Mass composition working group report

We present a summary of the measurements of mass sensitive parameters at the highest cosmic ray energies done by several experiments. The Xmax distribution as a function of energy has been measured with fluorescence telescopes by the HiRes, TA and Auger experiments and with Cherenkov light detectors by Yakutsk. The 〈Xmax〉 or the average mass (〈lnA〉) has been also inferred using ground detectors, such as muon and water Cherenkov detectors. We discuss the different data analyses elaborated by each collaboration in order to extract the relevant information. Special attention is given to the different approaches used in the analysis of the data measured by fluorescence detectors in order to take into account detector biases. We present a careful analysis of the stability and performance of each analysis. The results of the different experiments will be compared and the discrepancies or agreements will be quantified.


INTRODUCTION
In preparation for this meeting, several working groups were formed to establish a common view on the experimental status of measurements at ultra-high energies. Here we report the findings of the mass composition working group consisting of members of the Auger, HiRes, Telescope Array (TA) and Yakutsk collaborations. The aim was to understand and quantify potential differences between different experimental results and to try to discuss the measurements in terms of the cosmic ray mass composition.
A current issue in the field of ultra high energy cosmic rays is that the measurements [1] of the depth of shower maximum from Auger are shallower and less fluctuating than predictions from air shower simulations for a pure proton composition at energies 10 19 eV. On the other hand, the HiRes and TA results [3,4] in the same energy regions are consistent with QGSJet-II simulations assuming a constant composition dominated by light elements.

EPJ Web of Conferences
However, there are important differences in the analysis of the data between Auger, HiRes and TA. In this note we will review the differences between the analyses, compare the different experiment results, and evaluate their compatibility within the experimental statistical and systematic uncertainties.
The main focus of this work is the understanding of the apparent differences between the Auger, HiRes and TA X max results. Nevertheless, we also discuss in this paper other shower observables which are sensitive to the mass of the primary cosmic rays. The Yakutsk experiment has muon detectors operating and this information is used to infer lnA in a complementary way [5]. In a similar way, the Auger collaboration operates an array of water Cherenkov detectors [6] that are able to measure the muonic and electromagnetic signal at the ground level. This signal is used to infer lnA .
The challenge for determining the cosmic ray mass composition at the highest energy is our limited knowledge of the hadronic interaction properties. The current hadronic interaction models extrapolate interaction properties measured in particle accelerators at energies more than two orders of magnitude smaller. If we knew the hadronic interaction properties at these higher energies, the interpretation of the data in terms of the primary cosmic ray composition would be straightforward. Despite this limitation, we used the hadronic interaction models to convert the different observables measured by each experiment into lnA to allow direct comparison of the measurements.

DIFFERENT APPROACHES TO ANALYSE X MAX OBSERVATIONS
The fluorescence detectors (FDs) of the Auger [7], HiRes [8] and TA [9] experiments have a limited field of view (FOV) ranging from about 3 • to 30 • in elevation. This limited FOV introduces a detector bias depending on the shower geometry: For close by showers, the lack of a high elevation FOV prevents the observation of shallow showers and, therefore, deeper showers are favored in the event selection. Moreover, deep near-vertical showers may reach their maximum below the FOV range (or even below ground level) and therefore the ensemble of selected showers is artificially enriched with shallow showers. To address this problem the collaborations follow two different approaches.

