1.2. Primary species
When a CR enters the Earth atmosphere it collides with a nucleus of a an air atom, producing a roughly conical cascade of billions of elementary particles which reaches the ground in the form of a giant "saucer" traveling at nearly the speed of light.
Unfortunately, because of the highly indirect method of measurement, extracting information from EASs has proved to be exceedingly difficult. The most fundamental problem is that the first generations of particles in the cascade are subject to large inherent fluctuations and consequently this limits the event-by-event energy resolution of the experiments. In addition, the center-of-mass energy of the first few cascade steps is well beyond any reached in collider experiments. Therefore, one needs to rely on hadronic interaction models that attempt to extrapolate, using different mixtures of theory and phenomenology, our understanding of particle physics. At present, the different approaches used to model the underlying physics of p collisions show clear differences in multiplicity predictions which increase with rising energy [28, 29, 30]. Therefore, distinguishing between a proton and a nucleus shower is extremely difficult at the highest energies [31, 32].
Fortunately, photon and hadron primaries can be distinguished by comparing the rate of vertical to inclined showers, a technique which exploits the attenuation of the electromagnetic shower component for large slant depths. Comparing the predicted rate to the rate observed by Haverah Park for showers in the range 60° < < 80°, Ave et al.  conclude that above 1019 eV, less than 48% of the primary CRs can be photons and above 4 × 1019 eV less than 50% can be photons. Both of these statements are made at the 95% CL.
The longitudinal development has a well defined maximum, usually referred to as Xmax, which increases with primary energy as more cascade generations are required to degrade the secondary particle energies. Evaluating Xmax is a fundamental part of many of the composition studies done by detecting air showers. For showers of a given total energy, heavier nuclei have smaller Xmax because the shower is already subdivided into A nucleons when it enters the atmosphere. Specifically, the way the average depth of maximum <Xmax> changes with energy depends on the primary composition and particle interactions according to
where De is the so-called "elongation rate" and E0 is a characteristic energy that depends on the primary composition . Therefore, since <Xmax> and De can be determined directly from the longitudinal shower profile measured with a fluorescence detector, E0 and thus the composition, can be extracted after estimating E from the total fluorescence yield. Indeed, the parameter often measured is D10, the rate of change of <Xmax> per decade of energy.
Another important observable which can be related to primary energy and chemical composition is the total number of muons Nµ reaching ground level. For vertical proton showers, numerical simulations  indicate that the muon production is related to the energy of the primary via 
Thus, modeling a shower produced by a nucleus with energy EA as the collection of A proton showers, each with energy A-1 of the nucleus energy, leads - using Eq. (2) - to NAµ A(EA / A)0.93 . Consequently, one expects a CR nucleus to produce about A0.07 more muons than a proton. This implies that an iron nucleus produces a shower with around 30% more muons than a proton shower of the same energy.
The analysis of the elongation rate and the spread in Xmax at a given energy reported by the Fly's Eye Collaboration suggests a change from an iron dominated composition at 1017.5 eV to a proton dominated composition near 1019 eV [37, 38]. Such behavior of De is in agreement with an earlier analysis from Haverah Park . However, the variation of the density of muons with energy reported by the Akeno Collaboration favours a composition that remains mixed over the 1018 -1019 eV decade . More recently, Wibig and Wolfendale  reanalyzed the Fly's Eye data considering not only proton and iron components (as in ) but a larger number of atomic mass hypotheses. Additionally, they adopted a different hadronic model that shifts the prediction of Xmax for primary protons of 1018 eV from 730 g cm-2  to 751 g cm-2. The difference, although apparently small, has a significant effect on the mass composition inferred from the data. The study indicates that at the highest energies (1018.5 -1019 eV and somewhat above) there is a significant fraction of primaries with charge greater than unity. This result is more in accord with the conclusions of the Akeno group than those of the Fly's Eye group. Very recently, the Volcano Ranch data was re-analyzed taking into account a bi-modal proton-iron model . The best fit gives a mixture with 89 ± 5% of iron, with corresponding percentage of protons. A summary of the diferent bi-modal analyses is shown in Fig. 1. Within statistical errors and systematic uncertainties introduced by hadronic interaction models, the data seem to indicate that iron is the dominant component of CRs between ~ 1017 eV and ~ 1019 eV. Nonetheless, in view of the low statistics at the end of the spectrum and the wide variety of uncertainties in these experiments, one may conservatively say that this is not a closed issue.
Figure 1. Predicted fraction of iron nuclei in the CR beam at the top of the atmosphere from various experiments: Fly's Eye (), AGASA A100 (), AGASA A1 () using SIBYLL 1.5 as the hadronic interaction event generator  and Haverah Park , using QGSJET98 () and QGSJET01 () to process the hadronic collisions. The solid (dashed) line rectangle indicates the mean composition with the corresponding error estimated using the Volcano Ranch data and QGSJET98 (QGSJET01); the systematic shift in the fraction of iron induced by the hadronic event generator is 14% .