Search

Functional magnetic resonance imaging or functional MRI (fMRI) is a very popular tool used for differing brain regions by measuring brain activity. It is affected by physiological noise, such as head and brain movement in the scanner from breathing, heart beats, or the subject fidgeting. The purpose of this paper is to propose a novel approach to handling fMRI data for infants with high volatility caused by sudden head movements. Another purpose is to evaluate the volatility modelling performance of multiple dependent fMRI time series data. The models examined in this paper are AR and GARCH and the modelling performance is evaluated by several statistical performance measures. The conclusions of this paper are that multiple dependent fMRI series data can be fitted with AR + GARCH model if the multiple fMRI data have many sudden head movements. The GARCH model can capture the shared volatility clustering caused by head movements across brain regions. However, the multiple fMRI data without many head movements have fitted AR + GARCH model with different performance. The conclusions are supported by statistical tests and measures. This paper highlights the difference between the proposed approach from traditional approaches when estimating model parameters and modelling conditional variances on multiple dependent time series. In the future, the proposed approach can be applied to other research fields, such as financial economics, and signal processing. Code is available at \url{https://github.com/13204942/STAT40710}., Comment: 18 pages

Many quantities we are interested in predicting are geometric tensors; we refer to this class of problems as geometric prediction. Attempts to perform geometric prediction in real-world scenarios have been limited to approximating them through scalar predictions, leading to losses in data efficiency. In this work, we demonstrate that equivariant networks have the capability to predict real-world geometric tensors without the need for such approximations. We show the applicability of this method to the prediction of force fields and then propose a novel formulation of an important task, biomolecular structure refinement, as a geometric prediction problem, improving state-of-the-art structural candidates. In both settings, we find that our equivariant network is able to generalize to unseen systems, despite having been trained on small sets of examples. This novel and data-efficient ability to predict real-world geometric tensors opens the door to addressing many problems through the lens of geometric prediction, in areas such as 3D vision, robotics, and molecular and structural biology., Comment: 10 pages, 8 figures

Designing Artificial Intelligence (AI) solutions that can operate in real-world situations is a highly complex task. Deploying such solutions in the medical domain is even more challenging. The promise of using AI to improve patient care and reduce cost has encouraged many companies to undertake such endeavours. For our team, the goal has been to improve early identification of deteriorating patients in the hospital. Identifying patient deterioration in lower acuity wards relies, to a large degree on the attention and intuition of clinicians, rather than on the presence of physiological monitoring devices. In these care areas, an automated tool which could continuously observe patients and notify the clinical staff of suspected deterioration, would be extremely valuable. In order to develop such an AI-enabled tool, a large collection of patient images and audio correlated with corresponding vital signs, past medical history and clinical outcome would be indispensable. To the best of our knowledge, no such public or for-pay data set currently exists. This lack of audio-visual data led to the decision to conduct exactly such study. The main contributions of this paper are, the description of a protocol for audio-visual data collection study, a cloud-architecture for efficiently processing and consuming such data, and the design of a specific data collection device.

Computational methods that operate on three-dimensional molecular structure have the potential to solve important questions in biology and chemistry. In particular, deep neural networks have gained significant attention, but their widespread adoption in the biomolecular domain has been limited by a lack of either systematic performance benchmarks or a unified toolkit for interacting with molecular data. To address this, we present ATOM3D, a collection of both novel and existing benchmark datasets spanning several key classes of biomolecules. We implement several classes of three-dimensional molecular learning methods for each of these tasks and show that they consistently improve performance relative to methods based on one- and two-dimensional representations. The specific choice of architecture proves to be critical for performance, with three-dimensional convolutional networks excelling at tasks involving complex geometries, graph networks performing well on systems requiring detailed positional information, and the more recently developed equivariant networks showing significant promise. Our results indicate that many molecular problems stand to gain from three-dimensional molecular learning, and that there is potential for improvement on many tasks which remain underexplored. To lower the barrier to entry and facilitate further developments in the field, we also provide a comprehensive suite of tools for dataset processing, model training, and evaluation in our open-source atom3d Python package. All datasets are available for download from https://www.atom3d.ai ., Comment: NeurIPS 2021 Datasets and Benchmarks Track

Proteins are miniature machines whose function depends on their three-dimensional (3D) structure. Determining this structure computationally remains an unsolved grand challenge. A major bottleneck involves selecting the most accurate structural model among a large pool of candidates, a task addressed in model quality assessment. Here, we present a novel deep learning approach to assess the quality of a protein model. Our network builds on a point-based representation of the atomic structure and rotation-equivariant convolutions at different levels of structural resolution. These combined aspects allow the network to learn end-to-end from entire protein structures. Our method achieves state-of-the-art results in scoring protein models submitted to recent rounds of CASP, a blind prediction community experiment. Particularly striking is that our method does not use physics-inspired energy terms and does not rely on the availability of additional information (beyond the atomic structure of the individual protein model), such as sequence alignments of multiple proteins.

