Characterization of a CZT-based spectrometer for underwater operation via simulations and experiments

The harsh environmental conditions in the marine environment pose various constraints on developing efficient instruments to carry out long-term, in situ radioactivity measurements. In addition, the strong attenuation of γ-rays in the water medium, makes remote sensing of such radiation a challenging task. In the present work, we report on the efforts to find the optimal characteristics and deployment scenarios of a new prototype γ-ray instrument based on a small-size CZT crystal enclosed in seal-tight housing to be deployed for operation in large depths. Lab experiments and detailed Monte Carlo simulations were combined to validate the actual crystal dimensions, determine its efficiency and energy resolution, as well as establish the minimum detectable activity values of the instrument in different configurations and scenarios.


Introduction
Radioactivity monitoring in the oceans can become very beneficial for both the human population and the environment, as its applications have a strongly interdisciplinary character [1,2].For example, naturally occurring radioisotopes in the seawater, such as the progenies of the long-lived decay chains of 238 U and 232 Th, can act as powerful tracers for studying geochemical processes in the aquatic environment [3].On the other hand, artificially produced radionuclides resulting from the operation of nuclear power plants (NPP), although released in a careful and controlled manner, may additionally have an environmental impact on the natural ecosystems.Furthermore, emergency situations, such as the incident in Fukushima Daiichi nuclear power plants, can lead to significant releases of radionuclides (e.g. 137Cs and 134 Cs) into the ocean [2,4].Special attention should be paid to thoroughly evaluate the effect of such releases on the local ecosystems.
Monitoring based on in-situ -ray spectroscopy in the marine environment is a quite challenging task, mainly due to the harsh environmental conditions and the increased attenuation from water, leading to significant limitations in the radiation detection process.Despite the above, it offers many advantages when compared to typical measurements of water or sediment samples in the lab, especially when continuous monitoring of large seabed areas is required in a direct manner [5].A lot of efforts throughout the years have proposed various solutions, including different types of instrumentation and approaches.The most common ones involve the deployment of NaI(Tl) scintillators (see e.g.[6][7][8][9][10][11]), with the main limitation of such approaches residing in the generally poor energy resolution of the scintillator.
Improving on energy resolution, different types of scintillators have been tested by various groups worldwide, mainly LaBr3(Ce) [12][13][14] and CeBr3 [15].Even though these detectors have generally superior resolution compared to NaI(Tl), intrinsic radioactivity emitted by their own crystal material or their bulky sizes can be an issue when deployed for applications in the marine environment.
-1 -Intrinsic radioactivity is due to the permanent presence of one or more photopeaks in the spectrum due to stoichiometrical presence of radioactive isotopes in the crystal.Past simulations of a 3"×3" LaBr3(Ce) crystal [16,17] show that as a result, the identification of a radioactivity hotspot in the ocean, becomes even harder, due to the low-counting rate usually recorded in such environments, shadowed by the intrinsic radioactivity.
Despite the high-energy resolution HPGe semiconducting crystals provide, their need for continuous cooling to reach low temperatures restricts the overall duration of unsupervised monitoring in the ocean, especially when deep depths are targeted.The challenge to deploy an HPGe detector for in situ studies near the seabed was eventually overcome by Povinec et al. [18] who coupled it with a NaI(Tl) scintillator.
An alternative to the aforementioned detector types are those operating on CdZnTe (CZT) crystals.The CZT crystals offer room temperature operation (wide band gap) and high efficiency due to their high atomic number and density [19].Furthermore, they have good energy resolution, exhibit no intrinsic radioactivity, and are offered at quite portable sizes with low power consumption.Thus, CZT-based spectrometers are ideal for integration with deep-sea underwater housings, to perform measurements in the aquatic environment.
Up till now, applications of these instruments included radiation mapping and monitoring [20-23] using aerial unmanned vehicles or robotic vehicles.In addition, they have also been tested and evaluated for imaging purposes in nuclear medicine applications [24][25][26], as well as in applications for astrophysics studies [27,28].Even though, CZT detectors have recently been tested for radioactivity measurements in the South China Sea [29], the current study is a main part of the first attempt to deploy them in European waters aboard both mobile and stationary platforms.The advantages discussed above regarding the detection efficiency, resolution and power consumption, will allow for long-term spectroscopy studies of the seabed and the water column.
The detector studied in this content is a Kromek GR1 spectrometer housing a 1 cm 3 crystal and nominal detection range of 30-3000 keV [30].The harsh environmental conditions and inaccessibility of the oceans, as well as the challenge to achieve optimal performance in a generally low-level natural radioactivity environment make it more than necessary to characterize the response of the detector for a variety of scenarios and setups.
Firstly, the geometry of the detector was evaluated with the main focus being on the CZT crystal characteristics.Results from this stage were reproduced combining experimental data and Monte Carlo simulations.Furthermore, the efficiency and energy resolution of the instrument were calculated followed by estimation of its angular response and Minimum Detectable Activity (MDA) values were achieved for certain setups and isotopes.Then, preliminary studies regarding the housing of the detector and the effect of the water volume on the overall detection response were carried out.Lastly, simulations with two different Monte Carlo packages focused on modelling the spectrometer, but also to provide the most suitable material for the spectrometer's sealtight enclosure via a more sophisticated approach.

