Extreme Precipitation Modelling Using Geostatistics and Machine Learning Algorithms

The paper presents an approach for mapping of precipitation data. The main goal is to perform spatial predictions and simulations of precipitation fields using geostatistical methods (ordinary kriging, kriging with external drift) as well as machine learning algorithms (neural networks). More practically, the objective is to reproduce simultaneously both the spatial patterns and the extreme values. This objective is best reached by models integrating geostatistics and machine learning algorithms. To demonstrate how such models work, two case studies have been considered: first, a 2-day accumulation of heavy precipitation and second, a 6-day accumulation of extreme orographic precipitation. The first example is used to compare the performance of two optimization algorithms (conjugate gradients and Levenberg-Marquardt) of a neural network for the reproduction of extreme values. Hybrid models, which combine geostatistical and machine learning algorithms, are also treated in this context. The second dataset is used to analyze the contribution of radar Doppler imagery when used as external drift or as input in the models (kriging with external drift and neural networks). Model assessment is carried out by comparing independent validation errors as well as analyzing data patterns.

Characterization of pancreatic NMDA receptors as possible drug targets for diabetes treatment.

In the nervous system, NMDA receptors (NMDARs) participate in neurotransmission and modulate the viability of neurons. In contrast, little is known about the role of NMDARs in pancreatic islets and the insulin-secreting beta cells whose functional impairment contributes to diabetes mellitus. Here we found that inhibition of NMDARs in mouse and human islets enhanced their glucose-stimulated insulin secretion (GSIS) and survival of islet cells. Further, NMDAR inhibition prolonged the amount of time that glucose-stimulated beta cells spent in a depolarized state with high cytosolic Ca(2+) concentrations. We also noticed that, in vivo, the NMDAR antagonist dextromethorphan (DXM) enhanced glucose tolerance in mice, and that in vitro dextrorphan, the main metabolite of DXM, amplified the stimulatory effect of exendin-4 on GSIS. In a mouse model of type 2 diabetes mellitus (T2DM), long-term treatment with DXM improved islet insulin content, islet cell mass and blood glucose control. Further, in a small clinical trial we found that individuals with T2DM treated with DXM showed enhanced serum insulin concentrations and glucose tolerance. Our data highlight the possibility that antagonists of NMDARs may provide a useful adjunct treatment for diabetes.

A comparison of alpha and gamma spectrometry for environmental natural radioactivity surveys.

An alpha-spectrometry, using automated borate fusion and sequential extraction and exchange chromatography, was used to determine the uranium and thorium based on environmental radioactivity of 20 soil samples. The same set of the samples was analysed using gamma-spectrometry with an HPGe detector. The two data sets were checked for coherence using Z-score and chi2 statistical tests. We show that gamma-spectrometry is a valid alternative to time-consuming alpha-spectrometry for the determination of natural uranium and thorium activity in soil (activity range: 12.5-58.2 Bq/kg). The measured activities were compared with the theoretical activities to ensure secular equilibrium in the 238U and 232Th series. For 226Ra, a special study was made on deconvolution of the 186 keV multiplet with the Levenberg-Marquardt algorithm. Finally, the combined use of Z-score and chi2-tests was found to be a powerful tool for comparing the results obtained with two different methods.

Développement de l'interprétation quantitative de la sismique réflexion lacustre : application de l'AVO aux sédiments quaternaires dans le lac Léman

