Validation of a numerical simulation of the North Atlantic Ocean with ARGO Data

VALIDATION OF A NUMERICAL SIMULATION OF THE NORTH ATLANTIC OCEAN WITH ARGO DATA Degree in Marine Sciences Laura Gómez Navarro Tutor: Alonso Hernández Guerra Co-tutor: Joaquim Ballabrera Poy June 2014 Validation of a numerical simulation of the North Atlantic Ocean with Argo data. Abstract The outputs of a numerical simulation have been validated with Argo data; with the idea of being able to use these outputs to assimilate SSS information retrieved by the satellite SMOS. In addition to Argo data, the model is also compared against the WOA09 climatology. There are a considerable amount of valid profiles available during year 2012, and the coverage does not have temporal biases, as the amount of data is homogeneous throughout the year. This demonstrates that choosing Argo to validate the simulation was a correct choice. The results show that there exists a reasonable agreement between model and Argo data in the open Ocean. However, it has been observed that the simulation still contains spatial and temporal errors probably due to a bad representation of certain oceanic processes. Nevertheless, as the errors are easily identified and do not cover the open Ocean in the tropical and subtropical regions, the simulation could still be considered as an appropriate tool to improve the remote sensing of SSS data. Laura Gómez Navarro 4 Table of Contents 1. Introduction ................................................................................................................. 6 2. Data .............................................................................................................................. 8 2.1. Model and simulation description .......................................................................... 8 2.2. World Ocean Atlas ................................................................................................. 9 2.3. Argo ..................................................................................................................... 10 3. Methodology and results ............................................................................................ 14 3.1. Preliminary data exploration. ............................................................................... 14 3.2. The 2012 North Atlantic numerical simulation ................................................... 19 3.3. The 2012 North Atlantic Argo data ..................................................................... 23 3.4. Simulation-Argo differences. ............................................................................... 28 4. Conclusions ............................................................................................................... 34 5. References ................................................................................................................. 37 6. Appendices ................................................................................................................. 39 6.1. Appendix 1 ........................................................................................................... 39 6.2. Appendix 2 .......................................................................................................... 40 6.3. Appendix 3 ........................................................................................................... 41 7. Additional information about the development of this study .................................... 42 7.1. Detailed description of the activities carried out. ................................................ 42 7.1.1. Statistical characterization of the simulation’s outputs. ................................ 42 7.1.2. Recovery and analysis of the data measured by Argo floats. ................... 43 7.1.3. Global comparison of the numerical model and Argo data. ......................... 44 7.2. Training received ................................................................................................. 45 7.3. Level of integration and involvement in the department and relationship with personnel. .................................................................................................................... 45 7.4. Positive and negative aspects related to the development of this study. ............. 46 7.5. Personal appraisal of the learning achieved during this study. ............................ 46 Validation of a numerical simulation of the North Atlantic Ocean with Argo data. 5 List of figures Figure 1. Schematic of the surface circulation of the Atlantic Ocean ......................................................... 6 Figure 2. World distribution of the upper Oceans’ water masses ................................................................ 7 Figure 3. Argo floats distribution on 11th of May 2014 ............................................................................ 10 Figure 4. Description of Argo floats’ cycle ............................................................................................... 11 Figure 5. Different parts of one of the Argo floats models ....................................................................... 12 Figure 6. Sea Surface Height (m) in 2008 ................................................................................................. 14 Figure 7. Histograms of potential temperature data at layer 1 ................................................................... 16 Figure 8. Change of mean and median values with depth ......................................................................... 16 Figure 9. Contour map of salinity (psu) at layer 1 with the peripheral regions marked. ........................... 