The auger approach
The goal of the Auger Collaboration is to publish X max and RMS(X max ) values with minimal detector bias, that are close to the moments of the undistorted distribution. This procedure is only limited by systematic uncertainties. The advantage of this proposal is that the published results can be compared directly with model expectations, without the need of any knowledge of the detector and, therefore, also to simulations with hadronic interaction models and or composition hypotheses that were not available at the time of publication.
In order to measure an unbiased X max distribution, it is necessary to select events based only on their arriving geometry and energy (i.e. not using any information of their X max value). The basic idea is to select showers with geometries that will allow X max to be inside a detector FOV that is "wide enough" to cover the true X max distribution. In order to choose the appropriate geometries to be considered in the X max analysis, it is necessary to have a rough idea of what the range of the X max distribution is. The idea is to use the data themselves to determine this range for each energy bin.
Given the geometry and energy of an event, the range in X max for which a shower can be detected with good quality can be predicted with a semi-analytical calculation that takes into account the atmospheric attenuation and the optical efficiency of the detector. The result of such a calculation is an X max -range, X min fov to X max fov , for which a shower will be accepted for the data analysis defining the effective (as opposed to geometrical) FOV of the telescopes. Fig. 1 shows the scatter plots of the measured X max between 10 18.0 and 10 18.2 eV as a function of the estimated X min fov and X max fov as well as the dependence of the average X max values on these variables. The two plots clearly illustrate the resulting bias that is introduced when X min fov is too deep (panel on the left) or when X max fov is too shallow (panel on the right). The Auger Collaboration analysis procedure uses  Figure 1. Scatter plots of the measured X max between 10 18.0 and 10 18.2 eV as a function of the event-by-event field of view boundaries X min fov and X max fov . The mean X max values as a function of X min fov and X max fov are superimposed as large black dots. The horizontal lines denote the asymptotic X max value far away from the field of view boundary. The range where the X max values are consistent with this line defines the unbiased region. The vertical lines indicate the limits of fiducial field of view that result in an unbiased X max measurement.  (Fig. 6). In this table we have included the number of reconstructed Auger events that survived all the quality cuts (i.e. number of events prior to the application of the field-of-view cuts). The energy distribution of these data is not shown in Fig. 6. The total number of events that the HiRes collaboration has used for the X max analysis above 10 18.2 eV is 815. However, after the application of the energy normalization (normalized to the TA energy scale) across experiments, 798 HiRes events remained with energies above 10 18 these graphs for each energy bin in order to determine what ranges of X min fov and X max fov values allow for an unbiased sampling of the X max distributions as a function of energy [10,11]. The vertical lines in Figure 1 indicates the appropriate limits for X min fov and X max fov . These limits determine the fiducial field-ofview cuts. These fiducial field-of-view cuts are optimized independently for the data and for each Monte Carlo (MC) composition. Different MC compositions (i.e. different X max distributions) have been used to test this algorithm [10,11] and the reconstructed X max values were found consistent with the MC input within statistical uncertainties.
The Auger collaboration applies identical quality cuts to data and Monte Carlo, which is similar to the approach shared by HiRes and TA (in the analysis described below) in which equivalent cuts are applied to data and Monte Carlo throughout. The additional field-of-view cuts -for which there is no direct analog in the HiRes or TA analysis -are optimized independently for the data and for each MC composition, using the same algorithm for this optimization. The motivation behind this choice is to allow optimization of the field-of-view cuts without making any a priori assumptions as to the range of the X max distribution.
Above 10 18.2 eV, the application of the field-of-view cuts reduce the Auger statistics by half ( Table 1). The severity of these field-of-view cuts (for reducing the statistics) puts a constraint on the minimum number of events required for this approach.

The HiRes and TA approach
The HiRes and TA collaborations do not apply field-of-view cuts. This means that they do not attempt to measure the unbiased X max in the atmosphere, instead they quote the X max as measured in the 01006-p.3 EPJ Web of Conferences detector (we will refer it as X meas max ). These two X max values can differ significantly depending on the intrinsic X max distribution, or depending on the zenith angle distribution of the events (as shown in Figs. 8(b) and 8(c)). However, this should not affect the HiRes/TA X max composition analysis, because an accurate detector modeling is used for predicting the X meas max observations for a given composition.
In order to understand the acceptance and reconstruction biases arising from the inherent field-ofview limitations of a fluorescence detector, both the HiRes and TA collaborations focus on accurate detector modeling through the use of a detailed detector Monte Carlo. Air showers are generated using CORSIKA [12] and several hadronic interaction models including QGSJet01 [13], QGSJet-II [14] and SIBYLL [15,16]. Shower libraries are created in which the number of particles as a function of slant depth is recorded for a large number of air showers induced by different primary masses.
In the detector simulation, an event is drawn from the library and assigned a random core location, zenith and azimuthal angle. The fluorescence light is propagated from the shower to the detector, with attenuation simulated via the use of an empirically determined atmospheric database. Ray tracing is performed to determine the photoelectron response of individual PMT's in the fluorescence camera. The trigger algorithms are simulated, and if the trigger conditions are satisfied the Monte Carlo event is written to disk in the identical format as real data, allowing study by the same analysis chain.
A number of control distributions are checked for agreement between data and Monte Carlo in order to assure that all detector effects are accurately described by the simulation. Particular attention is paid to those distributions -e.g. first and last viewed depth of the shower -which touch closely on the detector biasing issue. Finally, to extract information about composition the observed X max distributions in the data are compared to the "observed" X max distributions for Monte Carlo events which pass identical event selection criteria.
In summary, in order to estimate an average mass composition, the X max measured by Auger can be compared directly with the predictions from air shower simulations (within remaining systematic uncertainties). In the case of HiRes and TA, the measured X meas max should be compared with the X meas max obtained from a convolution of simulated showers with a model of the detector, atmosphere and reconstruction.
It is important to note that the X max measurements by Auger and Yakutsk cannot be compared directly with the X meas max published by HiRes and TA. In Sec. 7 we will transform these X max and X meas max to lnA to compare the different experiment results.

