount of data and the level of coarse-graining. The issues of model complexity and model selection are seamlessly addressed by employing a hierarchical prior that favors the discovery of sparse solutions, revealing the most prominent features in the coarse-grained model. A flexible and parallelizable Monte Carlo - Expectation-Maximization (MC-EM) scheme is proposed for carrying out inference and learning tasks. A comparative assessment of the proposed methodology is presented for a lattice spin system and the SPC/E water model.

Most current semantic segmentation methods rely on fully convolutional networks (FCNs). However, their use of large receptive fields and many pooling layers cause low spatial resolution inside the deep layers. This leads to predictions with poor localization around the boundaries. Prior work has attempted to address this issue by post-processing predictions with CRFs or MRFs. But such models often fail to capture semantic relationships between objects, which causes spatially disjoint predictions. To overcome these problems, recent methods integrated CRFs or MRFs into an FCN framework. The downside of these new models is that they have much higher complexity than traditional FCNs, which renders training and testing more challenging. In this work we introduce a simple, yet effective Convolutional Random Walk Network (RWN) that addresses the issues of poor boundary localization and spatially fragmented predictions with very little increase in model complexity. Our proposed RWN jointly optimizes the objectives of pixelwise affinity and semantic segmentation. It combines these two objectives via a novel random walk layer that enforces consistent spatial grouping in the deep layers of the network. Our RWN is implemented using standard convolution and matrix multiplication. This allows an easy integration into existing FCN frameworks and it enables end-to-end training of the whole network via standard back-propagation. Our implementation of RWN requires just $131$ additional parameters compared to the traditional FCNs, and yet it consistently produces an improvement over the FCNs on semantic segmentation and scene labeling.

Biological systems are often modelled at different levels of abstraction depending on the particular aims/resources of a study. Such different models often provide qualitatively concordant predictions over specific parametrisations, but it is generally unclear whether model predictions are quantitatively in agreement, and whether such agreement holds for different parametrisations. Here we present a generally applicable statistical machine learning methodology to automatically reconcile the predictions of different models across abstraction levels. Our approach is based on defining a correction map, a random function which modifies the output of a model in order to match the statistics of the output of a different model of the same system. We use two biological examples to give a proof-of-principle demonstration of the methodology, and discuss its advantages and potential further applications.

Increases in atmospheric CO2 and CH4 result from a combination of forcing from anthropogenic emissions and Earth System feedbacks that reduce or amplify the effects of those emissions on atmospheric concentrations. Despite decades of research carbon-climate feedbacks remain poorly quantified. The impact of these uncertainties on future climate are of increasing concern, especially in the wake of recent climate negotiations. Emissions, long concentrated in the developed world, are now shifting to developing countries, where the emissions inventories have larger uncertainties. The fraction of anthropogenic CO2 remaining in the atmosphere has remained remarkably constant over the last 50 years. Will this change in the future as the climate evolves? Concentrations of CH4, the 2nd most important greenhouse gas, which had apparently stabilized, have recently resumed their increase, but the exact cause for this is unknown. While greenhouse gases affect the global atmosphere, their sources and sinks are remarkably heterogeneous in time and space, and traditional in situ observing systems do not provide the coverage and resolution to attribute the changes to these greenhouse gases to specific sources or sinks. In the past few years, space-based technologies have shown promise for monitoring carbon stocks and fluxes. Advanced versions of these capabilities could transform our understanding and provide the data needed to quantify carbon-climate feedbacks. A new observing system that allows resolving global high resolution fluxes will capture variations on time and space scales that allow the attribution of these fluxes to underlying mechanisms.

Semantic segmentation is critical to image content understanding and object localization. Recent development in fully-convolutional neural network (FCN) has enabled accurate pixel-level labeling. One issue in previous works is that the FCN based method does not exploit the object boundary information to delineate segmentation details since the object boundary label is ignored in the network training. To tackle this problem, we introduce a double branch fully convolutional neural network, which separates the learning of the desirable semantic class labeling with mask-level object proposals guided by relabeled boundaries. This network, called object boundary guided FCN (OBG-FCN), is able to integrate the distinct properties of object shape and class features elegantly in a fully convolutional way with a designed masking architecture. We conduct experiments on the PASCAL VOC segmentation benchmark, and show that the end-to-end trainable OBG-FCN system offers great improvement in optimizing the target semantic segmentation quality.