17 Figure 10. T-S diagram of all the simulation data. .................................................................................... 18 Figure 11. Location of Sm values lower than 30 psu. ............................................................................... 19 Figure 12. Maximum values’ distribution in metres for SSH and MLD ................................................... 20 Figure 13. SSH’s statistical parameters variation with time. ..................................................................... 20 Figure 14. MLD’s statistical parameters variation with time. ................................................................... 21 Figure 15. Salinity 1D plot at layer 15 for year 2012. .............................................................................. 22 Figure 16. Minimum salinity values’ distribution for 2012 layer 5 and WOA09 salinity values. ............. 23 Figure 17. Minimum salinity values’ distribution for 2012 layer 14 and WOA09 salinity values . .......... 23 Figure 18. Argo profiles for 2012 .............................................................................................................. 25 Figure 19. Potential temperature (ºC) of the Argo profiles for layer 1 and layer 2. .................................. 26 Figure 20. Salinity (psu) of the Argo profiles for layer 1 and layer 2. ...................................................... 27 Figure 21. Ta and Sa statiscal parameters variation with time for layer 3 ................................................. 27 Figure 22. Scatter map of the ΔT (ºC) for layer 2, layer 14 and layer 29. ................................................. 29 Figure 23. Scatter map of ΔS (psu) for layer 2 .......................................................................................... 30 Figure 24. ΔT and ΔS values for every buoy. ............................................................................................ 30 Figure 25. Location of the Argo buoys with an absolute value of ΔT and ΔS .......................................... 31 Figure 26. Variation of ΔT and ΔS with time for layer 2. ......................................................................... 31 Figure 27. Positions of the selected buoys for the T-S diagrams ............................................................... 32 Figure 28. T-S diagrams of the Argo and simulation data ......................................................................... 33 Figure 29. Salinity values at the Gulf of Saint Lawrence, for the 15th of November at 30m.................... 35 List of tables Table 1. Advantages of the Argo dataset in comparison to other sources of data ..................................... 12 Table 2. Argo quality control flag scale .................................................................................................... 13 Table 3. Description of the statistical parameters calculated. .................................................................... 15 Table 4. Statistical parameters of SSH and MLD ...................................................................................... 17 Table 5. Dates of the 12 profiles selected for the T-S diagrams. ............................................................... 32 Laura Gómez Navarro 6 1. Introduction The Atlantic Ocean has been thoroughly studied for more than a century, and it is still a subject of great focus for various reasons. It has a significant role in regulating the climate, of the northern hemisphere (for example, the heat transported by the water of the Gulf Stream contributes to warm the European Subcontinent). In addition, deep convection occurs in the North Atlantic Ocean. Finally, it has a human importance in maritime security, economic, social and military activities. All of these roles of the Atlantic Ocean are due by the wind and heat forcing, but regulated by its water masses and circulation which can be seen in Figure 1. Figure 1. Schematic of the surface circulation of the Atlantic Ocean. Continuous lines are warm and discontinuous cold currents (Brown et al., 1988). The Ocean circulation is modulated by local density, which is a function of temperature and salinity. According to Emery and Thomson (2004), in oceanography, water masses are defined in diverse ways; “real, objective physical entities, building blocks from which the oceanic stratification (vertical structure) is constructed”, “mainly descriptive words, summary shorthand for pointing to prominent features in property distributions” or “a single point on a characteristic diagram such as a TemperatureSalinity (T-S) curve”. Validation of a numerical simulation of the North Atlantic Ocean with Argo data. 7 Thus, the analysis of the characteristics of the water masses present in a given region is important as it provides elements helping to describe the Oceans’ circulation. In the Atlantic Ocean different water masses are found:  North Atlantic Central Water (NACW) o North Atlantic East Central Water (NAECW) o North Atlantic West Central Water (NAWCW)  South Atlantic Central Water (SACW)  Antarctic Intermediate Water (AIW)  Arctic Intermediate Water (AAIW)  Mediterranean Intermediate Water (MIW)  North Atlantic Deep Water (NADW) o North Atlantic East Deep Water (NAEDW) o North Atlantic West Deep Water (NAWDW)  Antarctic Bottom Water (ABW) Each one can be identified by their particular temperature, salinity and depth. For example the MIW, is identified easily by its high salinity and temperature. In addition, different regions have different prominent water masses (Figure 2). Figure 2. World distribution of the upper