difference between backward trajectories using pysplit and web-based hysplit

[mihaejeon]

I ran pysplit on mac with GDAS1 datasets. I got GDAS files from ARL FTP(/archives/gdas1) and generated bulk trajectories using generate_bulktraj as shown below. But I realized that the result is somewhat different with the result obtained from web-based hysplit, although I used the same condition (years, months, hours, altitudes, coordinates, runtime,...). For example, the 1500m trajectory obtained using the Python-based approach shows an increasing trend above 1000 after March 5th, while the web-based 1500m trajectory shows a decreasing trend below 500 after March 5th.

I am curious about the reasons for this discrepancy.