Internal structure determination
Technical details of the CZT (external detector dimensions, nominal crystal volume etc) were provided by the manufacturer [30].However, due to the challenges posed by the targeted application the main -2 -intention was to take the study a step further and reach a higher level of detail, concerning physical and operating properties of the detector.These data can feed a series of Monte Carlo simulations to extract a higher level of detail about the detector and its properties.
At first, the detector was placed inside a 128-head medical CT scanner (X-ray Computed-aided Tomography) at 130 kV operating voltage to obtain cross sectional scans of the CZT internal structure.In figure 1 some representative CT images are illustrated.Vital information were retrieved for the overall configuration of the electronic components and supporting structures of the detector, but most importantly about the physical dimensions of the CZT crystal, which were used to define the simulation model.CZT crystals often exhibit the phenomenon of inactive layer [31].This refers to an heterogeneous response to the incident -rays from the volume of the crystal.A CT scan provides information about the attenuation introduced from the various parts of the detector, but is unable to unveil its operational characteristics.Hence, a gross estimation of the effective volume of the CZT crystal was proposed through the measurements below.Our goal was to irradiate only a thin slice of the crystal each time and calculate its response regarding the position of this slice.These measurements will be referred to as narrow beam measurements.
To do so, a point isotropic source of 22 Na was collimated using standard lead bricks thus creating a beam of 2 mm in width.At first, the detector was positioned with its main axis parallel to the beam direction.Measurements of 400 s each were performed, by moving the detector with a step of 1 mm towards the corresponding orientation.Additionally, measurements at the edges of the detector were performed (1 st setup).This exact procedure was repeated by rotating the detector by 90 • around its longitudinal axis (2 nd setup).Then, the detector was positioned with its main axis perpendicular to the beam direction (3 rd setup).The total integral of the spectrum for each step-measurement was calculated and correlated with the part of the detector irradiated.In figure 2 the setups used are depicted, with the yellow dashed arrows indicating the direction along which the detector was moved between measurements.Results from this part are presented in section 3.
-3 - Regarding the computational part of this study, two widely used Monte Carlo simulation toolkits were exploited, MCNP5 [32] (v.1.40)and Geant4 (v.4.10.06.) [33][34][35].Both are used for the simulation of the passage of particles through matter employing Monte Carlo methods.The details regarding the simulations will be given in section 4.