Oceans’ water masses (Chen, 2009). Laura Gómez Navarro 8 In this study, the outputs of a numerical simulation of the North Atlantic Ocean are validated. This work represents a first step towards the goal of producing a better Seas Surface Salinity (SSS) remotely-sensed product in the North Atlantic Ocean. Indeed, since the launch of the Soil Moisture/Ocean Salinity (SMOS) satellite, members of the Institut de Ciències del Mar (CSIC) are working to produce worldwide maps of SSS from the information gathered by the satellite. However, the SSS retrievals are particularly noisy in the Atlantic Ocean as the satellite measures (in the L-band, 1.4GHz) are contaminated by many radio frequency interferences in this region. The numerical model used here is going to be used to help reduce the noise to signal ratio of the SMOS SSS in the North Atlantic using various data assimilation techniques that will use the model as a dynamical interpolator. 2. Data 2.1. Model and simulation description The numerical model solves the so called Primitive Equations with a free surface formulation (Brodeau et al., 2010). The regional, eddy-permitting configuration corresponds to the North Atlantic Ocean. The southern and northern open boundaries are relaxed towards the World Ocean Atlas 09 (WOA09) climatology data, to represent more realistically the Meridional Overturning Circulation (Hoareau et al., 2014). This configuration couples the Nucleus for European Modelling of the Ocean (NEMO) oceanic model and the Louvain-la-Neuve Ice Model (LIM2.0), which is inside the ORCA4 ¼º grid global model. The horizontal domain of this grid includes the Atlantic Ocean’s basin from 80ºN to 20ºS, the Nordic Seas, the Denmark Strait and part of the Mediterranean Sea, up to 23ºE (Minvielle et al., 2011). The spatial resolution is 0.25 degrees. Its vertical domain is of 45 geopotential levels, and the grid spacing is 6 m at the surface and 250 m at the bottom (Hoareau et al., 2014). River runoff is simulated by adding a flux of water with a salinity of 0 psu and with the same temperature as the one at its mouth. The solutions are advanced in time using a Leap-Frog method, combined with a Robert-Asselin filter to ensure the stability of the scheme. The time step of the simulation is 40 minutes. Validation of a numerical simulation of the North Atlantic Ocean with Argo data. 9 A 2001-2012 simulation using realistic forcing fields has been carried out by Ms. Nina Hoareau. The simulation requires about 6 hours for each simulated year when 60 Central Processing Units (CPUs) are used. Each output file of the simulation was saved in NetCDF format, occupied 106 Mb, and includes the five-day average of the stored parameters. 2.2. World Ocean Atlas Climatology is, as defined by the National Weather Service of the National Oceanic and Atmospheric Administration (NOAA), “a quantitative description of climate showing the characteristic values of climate variables over a region” (www.nws.noaa.gov/climate/help/glossary.php). Climatologies are static, averaged values of a given parameter, calculated during a certain period of time. Therefore, it does not provide, in general, an observable value, but it should give an idea of the most probable value to be measured. In particular, temperature and salinity climatologies in a given area are usually estimated by the average values of all observations ever measured at that area, and thus, they would define the mean state of the Ocean if the sampling were appropriate. The T and S observations are collected from measured surface data (as thermosalinographers) and/or profiles. However, as measurements are neither constant in time nor space, they require time-spatial averaging and smoothing (Higgison et al., 2009). The World Ocean Atlas, WOA (Antonov et al., 2010; Locarnini et al., 2010) is the most widely used climatology of oceanic temperature and salinity. This World Ocean climatology was first published by Levitus in 1982, and throughout the years it has been improved by including more data and using new methods to calculate climatology, and therefore releasing new atlases every few years (Chatterjee et al., 2012). The uses of these climatology atlases are diverse: boundary and/or initial conditions in Ocean circulation numerical models and Atmosphere-Ocean models, verification of numerical simulations of the ocean, as a form of “sea truth” for satellite Laura Gómez Navarro 16 temperature, it is interesting to notice the reduction of the differences between the mean and the median. The larger the differences between the mean and the median, the larger is the impact of the tails of the distribution, and the less significant are the values of the statistical parameters listed in Table 3. On the other hand, the results shown in Table 4 indicate that the statistical description of SSH is little affected by the latitude limits, but the MLD is. Figure 7. Histograms of potential temperature data at layer 1 (3m). Left: Original data. Right: With latitude limits and land values changed. Figure 8. Change of mean (red line) and median (blue line) values with depth for the original data (left) and with latitude limits and land values changed (right). -5 0 5 10 15 20 25 30 35 0 0.5 1 1.5 2 2.5 3x 104 Potential temperature (ºC) Frequency -5 0 5 10 15 20 25 30 35 0 0.5 1 1.5 2 2.5 3x 104 Potential temperature (ºC) Frequency 010 20 30 0 1000 2000 3000 4000 5000 6000 Potential temperature (ºC) Depth (m) 010 20 30 0 1000 2000 3000 4000 5000 6000 Potential temperature (ºC) Depth (m) Validation of a numerical simulation of the North Atlantic Ocean with Argo data. 17 Table 4. Statistical parameters of SSH and MLD for the original