I've used HYSPLIT to run back trajectory for 10 years.

If you run back trajectory, starting height is the arriving altitude.

In general, people run back trajectory with starting height of 0.5 PBL, some run with fixed starting height like 500 or 1000m agl, if you your receptor site is near surface, 500 or 1000 m should work, but it still depends on your site location, such as inland, coastal, high elevation.

You can find in the attachment how to run the back trajectory with starting height of 0.5PBL

thank you,

First of all, I've chosen 500 and 1000 m agl as starting height for 4 years in order to check for significant differences in trajectory calculation results. But no remarkable results have be done.

If the starting height is equivalent to 0.5 PBL. how can I calculate the altitude of PBL? HYSPLIT can calculate only the mixing layer height. 

I've done this using 500/1000 m for more than 10 sites in US including urban, rural, coastal, high elevations. I will say no significant difference for most cases. In general, PBL is about 1-2 km for most areas in US during daytime, but it might collapse to less than 100m for some regions. 

PBL=mixing layer height. All you have to do is to add a setup_cfg_frac_pbl.txt (see the file I attached in previous answer) in your working folder. 

We tend to use heights of between 200-500m for our back trajectory analyses. These elevations are well within the convective boundary layer during the day, but above the nocturnal boundary layer under most conditions. This means that, if you are trying to look at source/fetch regions corresponding to measurements made near the surface, then relating back trajectories to observations made between about 10am and 5pm each day (when mixing is fairly strong, and near surface measurements are representative of most of the CBL) is OK. At night, however, under all but the most windy and well mixed of conditions, any near surface measurements you make are likely to be partially (or completely) decoupled from what is happening aloft in the residual layer (200-500m). Under these situations, your measurements will have a fairly restricted "footprint" and the trajectories you calculate may not be related to this fetch at all (particularly in regions of uneven terrain where nocturnal drainage flows may occur under stable conditions). Cheers, Scott.

Would the back-trajectory analysis at 100 m AGL be justifiable when looking for transport across regional(200-500 km). According to me, it would simulate the near surface effects better. Can someone please confirm on this and let me know if there is some concrete literature supporting this.