1 Bioprocess Engineering Group, IIM-CSIC, Eduardo Cabello 6, 36208 Vigo, Spain 2 European Molecular Biology Laboratory, European Bioinformatics Institute (EMBL-EBI), Wellcome Trust Genome Campus, Hinxton, Cambridge CB10 1SD, UK 3 School of Computer Science, Manchester Centre for Integrative Systems Biology, The University of Manchester, 131 Princess Street, Manchester, M1 7DN, United Kingdom 4 Insilico Biotechnology AG, Meitnerstrase 8, 70563 Stuttgart, Germany 5 EMBL/CRG Research Unit in Systems Biology, Centre for Genomic Regulation (CRG), Dr. Aiguader 88, 08003 Barcelona, Spain 6 Universitat Pompeu Fabra (UPF), Placa de la Merce, 10, 08002 Barcelona, Spain

1 Bioprocess Engineering Group, IIM-CSIC, Eduardo Cabello 6, 36208 Vigo, Spain 2 European Molecular Biology Laboratory, European Bioinformatics Institute (EMBL-EBI), Wellcome Trust Genome Campus, Hinxton, Cambridge CB10 1SD, UK 3 School of Computer Science, Manchester Centre for Integrative Systems Biology, The University of Manchester, 131 Princess Street, Manchester, M1 7DN, United Kingdom 4 Insilico Biotechnology AG, Meitnerstrase 8, 70563 Stuttgart, Germany 5 EMBL/CRG Research Unit in Systems Biology, Centre for Genomic Regulation (CRG), Dr. Aiguader 88, 08003 Barcelona, Spain 6 Universitat Pompeu Fabra (UPF), Placa de la Merce, 10, 08002 Barcelona, Spain

Choosing a method to tune regularization is non-trivial. Bauer and Lukas[1] compared 17 regularization parameter choice methods for a large set of linear least squares problems. Apart from the numerical case studies, they also provided a unified framework to discuss the numerical aspects of each method, and the performance of the regularization methods from the estimated parameters point of view. We can see that most of the methods work very well in terms of achieving a small mean square error in the estimated parameters. However, small parameter estimation error do not imply small prediction errors. Most of the regularization tuning methods can be used together with nonlinear least squares (NLS) methods, such as e.g. the Landweber algorithm, the iteratively regularized Gauss Newton method or conjugate gradient method [2]. These optimization methods iteratively update the parameter estimates by a suitable step for which the residual vector or the objective function (depending on the method) is linearized in each step and the regularization is applied to the steps. Our approach is different, since we apply the regularization directly to the parameter vector instead of the updating steps, which allow us to use the regularization in both the global and the local optimization phases. In the following section we shortly summarize the regularization tuning methods considered and their computational details, which we utilized for the Tikhonov regularization of the hybrid optimization algorithm presented in the main text. In the following sections we also report the results of the regularization tuning methods for each case study. For each problem, the regularization parameter selected by the regularization tuning method is compared against the bias-variance trade-off curves (both for the bias-variance in the estimated parameters and in the estimated predictions). Note that the performance of a tuning method is evaluated based on the distance of the selected regularization parameter from the location of the minimum of the bias-variance trade-off curves.

As the structure of biofilms plays a key role in their resistance and persistence, this work presents for the 11 first time the numerical characterization of the temporal evolution of biofilm structures formed by three 12 Listeria monocytogenes strains on two types of stainless-steel supports, AISI 304 SS No. 2B and AISI 13 316 SS No. 2R. 14 Counting methods, motility tests, fluorescence microscopy and image analysis were combined to study 15 the dynamic evolution of biofilm formation and structure. Image analysis was performed with several 16 well-known parameters as well as a newly defined parameter to quantify spatio-temporal distribution. 17 The results confirm the interstrain variability of L. monocytogenes species regarding to biofilm structure 18 and structural evolution. Two types of biofilm were observed: homogeneous or flat and heterogeneous or 19 clustered. Differences in clusters and in attachment and detachment processes were due mainly to the 20 topography and composition of the two surfaces although an effect due to motility was also found. 21

