Atmospheric and Oceanic Remote Sensing Applications

A new algorithm based on the multiparameter neural network is proposed to retrieve wind speed (WS), sea surface temperature (SST), sea surface air temperature, and relative humidity (RH) simultaneously over the global oceans from Special Sensor Microwave Imager (SSM/I) observations. The retrieved geophysical parameters are used to estimate the surface latent heat flux and sensible heat flux using a bulk method over the global oceans. The neural network is trained and validated with the matchups of SSM/I overpasses and National Data Buoy Center buoys under both clear and cloudy weather conditions. In addition, the data acquired by the 85.5-GHz channels of SSM/I are used as the input variables of the neural network to improve its performance. The root-mean-square (rms) errors between the estimated WS, SST, sea surface air temperature, and RH from SSM/I observations and the buoy measurements are 1.48 m s−1, 1.54°C, 1.47°C, and 7.85, respectively. The rms errors between the estimated latent and sensible heat fluxes from SSM/I observations and the Xisha Island (in the South China Sea) measurements are 3.21 and 30.54 W m−2, whereas those between the SSM/I estimates and the buoy data are 4.9 and 37.85 W m−2, respectively. Both of these errors (those for WS, SST, and sea surface air temperature, in particular) are smaller than those by previous retrieval algorithms of SSM/I observations over the global oceans. Unlike previous methods, the present algorithm is capable of producing near-real-time estimates of surface latent and sensible heat fluxes for the global oceans from SSM/I data.

Snow plays a vital role in the energy and water budgets of drainage basins of western Canada. Various remote sensors such as Moderate Resolution Imaging Spectroradiometer (MODIS), Advanced Microwave Scanning Radiometer (AMSR-E) and Special Sensor Microwave/Imager (SSM/I) have been launched to map the snow cover extent (SCE), snow cover fraction (SCF), and snow water equivalent (SWE) across the globe. However, the distribution and variability of snow inferred from remote sensing products have not been comprehensively investigated in western Canada owing to its complex topography and harsh environment. So far, little research has been conducted on SCE-streamflow and SCE-SWE-runoff models focusing on Canadian watersheds where snow cover is very important for human well being. Although microwave remote sensing of snow is widely developed and applied in Canada, the retrieval of SWE in western Canada is not as well established owing to the complex topography in this area. Therefore, the Quesnel River Basin (QRB) of British Columbia is selected as a primary test site to develop and test SCE-streamflow and SCE-SWE-runoff models. Then the Mackenzie River Basin (MRB) is chosen as a secondary test site to apply the Environment Canada (EC) SWE retrieval algorithms to upscale the hydrometeorological research. In this thesis, a new approach referred to as the spatial filter (SF) method is developed to decrease the cloud coverage in the MODIS snow products. At the same time, the new snow products are evaluated based on in-situ observations of snow depth in the QRB. Then the relationships between SCF from MODIS, topography, and hydrometeorology of the QRB are explored. In addition, various retrieval algorithms of SWE from microwave remote sensing are tested in the QRB. At last, the Environment Canada algorithms of SWE from SSM/I are adopted to produce new SCF products evaluated with the MODIS snow products. The relationships between SWE and SCF from SSM/I and hydrometeorology are also investigated in the MRB ...The stud

The Sixth International Symposium on Neural Networks (ISNN 2009)

Results are presented of calibration/validation studies showing the ability of the Special Sensor Microwave/Imager (SSM/I) to measure total precipitable water in the atmosphere over the ocean. Comparisons between radiosondes and the SSM/I are presented for three different algorithms. The results show the possibility of a distinct improvement in the retrieval of total precipitable water over the ocean. The global, nonlinear algorithm is more sensitive to cloud liquid water content, rain, and sea ice. The additional sensitivity is due to the screening of rain and sea ice from the dependent data and the squared term in the retrieval algorithm. Thus, it will be very important to have good screening procedures for identifying these conditions. The linear algorithm overestimates in the mid-range and underestimates at large values of total precipitable water. The explanation for this effect is probably related to the selection of the center of the water vapor line as the operating frequency of the SSM/I water vapor channel. The line center is most likely to exhibit a saturation effect at large water vapor amounts, and pressure and temperature effects can also be important, depending on the distribution of water vapor in the atmosphere.< <ETX xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">&gt;</ETX>

