This is probably a matter of calibration. If this is not the raw data (actual voltage curves), look at that.
As an area below 0 is not possible, it has to be a calibration issue, either in the machine, or in the software.
the -H plots are about the maximum height of the voltage spike in the side or forward light detector, while the -W plots are about the width (using some definition of flank-detection, so variation prone).
The area is also computed in various ways: using something like (H*W)/c (with c something like 2) or with an actual computation of the area of the spike, but again this has all kinds of predefinitions in it (what's the beginning/end of a spike, what's the background-level, ...) Some of these predefinitions you can alter in software, some are hardcoded or even part of the functionality of the hardware, so it might be easy to impossible to get rid of the negative values.
The FSC-H values corresponding to the FSC-A values are in your plot, and positive, so the most obvious solution (a bad algorithm that calculated Area as Height times Width times Magic and did not account for negative Heights (which depending on spike definition, you might get)) is out.