Fully convolutional networks (FCNs) have been proven very successful for semantic segmentation, but the FCN outputs are unaware of object instances. In this paper, we develop FCNs that are capable of proposing instance-level segment candidates. In contrast to the previous FCN that generates one score map, our FCN is designed to compute a small set of instance-sensitive score maps, each of which is the outcome of a pixel-wise classifier of a relative position to instances. On top of these instance-sensitive score maps, a simple assembling module is able to output instance candidate at each position. In contrast to the recent DeepMask method for segmenting instances, our method does not have any high-dimensional layer related to the mask resolution, but instead exploits image local coherence for estimating instances. We present competitive results of instance segment proposal on both PASCAL VOC and MS COCO.

We introduce a guide to help deep learning practitioners understand and manipulate convolutional neural network architectures. The guide clarifies the relationship between various properties (input shape, kernel shape, zero padding, strides and output shape) of convolutional, pooling and transposed convolutional layers, as well as the relationship between convolutional and transposed convolutional layers. Relationships are derived for various cases, and are illustrated in order to make them intuitive.

TensorFlow is an interface for expressing machine learning algorithms, and an implementation for executing such algorithms. A computation expressed using TensorFlow can be executed with little or no change on a wide variety of heterogeneous systems, ranging from mobile devices such as phones and tablets up to large-scale distributed systems of hundreds of machines and thousands of computational devices such as GPU cards. The system is flexible and can be used to express a wide variety of algorithms, including training and inference algorithms for deep neural network models, and it has been used for conducting research and for deploying machine learning systems into production across more than a dozen areas of computer science and other fields, including speech recognition, computer vision, robotics, information retrieval, natural language processing, geographic information extraction, and computational drug discovery. This paper describes the TensorFlow interface and an implementation of that interface that we have built at Google. The TensorFlow API and a reference implementation were released as an open-source package under the Apache 2.0 license in November, 2015 and are available at www.tensorflow.org.

The Campi Flegrei volcanic field (Italy) poses very high risk to the highly urbanized Neapolitan area. Eruptive history was dominated by explosive activity producing pyroclastic currents (PDCs; (Proclastic Density Currents) ranging in scale from localized base surges to regional flows. Here we apply probabilistic numerical simulation approaches to produce PDC hazard maps, based on a comprehensive spectrum of flow properties and vent locations. These maps and provide all probable Volcanic Explosivity Index (VEI) scenarios from different source vents in the caldera, relevant for risk management planning. For each VEI scenario, we report the conditional probability for PDCs (i.e., the probability for a given area to be affected by the passage of PDCs) and related dynamic pressure. Model results indicate that PDCs from VEI<4 events would be confined within the Campi Flegrei caldera, PDC propagation being impeded by the northern and eastern caldera walls. Conversely, PDCs from VEI 4-5 events could invade a wide area beyond the northern caldera rim, as well as part of the Naples metropolitan area to the east. A major controlling factor of PDC dispersal is represented by the location of the vent area. PDCs from the potentially largest eruption scenarios (analogous to the ~15 ka, VEI 6 Neapolitan Yellow Tuff or even the ~39 ka, VEI 7 Campanian Ignimbrite extreme event) would affect a large part of the Campanian Plain to the north and the city of Naples to the east. Thus, in case of renewal of eruptive activity at Campi Flegrei, up to 3 million people will be potentially exposed to volcanic hazard, pointing out the urgency of an emergency plan. Considering the present level of uncertainty in forecasting the future eruption type, size and location we suggest that appropriate planning measures should face at least the VEI 5 reference scenario (at least 2 occurrences documented in the last 10 ka)