data and columns marked with* are for data between 60ºN and -5ºN. Parameters SSH (m) SSH* (m) MLD (m) MLD* (m) Maximum 0,79 0,79 393,84 249,92 Minimum -0,94 -0,94 12,84 12,84 Range 1,74 1,74 381,00 237,08 Percentile 99 0,53 0,55 189,89 139,91 Percentile 1 -0,87 -0,89 12,84 12,84 Robust range 1,40 1,44 177,05 127,08 Percentile 75 (Upper quartile) 0,13 0,17 74,89 77,93 Percentile 25 (Lower quartile) -0,38 -0,24 27,09 28,89 Interquartile range 0,51 0,41 47,79 49,04 Median -0,12 0,03 45,82 59,89 Mean -0,12 -0,04 56,28 57,62 Variance 0,11 0,10 1530,65 1010,65 Standard deviation 0,33 0,32 39,12 31,79 Even though the presence of extreme values has been reduced, some histograms remain bimodal (see Figure 7). With the help of salinity maps (like the one in Figure 9), some regions could be identified, where there was still anomalous data. These corresponded to peripheral regions, specifically the Mediterranean Sea, the North and Baltic Seas and the Hudson Bay. Figure 9 shows one of the contour maps used to establish the peripheral regions, particularly in this case the Baltic Sea and Hudson Bay due to their low salinity. Figure 9. Contour map of salinity (psu) at layer 1 (3m) with the peripheral regions marked. Longitude (ºE) Latitude (ºN) -100 -80 -60 -40 -20 0 20 -10 0 10 20 30 40 50 60 5 10 15 20 25 30 35 Laura Gómez Navarro 18 Figure 10 shows a T-S diagram for the whole region (including peripheral regions). Some characteristic water masses stand out from the rest of the system. The Mediterranean water (12-20ºC S>35), the Baltic and North seas’ water (T<10, wide range of S), and the Hudson Bay water (T<3ºC, 15<S<33), all appear concentrated around localized regions of the TS plot. Moreover, the circles circled in purple correspond to the Chesapeake Bay (15<T<18ºC, 25<S<35) that also stands out from the rest of the system. Figure 10. T-S diagram of all the simulation data. The T-S diagram also allows to identify water from particular regions: East coast of Canada (from the Gulf of Saint Lawrence up to Ungava Bay), Mississippi-Alabama Shelf and Marshland Island, Gulf of Paria (Trinidad Island), Amazon River delta, Gulf of Guinea, particularly at the Bight of Bonny and Chesapeake Bay already identified as a region of anomalous salinity (Figure 11). Validation of a numerical simulation of the North Atlantic Ocean with Argo data. 19 Figure 11. Location of Sm values lower than 30 psu. 3.2. The 2012 North Atlantic numerical simulation After the first set of preliminary experiments was done, all the files corresponding to the year 2012 were processed. This corresponds to a total of 73 files in NetCDF format. The procedure carried out was the same as the previous one with the exception that the statistical parameters were calculated both in space (2D) and time (1D). For example, at each grid-point, the maximum value reached during the whole 2012 can be estimated, and a (2D) map of the maximum values can be constructed. Alternatively, the maximum value at each snapshot can be calculated over the region of interest, and a (1D) time series of regional maximum values can be created. Figure 12 displays the map showing the maximum 2012 values of the sea level and the mixed layer depth. Although the graph does not include information about the time of occurrence of the maximum value, it clearly is a tool to delineate the different dynamical provinces of the North Atlantic Ocean. It also helps to illustrate the main mechanisms at work in the region as the 1 m sea level near the western boundary due to the wind driven oceanic circulation (combined with thermal effects), as well as the relative size of the resulting meridional and zonal gradients at the region. On the other hand, the largest values of the mixed layer depth are found at the western Atlantic Ocean, near the path of the Gulf Stream. The lowest values are concentrated at the 100oW 80oW 60oW 40oW 20oW 0o 20oE 40oE 0o 15oN 30oN 45oN 60oN Laura Gómez Navarro 20 north-eastern coast of Canada and at the Gulf of Guinea, two regions strongly influenced by the presence of surface fresh water fluxes. Figure 12. Maximum values’ distribution in metres for SSH (left) and MLD (right). Figure 13. SSH’s statistical parameters variation with time. Longitude (ºE) Latitude (ºN) -100 -80 -60 -40 -20 0 20 -10 0 10 20 30 40 50 60 -0.8 -0.6 -0.4 -0.2 0 0.2 0.4 0.6 0.8 Longitude (ºE) Latitude (ºN) -100 -80 -60 -40 -20 0 20 -10 0 10 20 30 40 50 60 20 40 60 80 100 120 140 160 180 200 220 240 Validation of a numerical simulation of the North Atlantic Ocean with Argo data. 21 Figure 14. MLD’s statistical parameters variation with time. An example of time series of the statistical properties of the system is given in Figure 13, where the time evolution of the minimal, central, and largest sea level values is shown. The mean value remains near to zero as the model uses a volume conservation constrain in the parameterization of the open boundaries. The departures of zero are due to the non-accounted contribution of the peripheral seas. The time evolution of the robust maximum (the 99 percentile) reaches its largest value during the month of August, indicating the seasonal warming of the upper Ocean. Contrarily, the time evolution of the absolute maximum has two maximum values (at the beginning of the year and towards the end of the month of October). The more reasonable evolution of the robust maximum illustrates the need of removing extreme values before the statistical exploration of the data. Figure 14 helps to visualize the strong seasonal variation of the MLD. As it would be expected, the maximum values occur during winter due to the combined effect of low stratification