YAKUTSK MEASUREMENT OF THE X MAX DISTRIBUTIONS
The determination of X max in individual showers is based on the measurement of the Cherenkov light flux at different core distances Q(r): 1. by the parameter p = lg(Q(200)/Q(500)); 2. reconstruction of a shower development curve from the lateral distribution of Cherenkov light [17]; 3. measurement of the Cherenkov light pulse width at a fixed core distance; 4. by recording a Cherenkov track with a differential detector based on camera obscura.
The sensitivity of these techniques is described in [18]. The accuracy of X max determination in individual showers was estimated in a simulation of EAS characteristics measurements at the array involving MC methods and amounted to 30-45 g/cm 2 , 35-55 g/cm 2 , 15-25 g/cm 2 , 35-55 g/cm 2 respectively for the first, second, third and fourth methods. The total error of X max estimation included errors associated with core location, atmospheric transparency during the observational period, hardware fluctuations and mathematical methods used to calculate a main parameters.   Figure 2 shows the X max measured by Auger [19] and Yakutsk [20], together with the X meas max as measured by HiRes [3] and TA [4]. The observed agreement between the measured X max and X meas max is not expected.

COMPARING DIFFERENT X MAX MEASUREMENTS
At this meeting, the energy spectrum working group has compared the shape of the energy spectrum from the Auger, Yakutsk, HiRes and TA experiments and has produced a table with normalization factors [22]. For the plots presented here, we have normalized the energy scales to an energy scale that is half way between the Auger and TA energy scales. The normalization factors that we have used are 1.102 for Auger, 0.55 for Yakutsk, 0.883 for HiRes and 0.908 for TA. Later in Sec. 7 we will evaluate the compatibility of the different results. We will transform X max and X meas max to lnA for meaningful comparisons.  The HiRes collaboration chooses a fluctuation estimator that differs from the one published by Auger and Yakutsk. Whereas the latter use simply the standard deviation (denoted by RMS(X max )), HiRes uses the width of an unbinned likelihood fit with a Gaussian to the distribution truncated at 2 × RMS, denoted by X . Figure 4 shows the X meas max and X as measured by HiRes. The lines are the corresponding X meas max and X expectations for proton and iron compositions. The different line types correspond to different models (QGSJet-01, QGSJet-II, SIBYLL2.1). Figure 5 shows the corresponding X meas max observation and expectation for the TA experiment. Currently the TA experiment does not have enough statistics to quantify the width of the X max distributions at the highest energies. Figure 6 shows the energy distributions and total number of events that survived the selection cuts at each experiment. For this Figure, the energy scales have been normalized to the TA energy scale. A summary of Figure 6 is shown in Table 1.
The X max measurements from Yakutsk (Fig. 3), HiRes (Fig. 4) and TA (Fig. 5) experiments are consistent with the QGSJet predictions for a constant proton composition at all energies above 10 18 eV, whereas the X max measurements from the Pierre Auger Observatory are significantly shallower than these predictions above a few EeV (cf. left panel Fig. 3).
At the same time, the width of the X max distribution measured by Auger gets narrower above a few EeV and the Yakutsk measurements of the fluctuations are consistent with the Auger up to about 10 19 eV. Yakutsk has one measurement of the X max width above 10 19 eV and it is wider than the Auger one by about 3 standard deviations (right panel Fig. 3). The X max distribution widths measured by HiRes at energies above 2 × 10 19 eV, while consistent with pure proton composition have large statistical errors and do not definitely exclude a heavier composition (right panel Fig. 4).