Characterization of the spectrometer
After getting a better insight on the geometry of the spectrometer and the characteristics of the crystal, basic features of the spectrometer, including its efficiency and energy resolution, were studied for a broad energy range covering a variety of radionuclides.Both were achieved by using standard calibrated point sources of radionuclides ( 60 Co, 152 Eu, 137 Cs and 22 Na).Spectra for each source were acquired with a measurement duration of 1200 s, while keeping the distance between the source and the front window of the detector constant at 10 cm.The analysis of the recorded spectra was performed by using the SPECTRW analysis software [36] for the efficiency calculations and the Fityk analysis software [37] for the FWHM values, as the latter provided more flexibility in the fitting functions.Regarding the efficiency of the detector, the full-energy peak efficiency (FEPE) was expressed as a function of energy, using the following equation [38]: where  denotes the full-energy peak count rate,  is the source activity in Bq and finally   corresponds to the probability (relative intensity) of emission of the gamma-ray studied.The efficiency curve created will serve as a baseline for the detector and will enable monitoring of possible performance deterioration after long-term use.On the other hand, resolution estimation apart from the obvious benchmarking, will also allow for a more accurate simulation description.Optimal placement of the detector is crucial to ensure maximum response during a measurement, since in most cases the incident -rays reaching the detector are expected to be anisotropic.For this reason, the angular response of the detector was investigated using a 137 Cs calibrated point source, -4 -which was aligned with the center of the detector's front window at a distance of 10 cm (see figure 3).Measurements with a varying angle of irradiation on the horizontal plane were carried out.The angles examined were 0 • , ±30 • , ±45 • , ± 60 • , ±90 • and ± 135 • , respectively.Each measurement lasted 300 s and the absolute efficiency of the detector was calculated accordingly.The absolute efficiency is defined as the ratio of the total counts registered by the sensor over the total number of -rays emitted by the source [38].Experimental results were also cross-checked with simulations using the MCNP5 toolkit.
Next goal was to check for consistency regarding the angular response of the detector for both planes perpendicular to the detector's face.To achieve that, the detector was rotated by 90 • around its longitudinal axis, and measurements were repeated for three representative angles.These measurements allowed for evaluating the attenuation introduced due to the external shielding of the detector, as well as potential existing asymmetries of the crystal, and consequently identify the most effective angles of irradiation.In general, radioactivity studies in the aquatic environment allow only for low count rate measurements.The natural background level plays a major role on the ability to identify and detect a certain radionuclide.To that extend, data to extract MDA were acquired for two radionuclides of interest.A natural radioisotope ( 40 K) found in high concentrations in seawater and an artificial one ( 137 Cs) related with nuclear accidents and nuclear waste disposals in the oceans.MDA values were derived using the following equation [38,39]: where  is the background area estimation under the studied photopeak region,  is the number of channels of the photopeak,  is the number of channels on the left and the right region from the -5 -photopeak used for background estimation, and  is the live time of the measurement.Finally,  is a coefficient determining the confidence level to accompany the MDA estimation.At first, the MDA was calculated in our dry setup, using a calibrated point source of 137 Cs.Calculations were carried out for different live times and distances between the source and the detector.In addition, the variation of the MDA value was studied when different confidence levels (through the  coefficient) were chosen.Then, the MDA was calculated for 40 K from a measurement acquired within a wet setup.The detector was encapsulated inside an improvised plastic housing and was submerged into water with a volume of 38.79 L (see figure 4).A potassium-rich fertilizer containing quantities of 40 K was diluted in the water, so that a homogeneous concentration could be accomplished.Prior to this measurement, a sample of the same fertilizer was weighed on a precision scale and then measured using a fully-shielded HPGe spectrometer.An IAEA reference material [40] of known 40 K specific activity was used to deduce by comparison the activity of 40 K contained in the fertilizer, which was found equal to 10.27±0.24kBq/kg.The total activity of 40 K eventually diluted in the water was -6 -0.33±0.01kBq/L.The uncertainty of the activity concentration was calculated via error propagation including contribution from the uncertainty in the fertilizer mass, the specific activity of the reference sample and the net area of the photopeak studied both in the fertilizer and the reference sample.