and strong wind stirring, and the lowest values occur during the summer season due to the stabilizing effect of the surface seasonal warming. Figure 15 shows the corresponding time series for the salinity at layer 15 (centred on 228 m depth). The most striking feature is the anomalous behaviour of the minimum salinity towards the end of the year. Further analysis have shown that such Laura Gómez Navarro 22 sudden reduction of salinity happens in some isolated grid-points, and that are not linked to any physical process. The analysis was repeated to the year 2011, and it was found that some grid-points, located at different depths, had similar errors. All points affected by this behaviour are located at the bathymetry, indicating a problem in the parameterization of the tracers in the bottom mixed layer. Figure 15. Salinity 1D plot at layer 15 (228m) for year 2012. The spatial distribution of the minimum salinity during the year 2012 in the Gulf of Saint Lawrence are shown in Figures 16 (32 m depth) and 17 (191 m depth). The model values are compared with the WOA09 salinity data. The model surface salinity can be much lower than the climatological salinity (more than 10 psu fresher). At depth, the differences are of about 1 psu. Such a 10 psu surface salinity differences between the model and climatology indicates deficiencies of the model in obtaining realistic fresh-water balances in such region dominated by local water runoff and shallow bathymetry. Validation of a numerical simulation of the North Atlantic Ocean with Argo data. 23 Figure 16. Left: Minimum salinity values’ distribution for 2012 layer 5 (32m). Right: WOA09 Salinity values for December 2012 layer 4 (30m). Figure 17. Left: Minimum salinity values’ distribution for 2012 layer 14 (191m). The pink point is where the lowest value is located. Right: WOA09 Salinity values for December 2012 layer 10 (200m). 3.3. The 2012 North Atlantic Argo data The Argo data corresponding in time and space to the simulation was obtained from the Barcelona Expert Centre (BEC) database which is a mirror of the Coriolis database. Only the profiles which were found in the Atlantic Ocean, between 60ºN and 5ºS and of the year 2012, were retrieved from the database. Only Delayed mode data and profiles outside the peripheral regions have been considered in this part of the work. The hydrographic profiles where then processed following these steps: Longitude (ºE) Latitude (ºN) -65 -60 -55 -50 -45 -40 42 44 46 48 50 52 54 56 58 60 18 20 22 24 26 28 30 32 34 36 Longitude (ºE) Latitude (ºN) -65 -60 -55 -50 -45 -40 42 44 46 48 50 52 54 56 58 60 18 20 22 24 26 28 30 32 34 36 Longitude (ºE) Latitude (ºN) -65 -60 -55 -50 -45 -40 42 44 46 48 50 52 54 56 58 60 33 33.5 34 34.5 35 35.5 36 Longitude (ºE) Latitude (ºN) -65 -60 -55 -50 -45 -40 42 44 46 48 50 52 54 56 58 60 33 33.5 34 34.5 35 35.5 36 Laura Gómez Navarro 24 a. Adjusted pressure, temperature and salinity variables were read. b. If the number of measurement levels is less than 20, the whole profile is dismissed c. If Z_QC at a point was not equal to 1, the corresponding pressure, temperature and salinity values were set to 99999 (missing value). d. If T_QC of a point was not equal to 1, the corresponding temperature and salinity values were set to missing. e. If S_QC of a point was not equal to 1, the corresponding salinity value was set to missing. This is done as the salinity values depend on both temperature and pressure, while temperature value depends only on pressure. f. For each retained Argo profile, a profile is created in such a way that only valid salinity data are stored. g. The number of data in each profile for these new variables was counted. If there were not at least 20 values or if more than 40% of the data in the original profile was discarded, the whole profile was discarded. Also, for this data filter, all the profiles which their first valid salinity measure was at depth lower than 100m were discarded too. h. As the temperature measured by the buoys is in situ the potential temperature was calculated. See Appendix 1 to see the programme used for its calculation. i. The potential temperature values were vertically interpolated. j. The data was interpolated using three interpolation methods:  Akima splines  Cubic splines  Third-order polynomial fitting k. When the Argo data was interpolated to the simulation’s vertical levels, any level of the simulation deeper than the last depth of the profile and shallower than the first valid depth of the profile was changed to missing. l. After the interpolation the mean was calculated. m. If the absolute difference between each individual interpolation and the mean was greater than 5% of the mean, the corresponding potential temperature value was changed to missing. n. Steps g) to k) were repeated for salinity. Validation of a numerical simulation of the North Atlantic Ocean with Argo data. 25 The final results of potential temperature and salinity of the Argo data hereafter referred to as Ta and Sa respectively, were saved in a NetCDF file. In this file, information about each profile was saved too: its longitude, latitude, date and the ID of the profile (Did) in the BEC database. In addition, the additional parameters were calculated and included in this file. The indices ipos and jpos are the values of the x and y dimension respectively, of the point in the model’s grid which is closest to the Argo profile’s position. The tpos variable points to the model snapshot that corresponds to the Argo measurement. In addition, the number of valid Ta and Sa values for 3 different surface depths; 10 m, 25 m and 50 m, and in different points of the data processing (after having applied different filters). This was done to know how much data was really available and valid, as particularly in the surface, there are many Argo data problems, but on the other hand, it‘s the surface data which is of greater interest to validate, for its use for SMOS SSS. From a total of 4121 Argo profiles, only 3117 profiles have passed the Quality Control described above (Figure 18). Some regions, as the North Atlantic subtropical gyre, show a clear lack of Argo data, due to the free advective nature of the buoys. Figure 18. Argo profiles for 2012. The valid profiles are marked in blue and the invalid in red. The spatial distribution of the Ta (Figure 19) and Sa (Figure 20) is shown for layers 1 (3 m) and 2 (9 m). Notice the lack of Argo data for the first layer of the model. 