Summary: Strandings provide valuable information on the presence and relative abundance of cetaceans in an area. The spatial, inter-annual and seasonal variation in the number of strandings may reflect the variability in cetacean abundance. However, this signal can be masked by other factors which are responsible both for the mortality of the dolphins and the probability that they reach the shore once they are dead. We analysed the time series of cetacean strandings in Galicia (NW Spain), available since 1990, with the aim of determining the trends and variability of strandings. We found that seasonal and inter-annual variation in the number of strandings is highly correlated with the oceanographic conditions most prevalent in each period. Moreover, a more detailed analysis of these effects should be carried out, and the use of further applicable indices (fishing effort rather than landings) may improve the explained part of the models, thus allowing us to disentangle oceanographic effects and to obtain relative seasonal and annual abundances.

As part of the VACLAN (Climate Variability in the North Atlantic) project, a section covering the Bay of Biscay was sampled in September 2005. This work estimates the distribution of the different water masses in the region using an extended optimum multiparametric method and analyzes water mass distribution of anthropogenic carbon as calculated using two different approaches. The Eastern North Atlantic Central Water layer is mainly constituted by its subpolar component and Mediterranean Water appears very diluted, its dilution increasing northeastward. In relation to the anthropogenic carbon inventory, small differences were found between the two different methods used, 95 vs 87 mol C m -2 , though both show the same distribution pattern, the concentration decreasing with depth. Eastern North Atlantic Central Water presents the highest anthropogenic carbon inventory, supporting more than 50% of the total column (52%). This work confirms the relevant

The carbon system in the water masses of the Iberian Basin (North Atlantic Ocean) has been affected over the last two decades by the increase in anthropogenic CO2 (Cant). In order to study the storage of Cant in the Iberian Basin, variables of the carbonic system (i.e., pH, total inorganic carbon, and total alkalinity), among others, were measured during the CAIBOX cruise conducted between July and August 2009 within the framework of the CAIBEX project (Shelf-Ocean Exchanges in the Canary-Iberian Large Marine Ecosystem). The storage of Cant was estimated using two different back-calculation techniques (i.e., the ϕCoT and TrOCA methods) and for six layers of the water column corresponding to the approximate locations of the characteristic water masses of the region and the mixed layers. For the whole water column and for the year 2009 the Cant storage values determined by the ϕCoT and TrOCA methods were 88.1 ± 3.8 and 93.7 ± 3.7 molC m -2 , respectively. Moreover, the Cant storage rate from 1993 to 2009 was also estimated considering data from three additional cruises (OACES 1993, CHAOS 1998, and OACES 2003). The Cant storage rates were 1.41 ± 0.25 and 1.67 ± 0.13 molC m -2 yr -1 with the ϕCoT and TrOCA methods, respectively. An increase in anthropogenic CO2 uptake by the ocean can be seen when compared with previous published results. Between the periods 1977-1997 and 1993-2009, the Cant concentration increased around 28-49% in the first 2000 m.

8 This work estimates new regionalized empirical parameterizations for preformed alkalinity ( ) and 9 the CO2 air-sea disequilibrium (∆Cdis). Both are key terms for the computation of anthropogenic CO2 10 in the back-calculation methods. Data from the subsurface layer (75–180m depth range) covering an 11 area from North to South and from 19oE to 67.5oW (Pacific and Indian oceans) were taken from 12 GLODAP (The Global Ocean Data Analysis Project) database. The subsurface layer is proved as a 13 reliable reference for representing the main characteristics of the different water masses of the 14 oceans. Besides, handing data from the two ocean basins altogether makes the new parameterizations 15 of and ∆Cdis to be more globally consistent. Nevertheless, each ocean basin, at least in some 16 regions, has different oceanographic characteristics based on its proper dynamical processes and 17 water masses formation. In order to maintain each ocean basin „identity‟ the whole domain was 18 divided in six different regions (two of them sharing waters from Pacific and Indian oceans) and 19 parameterizations in each region for both terms were obtained. Previously, data were transformed 20 into a grid of 4olat. x 5olon. and the results obtained from the parameterizations were visualized and 21 compare with pCO2 climatologies. From the comparisons with previous ∆Cdis estimations good 22 results are obtained showing the reliability and robustness of the new regionalized empiric 23 parameterizations. 24 25 *Manuscript Click here to download Manuscript: Pardoetal_aftrev.docx Click here to view linked References