Résumé Des développements antérieurs, au sein de l'Institut de Géophysique de Lausanne, ont permis de développer des techniques d'acquisition sismique et de réaliser l'interprétation des données sismique 2D et 3D pour étudier la géologie de la région et notamment les différentes séquences sédimentaires du Lac Léman. Pour permettre un interprétation quantitative de la sismique en déterminant des paramètres physiques des sédiments la méthode AVO (Amplitude Versus Offset) a été appliquée. Deux campagnes sismiques lacustres, 2D et 3D, ont été acquises afin de tester la méthode AVO dans le Grand Lac sur les deltas des rivières. La géométrie d'acquisition a été repensée afin de pouvoir enregistrer les données à grands déports. Les flûtes sismiques, mises bout à bout, ont permis d'atteindre des angles d'incidence d'environ 40˚ . Des récepteurs GPS spécialement développés à cet effet, et disposés le long de la flûte, ont permis, après post-traitement des données, de déterminer la position de la flûte avec précision (± 0.5 m). L'étalonnage de nos hydrophones, réalisé dans une chambre anéchoïque, a permis de connaître leur réponse en amplitude en fonction de la fréquence. Une variation maximale de 10 dB a été mis en évidence entre les capteurs des flûtes et le signal de référence. Un traitement sismique dont l'amplitude a été conservée a été appliqué sur les données du lac. L'utilisation de l'algorithme en surface en consistante a permis de corriger les variations d'amplitude des tirs du canon à air. Les sections interceptes et gradients obtenues sur les deltas de l'Aubonne et de la Dranse ont permis de produire des cross-plots. Cette représentation permet de classer les anomalies d'amplitude en fonction du type de sédiments et de leur contenu potentiel en gaz. L'un des attributs qui peut être extrait des données 3D, est l'amplitude de la réflectivité d'une interface sismique. Ceci ajoute une composante quantitative à l'interprétation géologique d'une interface. Le fond d'eau sur le delta de l'Aubonne présente des anomalies en amplitude qui caractérisent les chenaux. L'inversion de l'équation de Zoeppritz par l'algorithme de Levenberg-Marquardt a été programmée afin d'extraire les paramètres physiques des sédiments sur ce delta. Une étude statistique des résultats de l'inversion permet de simuler la variation de l'amplitude en fonction du déport. On a obtenu un modèle dont la première couche est l'eau et dont la seconde est une couche pour laquelle V P = 1461 m∕s, ρ = 1182 kg∕m3 et V S = 383 m∕s. Abstract A system to record very high resolution (VHR) seismic data on lakes in 2D and 3D was developed at the Institute of Geophysics, University of Lausanne. Several seismic surveys carried out on Lake Geneva helped us to better understand the geology of the area and to identify sedimentary sequences. However, more sophisticated analysis of the data such as the AVO (Amplitude Versus Offset) method provides means of deciphering the detailed structure of the complex Quaternary sedimentary fill of the Lake Geneva trough. To study the physical parameters we applied the AVO method at some selected places of sediments. These areas are the Aubonne and Dranse River deltas where the configurations of the strata are relatively smooth and the discontinuities between them easy to pick. A specific layout was developed to acquire large incidence angle. 2D and 3D seismic data were acquired with streamers, deployed end to end, providing incidence angle up to 40˚ . One or more GPS antennas attached to the streamer enabled us to calculate individual hydrophone positions with an accuracy of 50 cm after post-processing of the navigation data. To ensure that our system provides correct amplitude information, our streamer sensors were calibrated in an anechoic chamber using a loudspeaker as a source. Amplitude variations between the each hydrophone were of the order of 10 dB. An amplitude correction for each hydrophone was computed and applied before processing. Amplitude preserving processing was then carried out. Intercept vs. gradient cross-plots enable us to determine that both geological discontinuities (lacustrine sediments/moraine and moraine/molasse) have well defined trends. A 3D volume collected on the Aubonne river delta was processed in order ro obtain AVO attributes. Quantitative interpretation using amplitude maps were produced and amplitude maps revealed high reflectivity in channels. Inversion of the water bottom of the Zoeppritz equation using the Levenberg-Marquadt algorithm was carried out to estimate V P , V S and ρ of sediments immediately under the lake bottom. Real-data inversion gave, under the water layer, a mud layer with V P = 1461 m∕s, ρ = 1182 kg∕m3 et V S = 383 m∕s.