Deep neural network application for 4D seismic inversion to changes in pressure and saturation: Optimizing the use of synthetic training datasets

In this work, we tackle the challenge of quantitative estimation of reservoir dynamic property variations during a period of production, directly from four-dimensional seismic data in the amplitude domain. We employ a deep neural network to invert four-dimensional seismic amplitude maps to the simultaneous changes in pressure, water and gas saturations. The method is applied to a real field data case, where, as is common in such applications, the data measured at the wells are insufficient for properly training deep neural networks, thus, the network is trained on synthetic data. Training on synthetic data offers much freedom in designing a training dataset, therefore, it is important to understand the impact of the data distribution on the inversion results. To define the best way to construct a synthetic training dataset, we perform a study on four different approaches to populating the training set making remarks on data sizes, network generality and the impact of physics-based constraints. Using the results of a reservoir simulation model to populate our training datasets, we demonstrate the benefits of restricting training samples to fluid flow consistent combinations in the dynamic reservoir property domain. With this the network learns the physical correlations present in the training set, incorporating this information into the inference process, which allows it to make inferences on properties to which the seismic data are most uncertain. Additionally, we demonstrate the importance of applying regularization techniques such as adding noise to the synthetic data for training and show a possibility of estimating uncertainties in the inversion results by training multiple networks.     

Geochemistry of groundwater in front of a warm‐based glacier in Southeast Greenland

Groundwater in front of warm‐based glaciers is likely to become a more integrated part of the future proglacial hydrological system at high latitudes due to global warming. Here, we present the first monitoring results of shallow groundwater chemistry and geochemical fingerprinting of glacier meltwater in front of a warm‐based glacier in Southeast Greenland (Mittivakkat Gletscher, 65° 41′ N, 37° 48′ W). The groundwater temperature, electrical conductivity and pressure head were monitored from August 2009 to August 2011, and water samples were collected in 2009 and analyzed for major ions and water isotopes (δD, δ¹⁸O). The 2 yrs of monitoring revealed that major outbursts of glacier water during the ablation season flushed the proglacial aquifer and determined the groundwater quality for the next 2–8 weeks until stable chemical conditions were reached again. Water isotope composition shows that isotopic fractionation occurs in both groundwater and glacier meltwater, but fractionation due to evaporation from near‐surface soil moisture prior to infiltration has the most significant effect. This study shows that groundwater in Low Arctic Greenland is likely to possess a combined geochemical and isotopic composition, which is distinguishable from other water sources in the proglacial environment. However, the shallow groundwater composition at a given time is highly dependent on major outbursts of glacier water in the previous months.     

Sensitivity Kernels for Time-Distance Inversion

Inversion of local-area helioseismic time-distance data has so far only been done in the ray approximation (Kosovichev, 1996). Since this is a high-frequency approximation its applicability can be questioned for the solar case. Bogdan (1997) showed that for a simple solar model the localized wave packets do follow, but are not confined to, the ray path. We use an approximation based on the first Fresnel zone that has been developed in geophysics by Snieder and Lomax (1996) to go beyond the ray approximation in the inversions. We have calculated sensitivity kernels using both approximations. To test them we use a finite-difference forward modeling of the whole wave field in an acoustic medium reminiscent of the Sun. We use the finite-difference modeling to calculate sensitivity kernels for the full wave field and compare this with the other kernels. The results show that the Fresnel-zone-based kernels are in good agreement with the sensitivity obtained from the modeling. Thus these new kernels represent a significant step forward in the inversion of time-distance data.     