Preparation steps for the marinization of the spectrometer
Since the ultimate goal is to employ this type of detectors for measuring radioactivity in the aquatic environment, the detector needs to be encapsulated within an appropriate housing designed to operate under high pressure conditions (1 bar/10 m of depth).Consequently, different housing materials and their thicknesses had to be evaluated in terms of attenuation introduced as well as their possible effect on the spectrum obtained.Before running the necessary simulations, a simplified geometry in the lab was implemented to extract the first results (see figure 5).The detector and a point source ( 137 Cs) were set at a fixed distance (30 cm) and an attenuator (located 20 cm from the source) was interfered between them.The materials studied were Plexiglas™, copper, aluminum and lead.Then, the absolute efficiency of the detector for each measurement was estimated and plotted as a function of each material's thickness.Finally, the water-induced attenuation at the spectrum obtained from the detector was quantified.Measurements of 600 s each were conducted for different levels of water (0, 5, 10, 15, 20 and 24 cm) between the detector and the source.This was achieved with a geometry where the detector was placed vertically over a tap water column of variable height, with the 137 Cs source placed beneath the column (see figure 6).Following our results, which are presented in section 3.3, Plexiglas™is a suitable housing material.For that reason, complementary measurements to study the extra attenuation introduced were performed with a Plexiglas dome on top of the tap water column.

Internal structure determination
As mentioned in section 2.1, specific measurements were carried out to estimate the active volume of the CZT crystal.Three different configurations were used aiming to investigate the crystal's response when placed along three different orientations.Figure 7 represents the variation of the total count integral as a function of the relative position of the irradiated slice.
-7 - More specifically, in figure 7(a) the graph corresponds to a parallel positioning of the detector with respect to the beam direction.The red scatter is the result of a 90  rotation of the detector around its long axis.The x axis of this graph is the distance from the beam path up to the center of the detector's front window.A rather symmetrical behavior is observed around the center of the crystal (indicated with the blue dashed line).On the other hand, figure 7(b) is the result of placing the detector perpendicularly to the beam direction.The x axis in this case, is the distance from the beam path till the level of the detector's front window.The initial information extracted from this graph is twofold.First, the distance between the front shielding of the detector and the crystal itself is estimated to be 5 mm and secondly, the highest response is observed in the distance range 9-16 mm.

JINST 19 P05008
Nevertheless, a more detailed evaluation of the above experimental results requires fitting with an appropriate function.Ideally, if the narrow beam width was orders of magnitude smaller than the detector size, this would be achieved by the product of two Heaviside step functions: Eq. (3.1) corresponds to 0 for values below  1 or above  2 , and 1 for values between  1 and  2 , thus, is a suitable function to describe the response of the detector.However, since the width of the beam is comparable to the width of the detector, a more gradual response is expected and a function with a more smooth (continuous 1 st derivative) increase is required.This is achieved by replacing the  () function with the (1 +  −2  ) −1 approximation, where k defines the slope from minimum to maximum (and vice versa) and it depends on the relative sizes of the beam collimator and active volume examined.It should be mentioned that this method applies only when the beam width is smaller than the dimension under investigation.By doing so, eq.(3.1) is transformed to: Here,  1 and  2 correspond to the physical limits of the detector's active volume, and can provide its corresponding center () and active dimension () for each scan, by applying the following transformations: By substituting eqs.(3.3) to eq. (3.2), and adding additional variables for the amplitude () of the response and a constant background (), we get: Eq. (3.4) was used to evaluate the results obtained from the narrow beam experiments.What can be deduced from the results presented in the next table, is a clear asymmetry between the center of the crystal's effective volume and physical volume.Additionally, differences are observed regarding the dimensions of the crystal.

