The determination of the precise spectral type of stars with circumstellar material is extremely useful. The immediate result of the classification is a derivation of the effective temperature that allows us to fix the position of the star in the $T_{\rm eff}$ axis of a luminosity-effective temperature HR diagram. In addition, the comparison of the target spectra with spectra of standard stars may reveal the presence of non-standard features, which provide valuable information on the physical conditions of the photospheres and the surrounding medium.

4.1 Methodology

The spectral classification was done mainly by comparing the INT spectra of the target stars with spectra of standard stars obtained with the same or a very similar instrumental configuration.

Two sets of standard stars were used for the spectral classification: 28 standards observed with the INT during the four EXPORT campaigns; and a library of spectra of 116 late-type (F to M) field stars (Montes et al. 1997).

The instrumental configuration used to observe the standards with the INT was the same as the one used for the target stars. The spectra contained in the library compiled by Montes et al. (1997) were observed with different telescopes - the INT was one of them - and cover different spectral intervals with different resolutions. The closest configuration to those used during the EXPORT runs is the one labelled with number "8'' in the paper by Montes et al.; it covers the range 5626 to 7643 Å, therefore, it overlaps the spectral interval of the target stars completely, although the spectral resolution is lower (1.58 Å per pixel).

The first step in the classification procedure was a visual inspection of the average spectrum and each one of the individual target spectra. This analysis allowed us to identify significant changes in the spectral type, which lines are variable, the presence of permitted and forbidden lines appearing in emission, and lines caused by interstellar absorption.

The spectra of the standard stars were broadened with the rotation profile corresponding to the target star, and then a comparison among them was carried out. When available, values of $v \sin i$ obtained by us (see Sect. 5) were used; otherwise, published values were taken at this point. This usually provided quite an exact determination of the spectral type. In some cases the classification was assigned by bracketing the spectrum of the target star between those of two standards with slightly different spectral types, taking the mean as the correct result. In Fig. 3 we can see an example of the efficiency of this step. In the upper panel the spectrum of the target star HD 199143 (solid line) is shown along with the spectrum of the slow-rotator ( $v\sin i < 5$ km s^-1) standard star HR 4606 (F6 V) (dotted line); in the lower panel, the spectrum of the standard has been broadened with a rotation profile $v\sin i= 155$ km s^-1. The match with the spectrum of the problem star is remarkable.

$\begin{figure} \par\includegraphics[width=8.6cm,clip]{MS1324f3.eps}\par\end{figure}$

Figure 3: Upper panel: the spectrum of HD 199143 (solid line) and the spectrum of a slow-rotator standard star HR 4606 (F6 V) (dotted line). Lower panel: the spectrum of the standard has been broadened with $v\sin i= 155$ km s^-1.

The presence of veiling in some of the stars in the sample, mainly CTTs, poses an additional difficulty in the determination of spectral types. Since the effect of veiling is a global weakening, or even the disappearance of the absorption features in the spectral region affected, the comparison of a spectrum affected by veiling with the grid of spectra of normal stars would lead to the derivation of an earlier spectral type to the target star instead of the correct one, therefore this effect must be treated properly. We illustrate this problem with a typical case, that of DR Tau, in Sect. 4.4.

Some additional tools may be applied for confirming the spectral classification of these stars. One of them is the use of ultraviolet spectra. For those objects observed with IUE (International Ultraviolet Explorer), a comparison with synthetic energy distributions computed using the ATLAS9 code (Kurucz 1993) was done. The ultraviolet spectra provided by IUE are calibrated in absolute flux, so a direct comparison, after correcting for reddening, with the synthetic models was feasible.

The IUE spectra themselves have been used to estimate $E(B\!-\!V)$ , taking advantage of the wide feature in absorption around 2200 Å caused by the interstellar medium. The process consisted of correcting the spectra for reddening with different values of $E(B\!-\!V)$ until the absorption feature disappeared; at this point that particular value of $E(B\!-\!V)$ was assigned to the star. The Galactic extinction law by Howarth (1983) was used. This method is applicable only when there is a substantial amount of flux in the continuum around 2200 Å, i.e. it is only suitable for stars hotter than F0. Otherwise, published values of the colour excesses were used. In Table 3 we give the values of the colour excesses and their uncertainties determined in this way; in the last column values of $E(B\!-\!V)$ computed as $(B\!-\!V)_{\rm target \,star}-(B\!-\!V)_{\rm st}$ are given, being $(B\!-\!V)_{\rm st}$ the colour index of a standard with the same spectral type as the target star. An uncertainty of 0.05 in the colour index leads to an error in the classification of three subtypes, so the more dubious confirmation of the spectral type following this method would be for VX Cas.

