The following information is taken from Waters et al. (2016), which should be cited appropriately.
Weight images are variance maps. For single epoch PS1 Warp Images, these contain the read noise, Poisson noise, and other noise sources. For PS1 Stack Images, the noise is propagated from the individual input warps. These weight images can be used for estimates of the uncertainties in the photometry. We note that the deprojection of the chip images into the warps correlates the pixels, introducing covariance. Therefore the derived uncertainties might underestimate the true noise.
Description
Weight images contain the estimated pixel-level variance for images, represented as 32-bit floating point values. They are automatically generated from images during processing by the Pan-STARRS Image Processing Pipeline photometry software, psphot, and are based on detector characteristics (such as gain and read noise) and observed signal levels. The variance model includes expected Poissonian variance, proportional to measured signal scaled by gain, as well as additive noise terms such as read noise. For stack images, the total variance is a combination of the multiple input images. There may be a variable number of input measurements in this case due to masking and dithering.
These weight images are used throughout psphot, including during the peak detection process to test the validity of candidate sources. See Magnier et al. (2016) for more details about the psphot software.
Example of a warp weight image for a skycell of exposure o5104g0266o, as seen in Waters et al. (2016). This variance map shows streaks where higher flux levels were found and other corrections were made.