A toy MC to evaluate the performance of different fluctuation measures
The X max distribution is expected to have an asymmetric long tail due to the exponential nature of the depth of the first interaction. Both, RMS(X max ) and X , could be under-estimated if they are sampled with a small number of events. We have used a toy MC simulation to evaluate the performance of both methods for the quantification of the shower-to-shower fluctuations of X max . We have generated random samples of "N" events with X max distributed following the expectation for 10 19 eV proton and iron showers using the QGSJet-II model. Fig. 7 (left panel) shows the measured RMS and X as a function of the number of events "N". As can be seen, both approaches show a bias in the measured X max distribution width when the number of events in the sample is small. For samples with at least 30 events, this bias is negligible. Even for smaller statistics, the magnitude of the bias from under-sampling is negligible compared with the statistical uncertainty. The statistical uncertainties for the measured RMS and X are shown in the middle panel of Fig. 7.
When using the RMS to quantify the X max distribution width, the separation between proton and iron compositions is larger. On the other hand, the associated statistical uncertainties are smaller when using X . In order to evaluate the performance of using the RMS or the X to quantify the width of the X max distribution, we have computed a performance indicator defined as:  Figure 8. The X max and X meas max for Auger and HiRes using showers from different zenith angle ranges.
where width(p/Fe) denotes the average difference between the X max distribution width for proton and iron (measured using the RMS(X max ) or X respectively), and p and f e are the corresponding statistical uncertainties of the fluctuation measurements. Figure 7 shows the computed sensitivities (right panel). It turns out that the sensitivity of both approaches is basically equivalent at all ranges of number of events. We have also introduced a 20 g/cm 2 X max resolution effect to compute the sensitivity. As expected, this reduces the sensitivity in both cases, but does not change the equivalence of both approaches.

STABILITY OF THE X MAX OBSERVATIONS AND CROSS CHECKS
In this section we want to show how stable the X max distribution measurements are.
We have checked whether the measured X max distributions depend on the the zenith angle. Vertical showers are more affected by the ground level truncation of the distribution and, moreover, the fluorescence light has to traverse denser regions of the atmosphere to the detector.
Due to the analysis strategy used in Auger, there is no significant difference between the vertical and inclined X max measurements, as is illustrated in Fig. 8(a)).
In the case of HiRes, there is about 40 g/cm 2 difference between the X max measured in two zenith angle intervals, as can be seen by comparing Figures 8(b) and 8(c)). This difference is however well reproduced by the detector simulation and for both zenith angle intervals the data are compatible with the proton prediction from QGSJet-II.
The Auger collaboration has used MC data to evaluate the flatness of the detector acceptance as a function of the depth of X max . Reference [24] shows that this acceptance becomes flat after the application of the field-of-view cuts.
Another way to cross check that there is a homogeneous acceptance in the Auger X max analysis, which is independent of the shower composition, is by comparing data and MC energy distributions resulting after the fiducial cuts. Figure 9 shows the energy distribution of the Auger data from [1] compared with the distributions for proton and iron. For Fig. 9(a), the MC distributions were obtained applying quality cuts only (i.e. without applying fiducial cuts). The MC events were re-weighted at generator level to match the spectral shape of the CR flux measured by the Auger surface detector [23] and normalized to the data at 10 18.6 eV. As expected, the spectral shape of the MC without fiducial selection does not match the data. It is because the acceptance depends on the composition (or, more precisely, on the distribution of shower maxima in the atmosphere). After application of the fiducial field-of-view cuts, the spectral shapes of both, the proton and iron simulations, agree well with the data (Fig. 9(b)).
To check if there could be a difference in interpreting the data due to different analysis strategies, currently the TA and Auger collaborations are separately working on their data interpretation using both analysis strategies.

VALIDITY OF THE DETECTOR MONTE CARLO
The X max analysis approaches followed by Auger, TA and HiRes require some information from detector Monte Carlo simulations.

Auger
For the Auger approach, the detector MC simulations are used to estimate the average X max reconstruction bias and the average X max resolution as a function of energy. They are used to correct the observed X max and RMS(X max ) values respectively. After applying fiducial volume cuts, the correction on X max is smaller than 4 g/cm 2 , and the average X max resolution is about 20-25 g/cm 2 .
The Auger collaboration has used stereo events to cross check the validity of its detector simulations. Stereo events have been simulated, reconstructed and selected in the same way as data. The advantage of using stereo events is that showers are reconstructed almost independently using each of the FD observations (they are not completely independent because they use the same surface station for the hybrid reconstruction of the geometry). As a result we obtain two measurements of the shower parameters. From the comparison of these two sets of shower parameters the corresponding resolutions are estimated. The resolution in X max depends on the characteristics of the showers (such as geometry and energy). So, the resolution obtained using stereo events is not a representative X max resolution of regular hybrid events (that are on average of lower energy than stereo events). However, the X max resolution obtained with data and MC stereo events has to be consistent if the detector simulation is working correctly. Figure 10(a) shows the consistency of the X max resolution obtained using data and simulated stereo events.