The application of artificial neural networks to the estimation of bioprocess variables will be discussed. In fermentation processes, direct on-line measurements of primary process variables usually are unavailable. The state of the cultivation, therefore, has to be inferred from measurements of secondary variables and any previous knowledge of process dynamics. This research investigates the learning, recall and generalization characteristics of neural networks trained to model the nonlinear behavior of a fermentation process. Two different neural network methodologies are discussed, namely, feed-forward and recurrent neural networks, which differ in their treatment of time dependence. The neural networks are trained by backpropagation using a conjugate gradient technique, which provides a dramatic improvement in the convergence speed. The objective is to use environmental and physiological information available from on-line sensors to estimate concentrations of species in a bioreactor. Results of the neural network estimators are presented, based on experimental data available from the ethanol production by Zymomonas mobilis fermentation. The feed-forward and recurrent neural network methodologies are demonstrated to perform suitably as unmeasurable state estimators. Both networks offer comparable abilities of recall, but recurrent networks perform better than feed-forward networks in generalization.

Minimization techniques have long been used by the remote sensing community to obtain relationships between observed radiances and retrieved geophysical quantities. Originally, special sensor microwave imager (SSM/I) passive microwave wind observations were regressed against coincident wind speed observations from buoys. More recent research has shown that a wind direction signal may also be present in SSM/I observations. This wind direction signal, however, is not very large, meaning that accounting for the error characteristics of this signal is very important, and it is believed that the signal is somewhat nonlinear. Thus, more advanced minimization techniques are required to extract this wind direction signal. The paper discusses neural networks (NN) which are a nonlinear minimization technique and they can be used either in simulation mode, where the net is trained using simulated radiances assumed to be the truth, or in observation mode, where the observed radiances are matched with in situ observations.

This paper explores the use and applications of neural networks in the construction of a text markup program. This research paper describes various tasks that require textual data analysis and discusses the problems, issues, and solutions that accompany them. Interest in neural networks has increased in recent years, and they are find­ing applications in a wide variety of fields, such as business, medicine, engineering, geology, and physics. Neural networks have made great strides in forecasting, planning, and management. There are several reasons for this situation. Neural networks are very powerful modeling systems capable of creating complex dependencies. Neural networks can be widely used in areas such as text/speech recognition, semantic search, decision support/expert systems, inventory forecasting, data storage systems and content analysis. The object of the work is the process of functioning of neural networks and an algorithm for text markup recognition. The purpose of the scientific paper is the application of neural networks in the construction of the program for the markup of the text. In order to achieve the goal, the following tasks were put forward: a) study existing neural networks, b) choose a neural network to cre­ate a model and study its structure, c) convert input data to feed it into a neural network model. Representation and analysis of symbolic structures in neural networks seems to be an interesting and useful direction in neural network theory. In this paper, we have reviewed some neural network architectures that may merit further consideration in these circumstances. In what follows, we will focus on specific experiments in this area. Thus, this paper outlines a problem area for further research and testing.

When operating gas transfer units with gas turbine drive a significant share (about 9%) of natural gas is spent on compressor stations own needs. A considerable expense of pumped gas for gas transfer units own needs at compressor stations determines the actuality of realization of resources saving technologies. When performing energy saving measures it is required to control the resulting effect. Control of energy consumption is based on energy efficiency indicators of gas transfer units which include efficiency and gas fuel rate. The aim of this paper was to develop a universal method of calculating energy efficiency indicators using intelligent methods which exclude risks of reducing accuracy of results. In this paper we proposed and justified method of calculating energy efficiency indicators on the base of monitored parameters of gas transfer units work with application of neural networks. The research made it possible to receive and to substantiate the method of determining the specific consumption of fuel gas by the parameters of HPA controlled standard automation system based on the application of intelligent neural networks. Neural network model can be integrated in the station monitoring system compressor station need only advance to pick up army of weights to train a neural network) for a particular type HPA. When training a neural network according to the variation of the parameters in the training set and 20% teaching error does not exceed 1%. When using neural networks for the calculation of energy efficiency indicators for the unit for which there was training for other units of the same type, the mean square error does not exceed 5%.