Characterization of the spectrometer
The efficiency of the detector and more specifically the full-energy peak efficiency (FEPE) was calculated for the air and at a distance of 10 cm from the detector's front window.Experimental data were fitted using the equation of Debertin [41]: To present the variation of the system's energy resolution against the -ray energy, a Gaussian distribution was fitted for each photopeak studied and the FWHM was calculated in keV.Experimental data were fitted using the following equation [42]: The detector's energy resolution was found to be 2.4% for the energy of 661.66 keV, which corresponds to the photopeak of 137 Cs.This value agrees with the nominal value, < 2.5%, quoted by the manufacturer in the original technical data sheet of the detector.An issue exhibited by the CZT crystals is the phenomenon of incomplete charge collection which is expressed in the acquired spectrum as a low-energy tail at each photopeak [43].Additionally, a high-energy tail was also observed.These tailing effects have been approached by the combination of computational tools [44], however in the scope of this study, a simple method of elimination was proposed.A new curve of FWHM as a function of energy was created.For each photopeak studied, a double Gaussian composition was fitted.As can be inferred, the first Gaussian was quite narrow, while the second one much broader so that tailing could be approached (see figure 9).By using only the FWHM value of the first peak, the plotted curve was free from these tailing effects.
In section 2.2 a set of measurements was described for determining the angular effect on the detector's efficiency.Several setups were used for varying angle of irradiation.The results are shown in figure 10(a).The minimum value for the detection efficiency corresponds to the setup where the source is aligned with the long axis of the detector.With increasing the angle of irradiation, the efficiency seems to rise in a symmetrical way for both positive and negative values.This behavior can be explained from the position of the CZT crystal against the external shielding.More specifically, the greater the irradiation angle, the smaller the distance between the source and the crystal.Hence, the rise in the efficiency of detection.Comparison of experimental and simulated results showed a good agreement.Differences observed for the irradiation angles of ± 135 • can be most likely attributed to the simplified simulation of the electronic circuits lying behind the CZT crystal.Additionally, -10 - complementary measurements were conducted by rotating the detector by 90 • about its long axis.From figure 10(b) a good agreement between the efficiency values, before and after rotation, can be observed.Finally, MDA was calculated for two radionuclides of particular interest.The procedure followed and measurement setup are described in the section 2.2.Initially, calculations of the MDA in air for 137 Cs were performed for various distances and live time values (see figure 11).
By using an appropriate equation ( =   ) to fit the experimental data, MDA was found to depend on the inverse square root of the live time value.This finding is of statistical nature and relies on the Poisson distribution describing the process of radioactive decay.On the other hand, a linear relationship was observed between the MDA value and the distance between the source and the detector.This can be justified based on two cases: the square root of the background is inversely proportional to the distance source-detector and FEPE was not being subject to correction for the geometry factor.Next a measurement with a total duration of 8•10 4 s was performed in an aquatic environment and the MDA for 40 K was estimated.A value of 56.1 Bq/L was found for a confidence level of 95% and 43.3 Bq/L for a confidence level of 90%.According to [3], the concentration of 40 K in seawater is 12.2 Bq/L, thus a measurement of longer duration is required or a different confidence level should be preferred.

Initial preparation steps for the spectrometer's marinization
Another important aspect that had to be studied is the material of the detector's housing thus to be able to submerge into the aquatic environment.At first, a simplified geometry (see section 2.3) was prepared so that various materials could be studied in terms of their effect on the detector's efficiency.
Our goal was to run a preliminary test for candidate materials for the detector housing.The results of this test are presented in figure 12, where for each material different thicknesses were examined.What can be inferred from the above plot is that materials with lower density provide the highest detection efficiency.Moreover, as expected, the larger the material thickness, the lower the efficiency -12 -value.What seems peculiar at first is the efficiency value for lead being higher than the one from copper, even though lead has a higher density.That can be explained from the build-up effect caused by lead which results in the increase of counts in the detector.From the above plot, it can be concluded that one of the best candidates that can support the needs of the targeted application is aluminum.
A first insight on the attenuation introduced by water and the effect it has on the acquired CZT spectrum was made possible through a simple test.Different levels of tap water were inserted between a radioactive source and the detector (see figure 6).For the same levels of water, measurements were repeated by introducing a hemispherical dome made of Plexiglas™. Figure 13 represents a comparison of spectra obtained for both parts of this test.In figure 13(a) it is more than profound the decrease in counts of the photopeak due to absorption and scattering introduced from water.Additionally, the overlapping X-rays [45] from the deexcitation (K1, K2 and K'1) of the atom of 137 Ba is completely cut-off in the presence of water.On the other hand, from figure 13(b), the dome does seem to have little impact on the spectrum.More specifically, the ratio of the total area from the spectrum with the dome to the spectrum without the dome was estimated to be 0.96 ± 0.03.For comparison reasons the equivalent ratio from the first spectra (figure 13(a)) was 0.55 ± 0.02.