Some of the natural variability in climate is understood to come from changes in the Sun. A key route whereby the Sun may influence surface climate is initiated in the tropical stratosphere by the absorption of solar ultraviolet (UV) radiation by ozone, leading to a modification of the temperature and wind structures and consequently to the surface through changes in wave propagation and circulation. While changes in total, spectrally-integrated, solar irradiance lead to small variations in global mean surface temperature, the `top-down' UV effect preferentially influences on regional scales at mid-to-high latitudes with, in particular, a solar signal noted in the North Atlantic Oscillation (NAO). The amplitude of the UV variability is fundamental in determining the magnitude of the climate response but understanding of the UV variations has been challenged recently by measurements from the SOlar Radiation and Climate Experiment (SORCE) satellite, which show UV solar cycle changes up to 10 times larger than previously thought. Indeed, climate models using these larger UV variations show a much greater response, similar to NAO observations. Here we present estimates of the ozone solar cycle response using a chemistry-climate model (CCM) in which the effects of transport are constrained by observations. Thus the photolytic response to different spectral solar irradiance (SSI) datasets can be isolated. Comparison of the results with the solar signal in ozone extracted from observational datasets yields significantly discriminable responses. According to our evaluation the SORCE UV dataset is not consistent with the observed ozone response whereas the smaller variations suggested by earlier satellite datasets, and by UV data from empirical solar models, are in closer agreement with the measured stratospheric variations. Determining the most appropriate SSI variability to apply in models...

In this paper we study predictive pattern mining problems where the goal is to construct a predictive model based on a subset of predictive patterns in the database. Our main contribution is to introduce a novel method called safe pattern pruning (SPP) for a class of predictive pattern mining problems. The SPP method allows us to efficiently find a superset of all the predictive patterns in the database that are needed for the optimal predictive model. The advantage of the SPP method over existing boosting-type method is that the former can find the superset by a single search over the database, while the latter requires multiple searches. The SPP method is inspired by recent development of safe feature screening. In order to extend the idea of safe feature screening into predictive pattern mining, we derive a novel pruning rule called safe pattern pruning (SPP) rule that can be used for searching over the tree defined among patterns in the database. The SPP rule has a property that, if a node corresponding to a pattern in the database is pruned out by the SPP rule, then it is guaranteed that all the patterns corresponding to its descendant nodes are never needed for the optimal predictive model. We apply the SPP method to graph mining and item-set mining problems, and demonstrate its computational advantage.

We use numerical climate simulations, paleoclimate data, and modern observations to study the effect of growing ice melt from Antarctica and Greenland. Meltwater tends to stabilize the ocean column, inducing amplifying feedbacks that increase subsurface ocean warming and ice shelf melting. Cold meltwater and induced dynamical effects cause ocean surface cooling in the Southern Ocean and North Atlantic, thus increasing Earth's energy imbalance and heat flux into most of the global ocean's surface. Southern Ocean surface cooling, while lower latitudes are warming, increases precipitation on the Southern Ocean, increasing ocean stratification, slowing deepwater formation, and increasing ice sheet mass loss. These feedbacks make ice sheets in contact with the ocean vulnerable to accelerating disintegration. We hypothesize that ice mass loss from the most vulnerable ice, sufficient to raise sea level several meters, is better approximated as exponential than by a more linear response. Doubling times of 10, 20 or 40 years yield multi-meter sea level rise in about 50, 100 or 200 years. Recent ice melt doubling times are near the lower end of the 10-40 year range, but the record is too short to confirm the nature of the response. The feedbacks, including subsurface ocean warming, help explain paleoclimate data and point to a dominant Southern Ocean role in controlling atmospheric CO2, which in turn exercised tight control on global temperature and sea level. The millennial (500-2000 year) time scale of deep ocean ventilation affects the time scale for natural CO2 change and thus the time scale for paleo global climate, ice sheet, and sea level changes, but this paleo millennial time scale should not be misinterpreted as the time scale for ice sheet response to a rapid large human-made climate forcing.

When learning a hidden Markov model (HMM), sequen- tial observations can often be complemented by real-valued summary response variables generated from the path of hid- den states. Such settings arise in numerous domains, includ- ing many applications in biology, like motif discovery and genome annotation. In this paper, we present a flexible frame- work for jointly modeling both latent sequence features and the functional mapping that relates the summary response variables to the hidden state sequence. The algorithm is com- patible with a rich set of mapping functions. Results show that the availability of additional continuous response vari- ables can simultaneously improve the annotation of the se- quential observations and yield good prediction performance in both synthetic data and real-world datasets.

