Because of the nature of how ASTER data are stored in the HDF file (files with different spatial resolutions), I think it is better to orthorectify the data prior to processing it in FLAASH. This is because you will need to orthorectify each of the datasets (VNIR and SWIR) separately, then stack them into a single file for processing in FLAASH. When orthorectifying the data, use a nearest neighbor resampling to retain the spectral values. It is not uncommon to see negative reflectance values in some pixels for any dataset, including ASTER. Some pixels are not well modelled in FLAASH due to various things such as poor response from the sensor or over-saturation of a pixel. You should not have *most* of the pixels negative though. For negative pixels, you can generate a mask of values of less than 0 to use in other processing routines.