In 27 stars, this confirmed the spectral type obtained from the optical spectra. The use of IUE spectra is particularly powerful for Vega-type and A-shell stars since the spectral region covered ( $\sim$ 1150-3350 Å) is not affected by excesses or veiling. In the remaining cases, an accurate determination of the spectral type in the ultraviolet requires the knowledge of these peculiar features (Catala & Bertout 1990; Valenti et al. 2000) and also of the potential variability of the object in this range. In any case, the fact must be taken into account that the IUE spectra are not simultaneous with the EXPORT observations. Different determinations of the spectral type from optical and ultraviolet observations might be attributable to long-term intrinsic variability of the stars.

Table 3: $E(B\!-\!V)$ values from IUE spectra.
Star $E(B\!-\!V)_{IUE}$ $E(B\!-\!V)_{\rm target-st}$

HD 58647 0.16 $\pm$ 0.02 0.16

HD 179218 0.10 $\pm$ 0.05 0.08

VX Cas 0.30 $\pm$ 0.10 0.26

17 Sex 0.07 $\pm$ 0.02 0.02

WW Vul 0.20 $\pm$ 0.05 0.30

**Table 3:** $E(B\!-\!V)$ values from IUE spectra.
Star	$E(B\!-\!V)_{IUE}$	$E(B\!-\!V)_{\rm target-st}$
HD 58647	0.16 $\pm$ 0.02	0.16
HD 179218	0.10 $\pm$ 0.05	0.08
VX Cas	0.30 $\pm$ 0.10	0.26
17 Sex	0.07 $\pm$ 0.02	0.02
WW Vul	0.20 $\pm$ 0.05	0.30

$\begin{figure} \par\includegraphics[width=8.8cm,clip]{MS1324f4.eps}\end{figure}$

Figure 4: An example of fitting of synthetic energy distributions (dashed lines) to the IUE spectrum of a Vega type star, HD 109085 (solid line). The two models correspond to spectral types F1 V (top) and F3 V (bottom). The original models were normalized to the target spectrum and shifted for clarity. The spectral type derived for HD 109085 from optical spectra is F2 V.

In Fig. 4 we can see an example of a fit, in the ultraviolet range, of two synthetic Kurucz's models to the spectrum of HD 109085, a Vega-type star with a spectral type F2 V as determined from the comparison with optical spectra of standards. We show the ordinates in logarithmic units. The two models (dashed lines) correspond to spectral types F1 V (top) and F3 V (bottom) and were computed with the solar abundance. The synthetic energy distributions were normalized to the target spectrum (solid line) and shifted 0.7 dex up and down for clarity. The similarity of the synthetic spectra with the energy distribution of the target star is remarkable.

The derivation of the luminosity class was also done by comparing the target spectra with those of standards with different luminosity classes. A consistent check of this parameter can also be obtained by estimating, if possible, the absolute magnitude M_V of each star, and then comparing it with standard values of that magnitude for standard stars of different luminosity classes. One must be cautious with the results obtained following this method since we are comparing magnitudes of stars with peculiar features with magnitudes of standard, normal stars. M_V can be easily found from V and the distance (parallax), provided the colour excess $E(B\!-\!V)$ is known. A total of 38 stars have parallaxes available from Hipparcos, these have been retrieved from SIMBAD.

4.2 Spectral ranges and lines used in the classification

In general, the comparison between the spectra of the target stars and the standards has made use of all the lines present, excluding only those whose origin was clearly circumstellar or interstellar. Particular care was taken in excluding the so-called DIB's (diffuse interstellar bands) at the following wavelengths: 5780, 5797, 5850, 6196, 6234, 6270, 6284, 6369, 6376 and 6614 Å (Schmidt-Kaler 1982; Moutou et al. 1999).

Table 4: Useful lines to classify cool stars.
Line identification Spectral types

Ca I 6102 Å

Ca I 6122 Å K0 V-K5 V

Fe II 6129 Å

Fe I 6546 Å F0 V-F9 V

N II 6596 Å

Ca I 6162 Å

Fe I 6136 Å F0 V-M0 V

Fe I 6677 Å

Ca I/Fe I 6494 Å

**Table 4:** Useful lines to classify cool stars.
Line identification	Spectral types
Ca I 6102 Å
Ca I 6122 Å	K0 V-K5 V
Fe II 6129 Å
Fe I 6546 Å	F0 V-F9 V
N II 6596 Å
Ca I 6162 Å
Fe I 6136 Å	F0 V-M0 V
Fe I 6677 Å
Ca I/Fe I 6494 Å

In Table 4 we list some specific lines especially useful for classifying late-type stars (F to M) due to their gradual variation in intensity as one moves from earlier to later spectral types. We show in the table the interval of spectral types where the changes in intensity are significant for discriminating between similar types. In addition to these lines, the molecular bands TiO and VO appear in the spectra of late-type stars cooler than K7 V and their depth increases for later spectral types. Further information on lines useful for classifying this kind of object in other spectral intervals can be found in Cohen & Kuhi (1979), Schmidt-Kaler (1982) and Jaschek & Jaschek (1990).