100oW 80oW 60oW 40oW 20oW 0o 20oE 40oE 0o 15oN 30oN 45oN 60oN Laura Gómez Navarro 32 These 4 regions are the ones shown in Figure 2 (Chen, 2009). The dates of the 12 randomly selected profiles are shown in Table 5. Table 5. Dates of the 12 profiles selected for the T-S diagrams. Buoy number Day Month Year 1 12 10 2012 3 6 1 2012 2 11 10 2012 4 6 2 2012 5 1 2 2012 6 3 7 2012 7 23 11 2012 8 25 3 2012 9 4 4 2012 10 23 7 2012 11 20 2 2012 12 8 10 2012 Figure 27. Positions of the selected buoys for the T-S diagrams. The blue circles indicate the region. The T-S diagrams in Figure 28 show that, for region 4 (bottom), both T-S diagrams are similar. Similarly, in region 2 (second row) the model and Argo agree with the exception of the surface layers. On the other, for regions 1 (top) and 3 (third row) the T-S diagrams for some of the profiles are quite different, especially near the surface. Again, these T-S diagrams provide further evidence that the numerical simulation is less accurate at high latitudes, as the ones with the greatest differences are those of regions 1 and 3, which are the regions with buoys selected at higher latitudes. 100oW 80oW 60oW 40oW 20oW 0o 20oE 40oE 0o 15oN 30oN 45oN 60oN 1 2 3 4 5 6 7 8 9 10 11 12 3 4 2 1 Validation of a numerical simulation of the North Atlantic Ocean with Argo data. 33 Figure 28. T-S diagrams on the left correspond to the Argo data, and on the right to the simulation’s outputs. From bottom to top they correspond to points from region 4 to 1, and following the colour order blue, red, black, they correspond to the points 1 to 12, as labelled on Figure 27. The green, dashed lines are the isopycnals. 22 23 24 25 26 27 28 29 Salinity (psu) Temperature (C) 32 34 36 38 0 5 10 15 20 25 30 22 23 24 25 26 27 28 29 Salinity (psu) Temperature (C) 32 34 36 38 0 5 10 15 20 25 30 22 23 24 25 26 27 28 29 Salinity (psu) Temperature (C) 32 34 36 38 0 5 10 15 20 25 30 22 23 24 25 26 27 28 29 Salinity (psu) Temperature (C) 32 34 36 38 0 5 10 15 20 25 30 22 23 24 25 26 27 28 29 Salinity (psu) Temperature (C) 32 34 36 38 0 5 10 15 20 25 30 22 23 24 25 26 27 28 29 Salinity (psu) Temperature (C) 32 34 36 38 0 5 10 15 20 25 30 22 23 24 25 26 27 28 29 Salinity (psu) Temperature (C) 32 34 36 38 0 5 10 15 20 25 30 22 23 24 25 26 27 28 29 Salinity (psu) Temperature (C) 32 34 36 38 0 5 10 15 20 25 30 · . '\ '" ... 1// Laura Gómez Navarro 34 4. Conclusions The analysis of the outputs of the numerical simulation has indicated that north and south regions, close to the open boundaries of the model, together with the peripheral regions are identified as problematic. As described in Section 2.1, the open boundaries of the simulation are relaxed towards climatology data. Therefore, its variability is not reflected accurately. In addition, at high latitudes, an important process is not represented at all by the simulation: the interannual variation of the cold, freshwater input from the Arctic Ocean. In the peripheral regions, the characteristics of the water masses are very different than the characteristics of the pelagic Ocean that is the main objective of the simulation. The characteristics of the peripheral regions arise from the complex interaction between large river runoff, semi-closed domain, shallow bathymetry and atmospheric forcing. Errors in any of these factors (for example, the river runoff) may strongly affect the realism of the simulation. As there are very few, or none, Argo profile sampling these regions, they are eliminated from our analysis. The eastern Canadian coast could have also been considered a peripheral region as it clearly follows a different data distribution, than the rest of the system. The simulation at the Gulf of Saint Lawrence has been found excessively fresh by respect to the climatology data and by respect to published data from the region (Petrie et al., 2006). The minimum salinity observed in the Atlas at a depth of 30 m during the winter is of 29 psu (Figure 29), whilst the corresponding value of the simulation is 18 psu. This is a very important difference in salinity values, which could not be detected with the Argo data as this is one of the regions which are poorly represented by the Argo network (Figure 17). Validation of a numerical simulation of the North Atlantic Ocean with Argo data. 35 Figure 29. Salinity values at the Gulf of Saint Lawrence, for the 15th of November at 30m (Petrie et al., 2010). The region of the Gulf of Saint Lawrence is a particular case, as freshwater seems to accumulate there. This could be due to the combined effect of the river runoff, the presence of ice during certain times of the year and due to bathymetry errors. As stated in Buongiorno (2012), temperature and salinity values, especially the surface ones, are sensible to external forces, like freshwater inputs from rivers or precipitation, and to the Ocean’s