Learning on 3D structures of large biomolecules is emerging as a distinct area in machine learning, but there has yet to emerge a unifying network architecture that simultaneously leverages the graph-structured and geometric aspects of the problem domain. To address this gap, we introduce geometric vector perceptrons, which extend standard dense layers to operate on collections of Euclidean vectors. Graph neural networks equipped with such layers are able to perform both geometric and relational reasoning on efficient and natural representations of macromolecular structure. We demonstrate our approach on two important problems in learning from protein structure: model quality assessment and computational protein design. Our approach improves over existing classes of architectures, including state-of-the-art graph-based and voxel-based methods. We release our code at https://github.com/drorlab/gvp., Comment: Presented at ICLR 2021

Predicting the structure of multi-protein complexes is a grand challenge in biochemistry, with major implications for basic science and drug discovery. Computational structure prediction methods generally leverage pre-defined structural features to distinguish accurate structural models from less accurate ones. This raises the question of whether it is possible to learn characteristics of accurate models directly from atomic coordinates of protein complexes, with no prior assumptions. Here we introduce a machine learning method that learns directly from the 3D positions of all atoms to identify accurate models of protein complexes, without using any pre-computed physics-inspired or statistical terms. Our neural network architecture combines multiple ingredients that together enable end-to-end learning from molecular structures containing tens of thousands of atoms: a point-based representation of atoms, equivariance with respect to rotation and translation, local convolutions, and hierarchical subsampling operations. When used in combination with previously developed scoring functions, our network substantially improves the identification of accurate structural models among a large set of possible models. Our network can also be used to predict the accuracy of a given structural model in absolute terms. The architecture we present is readily applicable to other tasks involving learning on 3D structures of large atomic systems., Comment: 11 pages, 5 figures + SI: Updated based on the published version in PROTEINS. Presented at NeurIPS 2019 workshop Learning Meaningful Representations of Life

Despite an explosion in the number of experimentally determined, atomically detailed structures of biomolecules, many critical tasks in structural biology remain data-limited. Whether performance in such tasks can be improved by using large repositories of tangentially related structural data remains an open question. To address this question, we focused on a central problem in biology: predicting how proteins interact with one another---that is, which surfaces of one protein bind to those of another protein. We built a training dataset, the Database of Interacting Protein Structures (DIPS), that contains biases but is two orders of magnitude larger than those used previously. We found that these biases significantly degrade the performance of existing methods on gold-standard data. Hypothesizing that assumptions baked into the hand-crafted features on which these methods depend were the source of the problem, we developed the first end-to-end learning model for protein interface prediction, the Siamese Atomic Surfacelet Network (SASNet). Using only spatial coordinates and identities of atoms, SASNet outperforms state-of-the-art methods trained on gold-standard structural data, even when trained on only 3% of our new dataset. Code and data available at https://github.com/drorlab/DIPS., Comment: NeurIPS 2019

One of the more disruptive reforms associated with the modern Internet is the emergence of online communities working together on knowledge artefacts such as Wikipedia and OpenStreetMap. Recently it has become clear that these initiatives are vulnerable because of problems with membership turnover. This study presents a longitudinal analysis of 891 WikiProjects where we model the impact of member turnover and social capital losses on project productivity. By examining social capital losses we attempt to provide a more nuanced analysis of member turnover. In this context social capital is modelled from a social network perspective where the loss of more central members has more impact. We find that only a small proportion of WikiProjects are in a relatively healthy state with low levels of membership turnover and social capital losses. The results show that the relationship between social capital losses and project performance is U-shaped, and that member withdrawal has significant negative effect on project outcomes. The results also support the mediation of turnover rate and network density on the curvilinear relationship.