During the last two decades, important advances have been made in the field of bilateral teleoperation. Different techniques for performing stable teleoperation in non-ideal conditions have been developed, especially in a passivity framework. Until recently, however, no robust solutions for addressing this problem with variable delays and other drawbacks of packet-switched networks have been developed. The requirement of maintaining passivity in these circumstances degrades performance, due to the loss of energy that it involves. In this paper an arrangement is proposed which is capable of eliminating position errors, while maintaining passivity of an internet-like channel. The behaviour of this new controller is studied by Lyapunov analysis, compared to previous methods, and validated through numerical simulations.

The “MedBox” region, comprising the Strait of Gibraltar to 22oW and from 24oN to 41oN, has been defined using thermohaline and chemical data from three WOCE cruises conducted in 1997/98. The water mass structure of the region was objectively solved by an extended optimum multiparameter (OMP) analysis. Volume transports were estimated from field data and an inverse box model that ensures volume conservation and no deep water formation within the region. The combination of the volume transports, the OMP water mass analysis, and the distributions of chemical parameters allowed for the assessment of the oxygen, inorganic carbon, nitrate, phosphate, and silicate transport mechanisms, their vertical and lateral variability, and the relative contribution of the different water masses in the mid-latitudes of the eastern North Atlantic to these transports.

Certain heavy minerals enriched in yttrium, such as monazite and xenotime, have previously been indexed in the Vigo area. The levels and spatial distribution of this element were studied in 50 surface sediment samples from the Vigo Ria. The highest levels were found along the southern margin in the middle of the ria, coinciding with the geological source of the Galineiro Complex. From this zone, a dilution of yttrium in the sediments was observed seaward due to the presence of calcium carbonate (outer ria) and upstream due to the input of migmatites, anatexites, and two-mica granites (inner ria). The estimated background concentration in the Vigo Ria was 14.6 ± 2.6 mg kg –1 (Sc normalized), which is close to the average concentration found in surface sediments (13.4 ± 5.4 mg kg –1 ). This suggests that yttrium in the Vigo Ria sediments is mainly of lithogenic origin.

The anthropogenic CO2 (Cant) estimates from cruises spanning more than two decades (1981-2006) in the Irminger Sea area of the North Atlantic Sub- polar Gyre reveal a large variability in the C ant stor- age rates. During the early 1990's, the C ant storage rates (2.3±0.6 mol C m 2 yr 1 ) doubled the average rate for 1981-2006 (1.1±0.1 mol C m 2 yr 1 ), whilst a remark- able drop to almost half that average followed from 1997 onwards. The Cant storage evolution runs parallel to chlorofluorocarbon-12 inventories and is in good agreement with Cant uptake rates of increase calculated from sea sur- face pCO2 measurements. The contribution of the Labrador Seawater to the total inventory of Cant in the Irminger basin dropped from 66% in the early 1990s to 49% in the early 2000s. The North Atlantic Oscillation shift from a positive to a negative phase in 1996 led to a reduction of air-sea heat loss in the Labrador Sea. The consequent convection weakening accompanied by an increase in stratification has lowered the efficiency of the northern North Atlantic CO 2 sink.

Thermohaline and chemical data from three WOCE (World Ocean Circulation Experiment) cruises conducted in 1997 and 1998 define the MedBox region, bounded by the Strait of Gibraltar, 24uN, 41uN, and 22uW. The carbon budget indicates that the MedBox is a heterotrophic region, where carbon is mineralized at a net rate of 17 6 11 g C m22 yr21, supported by the input of allochthonous organic matter, mainly in the dissolved form, from the adjacent ocean. Dissolved organic carbon accounts for 90% of the organic carbon demand. In vitro measurements to estimate the net community production of the study area differ largely from our geochemical budget estimation, likely reflecting different spatial and temporal scales and/or terms or processes not taken into account by both methods. The nitrogen budget of the MedBox pointed to a significant atmospheric input via N2 fixation (3.4 6 3.1 g N m22 yr21).