Observations have long associated ice streams with the presence of meltwater at the bed. More recently, theoretical models have been able to reproduce ice-stream behaviour as a consequence of the coupled dynamics of ice and subglacial meltwater. In this paper we analyse the properties of ice streams that form in a coupled model of ice flow and subglacial hydrology. We see that there is a natural length scale defining ice stream separation and width. This arises as a result of the balance between effective pressure gradients driving meltwater away from ice streams and the enhanced water production in the streams due to the fast ice flow. We further discuss how the model interacts with topography and we show that small perturbations to a uniform bed have a strong effect on where ice streams emerge in the model. However, in many cases ice streams then evolve to be closer to the dimensions defined by the natural length scale of the unperturbed system. The non-dimensional parameter that defines this length scale is therefore of fundamental importance in the model.

The CHAMP magnetic field variations during international quiet days of low solar activity period 2008-2009 are investigated. The present paper reports the existence of frequency-peaks < 20 mHz in the compressional component of the magnetic field in almost all CHAMP passes. The magnetic field variations associated with these frequencies have amplitude of a few tens of nT during daytime. The geomagnetic activity and interplanetary magnetic field parameters were observed to be low during the period of study. The spectral powers of the observed frequencies show no dependence on solar wind velocity and cone angle; hence the reported frequencies are not related to the geomagnetic pulsations. For frequency-peaks <15 mHz, strong local-time dependence is observed with maximum power near noon and minimum at night. The longitudinal and seasonal variations of the powers of these frequency-peaks match well with those of the equator-to-middle latitude ionospheric currents derived by the earlier studies. As a polar Low-Earth-Orbiting (LEO) satellite spans the entire range of latitudes within few minutes, it monitors the geomagnetic field variations caused by the quiet-time ionospheric currents flowing at different latitudes. This can result in certain frequencies in the magnetic field recorded by LEO satellites. We demonstrate that the frequencies <10mHz are mainly due to the latitudinal structure of the equatorial electrojet. The observed frequencies in CHAMP data are therefore attributed to the latitudinal structures of the ionospheric currents that are monitored only by the polar LEO satellites and are found to alter the observations of geomagnetic pulsations (Pc4-5 and Pi2) significantly.

Deeper neural networks are more difficult to train. We present a residual learning framework to ease the training of networks that are substantially deeper than those used previously. We explicitly reformulate the layers as learning residual functions with reference to the layer inputs, instead of learning unreferenced functions. We provide comprehensive empirical evidence showing that these residual networks are easier to optimize, and can gain accuracy from considerably increased depth. On the ImageNet dataset we evaluate residual nets with a depth of up to 152 layers---8x deeper than VGG nets but still having lower complexity. An ensemble of these residual nets achieves 3.57% error on the ImageNet test set. This result won the 1st place on the ILSVRC 2015 classification task. We also present analysis on CIFAR-10 with 100 and 1000 layers. The depth of representations is of central importance for many visual recognition tasks. Solely due to our extremely deep representations, we obtain a 28% relative improvement on the COCO object detection dataset. Deep residual nets are foundations of our submissions to ILSVRC & COCO 2015 competitions, where we also won the 1st places on the tasks of ImageNet detection, ImageNet localization, COCO detection, and COCO segmentation.

We present a method for detecting objects in images using a single deep neural network. Our approach, named SSD, discretizes the output space of bounding boxes into a set of default boxes over different aspect ratios and scales per feature map location. At prediction time, the network generates scores for the presence of each object category in each default box and produces adjustments to the box to better match the object shape. Additionally, the network combines predictions from multiple feature maps with different resolutions to naturally handle objects of various sizes. Our SSD model is simple relative to methods that require object proposals because it completely eliminates proposal generation and subsequent pixel or feature resampling stage and encapsulates all computation in a single network. This makes SSD easy to train and straightforward to integrate into systems that require a detection component. Experimental results on the PASCAL VOC, MS COCO, and ILSVRC datasets confirm that SSD has comparable accuracy to methods that utilize an additional object proposal step and is much faster, while providing a unified framework for both training and inference. Compared to other single stage methods, SSD has much better accuracy, even with a smaller input image size. For $300\times 300$ input, SSD achieves 72.1% mAP on VOC2007 test at 58 FPS on a Nvidia Titan X and for $500\times 500$ input, SSD achieves 75.1% mAP, outperforming a comparable state of the art Faster R-CNN model. Code is available at https://github.com/weiliu89/caffe/tree/ssd .