We outline a model and algorithm to perform inference on the palaeoclimate and palaeoclimate volatility from pollen proxy data. We use a novel multivariate non-linear non-Gaussian state space model consisting of an observation equation linking climate to proxy data and an evolution equation driving climate change over time. The link from climate to proxy data is defined by a pre-calibrated forward model, as developed in Salter-Townshend and Haslett (2012) and Sweeney (2012). Climatic change is represented by a temporally-uncertain Normal-Inverse Gaussian Levy process, being able to capture large jumps in multivariate climate whilst remaining temporally consistent. The pre-calibrated nature of the forward model allows us to cut feedback between the observation and evolution equations and thus integrate out the state variable entirely whilst making minimal simplifying assumptions. A key part of this approach is the creation of mixtures of marginal data posteriors representing the information obtained about climate from each individual time point. Our approach allows for an extremely efficient MCMC algorithm, which we demonstrate with a pollen core from Sluggan Bog, County Antrim, Northern Ireland., Comment: 25 pages, 7 figures

The line profiles of the stars with v sin i below a few km/s can reveal direct signatures of local velocity fields (e.g. convection) in stellar atmospheres. This effect is well established in cool main sequence stars, and has been detected and studied in three A stars. This paper reports observations of main sequence B, A and F stars with two goals: (1) to identify additional stars having sufficiently low values of v sin i to search for spectral line profile signatures of local velocity fields, and (2) to explore how the signatures of the local velocity fields in the atmosphere depend on stellar parameters such as effective temperature T_eff and peculiarity type. For stars having T_eff below about 10000 K, we always detect local atmospheric velocity fields indirectly through a non-zero microturbulence parameter, but not for hotter stars. Among the A and F stars in our sample having the sharpest lines, direct tracers of atmospheric velocity fields are found in six new stars. The velocity field signatures identified include asymmetric excess line wing absorption, deeper in the blue line wing than in the red; line profiles of strong lines that are poorly fit by computed profiles; and strong lines that are broader than they should be for the v sin i values deduced from weak lines. These effects are found in both normal and Am stars, but seem stronger in Am stars. These data still have not been satisfactorily explained by models of atmospheric convection, including numerical simulations., Comment: Acepted for publication by Astronomy and Astrophysics

The final evaluation report of the Federally-funded Windham Central Career Education project (July 1, 1973-June 30, 1975) in Townshend, Vermont addresses the status of career education in the Windham Central District, the impact which two years of effort had, and assesses the future of career education in the district. The report provides information on the demographic and geographic characteristics of the area, the project's goals and objectives, and special factors affecting the career education project. The impact study of the first two years discusses the major problems involved in implementing career education throughout the Windham Supervisory Union. It describes, at the secondary and elementary level, the initial participant reaction, the on-going workshops, the utilization of materials, and commitments to the future. Initially planned as a three year project, budgetary, program, and personnel problems became a growing concern and the project was terminated June 30, 1975. The concluding analysis indicated that the selection of a new superintendent was a key factor in determining the future path of career education in Windham. A nine-page evaluation design is appended. (BP)

In October 1996, the Programme on Educational Building held a conference in Stockholm entitled "Under One Roof." This report draws on the proceedings. Organization for Economic Cooperation and Development countries have been attempting to facilitate the provision of a range of community services on school sites, including adult education and other social and welfare services. This report describes development of integrated service provision, i.e., the integrating on one site of multiple services that are usually provided separately, focusing on the implications for local government authorities, planners, and architects who are designing school sites that must meet the new requirements. Several case studies are presented from other countries that demonstrate the necessity of optimizing the use of buildings and facilities and making schools better at promoting service synergies with local communities and with other services. The case studies provide examples of solutions to the practical problems of service integration that caters to the needs of both children and adults. The report's four chapters provide policy background, the issues and themes emerging from the case studies, and some provisional conclusions, and from them, possible initiatives and recommendations for governments. (GR)

Data from the Moderate Resolution Imaging Spectro-radiometer (MODIS)on-board the Earth Observing System Terra and Aqua satellites are processed using a land water mask to determine when an algorithm no longer needs to be run or when an algorithm needs to follow a different pathway. Entering the fourth reprocessing (Collection 6 (C6)) the MODIS team replaced the 1 km water mask with a 500 m water mask for improved representation of the continental surfaces. The new water mask represents more small water bodies for an overall increase in water surface from 1 to 2 of the continental surface. While this is still a small fraction of the overall global surface area the increase is more dramatic in certain areas such as the Arctic and Boreal regions where there are dramatic increases in water surface area in the new mask. MODIS products generated by the on-going C6 reprocessing using the new land water mask show significant impact in areas with high concentrations of change in the land water mask. Here differences between the Collection 5 (C5) and C6 water masks and the impact of these differences on the MOD04 aerosol product and the MOD11 land surface temperature product are shown.