The dynamic optimization of microwave heating of foods is considered. Two classes of problems (maximum temperature uniformity and maximum quality uniformity) are stated here. Results obtained using a deterministic method, an stochastic method and new hybrid method are reported. It is shown that, for the models and food loads considered, it is theoretically possible to achieve a quite uniform final quality distribution using the combined optimal controls of microwave heating and oven (convection) heating. INTRODUCTION Most processes in the food industry are operated in batch or semi-continuous mode. In order to find the best operating policies, robust optimal control (dynamic optimization) methods are needed. For example, thermal sterilization of prepackaged foods, one of the most important preservation techniques, is often carried out in batch retorts. The determination of the optimal retort temperature profile is a complex dynamic optimization problem which has received great attention in the literature (Banga et al, 1991; Silva et al, 1993). In contrast, there is a lack of studies regarding the dynamic optimization of microwave heating of foods. This has been very likely due to the high complexity of this process, and therefore of its mathematical models, which has hindered the application of dynamic optimization methods. In this study, our objective is to obtain the optimal control policies (microwave power and oven temperature) for microwave combination ovens considering process models of medium complexity. The ultimate aim is to improve the final quality of the food by optimally operating the oven, avoiding some of the well-known problems of microwave heating, like cold spots and edges overheating. STATEMENT OF OPTIMAL CONTROL PROBLEMS Here, we present the general mathematical statements for two classes of optimal control problems concerning microwave heating in combination ovens: OCP-I Find the optimal controls (microwave field power P{ t} and oven temperature Toven{ t} ) to achieve maximum temperature uniformity at a ACoFoP IV (Automatic Control of Food & Biological Processes), Göteborg, Sweden, 21-23 September 1998. Correspondence to: Dr. Julio R. Banga, IIM-CSIC. Eduardo Cabello 6, 36208 Vigo, SPAIN. E-mail:julio@iim.csic.es 2 given final time. OCP-II Find the optimal controls, P{ t} and Toven{ t} , to achieve maximum uniformity of the final product quality (as measured by the “cook” Cvalue), while ensuring a minimum C-value. The corresponding mathematical statements (e.g. for cylindrical coordinates) are: OCP-I Maximum temperature uniformity Find {} t P and {} t Toven over [ ] f t , t t 0 ∈ to minimize [ ] [ ] } t ; z , r { T min } t ; z , r { T max J f food f food − = (1) subject to ƒ An inequality constraint on the minimum desired temperature to be achieved: [ ] SET min f food T } t ; z , r { T min ≥ (2) ƒ The system model (set of partial differential, ordinary differential and algebraic equality constraints, i.e. the equation for heat conduction with internal generation due to microwaves, its boundary and initial conditions, degradation kinetics, etc.) OCP-II Maximum C-value uniformity (free terminal time) Find {} t P , {} t Toven and f t over [ ] f t , t t 0 ∈ to minimize [ ] } t ; z , r { C max J f Tref = (3) subject to ƒ An inequality constraint on the minimum desired C-value to be achieved: [ ] SET min , Zref f Zref C } t ; z , r { C min ≥ (4) ƒ An inequality constraint on the maximum allowed temperature at the end of the process: [ ] SET max f food T } t ; z , r { T max ≤ (5) ƒ The system model (set of partial, ordinary differential and algebraic equality constraints, as mentioned above) In this latter statement, the C-value (“cook” value) is used as the performance index. The C-value has been used successfully to measure the product quality changes during thermal processing (Ohlsson, 1980; Ohlsson, 1986; Ohlsson, 1988). It is defined as the equivalent number of minutes the product would have to spend at a reference temperature (usually Tref=100°C) to achieve a given effect on the final product quality. Mathematically: dt C t Z / ) T T ( Z ref ref ref ∫ − =