For hot stars (B and A) the He I lines at 5875 and 6678 Å have been used to confirm the B types when these lines are clearly photospheric, i.e., when no significant changes were observed when comparing spectra of the same star obtained on different dates. The presence of variability in these lines probably indicates that they are formed in the circumstellar environment and therefore they are not appropriate for classification.

4.3 Sources of errors in the spectral classification

The classification is based on comparisons of the target spectra with those of standards broadened with the rotational profile corresponding to the target star. Therefore, when the projected rotational velocity is of the order of 200 km s^-1 or larger, the lines in the spectra are blended and it is not possible to carry out a comparison line by line of both sets of spectra, although a comparison of the broad features is still feasible. In that case, it is possible to give an estimate of the spectral type but with around five subtypes of error; the use of IUE spectra could help to classify these stars, since the shape of the continuum is not very much affected by the high rotational velocity, although, as we mention in Sect. 4.1, the lack of simultaneity between the ultraviolet and optical observations could lead to the assignment of different spectral types if some kind of variability is present. The uncertainties in the fitting of Kurucz's synthetic models to the ultraviolet spectra arise mainly from the fact that some ultraviolet excess may be present, as it is observed in some T Tauri and Herbig Ae/Be stars. In these cases we have not attempted any fit, thus avoiding a wrong classification of the star.

Another source of error in the determination of the spectral type of a young star is the potential variability in the equivalent width and shape of the lines and even the appearance and disappearance of some lines in intervals of a few days due to intense circumstellar activity. Therefore, it is very important to have several spectra of the same star, obtained with enough separation in time to distinguish which lines are photospheric and which are formed in the circumstellar environment.

4.4 The problem of veiling

The phenomenon of veiling can be defined as the superposition of a continuum on the stellar spectrum proper. The interpretation of this phenomenon is a matter of debate and different scenarios have been proposed in the last few years. It is not whithin the scope of this paper to discuss which one is the most physically sound.

It is clear that this effect must be taken into account when attempting to classify a PMS star. A star with strong veiling would have spectral lines attenuated with respect to the global continuum, therefore one would tend to use standard stars corresponding to an earlier spectral type - compared with the real one - since these show fewer and less intense lines than the spectra of late-type templates.

In a first step of our work, five stars in the sample, namely HR 26 B, HR 5422 B, BO Cep, CW Tau and DR Tau, were classified with spectral types much earlier than those reported in previous studies. This led us to consider the effect of veiling in their spectra. DR Tau has been studied extensively and a strong veiling has been reported, as we explain in detail below; CW Tau shows veiling in the near infrared range (Folha & Emerson 1999). For HR 26 B, HR 5422 B and BO Cep, it turned out that a classification taking into account veiling effects was much more accurate. In Table 5 we give the values for the veiling we have found for these five stars with their correspondent uncertainties. Note the variability in DR Tau from October 1998 to January 1999.

Table 5: Veiling values from INT spectra.
Star Veiling

HR 26 B 0.1 $\pm$ 0.1

HR 5422 B 0.3 $\pm$ 0.1

BO Cep 1.0 $\pm$ 0.4

CW Tau 0.7 $\pm$ 0.4

DR Tau (Oct) 1.0 $\pm$ 0.5

DR Tau (Jan) 3.0 $\pm$ 1.0

**Table 5:** Veiling values from INT spectra.
Star	Veiling
HR 26 B	0.1 $\pm$ 0.1
HR 5422 B	0.3 $\pm$ 0.1
BO Cep	1.0 $\pm$ 0.4
CW Tau	0.7 $\pm$ 0.4
DR Tau (Oct)	1.0 $\pm$ 0.5
DR Tau (Jan)	3.0 $\pm$ 1.0

Hessman & Guenther (1997) give a detailed description of how the veiling can be parametrized (see Sect. 4 of their paper). We have used as a working definition of veiling the ratio between the non-stellar to the stellar continuum. In order to classify the stars with veiling we have added a flat continuum to the spectra of the grid of standards with several values of the veiling, in the sense described above, and normalizing afterwards to unity again. Given the subtle difference among these spectra, it is impossible to decide only by visual inspection which one is the closest to the target spectrum. To overcome this difficulty we have carried out cross correlations between the target spectrum and the grid of standard stars, both unaffected and affected by veiling, in order to determine the correct spectral type.