Monitoring current forest characteristics in the taiga-tundra ecotone (TTE) at multiple scales is critical for understanding its vulnerability to structural changes. A 30 m spatial resolution Landsat-based tree canopy cover map has been calibrated and validated in the TTE with reference tree cover data from airborne LiDAR and high resolution spaceborne images across the full range of boreal forest tree cover. This domain-specific calibration model used estimates of forest height to determine reference forest cover that best matched Landsat estimates. The model removed the systematic under-estimation of tree canopy cover greater than 80% and indicated that Landsat estimates of tree canopy cover more closely matched canopies at least 2 m in height rather than 5 m. The validation improved estimates of uncertainty in tree canopy cover in discontinuous TTE forests for three temporal epochs (2000, 2005, and 2010) by reducing systematic errors, leading to increases in tree canopy cover uncertainty. Average pixel-level uncertainties in tree canopy cover were 29.0%, 27.1% and 31.1% for the 2000, 2005 and 2010 epochs, respectively. Maps from these calibrated data improve the uncertainty associated with Landsat tree canopy cover estimates in the discontinuous forests of the circumpolar TTE.

The compilation of global Landsat data-sets and the ever-lowering costs of computing now make it feasible to monitor the Earth's land cover at Landsat resolutions of 30 m. In this article, we describe the methods to create global products of forest cover and cover change at Landsat resolutions. Nevertheless, there are many challenges in ensuring the creation of high-quality products. And we propose various ways in which the challenges can be overcome. Among the challenges are the need for atmospheric correction, incorrect calibration coefficients in some of the data-sets, the different phenologies between compilations, the need for terrain correction, the lack of consistent reference data for training and accuracy assessment, and the need for highly automated characterization and change detection. We propose and evaluate the creation and use of surface reflectance products, improved selection of scenes to reduce phenological differences, terrain illumination correction, automated training selection, and the use of information extraction procedures robust to errors in training data along with several other issues. At several stages we use Moderate Resolution Spectroradiometer data and products to assist our analysis. A global working prototype product of forest cover and forest cover change is included.

Surface reflectance adjusted for atmospheric effects is a primary input for land cover change detection and for developing many higher level surface geophysical parameters. With the development of automated atmospheric correction algorithms, it is now feasible to produce large quantities of surface reflectance products using Landsat images. Validation of these products requires in situ measurements, which either do not exist or are difficult to obtain for most Landsat images. The surface reflectance products derived using data acquired by the Moderate Resolution Imaging Spectroradiometer (MODIS), however, have been validated more comprehensively. Because the MODIS on the Terra platform and the Landsat 7 are only half an hour apart following the same orbit, and each of the 6 Landsat spectral bands overlaps with a MODIS band, good agreements between MODIS and Landsat surface reflectance values can be considered indicators of the reliability of the Landsat products, while disagreements may suggest potential quality problems that need to be further investigated. Here we develop a system called Landsat-MODIS Consistency Checking System (LMCCS). This system automatically matches Landsat data with MODIS observations acquired on the same date over the same locations and uses them to calculate a set of agreement metrics. To maximize its portability, Java and open-source libraries were used in developing this system, and object-oriented programming (OOP) principles were followed to make it more flexible for future expansion. As a highly automated system designed to run as a stand-alone package or as a component of other Landsat data processing systems, this system can be used to assess the quality of essentially every Landsat surface reflectance image where spatially and temporally matching MODIS data are available. The effectiveness of this system was demonstrated using it to assess preliminary surface reflectance products derived using the Global Land Survey (GLS) Landsat images for the 2000 epoch. As surface reflectance likely will be a standard product for future Landsat missions, the approach developed in this study can be adapted as an operational quality assessment system for those missions.

The paper discusses the role of schools in Great Britain's preservice teacher education, noting the political history of the country's teacher education. Research shows that the result of partnerships between schools and institutions of higher education has been teachers who enter the field with satisfactory qualifications. (SM)

North American vegetation has been discovered to be a net carbon sink, with atypical behavior of drawing down more carbon from the atmosphere during the past century. It has been suggested that the Northern Hemisphere will respond favorably to climate warming by enhancing productivity and reducing the impact of fossil fuel emissions into the atmosphere. Many investigations are currently underway to understand and identify mechanisms of storage so they might be actively managed to offset carbon emissions which have detrimental consequences to the functioning of ecosystems and human well being. This paper used a time series of satellite data from multiple sensors at multiple resolutions over the past thlrty years to identify and understand mechanisms of change to vegetation productivity throughout North America. We found that humans had a marked impact to vegetation growth in half of the six selected study regions which cover greater than two million km2. We found climatic influences of increasing temperatures, and longer growing seasons with reduced snow cover in the northern regions of North America with forest fire recovery in the Northern coniferous forests of Canada. The Mid-latitudes had more direct land cover changes induced by humans coupled with climatic influences such as severe drought and altered production strategies of rain-fed agriculture in the upper Midwest, expansion of irrigated agriculture in the lower Midwest, and insect outbreaks followed by subsequent logging in the upper Northeast. Vegetation growth over long time periods (20+ years) in North America appears to be associated with long term climate change but most of the marked changes appear to be associated with climate variability on decadal and shorter time scales along with direct human land cover conversions. Our results document regional land cover land use change and climatic influences that have altered continental scale vegetation dynamics in North America.