Monte Carlo simulations
Simulations are divided into three stages.Initially, both MCNP5 and Geant4 were utilized to model the detector geometry and match its response between computational and experimental method.Cross sections for MCNP5 were derived by the ENDF/B-VI library, while for Geant4 the Low-Energy Electromagnetic Package was selected.All types of relevant interactions of photons and electrons/positrons with matter were considered utilising low-energy data packages and more specifically, for the simulations of this work, the G4EmLivermorePhysics package was implemented.During the second stage, with the aid of the MCNP5 toolkit, the angular response of the sensor was investigated (10 7 -13 -events for each run), using a different setup compared to the experiments described in section 2.2.Finally, various simulations with the Geant4 package only, were examined to provide information about possible housings that can be used in the field.All the simulations were performed for 10 8 events.
A detailed modeling of the detector geometry, firstly, was carried out using information from the CT scan images and the data from the manufacturer (see figure 14).The CZT crystal along with its shielding was modeled in terms of dimensions and material properties.The crystal density value was set to 5.78 g/cm 3 while its composition was Cd 0.9 Zn 0.1 Te [19].The electronic components and circuits were approximated from the CT scan.The dimensions of the crystal were chosen to be 10 × 10 × 10 mm 3 after confirmation based on analysis of the CT scan images.Next, the response of the sensor had to be evaluated and optimized.The main challenge was to incorporate the phenomena of inactive layer and incomplete charge collection exhibited by the CZT crystal in the simulations.The former was addressed by introducing an effective volume of detection within the physical volume of the crystal, the dimensions of which were derived by the narrow beam measurements (see table 2).However, successive approaches via simulations showed that the optimal effective volume was 9.5 × 9.5 × 9.0 mm 3 , well within uncertainty when compared to the experimental response of the detector.
Regarding the treatment for the incomplete charge collection, simulations were fed with the energy resolution curve (FWHM vs energy), created after the elimination method using a double Gaussian approach described in section 3.2.From figure 15 a good agreement was observed at the photopeak region of 137 Cs between the two Monte Carlo codes used.The difference observed in the peak height between experiment and calculations is justified by the presence of the tailing effect leading to a broader and hence shorter experimental peak.However, when the total peak integral was considered, the ratio of experimental to computational area was found to be 98.2%.This value is considered satisfactory for the purposes of this study.The discrepancy observed at the backscatter peak occurs due to simplified simulation of the electronic components of the sensor.
Next, the angular response of the sensor was investigated via simulations with the MCNP5 toolkit.However, during this simulation and in contradiction with the equivalent experiments conducted, a different setup was examined.More specifically, various angles of irradiation were studied while -14 - the distance from the source to the center of the CZT crystal was kept fixed (10 cm).The point of these simulations was to remove the effect of the square of the distance, and isolate the effects of the material thickness and detection solid angle.Results are presented in figure 16. -15 -As we can see, there are two families of results: first we have simulations at ±90 • and 0 • .For these angles of irradiation, the CZT crystal has a 100 mm 2 irradiation surface at 9.5 cm.The only difference for these three simulations regards the 0 • slightly elevated measurement, which corresponds nicely to the thinner, compared to the sides, front window.The second family of results are the angles of irradiation on the diagonals of the cubic crystal (±45 • ), and ±15 • .For these angles, the irradiation window is larger, peaking at √ 2 × 100 mm 2 , at the diagonals.This is clearly depicted in our results, where absolute efficiency values are in their majority larger than the first family of results.Moreover, for the subgroups of ±(30 • , 45 • , 60 • ), a decrease is observed in absolute efficiency as absolute degree increases.The main driver behind this decrease is that as degrees increase, less -rays pass through the thin window, and more through the thicker enclosure.
Due to the harsh conditions prevailing at the marine environment, and more specifically near the seabed, some restrictions arise regarding the geometries, the materials, as well as the thicknesses of the materials that can be used for the housing of the detector.For that reason, simulations were carried out by changing the aforementioned parameters to find the most appropriate combination.The first housing geometry that was simulated was like the one shown in figure 19(a).In this model, the CZT detector along with its housing were placed inside a cubic water tank of 27 m 3 filled with pure water (H 2 O).The simulations were carried out with a point-like 152 Eu source placed at 10 cm distance from the housing entrance window by varying, firstly, the material of the housing.The materials studied were Plexiglas™, 6082-T6 hard anodized aluminum type III and 6Al-4v grade 5 titanium with 4 mm thickness each.The latter are commonly used as marinization materials in oceanographic missions.-16 -From figure 17, it can be observed that the difference of the simulated counts between the Plexiglas™and the 6082-T6 anodized aluminum type III material can be considered negligible (∼2.8%) but this is not the case for the 6Al-4v grade 5 Titanium where the difference is around 6 times higher (∼17%).Thus, the most appropriate materials for our needs in order to eliminate the signal losses from the surrounding materials, is 6082-T6 anodized aluminum type III or Plexiglas™instead of 6Al-4v grade 5 Titanium.Between the two, the type of aluminum that was simulated is less prone to corrosion which is the most common damage of the materials inside the aquatic environment and for that reason it is more preferred, when compared to Plexiglas™.Moreover, according to the manufacturer [46], the aluminum housings have thinner walls and are rated for operation at larger depths.After the selection of the material for the detector housing, the next studied aspect was the geometry of the housing.Figure 19 depicts three simulated geometries.The first one is a cylinder with a diameter of 4 cm and length of 10 cm with a dome in front of the detector entrance window.The second one is a cylinder with a diameter of 12 cm and length of 10 cm with flat ends.In this geometry, the detector's entrance window is facing the curved part of the cylinder.The third configuration is a 64 cm 3 cubic box.From the results of the simulation depicted in figure 18(a), minor discrepancies between the three different geometries can be noticed.The higher counting rates were detected for the cylindrical geometry with the dome at the entrance window and as a result this will be preferred compared to the other two.The last aspect that had to be investigated was the thickness of the selected material and geometry.Figure 18(b) shows the simulated results for three different thicknesses by selecting the 6082-T6 anodized aluminum type III as housing material and the cylinder with the dome at the entrance window as housing's geometry.The difference of the simulated spectra for the four different thicknesses can be considered negligible (less than 3%) so any of these can be employed in our case.