$\begin{figure} \par\includegraphics[width=13cm,clip]{MS1324f5.eps} % \end{figure}$

Figure 5: Two examples of classification of stars with veiling. Upper panel: from top to bottom, the spectra of DR Tau obtained in January 1999 and October 1998, the spectrum of a standard K5 V with a veiling of 1.0 (see text for details) and a standard F6 V, which was the first choice for the spectral type before veiling was considered. Lower panel: the spectra of HR 26 B, a standard K0 V with a veiling of 0.1 and the spectrum of a G7 V standard.

We show in Fig. 5 two examples of this process. In the upper panel the average INT spectrum of DR Tau obtained in October 1998 (thick line) is compared with both the spectrum of a K5 V standard affected by a veiling of 1.0 (i.e. a flat continuum has been added to the spectrum with the same level as the stellar continuum), and the spectrum of a F6 V standard which could lead to erroneous classification if veiling is not considered. The three spectra are normalized and the standards have been shifted downwards for clarity. At the very top of the graph the spectrum of DR Tau obtained in January 1999 is shown, shifted upwards. A similar comparison is done for the star HR 26 B in the lower panel. It can be seen how subtle the differences are, especially in the case of HR 26 B, which can easily lead, if one is not careful enough, to a wrong determination of the spectral type.

$\begin{figure} \par\includegraphics[width=13.5cm,clip]{MS1324f6.eps} % \end{figure}$

Figure 6: Five examples of the final results of the spectral classification. The spectra of the target stars are drawn with thicker lines, the spectra of the standards have been shifted down 0.35 units for clarity. From top to bottom, BD+31 643 (Vega type) and the standard HR 930 (B5 V) broadened with $v \sin i=162$ km s^-1, HR 10 (A-shell) and the standard HR 4633 (A4 V) broadened with $v \sin i=110$ km s^-1, HD 190073 (Herbig Ae/Be) and the standard HR 4670 (A2 IV), CO Ori (ETT) and the standard HR 4606 (F6 V), and LkH $\alpha$ 200 (CTT) and the standard HD 166620 (K2 V).

4.5 Results of the spectral classification

We have very accurately determined the spectral type of the 70 stars observed with the INT. Out of them, 58 stars (83%) have been classified with an error less than two spectral subtypes and 12 (17%) with an error of up to five spectral subtypes. 31 stars (43%) now have a new spectral type from our work, this means that the former spectral types are outside the error bars of our new determination or that we have different or new information about the luminosity class.

In Fig. 6 we show some representative examples of the spectral classification of several types of stars in the sample. From top to bottom we can see the spectra of BD+31 643 (Vega type), HR 10 (A-shell star), HD 190073 (Herbig Ae/Be), CO Ori (early T Tauri) and LkH $\alpha$ 200 (classical T Tauri) (thick lines) and the spectra of the standards, broadened with the corresponding rotation profiles, (thin lines, shifted 0.35 units down in the intensity scale), that give the best fits to the target spectra.

In Table 6 we give the result of the spectral classification for the 70 target stars. Column 1 is the star identification, Col. 2 gives the class of star, in Cols. 3 and 4 we give the result of our classification and those from previous work, respectively. The class of star (Col. 2) for each object has been assigned according to the spectral types found in this work. We have adopted the criterion of assigning a main-sequence type to the primary stars of the Lindroos' binaries and a PTT type to the secondary stars.

Almost all the PMS in the sample present spectral variability. Some non-photospheric features appearing in the spectra are: H $\alpha$ in emission, forbidden emission lines ([O I] 6300 Å, [O I] 6363 Å, [S II] 6716 Å, [S II] 6731 Å, [N II] 6583 Å, [N II] 6548 Å and [N II] 6527 Å) which may be related to winds; and variable lines of He I 5875 Å, 6678 Å and Na I D₁ D₂ 5890, 5896 Å that may be produced in the circumstellar environment.

4.6 Comments on particular stars

In this subsection we comment on the stars marked with an asterisk in Table 6 and also on those showing some peculiarity.

HD 58647, HD 141569, HD 163296, HR 2174 and VX Cas have very high projected rotational velocities and show many non-photospheric features in their spectra. VY Mon also shows a lot of circumstellar or interstellar lines contaminating its spectrum; V1686 Cyg and VV Ser show very variable emission lines; MWC 297 has very few useful lines in absorption for classification due to the blends caused by its high $v \sin i$ ( $350\pm 50$ km s^-1, Drew et al. 1997); HD 123160, RR Tau and LkH $\alpha$ 262 show changes in the continuum in time spans of days to months; HR 26 B, HR 5422 B, BO Cep, CW Tau and DR Tau show veiling that attenuate the absorption lines.

In addition, HK Ori shows a composite spectrum from a non-resolved binary system; HR 10 has an INT spectrum classified as a A0 Vn but shows an IUEspectrum only compatible with a G1 V Kurucz model.