The northern regions of the earth are currently experiencing rapid change in temperature and precipitation. This region contains -40% of carbon stored in the world's soil which has accumulated from the last ice age (over 10,000 years ago). The carbon has remained to this point due to reduced decomposition from the short growing seasons and subfreezing temperatures. The influence of climate upon plant growth can have significant consequences to the carbon cycle balance in this region and could potentially alter and release this long term store of carbon to the atmosphere, resulting in a negative feedback enhancing climate warming. These changes have the potential to alter ecosystems processes, which impact human well being. This paper investigated a global satellite record of increases in vegetation growth from 1982 to 2003 developed at GSFC. It was found that, Newfoundland's vegetation growth during the 1990s exceeded global measurements. A number of potential causes were investigated to understand the mechanistic environmental drivers that could alter the productivity of this ecosystem. Possible drivers of change included: human influence of land use change on vegetation cover; changes in precipitation; temperature; cloud cover; snow cover; and growing season length. We found that humans had a minimal influence on vegetation growth in Newfoundland. Less than 6% of the island was logged during the investigation. We found a strong correlation of vegetation growth to a lengthening of the growing season of -9 and -17 days from 1982-1990 and 1991-1999. A distinct drop in plant growth and air temperature was found in 1990 to 1991 from the volcanic eruption of Mt. Pinatubo that reduced global surface air temperatures. These results document the influences of air temperature upon northern forest plant growth and the cooling effects of major volcanic eruptions in this ecological system.

In an effort to more full explore the potential of commercial remotely sensed land data sources, the NASA Earth Science Enterprise (ESE) implemented an experimental Scientific Data Purchase (SDP) that solicited bids from the private sector to meet ESE-user data needs. The images from the Space Imaging IKONOS system provided a particularly good match to the current ESE missions such as Terra and Landsat 7 and therefore serve as a focal point in this analysis.

Landsat satellite images from the mid-1980s and early 1990s were used to map tropical forest extent and deforestation in approximately 800,000 sq km of Amazonian Bolivia. Forest cover extent, including tropical deciduous forest, totalled 472,000 sq km while the area of natural non-forest formations totalled 298,000 sq km. The area deforested totalled 15,000 sq km in the middle 1980s and 28,800 sq km by the early 1990s. The rate of tropical deforestation in the >1,000 mm/y precipitation forest zone of Bolivia was 2,200 sq km/y from 1985-1986 to 1992-1994. We document a spatially-concentrated "deforestation zone" in Santa Cruz Department where >60% of the Bolivian deforestation is occurring at an accelerating rate in areas of tropical deciduous dry forest.

The impact of images misregistration on the detection of changes in land cover was studied using spatially degraded Landsat MSS images. Emphasis is placed on simulated images of the Normalized Difference Vegetation Index (NDVI) at spatial resolutions of 250 and 500 m. It is pointed out that there is the need to achieve high values of registration accuracy. The evidence from simulations suggests that misregistrations can have a marked effect on the ability of remotely sensed data to detect changes in land cover. Even subpixel misregistrations can have a major impact, and the most marked proportional changes will tend to occur at the finest misregistrations.

Rainfall estimates, based on cold cloud duration estimated from Meteosat data, are compared with vegetation development depicted by data of the normalized difference vegetation index (NDVI) from the NOAA AVHRR for part of the Sahel. Decadal data from the 1985 and 1986 growing seasons are examined to determine the synergism of the datasets for rangeland monitoring. There is a general correspondence between the two datasets with a marked lag between rainfall and NDVI of between 10 and 20 days. This time lag is particularly noticeable at the beginning of the rainy season and in the more northern areas where rainfall is the limiting factor for growth. Principal component analysis was used to examine deviations from the general relationship between rainfall and the NDVI. Areas of low NDVI values for a given input of rainfall were identified: at a regional scale, they give an indication o areas of low production potential and possible degradation of ecosystems.