Conclusions
The characterization of a novel CZT based spectrometer intended for in-situ radioactivity monitoring at large depths in the marine environment was performed, as one of the main objectives of the EU H2020 RAMONES Research Programme [47].The results obtained from experimental studies of the crystal dimensions (CT scans and tomography) showed that a small fraction of the crystal mass does not participate in the detection of -rays.These findings were introduced into the detailed Monte Carlo simulations used to reproduce a more accurate model of the detector.
Experiments in the lab, using standard calibrated sources, provided information about the efficiency, energy resolution and angular response of the detector.FEPE and FWHM curves were plotted to provide a baseline for the performance monitoring of the spectrometer after long-term use.Regarding the angular response of the instrument, figure 16 indicates that the highest response is observed within a opening of 30 • -60 • angles for the XZ and YZ planes.Consequently, the optimal placement of the instrument on a benthic laboratory is by keeping its longest axis perpendicular to the seabed, in order to take full advantage of the enhanced response in combination with the thin window.Achieving the highest efficiency possible is crucial, not only for the validity of the measurement, but also on the total duration of the deployment of the instrument near the seabed.Finally, MDA calculations were performed for two radionuclides ( 137 Cs in the air and 40 K in an underwater measurement).A point source of 137 Cs was placed at various distances from the detector and measurements of variable duration were carried out.On the other hand, a uniform volume source of 40 K was created, after diluting a certain amount of fertilizer in tap water.
To begin the preparation for marinizing the instrument, a set of measurements was conducted to test materials as candidates for the spectrometer's housing.By considering only the attenuation introduced by the material, aluminum was found to be the best option.Moreover, the effect of the water volume on the spectrum acquired was studied by placing different levels of water column between the source and the detector.
Finally, Monte Carlo simulations were carried out with two simulation packages (MCNP5 and Geant4).The detector was modelled by taking into account data retrieved from the CT scan and the results from the narrow-beam measurements.Throughout successive simulations the effective volume of the crystal was found to be 2.1% larger than the one suggested by the narrow-beam measurements.This can be justified by uncertainties introduced from the experimental setup used and especially, on the placement of the detector during each measurement.Simulated results were validated against experiments made in the lab.
A more detailed study was performed with Geant4 focusing on the housing's material, aiming to complement the experimental results.It was concluded that aluminum is a good choice as it combines low levels of attenuation, on top of its known anti-corrosion properties.To that extend, not only the thickness of the housing was investigated, but also the most efficient shape for the underwater housing.