The Malaysian satellite RazakSAT-1 was designed to operate in a near-equatorial orbit (NEqO) and low earth orbit (LEO). However, after one year of operation in 2010, communication to the satellite was lost. This study attempted to identify whether space radiation sources could have caused the communication loss by comparing RazakSAT-1 with two functional satellites. Data on galactic cosmic rays (GCR), trapped protons, trapped electrons, and solar energetic particles (SEPs) obtained from Space Environment Information System (SPENVIS) was analyzed.

We present a deep learning framework for probabilistic pixel-wise semantic segmentation, which we term Bayesian SegNet. Semantic segmentation is an important tool for visual scene understanding and a meaningful measure of uncertainty is essential for decision making. Our contribution is a practical system which is able to predict pixel-wise class labels with a measure of model uncertainty. We achieve this by Monte Carlo sampling with dropout at test time to generate a posterior distribution of pixel class labels. In addition, we show that modelling uncertainty improves segmentation performance by 2-3% across a number of state of the art architectures such as SegNet, FCN and Dilation Network, with no additional parametrisation. We also observe a significant improvement in performance for smaller datasets where modelling uncertainty is more effective. We benchmark Bayesian SegNet on the indoor SUN Scene Understanding and outdoor CamVid driving scenes datasets.

For historical buildings and monuments, i.e. when considering time intervals of about a million year (we do not want to loose cultural heritage), the applicability of standard estimates of seismic hazard is really questionable. A viable alternative is represented by the use of the scenario earthquakes, characterized at least in terms of magnitude, distance and faulting style, and by the treatment of complex source processes. Scenario-based seismic hazard maps are purely based on geophysical and seismotectonic features of a region and take into account the occurrence frequency of earthquakes only for their classification into exceptional (catastrophic), rare (disastrous), sporadic (very strong), occasional (strong) and frequent. Therefore they may provide an upper bound for the ground motion levels to be expected for most regions of the world, more appropriate than probabilities of exceedance in view of the long time scales required for the protection of historical buildings. The neo-deterministic approach naturally supplies realistic time series of ground motion, which represent also reliable estimates of ground displacement readily applicable to seismic isolation techniques, useful to preserve historical monuments and relevant man made structures. This methodology has been successfully applied to many urban areas worldwide for the purpose of seismic microzoning, to strategic buildings, lifelines and cultural heritage sites; we will discuss its application to the cities of Rome and Florence.

Lichens are good bio-indicators of air pollution, but in most tropical countries there are few studies on the subject; however, in the city of San Jos\'e, Costa Rica, the relationship between air pollution and lichens has been studied for decades. In this article we evaluate the hypothesis that air pollution is lower where the wind enters the urban area (Northeast) and higher where it exits San Jos\'e (Southwest). We identified the urban parks with a minimum area of approximately 5 000m2 and randomly selected a sample of 40 parks located along the passage of wind through the city. To measure lichen coverage, we applied a previously validated 10 x 20cm template with 50 random points to five trees per park (1.5m above ground, to the side with most lichens). Our results (years 2008 and 2009) fully agree with the generally accepted view that lichens reflect air pollution carried by circulating air masses. The practical implication is that the air enters the city relatively clean by the semi-rural and economically middle class area of Coronado, and leaves through the developed neighborhoods of Escaz\'u and Santa Ana with a significant amount of pollutants. In the dry season, the live lichen coverage of this tropical city was lower than in the May to December rainy season, a pattern that contrasts with temperate habitats; but regardless of the season, pollution follows the pattern of wind movement through the city.