Images of the normalized difference vegetation index (NDVI) are examined with specific attention given to the effect of spatial scales on the understanding of surface phenomena. A scale variance analysis is conducted on NDVI annual and seasonal images of Africa taken from 1987 NOAA-AVHRR data at spatial scales ranging from 8-512 km. The scales at which spatial variation takes place are determined and the relative magnitude of the variations are considered. Substantial differences are demonstrated, notably an increase in spatial variation with coarsening spatial resolution. Different responses in scale variance as a function of spatial resolution are noted in an analysis of maximum value composites for February and September; the difference is most marked in areas with very seasonal vegetation. The spatial variation at different scales is attributed to different factors, and methods involving the averaging of areas of transition and surface heterogeneity can oversimplify surface conditions. The spatial characteristics and the temporal variability of areas should be considered to accurately apply satellite data to global models.

The adequacy of traditional approaches to the derivation of global information on land cover is examined. The contribution of coarse resolution satellite data from the NOAA series of satellites is discussed. The Moderate Resolution Imaging Spectrometer of the Earth Observing System is considered to be a substantially better instrument than the AVHRR in terms of several spectral, radiometric, and geometric properties. In terms of its total field of view and frequency of imaging, the instrument will be comparable to the AVHRR for global land cover monitoring.

Previous analysis (Townshend and Justice 1988) is extended to examine the spatial variations in images of the normalized difference vegetation index (NDVI) of seven areas. These images were derived by subtracting corresponding pixel values from pairs of registered MSS images. The changes depicted by these derived images were analyzed by scale variance analysis for pixel sizes between 4 km and 64 km. It is shown that for some areas substantial changes are detectable at these very coarse scales, although there is less contrast between the areas than at finer spatial resolutions below 1 km. In all the areas the total spatial variability of the images is contributed at a wide variety of spatial scales.

Consideration is given to a technique for spatially degrading high-resolution satellite data to produce comparable data sets over a range of coarser resolutions. Landsat MSS data is used to produce seven spatial resolution data sets by applying a spatial filter designed to simulate sensor response. Also, spatial degradation of coarse resolution data to provide data compression for the production of global-scale data sets is examined. NOAA AVHRR Global Area Coverage data is compared to other sampling procedures. It is found that sampling procedures that incorporate averaging result in decreased variance, while sampling procedures adopting single-value selection have higher variances and produce data values comparable with those from the original data.

Images using reflected visible and NIR data and images using passive microwave data were compared in terms of their usefulness for characterizing land-cover types in South America and Africa. The former images are of the normalized difference vegetation index (NDVI), and the latter images are of the microwave polarization difference temperature (MPDT). The combined use of MPDT and NDVI data sets show clear synergistic benefits in using the two data sets. However, the evidence suggests that for most cover types, increasing the temporal frequency of the NDVI images is more advantageous than incorporating MPDT data sets.

AVHRR normalized difference vegetation index (NDVI) data for a one-year period were compared with Scanning Multichannel Microwave Radiometer microwave polarization difference temperature (MPDT) data for the study of vegetation phenology. It is shown that the MPDT response differs considerably from the seasonal NDVI pattern. The results do not support the hypothetical relationship between MPDT and leaf water content. It is found that only vegetation types with a substantial seasonal variation in the areal extent of vegetated cover show strong seasonality in MPDT data.

Comparison between the microwave polarized difference temperature (MPDT) derived from 37 GHz band data and the normalized difference vegetation index (NDVI) derived from near-infrared and red bands, from several empirical investigations are summarized. These indicate the complementary character of the two measures in environmental monitoring. Overall the NDVI is more sensitive to green leaf activity, whereas the MPDT appears also to be related to other elements of the above-ground biomass. Monitoring of hydrological phenomena is carried out much more effectively by the MPDT. Further work is needed to explain spectral and temporal variation in MPDT both through modelling and field experiments.

An empirical investigation of the required spatial resolution for MODIS-N is outlined. It is based on 5 LANDSAT multispectral scanner system images of the normalized difference vegetation index degraded to resolutions between 250 m and 4000 m. Pairs of images from different dates were registered and difference images were generated. Fourier analysis indicates that resolutions finer than 1 km are highly desirable for change detection. A sensor with a resolution of 500 m is recommended as providing the best compromise between detail of changes detected and the size of the resultant data volume, but other options are also suggested.

The paper provides preliminary evidence for the spatial resolutions required to monitor land transformations at broad scales. This is obtained from simulations of imagery at various spatial resolutions between 125 and 4000 m derived from Landsat MSS imagery. Consideration is given to the various types of spatial images detectable by remotely-sensed systems, as well as to the difficulties associated in disentangling permanent land transformations from shorter term changes such as phenological and interannual changes.