Processing-in-memory (PIM) is a promising solution to address the "memory wall" challenges for future computer systems. Prior proposed PIM architectures put additional computation logic in or near memory. The emerging metal-oxide resistive random access memory (ReRAM) has showed its potential to be used for main memory. Moreover, with its crossbar array structure, ReRAM can perform matrix-vector multiplication efficiently, and has been widely studied to accelerate neural network (NN) applications. In this work, we propose a novel PIM architecture, called PRIME, to accelerate NN applications in ReRAM based main memory. In PRIME, a portion of ReRAM crossbar arrays can be configured as accelerators for NN applications or as normal memory for a larger memory space. We provide microarchitecture and circuit designs to enable the morphable functions with an insignificant area overhead. We also design a software/hardware interface for software developers to implement various NNs on PRIME. Benefiting from both the PIM architecture and the efficiency of using ReRAM for NN computation, PRIME distinguishes itself from prior work on NN acceleration, with significant performance improvement and energy saving. Our experimental results show that, compared with a state-of-the-art neural processing unit design, PRIME improves the performance by ~2360× and the energy consumption by ~895×, across the evaluated machine learning benchmarks.

Processing-in-memory (PIM) is a promising solution to address the "memory wall" challenges for future computer systems. Prior proposed PIM architectures put additional computation logic in or near memory. The emerging metal-oxide resistive random access memory (ReRAM) has showed its potential to be used for main memory. Moreover, with its crossbar array structure, ReRAM can perform matrix-vector multiplication efficiently, and has been widely studied to accelerate neural network (NN) applications. In this work, we propose a novel PIM architecture, called PRIME, to accelerate NN applications in ReRAM based main memory. In PRIME, a portion of ReRAM crossbar arrays can be configured as accelerators for NN applications or as normal memory for a larger memory space. We provide microarchitecture and circuit designs to enable the morphable functions with an insignificant area overhead. We also design a software/hardware interface for software developers to implement various NNs on PRIME. Benefiting from both the PIM architecture and the efficiency of using ReRAM for NN computation, PRIME distinguishes itself from prior work on NN acceleration, with significant performance improvement and energy saving. Our experimental results show that, compared with a state-of-the-art neural processing unit design, PRIME improves the performance by ~2360x and the energy consumption by ~895x, across the evaluated machine learning benchmarks.

As an important research direction in the field of artificial intelligence, neural networks have made significant progress in many fields such as image recognition and natural language processing in recent years. The mathematical principles behind it, particularly ordinary and partial differential equations, play a pivotal role in describing the dynamic behavior and learning processes of neural networks. Therefore, the study of the mathematical principles of these two and their application in neural networks has important theoretical significance and practical value. This paper first introduces the mathematical principles of ordinary differential equations and their applications in neural networks, then expounds the mathematical principles of partial differential equations and their applications in neural networks, and finally analyzes the limitations of existing research and looks forward to future development. By systematically reviewing the application of ordinary differentiation and partial differentiation in neural networks, this paper provides a new perspective and idea for related research. At the same time, the research in this paper is helpful to deepen the understanding of the dynamic behavior of neural networks, promote the development of neural network theory, and provide a theoretical basis for model design and optimization in practical applications.

This paper proposes the application of Artificial Neural Networks (ANN) in forecasting miners' occupational disease incidence, examines the effect of BP network topology and parameters on training time, describes a neural network that uses Legendre polynomials as nonlinear weights and tries to find the most suitable network model to solve the problem via analyzing the forecast result. Through computer simulation, it clarifies that this model could forecast miners' pneumoconiosis incidence commendably.