The simulation of Arctic clouds and their influence on the winter surface temperature in present-day climate in the CMIP3 multi-model dataset

We investigate the influence of clouds on the surface energy budget and surface temperature in the sea-ice covered parts of the ocean north of the Arctic circle in present-day climate in nine global climate models participating in the Coupled Model Intercomparison Project phase 3, CMIP3. Monthly mean simulated surface skin temperature, radiative fluxes and cloud parameters are evaluated using retrievals from the extended AVHHR Polar Pathfinder (APP-x) product. We analyzed the annual cycle but the main focus is on the winter, in which large parts of the region experience polar night. We find a smaller across-model spread as well as better agreement with observations during summer than during winter in the simulated climatological annual cycles of total cloudiness and surface skin temperature. The across-model spread in liquid and ice water paths is substantial during the whole year. These results qualitatively agree with earlier studies on the present-day Arctic climate in GCMs. The climatological ensemble model mean annual cycle of surface cloud forcing shows good agreement with observations in summer. However, during winter the insulating effect of clouds tends to be underestimated in models. During winter, most of the models as well as the observations show higher monthly mean total cloud fractions, associated with larger positive surface cloud forcing. Most models also show good correlation between the surface cloud forcing and the vertically integrated ice and liquid cloud condensate. The wintertime ensemble model mean total cloud fraction (69%) shows excellent agreement with observations. The across-model spread in the winter mean cloudiness is substantial (36?C94%) however and several models significantly underestimate the cloud liquid water content. If the two models not showing any relationship between cloudiness and surface cloud forcing are disregarded, a tentative across-model relation exists, in such a way that models that simulate large winter mean cloudiness also show larger surface cloud forcing. Even though the across-model spread in wintertime surface cloud forcing is large, no clear relation to the surface temperature is found. This indicates that other processes, not explicitly cloud related, are important for the simulated across-model spread in surface temperature.     

Single-Column Model Intercomparison for a Stably Stratified Atmospheric Boundary Layer

The parameterization of the stably stratified atmospheric boundary layer is a difficult issue, having a significant impact on medium-range weather forecasts and climate integrations. To pursue this further, a moderately stratified Arctic case is simulated by nineteen single-column turbulence schemes. Statistics from a large-eddy simulation intercomparison made for the same case by eleven different models are used as a guiding reference. The single-column parameterizations include research and operational schemes from major forecast and climate research centres. Results from first-order schemes, a large number of turbulence kinetic energy closures, and other models were used. There is a large spread in the results; in general, the operational schemes mix over a deeper layer than the research schemes, and the turbulence kinetic energy and other higher-order closures give results closer to the statistics obtained from the large-eddy simulations. The sensitivities of the schemes to the parameters of their turbulence closures are partially explored.     

Multi-species chemical data assimilation with the Danish Eulerian hemispheric model: system description and verification

Satellite retrievals of atmospheric composition provide a wealth of data on a global scale. These complement results from atmospheric chemistry-transport models (CTMs), and can be combined using data assimilation. We present two assimilation schemes coupled to the Danish Eulerian Hemispheric Model (DEHM), a three-dimensional, off-line CTM with full photochemistry: a variant on the ensemble Kalman filter and the three-dimensional variational scheme. The aim of this paper is to describe the two schemes and present an initial assessment of their impacts on model skill. Retrievals of multiple atmospheric trace gases are assimilated, namely: NO₂ tropospheric column densities, CH₄ total column densities, and partial column concentrations of O₃, CO and CH₄; these data are retrieved from four satellite sensors. Data for each species are assimilated independently of one another, and other species are only adjusted indirectly via the model’s chemistry and dynamics. Assimilation results are compared with measurements from surface monitoring stations and other satellite retrievals, and preliminary validation results are presented.Reference simulations (without assimilation) grossly underestimate surface CO concentrations, and both assimilation schemes eliminate this large and systematic model bias. The assimilation improves the spatial correlation of modelled CO with surface observations, and improves the spatial correlation between forecasts and retrievals for CO, NO₂ and O₃. Results for CH₄ show a loss of skill due to a mismatch in model bias between two assimilated CH₄ data-sets. Finally, we discuss differences in methodology and results between this paper and a recent study on multi-species chemical data assimilation. Joint optimisation of initial conditions and emission rates offers a promising direction for improving modelled boundary-layer concentrations.