Various methods are compared for carrying out land cover classifications of South America using multitemporal Advanced Very High Resolution Radiometer data. Fifty-two images of the normalized difference vegetation index (NDVI) from a 1-year period are used to generate multitemporal data sets. Three main approaches to land cover classification are considered, namely the use of the principal components transformed images, the use of a characteristic curves procedure based on NDVI values plotted against time, and finally application of the maximum likelihood rule to multitemporal data sets. Comparison of results from training sites indicates that the last approach yields the most accurate results. Despite the reliance on training site figures for performance assessment, the results are nevertheless extremely encouraging, with accuracies for several cover types exceeding 90 per cent.

LANDSAT Thematic Mapper and Multispectral Scanner data have been used to construct change detection images for three playas in south-central Tunisia. Change detection images have been used to analyze changes in surface reflectance and absorption between wet and dry season (intra-annual change) and between different years (inter-annual change). Change detection imagery has been used to examine geomorphological changes on the playas. Changes in geomorphological phenomena are interpreted from changes in soil and foliar moisture levels, differences in reflectances between different salt and sediments and the spatial expression of geomorphological features. Intra-annual change phenomena that can be detected from multidate imagery are changes in surface moisture, texture and chemical composition, vegetation cover and the extent of aeolian activity. Inter-annual change phenomena are divisible into those restricted to marginal playa facies (sedimentation from sheetwash and alluvial fans, erosion from surface runoff and cliff retreat) and these are found in central playa facies which are related to the internal redistribution of water, salt and sediment.

Images at a resolution of 8 km are currently being generated for the whole of Africa, displaying the normalized difference vegetation index (NDVI). These images have undergone a process of temporal compositing to reduce the effects of cloud cover and atmospheric variation. When the NDVI is plotted against time, different cover types are shown to have characteristic profiles corresponding closely with their phenology. The resultant pattern of NDVI values displayed on the images is analyzed in terms of the cover types present and local variations in rainfall. Comparison between images for 1983 and 1984 overall showed considerable similarities, but significant differences were observed in the northward extent of the greening wave in the Sahel, the greening up of the Kalahari Desert and East African communities. It is concluded that vegetation monitoring using NDVI images needs to be associated with scene stratification according to cover type.

Two sets of multitemporal data derived from NOAA world data product are analyzed by means of principal components analysis in order to examine their underlying multitemporal dimensionality. Specifically, images of the normalized difference vegetation index (NDVI) were analyzed for eight 3-week periods for Africa and ten 3-week periods for North America sampled from throughout the year extending from April 1982 to March 1983. The two multitemporal sets of images displayed remarkable similarities in terms of their first two components, the first corresponding very closely to the annualized integrated NDVI and the second to the seasonality of the NDVI. In particular, for the African data set the feature space defined by the first two components allows separation of the main cover types.

Four different approaches to the classification of land cover for whole continents using multitemporal images of the normalized difference vegetation index derived from the Advanced Very High Resolution Radiometer of the NOAA series of satellites are discussed. The first approach uses only two dates from different seasons and classification dependent upon subdivision of the resultant two-dimensional feature space by an analyst using a track ball. The second approach involves a similar method of partitioning the feature space, but with the two dimensions being the first and second principal components derived from 13 four-week composite images. The third approach uses the maximum likelihood rule to derive the classified map. In the fourth approach, the amount of deviation from characteristic curves is used as a basis for classification.

The principal objectives of the UK SATMaP program are to determine thematic mapper (TM) performance with particular reference to spatial resolution properties and geometric characteristics of the data. So far, analysis is restricted to images from the U.S. and concentrates on spectra and radiometric properties. The results indicate that the data are inherently three dimensional compared with the two dimensional character of MSS data. Preliminary classification results indicate the importance of the near infrared band (TM 4), at least one middle infrared band (TM 5 or TM 6) and at least one of the visible bands (preferably either TM 3 or TM 1). The thermal infrared also appears to have discriminatory ability despite its coarser spatial resolution. For band 4 the forward and reverse scans show somewhat different spectral responses in one scene but this effect is absent in the other analyzed. From examination of the histograms it would appear that the full 8 bit quantization is not being effectively utilized for all the bands.

Data from the advanced very high resolution radiometer sensor on the National Oceanic and Atmospheric Administration's operational series of meteorological satellites were used to classify land cover and monitor vegetation dynamics for Africa over a 19-month period. There was a correspondence between seasonal variations in the density and extent of green leaf vegetation and the patterns of rainfall associated with the movement of the Intertropical Convergence Zone. Regional variations, such as the 1983 drought in the Sahel of western Africa, were observed. Integration of the weekly satellite data with respect to time for a 12-month period produced a remotely sensed estimate of primary production based upon the density and duration of green leaf biomass. Eight of the 21-day composited data sets covering an 11-month period were used to produce a general land-cover classification that corresponded well with those of existing maps.