HiRes and TA
For the TA and HiRes approach, the detector MC simulations are used to estimate the expected X max distributions after considering the detector effects. These expectations are estimated for different cosmic ray primaries. Then, the expected and observed X max distributions are compared to infer the average cosmic ray composition. Figure 10(b) shows the R p distribution for TA data and for MC calculations. Figure 10(c) shows the X max difference between HiRes-II (X I I ) and HiRes-I (X I ) for HiRes stereo data (points) overlaid with QGSJet-II proton Monte Carlo calculations. The asymmetry is caused by HiRes-I covering only half the range in elevation angle. The Gaussian width of the peak is 44 g/cm 2 , setting an upper limit of 31 g/cm 2 for the HiRes-II X max resolution. Monte Carlo studies indicate that the actual HiRes-II X max resolution is better than 25 g/cm 2 over most of the HiRes energy range.

COMPARISON OF X MAX RESULTS FROM DIFFERENT EXPERIMENTS
In order to make sensible comparisons between experiments, we have used the observed X max values to infer the average logarithmic mass, lnA , using.
Similarly, one can transform X meas max into lnA by replacing X max with X meas max in this equation although this is only correct as a first order approximation. This is because X meas max does not correlate linearly with lnA as X max does [2].
When we transformed the measured X max into lnA , we used the expected X max values for proton and iron obtained directly from Conex simulations. On the other hand, when we transformed the measured X meas max values, we used the expected X meas max values for proton and iron extracted from the simulation including the detector. Figure 11 shows the lnA estimated using the QGSJet-II and SIBYLL interaction models. The shaded regions indicate the range of the corresponding systematic uncertainties that were propagated from the systematic uncertainties in X max (12 g/cm 2 for Auger and TA, 20 g/cm 2 for Yakutsk, and 6 g/cm 2 for HiRes).
HiRes quotes systematics broken down into a 3.4 g/cm 2 shift in the mean and an uncertainty of 3.2 g/cm 2 decade in the elongation rate. For the purposes of the present comparison, we have combined 01006-p. 10
(b) using SIBYLL model. Figure 11. Comparing the average composition ( lnA ) estimated using Auger, HiRes, TA and Yakutsk data. The shaded regions correspond to the systematic uncertainty ranges. To infer the average composition from X max , QGSJet-II and SIBYLL models have been used.
the two HiRes uncertainties into a single number by adding in quadrature the uncertainty in the mean and the shift due to a 1 variation in slope over 1.6 decades of energy.
All the systematic uncertainties (on the measured X max ) used in this work correspond to each experiment's quoted value. This working group has not attempted to validate those values.
At ultra-high energies, the Auger data suggest a larger lnA than all other experiments. The Auger results are consistent within systematic uncertainties with TA and Yakutsk, but not fully consistent with HiRes. HiRes is compatible with the Auger data only at energies below 10 18.5 eV when using QGSJet-II ( Fig. 11(a)), and when using SIBYLL model, Auger and HiRes become compatible within a larger energy range (Fig. 11(b)).
Comparing Figs. 11(a) and 11(b) we find that the level of incompatibility between Auger and HiRes data depends on the model used to interpret the X max observations. Different models predict different ranges of X max values for proton and iron cosmic rays, and depending on how these predictions compare with the range of X max values that could be inside the FOV of the detector, the X meas max (observed by HiRes) could be more or less different to the intrinsic X max , changing the interpretation of X meas max . The HiRes results are compatible in every way with the interpretation that the composition is light, i.e. lighter than the CNO group of elements. The Auger X max and RMS(X max ) results do not allow this interpretation. Figure 2 shows that the X max observed by Auger and the X meas max observed by HiRes and TA are similar. Is there any physical reason that the X max for Auger and the X meas max for HiRes and TA are all similar, or is it just coincidence?. A direct way of checking the Auger and HiRes/TA compatibility would be to simulate a hypothetical composition which had the same X max distributions as observed by Auger. Then this composition would be propagated through the HiRes and TA detector simulations and the expected X meas max computed. So, we could compare directly the expected and observed X meas max to evaluate the compatibility of the Auger and HiRes/TA observations (this is work in progress).
We have also evaluated how the average logarithmic mass estimated by the experiments evolves as a function of energy. Currently there are two different models suggested by the Auger and HiRes collaborations. The X max and RMS(X max ) observed by the Auger experiment suggest that the composition might be becoming lighter with energy up to 10 18.3 eV, and heavier above this energy. On the contrary, the X max and RMS(X max ) observed by the HiRes experiment is consistent with a constant composition (light composition) all along the observed energy range. We have evaluated both, the Auger and HiRes composition models using Auger, HiRes, TA and Yakutsk data (only statistical uncertainties were considered for this evaluation). The results are summarized in Tables 2 and 3.      Table 2). Figure 12(b) shows the test of the Auger model (a fit to a broken line). The fitted parameters are only the energy and lnA values at which the lines break. The slopes before and after the breaking point are fixed to the results of the Auger fit. The 2 /ndf values for these fits are small. However, the Auger energy and lnA for the break point is not statistically compatible with the break points fitted by HiRes, TA or Yakutsk (see Table 3). Further, studies (exploring the effect of different interaction models) and more statistics in the Northern Hemisphere are required to establish the level of compatibility between Southern and Northern Hemispheres.