internal dynamics, like convection, advection and mixing processes. Consequently, if these inflows of freshwater and internal processes are not accurately represented by the model, its effect can be clearly observed in the surface layers. As is the case of figure 10 where low values of salinity can be detected at the mouth of great rivers, as the model attributes a salinity of 0 psu to them, which is quite an underestimate. The anomalous results obtained from the simulation, as already discussed above, could have various reasons and explanations. In Thorpe et al., 2005 it is suggested that possible sources of error are low vertical resolution, poor representation of the bathymetry, and issues with the vertical mixing schemes (as described by Buongiorno, 2012), which lead to an unrealistic oceanic circulation, which can cause certain water masses to not be detected in simulations. The importance of the bathymetry used in simulations, is also highlighted in Minvielle et al. (2011) and Girton and Sanford (2003). They state that its importance is due to that it affects bottom layer processes which control the deep oceanic circulation. Laura Gómez Navarro 36 The T-S diagrams (Figure 28) and the difference between the model and Argo data (Figures 22, 23 and 24), show how the temperature of the model has a positive bias and the salinity has a negative one. The negative salinity tendency is much more marked than the positive one seen for the temperature values. This means that the model is too ‘fresh’, and could be due to all the reasons exposed above. They also show that the simulation is more inaccurate near the surface, but also it should be kept in mind that here is where the oceanic parameters show a greater variability. The shallowest depth at which the simulation could be validated was 9 m, due to the lack of valid Argo data near the surface, and also there were regions like the North Atlantic Subtropical gyre and the Gulf of St. Lawrence between others were there was a lack of data too. In the latter region, another source of data had to be used to validate the outputs of the simulation there. This lack of valid Argo data at the surface was expected, as in many occasions when the buoys are near the surface the hydraulic pump is stopped and so the salinity sensor stops measuring. More Argo data at the surface data could have been obtained if not only the profiles flagged with a QC = 1 have been considered, but also the ones flagged with a QC of 2, 5 or 8 (Table 2). Due to the limited amount of time for this study, only the data with QC=1 were selected. However, it has been noticed that the Argo observing system is able to represent seasonal variability of the Ocean (Figure 21). There is a considerable amount of valid profiles (3117) for year 2012, with the exception of some points and buoys the data they measure is very reasonable (Figure 20) and there are no temporal biases as they provide data throughout the year. This demonstrates that choosing Argo to validate the simulation was a correct choice. To conclude, in this work I have contributed to the first efforts of the Department of Physical and Technological Oceanography of the Institut de Ciències del Mar (CSIC) to the validation of their North Atlantic Ocean simulation. This simulation couples two of the state-of-art oceanic models: NEMO (oceanic dynamics) and LIM-2 (sea-ice dynamics). By comparing the outputs of the model against the Argo hydrographic profiles, I have found that the pelagic circulation of the model has some reasonable similarities with the observations. Although important deviations between the model and data exist near the meridional borders and off the Canadian Coast, our results indicate that the pelagic circulation of the model has some reasonable similarities Validation of a numerical simulation of the North Atlantic Ocean with Argo data. 37 with observations. This is important because the ability of retrieving SSS from remote sensing is extremely low in cold waters and near the coast (due to land sea abrupt emissivity transition). That is, the model provides the largest compatibility with data right in the region where it would be more useful to provide a coherent map of salinity in the North Atlantic Ocean. 5. References Antonov, J. I., Seidov, D., Boyer, T. P., Locarnini, R. A., Mishonov, A. V., Garcia, H. E., Baranova, O. K., Zweng, M. M. and Johnson, D. R. (2010): World Ocean Atlas 2009 Volume 2: Salinity. S. Levitus Ed. NOAA Atlas NESDIS 69, U.S. Gov. Printing Office, Washington, D.C., 184 pp. Brodeau, L., Barnier, B., Treguier, A. M., Penduff, T. and Gulev, S. (2010): An ERA40-based atmospheric forcing for global ocean circulation models. Ocean Modelling 31, 88–104. Brown, E., Colling, A., Park, D., Phillips, J. and Rothery, D. (1988): Ocean Circulation. The Open University, 287 pp. Buongiorno Nardelli, B. (2012): A Novel Approach for the High-Resolution Interpolation of In Situ Sea Surface Salinity. J. Atmos. Oceanic Technol., 29, 867– 879. Carval, T., Keeley, B., Takatsuki, Y., Yoshida, T., Loch, S., Schmid, C., Goldsmith, R., Wong, A., McCreadie, R., Thresher, A. and Tran, A. (2012): ARGO USER’S MANUALVersion 2.4, 85 pp. Chatterjee, A., Shankar1, D., Shenoi, S. S. C., Reddy, G. V., Michael, G. S., Ravichandran, M., Gopalkrishna, V. V., Rama Rao, E. P., Udaya Bhaskar, T. V. S., and Sanjeevan, V. N. (2012): A new atlas of temperature and salinity for the North Indian Ocean. J. Earth Syst. Sci.121, No. 3, 559–593. Chen, C. (2009): Ocean Water Masses-T-S Diagrams and Upper Ocean Waters. General Physical Oceanography, MAR 555, School for Marine Sciences and Technology, Umass-Dartmouth, 18 pp. Emery, W. J., and