We present a novel and practical deep fully convolutional neural network architecture for semantic pixel-wise segmentation termed SegNet. This core trainable segmentation engine consists of an encoder network, a corresponding decoder network followed by a pixel-wise classification layer. The architecture of the encoder network is topologically identical to the 13 convolutional layers in the VGG16 network. The role of the decoder network is to map the low resolution encoder feature maps to full input resolution feature maps for pixel-wise classification. The novelty of SegNet lies is in the manner in which the decoder upsamples its lower resolution input feature map(s). Specifically, the decoder uses pooling indices computed in the max-pooling step of the corresponding encoder to perform non-linear upsampling. This eliminates the need for learning to upsample. The upsampled maps are sparse and are then convolved with trainable filters to produce dense feature maps. We compare our proposed architecture with the widely adopted FCN and also with the well known DeepLab-LargeFOV, DeconvNet architectures. This comparison reveals the memory versus accuracy trade-off involved in achieving good segmentation performance. SegNet was primarily motivated by scene understanding applications. Hence, it is designed to be efficient both in terms of memory and computational time during inference. It is also significantly smaller in the number of trainable parameters than other competing architectures. We also performed a controlled benchmark of SegNet and other architectures on both road scenes and SUN RGB-D indoor scene segmentation tasks. We show that SegNet provides good performance with competitive inference time and more efficient inference memory-wise as compared to other architectures. We also provide a Caffe implementation of SegNet and a web demo at http://mi.eng.cam.ac.uk/projects/segnet/.

This paper considers environmental problems of natural and anthropogenic atmospheric aerosol pollution and its global and regional monitoring. Efficient aerosol investigations may be achieved by spectropolarimetric measurements. Specifically second and fourth Stokes parameters spectral dependencies carry information on averaged refraction and absorption indexes and on particles size distribution functions characteristics.

Information about the hydro-meteorological parameters during the extreme sea storms is of significant importance for the sustainable development in the context of flood risk for the coastal areas. Usually there is a lack of sufficiently long history of instrumental measurements of the extreme winds, waves and storm surges. Simulation of historical storms is an important tool to evaluate the potential coastal hazards. In the absence of measured data hindcasts can satisfy the need for historical data. The wave and storm-surge regional numerical simulations have been carried out for the ten most severe storms over the Bulgarian coast of the Black Sea from the period 1972-2012. The ERA-Interim and ERA-40 reanalysis of wind at 10 m and mean sea level pressure have been downscaled with a high resolution atmospheric model ALADIN to the horizontal and time scales suitable for precise evaluation of hydro-meteorological parameters during the storms. The downscaled fields of wind and sea level pressure have been used as input for the wave and storm surge models.

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.

In this study, we propose an automatic learning method for variables selection based on Lasso in epidemiology context. One of the aim of this approach is to overcome the pretreatment of experts in medicine and epidemiology on collected data. These pretreatment consist in recoding some variables and to choose some interactions based on expertise. The approach proposed uses all available explanatory variables without treatment and generate automatically all interactions between them. This lead to high dimension. We use Lasso, one of the robust methods of variable selection in high dimension. To avoid over fitting a two levels cross-validation is used. Because the target variable is account variable and the lasso estimators are biased, variables selected by lasso are debiased by a GLM and used to predict the distribution of the main vector of malaria which is Anopheles. Results show that only few climatic and environmental variables are the mains factors associated to the malaria risk exposure.

This is a proposal to create a research facility for the development of a high-parallel version of the "SP machine", based on the "SP theory of intelligence". We envisage that the new version of the SP machine will be an open-source software virtual machine, derived from the existing "SP computer model", and hosted on an existing high-performance computer. It will be a means for researchers everywhere to explore what can be done with the system and to create new versions of it. The SP system is a unique attempt to simplify and integrate observations and concepts across artificial intelligence, mainstream computing, mathematics, and human perception and cognition, with information compression as a unifying theme. Potential benefits and applications include helping to solve problems associated with big data; facilitating the development of autonomous robots; unsupervised learning, natural language processing, several kinds of reasoning, fuzzy pattern recognition at multiple levels of abstraction, computer vision, best-match and semantic forms of information retrieval, software engineering, medical diagnosis, simplification of computing systems, and the seamless integration of diverse kinds of knowledge and diverse aspects of intelligence. Additional motivations include the potential of the SP system to help solve problems in defence, security, and the detection and prevention of crime; potential in terms of economic, social, environmental, and academic criteria, and in terms of publicity; and the potential for international influence in research. The main elements of the proposed facility are described, including support for the development of "SP-neural", a neural version of the SP machine. The facility should be permanent in the sense that it should be available for the foreseeable future, and it should be designed to facilitate its use by researchers anywhere in the world.