OTHER OBSERVATIONS SENSITIVE TO MASS COMPOSITION
Apart from X max observations, other shower observables can also provide information of the average composition. Yakutsk uses an array of muon detectors [5] to measure muon signals at ground level. Auger uses its ground array of water Cherenkov tanks to measure the signal asymmetries around the 01006-p.12

UHECR 2012
lg(E/eV) 17 17 . Figure 13. Average composition estimated using other (other than X max ) shower observables. Open circles are using muon detectors from the Yakutsk experiment [5], solid circles use the observed shower asymmetries around the core with the Auger SD [21], and open crosses are using the estimated muon production depth maximum with the Auger SD.
shower core, and to estimate the muon production depth (MPD) maximum. These observations together with the assistance of Monte Carlo simulations of the detector and hadronic interaction models provide measurements of the average composition ( lnA ). Figure 13 shows the average composition as a function of energy estimated using the muon detectors from the Yakutsk experiment, and the Auger ground array. For comparison purposes we have also included the broken lines fitted to the composition estimated using the X max observations from Auger. For all these estimates of the average composition the model QGSJet-II has been used.
Despite some systematic difference between measurements from Auger X max and Yakutsk muons, both observations suggest that the composition becomes lighter up to about 10 18.5 eV and then it becomes heavier again above this energy. Measurements from Auger asymmetries also suggest that the composition becomes heavier above 10 18.5 eV. Measurements from Auger MPD only expand within a narrow energy range, and they do not provide much information regarding the evolution of the composition as a function of energy.

DISCUSSION
When comparing the lnA values estimated from the X max observations, results from Auger, TA and Yakutsk are compatible within systematic uncertainties. TA and Yakutsk are also compatible with HiRes. However, Auger and HiRes are not fully compatible within systematic uncertainties.
As shown in Section 7, the level of compatibility between Auger and HiRes (the two higheststatistics observatories) depends on the particular interaction model used to interpret the X max observations. Further experimental data on the high-energy hadronic interactions, e.g. from the LHC [25], would help to refine the current composition picture.
We need more statistics in the Northern Hemisphere (about 3 times the current statistics) in order to provide a conclusive statement to whether or not the composition is changing with energy in this Hemisphere. The current data, while completely consistent with a constant light composition, cannot definitively exclude a changing composition as suggested by Auger. More statistics are also necessary to establish whether there is indeed a difference in the RMS(X max ) at higher energies between Auger and Yakutsk (Fig. 3).

EPJ Web of Conferences
In the Northern Hemisphere HiRes has stopped data taking in 2006, however the hybrid TA observatory with a surface area of approximately 800 km 2 will be acquiring additional data for the next several years at least. Figure 13 shows lnA measurements as a function of energy using different techniques. Despite the systematic differences, the measurements suggest a composition getting lighter at energies up to about 10 18.5 eV and a composition getting heavier above this energy. The systematic differences between different type of measurements are very sensitive to the particular interaction model used for the interpretation. We showed the results for model QGSJet-II (in Fig. 13), because all experiments had results available using this model.