Thomson, R. E. (2004): Data analysis methods in physical oceanography 2nd edition. Elsevier Ed., 654 pp. Girton, J.B. and Sanford, T. B. (2003): Descent and modification of the overflow plume in the Denmark Strait. J. Phys. Oceanogr., 33, 1351–1364. Locarnini, R. A., Mishonov, A. V., Antonov, J.I., Boyer, T. P., Garcia, H. E., Baranova, O. K., Zweng, M. M. and Johnson, D. R. (2010): World Ocean Atlas 2009, Volume 1: Temperature. S. Levitus, Ed.,NOAA Atlas NESDIS 68, U.S. Government Printing Office, Washington, D.C., 184 pp. Laura Gómez Navarro 38 Higgison, S., Thompson, K. R. and Liu, Y. (2009): Estimating ocean climatologies for short periods: A simple technique for removing the effect of eddies from temperature and salinity profiles. GEOPHYSICAL RESEARCH LETTERS, 36, 4 pp. Hoareau, N., Portabella, M., Garcia-Ladona, E., Turiel, A. and Ballabrera-Poy, J. Meridional variability of North-Atlantic Sea Surface Salinity: Wavenumber spectra derived from climatology, ocean model and satellite observations. Submitted to J. Geophys. Res. MetOffice, United Kingdom: http://www.metoffice.gov.uk/weather/ marine/observations/gathering_data/argo.html Minvielle, M., Cassou, C., Bourdallé-Badie, R., Terray, L. and Najac, J. (2011): A statistical–dynamical scheme for reconstructing ocean forcing in the Atlantic. Part II: methodology, validation and application to high-resolution ocean models. J. Climate Dynamics, 36, 401-417. National Weather Service, NOAA: www.nws.noaa.gov/climate/help/glossary.php Petrie, B., Drinkwater, K., Sandström, A., Pettipas, R., Gregory, D., Gilbert, D. and Sekhon, P. (1996): Temperature, salinity and sigma-t Atlas for the Gulf of St. Lawrence. Canadian Technical Report of Hydrography and Ocean Sciences 178, 263 pp. Thorpe, S. E., Stevens, D. P. and Heywood, K. J. (2005): Comparison of two timevariant forced eddy-permitting global ocean circulation models with hydrography of the Scotia Sea. Ocean modeling, 9, 105-132. University of California, San Diego: http://www.argo.ucsd.edu/ Validation of a numerical simulation of the North Atlantic Ocean with Argo data. 39 6. Appendices 6.1. Appendix 1 SUBROUTINE sw_ptmp(n,S,T,P,PR,TP) IMPLICIT NONE INTEGER, INTENT(in) :: n real(kind=8), dimension(n), intent(in) :: T,P real(kind=8), intent(in) :: PR real(kind=8), dimension(n), intent(in) :: S real(kind=8), dimension(n), intent(out) :: TP ! ... Local variables real(kind=8), dimension(n) :: ADGT,del_P,del_th,q,th,PP ! ... theta1 del_P = PR - P CALL sw_adtg(n,S,T,P,ADGT) del_th = del_P * ADGT th = T + 0.5D0 * del_th q = del_th ! ... theta2 PP = P+0.5D0*del_P CALL sw_adtg(n,S,th,PP,ADGT) del_th = del_P * ADGT th = th + (1.0D0 -1.0D0/sqrt(2.0D0))*(del_th -q) q = (2.0D0-sqrt(2.0D0)) * del_th + (-2.0D0+3.0D0/sqrt(2.0D0))* q ! ... theta3 PP = P+0.5D0*del_P CALL sw_adtg(n,S,th,PP,ADGT) del_th = del_P * ADGT th = th + (1.0D0 +1.0D0/sqrt(2.0D0))*(del_th -q) q = (2.0D0+sqrt(2.0D0)) * del_th + (-2.0D0-3.0D0/sqrt(2.0D0))* q ! ... theta4 PP = P+del_P CALL sw_adtg(n,S,th,PP,ADGT) del_th = del_P * ADGT TP = th + (del_th -2.0D0*q)/6.0D0 END SUBROUTINE sw_ptmp !%%%%%%%%%%%%%%%%%% ! Adiabatic temperature gradient SUBROUTINE sw_adtg(n,S,T,P,ADGT) IMPLICIT NONE integer, intent(in) :: n real(kind=8), dimension(n), intent(in) :: S,T,P real(kind=8), dimension(n), intent(out) :: ADGT ! ... Parameters: real(kind=8), parameter :: a0 = 3.5803D-5 real(kind=8), parameter :: a1 = 8.5258D-6 real(kind=8), parameter :: a2 = -6.836D-8 real(kind=8), parameter :: a3 = 6.6228D-10 real(kind=8), parameter :: b0 = 1.8932D-6 real(kind=8), parameter :: b1 = -4.2393D-8 real(kind=8), parameter :: c0 = 1.8741D-8 real(kind=8), parameter :: c1 = -6.7795D-10 real(kind=8), parameter :: c2 = 8.733D-12 real(kind=8), parameter :: c3 = -5.4481D-14 real(kind=8), parameter :: d0 = -1.1351D-10 real(kind=8), parameter :: d1 = 2.7759D-12 real(kind=8), parameter :: e0 = -4.6206D-13 real(kind=8), parameter :: e1 = 1.8676D-14 Laura Gómez Navarro 40 real(kind=8), parameter :: e2 = -2.1687D-16 ADGT = a0 + (a1 + (a2 + a3 * T)*T)*T + (b0 + b1 *T) * (S -35.0D0) + ((c0 + (c1 + (c2 + c3 *T) *T)*T) + (d0 + d1 *T)*(S-35.0D0))*P + ( e0 + (e1 + e2 *T)*T) *P *P END SUBROUTINE 6.2. Appendix 2 SUBROUTINE percentiles (n,x,np,p,xp) IMPLICIT NONE integer, intent(in) :: n,np real(kind=8), dimension(np), intent(in) :: p real(kind=8), dimension(n), intent(in) :: x real(kind=8), dimension(np), intent(out) :: xp integer i,kk real(kind=8) rr,dd integer indx(n) CALL indexx(n,x,indx) DO i=1,np rr = p(i)*(n-1.0D0)/100.0D0 + 1.0D0 kk = FLOOR(rr) dd = rr - kk IF (kk.EQ.0) THEN xp(i) = x(indx(1)) ELSE IF (kk.EQ.n) THEN xp(i) = x(indx(n)) ELSE xp(i) = x(indx(kk)) + dd*(x(indx(kk+1))-x(indx(kk))) ENDIF ENDDO RETURN END Validation of a numerical simulation of the North Atlantic Ocean with Argo data. 6.3. Appendix 3 SUBROUTINE avevar (data,n,ave,var) IMPLICIT NONE integer, intent(in) :: n real(kind=8), dimension(n), intent(in) :: data real(kind=8), intent(out) :: ave,var ! ... Local variables: integer j real(kind=8) s,ep ave = SUM(data(1:n))/n var = 0.0d0 ep = 0.0d0 do j=1,n s = data(j) - ave ep = ep + s var = var + s*s ENDDO var = (var-ep**2/n)/(n-1) RETURN END ! (C) Copr. 1986-92 Numerical Recipes Software *5sV1. ! ... ! ===================================== ! ... SUBROUTINE avevar4 (data,n,ave,var) IMPLICIT NONE integer, intent(in) :: n real(kind=4), dimension(n), intent(in) :: data real(kind=4), intent(out) :: ave,var ! ... Local variables: integer j real(kind=8) s,ep,v8 ave = SUM(data(1:n))/n v8 = 0.0d0 ep = 0.0d0 do j=1,n s = DBLE(data(j) - ave) ep = ep + s v8 = v8 + s*s ENDDO v8 = (v8-ep**2/n)/(n-1) var = SNGL(v8) RETURN END ! (C) Copr. 1986-92 Numerical Recipes Software *5sV1.