The role of coarse resolution meteorological satellite data for monitoring and mapping of vegetation for global, continental and regional scales is outlined. In the NOAA products used the effects of cloud cover are reduced by the generation of temporal composites of images of the normalized difference vegetation index. Different land cover types are shown to have characteristic spectral phenological curves. Such data have the disadvantage of effectively increasing the apparent areal extent of small areas of green vegetation.

In assessing the accuracy of classification techniques for Thematic Mapper data the consistency of the detector-to-detector response is critical. Preliminary studies were undertaken, therefore, to assess the significance of this factor for the TM. The overall structure of the band relationships can be examined by principal component analysis. In order to examine the utility of the Thematic Mapper data more carefully, six different land cover classes approximately Anderson level 1 were selected. These included an area of water from the sediment-laden Mississippi, woodland, agricultural land and urban land. A plume class was also selected which includes the plume of smoke emanating from the power station and drifting over the Mississippi river.

The validity of using data from the Advanced Very High Resolution Radiometer (AVHRR) of the satellites NOAA-6 and NOAA-7 for land cover mapping is assessed by making comparisons with much higher resolution LANDSAT multi-spectral scanner (MSS) data. Near synchronous data for both systems are analysed for test sites in the Imperial Valley, California, the Nile Delta, and southern Italy. The results strongly indicate that despite the very coarse resolution of the AVHRR data compared with conventional MSS data they are sufficiently strongly correlated to suggest that the former have significant potential for land cover mapping, especially at small scales and for large areas. Hence the outstanding benefits of AVHRR data, namely their high temporal frequency, relative cheapness and low data volumes for image processing, may readily be taken advantage of for such tasks.

The use of the National Oceanic and Atmospheric Administration's advanced very high resolution radiometer satellite data for classifying land cover and monitoring of vegetation dynamics over an extremely large area is demonstrated for the continent of Africa. Data from 17 imaging periods of 21 consecutive days each were composited by a technique sensitive to the in situ green-leaf biomass to provide cloud-free imagery for the whole continent. Virtually cloud-free images were obtainable even for equatorial areas. Seasonal variation in the density and extent of green leaf vegetation corresponded to the patterns of rainfall associated with the inter-tropical convergence zone. Regional variations, such as the 1982 drought in east Africa, were also observed. Integration of the weekly satellite data with respect to time produced a remotely sensed assessment of biological activity based upon density and duration of green-leaf biomass. Two of the 21-day composited data sets were used to produce a general land cover classification. The resultant land cover distributions correspond well to those of existing maps.

The principal objectives of the UK SATMaP program are to determine thematic mapper (TM) performance with particular reference to spatial resolution properties and geometric characteristics of the data. So far, analysis is restricted to images from the U.S. and concentrates on spectra and radiometric properties. The results indicate that the data are inherently three dimensional compared with the two dimensional character of MSS data. Preliminary classification results indicate the importance of the near infrared band (TM 4), at least one middle infrared band (TM 5 or TM 6) and at least one of the visible bands (preferably either TM 3 or TM 1). The thermal infrared also appears to have discriminatory ability despite its coarser spatial resolution. For band 4 the forward and reverse scans show somewhat different spectral responses in one scene but this effect is absent in the other analyzed. From examination of the histograms it would appear that the full 8-bit quantization is not being effectively utilized for all the bands.

Two unsupervised classification procedures were applied to ratioed and unratioed LANDSAT multispectral scanner data of an area of spatially complex vegetation and terrain. An objective accuracy assessment was undertaken on each classification and comparison was made of the classification accuracies. The two unsupervised procedures use the same clustering algorithm. By on procedure the entire area is clustered and by the other a representative sample of the area is clustered and the resulting statistics are extrapolated to the remaining area using a maximum likelihood classifier. Explanation is given of the major steps in the classification procedures including image preprocessing; classification; interpretation of cluster classes; and accuracy assessment. Of the four classifications undertaken, the monocluster block approach on the unratioed data gave the highest accuracy of 80% for five coarse cover classes. This accuracy was increased to 84% by applying a 3 x 3 contextual filter to the classified image. A detailed description and partial explanation is provided for the major misclassification. The classification of the unratioed data produced higher percentage accuracies than for the ratioed data and the monocluster block approach gave higher accuracies than clustering the entire area. The moncluster block approach was additionally the most economical in terms of computing time.