Neural networks (NNs) are successfully used to solve inverse and other problems in geophysics. The aim of this work, which is a continuation of a series of works by a group of authors, is to improve the efficiency of the NN method for solving nonlinear inverse 3D problems of geoelectrics, based on the construction of the author’s convolutional neural network. The network includes a number of additional special transformations (data compression, suppression of the influence of an unknown background environment, etc.) preceding the training of a classical MLP neural network and adapted to the inverse problem that is being solved. This allows us to formally, excluding the human factor, solve inverse problems of geoelectrics of large dimensions without specifying a first approximation based on data measured in areas whose dimensions exceed the dimensions of the network training area. The inversion speed is a few tens of seconds and does not depend on the physical dimensionality (2D or 3D) of the data. The solution to the inverse problem found using a trained neural network can, if necessary, be refined using a random search method. Numerical results of solving 3D geoelectric problems on model and field data are presented, confirming the stated development parameters.

Weather Research and Forecasting (WRF-ARW) model and its three-dimensional variational data assimilation (3D-Var) system are used to investigate the impact of the Quick Scatterometer (QuikSCAT) near surface winds, Special Sensor Microwave/Imager (SSM/I)-derived Total Precipitable Water (TPW), and Meteosat-7-derived Atmospheric Motion Vectors (AMVs) on the track and intensity prediction of tropical cyclones over the North Indian Ocean. The case of tropical cyclone, Gonu (June 2007; Arabian Sea), is first tested and the results show significant improvements particularly due to the assimilation of QuikSCAT winds. Three other cases, cyclone Mala (April 2006; Bay of Bengal), Orissa super cyclone (October 1999; Bay of Bengal), and Very Severe Cyclonic storm (October 1999; Bay of Bengal), are then examined. The prediction of cyclone tracks improved significantly with the assimilation of QuikSCAT winds. The track improvement resulted from the relocation of the initial cyclonic vortices after the assimilation of QuikSCAT wind vectors. After the assimilation of QuikSCAT winds, the mean (for four cyclone cases) track errors for first, second, and third day forecasts are reduced to 72, 101, and 166 km, respectively, from 190, 250, and 381 km of control (without QuikSCAT winds) runs. The assimilation of QuikSCAT winds also shows positive impact on the intensity (in terms of maximum surface level winds) prediction particularly for those cyclones, which are at their initial stages of the developments at the time of data assimilation. The assimilation of SSM/I TPW has significant influence (negative and positive) on the cyclone track. In three of the four cases, the assimilation of the SSM/I TPW resulted in drying of lower troposphere over cyclonic region. This decrease of moisture in TPW assimilation experiment resulted in reduction of cyclonic intensity. In three of the four cyclones, the assimilation of Meteosat-7 AMVs shows negative impact on the track prediction.

For general users, training a neural network from scratch is usually challenging and labor-intensive. Fortunately, neural network zoos enable them to find a well-performing model for directly use or fine-tuning it in their local environments. Although current model retrieval solutions attempt to convert neural network models into vectors to avoid complex multiple inference processes required for model selection, it is still difficult to choose a suitable model due to inaccurate vectorization and biased correlation alignment between the query dataset and models. From the perspective of knowledge consistency, i.e., whether the knowledge possessed by the model can meet the needs of query tasks, we propose a model retrieval scheme, named Know2Vec, that acts as a black-box retrieval proxy for model zoo. Know2Vec first accesses to models via a black-box interface in advance, capturing vital decision knowledge from models while ensuring their privacy. Next, it employs an effective encoding technique to transform the knowledge into precise model vectors. Secondly, it maps the user's query task to a knowledge vector by probing the semantic relationships within query samples. Furthermore, the proxy ensures the knowledge-consistency between query vector and model vectors within their alignment space, which is optimized through the supervised learning with diverse loss functions, and finally it can identify the most suitable model for a given task during the inference stage. Extensive experiments show that our Know2Vec achieves superior retrieval accuracy against the state-of-the-art methods in diverse neural network retrieval tasks.