Figure 1 .
Figure 1.(a) Full CT topogram of the GR1 detector and (b) a CT slice with the CZT crystal and the external shielding.

Figure 2 .
Figure 2. (a) Detector positioned parallel (1 st and 2 nd setup) and (b) perpendicularly to the beam path (3 rd setup).Dashed arrows describe the movement of the CZT detector with respect to the source placed behind the collimation channel. axis is the symmetry axis of the longer dimension of the CZT detector.

Figure 3 .
Figure 3. Representation of the experimental setup to investigate the angular response of the detector (detector not in scale).

Figure 4 .
Figure 4. Representation of the measurement setup for acquiring an MDA value for 40 K in water.The underwater housing is illustrated with the red frame while the water is the light blue shaded area.

Figure 5 .
Figure 5. Representation (top view) of the measurement setup used to investigate candidate materials for the detector's housing (detector not in scale).

Figure 6 .Figure 7 .
Figure 6.Representation of the measurement setup used for studying the attenuation introduced from various water columns.

Figure 8 .
Figure 8.(a) Full-energy peak efficiency and (b) energy resolution (FWHM vs energy) curves acquired from several point-like sources ( 152 Eu, 137 Cs, 60 Co, 22 Na) placed at 10 cm distance from the detector end cap.Non-visible error bars are smaller than the symbols.Red lines indicate fitted equation.See text for more details.

Figure 9 .Figure 10 .
Figure 9.An example of the double Gaussian method used, as implemented in the Fityk analysis software[37].

Figure 11 .
Figure 11.MDA values for 137 Cs measured in air for varying (a) live time and (b) source to detector distance.

Figure 12 .
Figure 12.Absolute efficiency of detection as a function of thickness for various materials-attenuators.

Figure 13 .
Figure 13.Spectra acquired for a 137 Cs source (a) with and without any water between the source and the detector (b) as well as a comparison of the spectra including the same level of water with and without the hemispherical dome (total duration for each measurement was 600 s).

Figure 14 .
Figure 14.Simulated geometry of the detector using MCNP5.

Figure 15 .
Figure 15.Comparison between experimental and computational spectra for a 137 Cs source.

Figure 16 .
Figure 16.Angular dependency on the absolute efficiency of the sensor, studied via Monte Carlo simulations, under fixed source to crystal's center distance.

Figure 17 .
Figure 17.Simulated spectra in logarithmic scale for a point-like 152 Eu source placed at 10 cm distance from the housing's entrance window for three different materials with 4 mm thickness.

Figure 18 .
Figure 18.Simulated spectra in logarithmic scale for a point-like 152 Eu source placed at 10 cm distance from the housing's entrance window for (a) three different geometries of the same material and thickness and (b) same geometry and four different thicknesses.

Figure 19 .
Figure 19.Simulated setups for different geometries, materials and thicknesses for the detector's housing inside the marine environment.(a)-(b) Cylindrical geometry with a dome, (c)-(d) Cylindrical geometry with flat ends and (e)-(f) Cubic geometry.The red dot indicates a 152 Eu point source 10 cm away from the housing's window.

Table 2 .
Fitting parameters for narrow beam measurements.