Atmospheric Correction - piSOFIT model

piSOFIT is Pixxel’s customized implementation of NASA JPL’s open-source ISOFIT (Imaging Spectrometer Optimal Fitting) model, optimized for Firefly and Honeybee hyperspectral data.

Methodology

• Radiative Transfer Modeling: piSOFIT uses a neural-network-based MODTRAN emulator derived from 6S to model the interaction of solar radiation with the atmosphere and Earth’s surface.

• Optimal Estimation Inversion: The workflow iteratively adjusts atmospheric parameters (e.g., water vapor, aerosols) and surface reflectance to minimize residuals between observed at-sensor radiance and modeled radiance using a Levenberg-Marquardt optimization approach.

• Bayesian Priors & Uncertainty Quantification: Prior information about atmospheric and surface conditions is incorporated into the inversion, providing improved retrieval stability and per-pixel uncertainty estimates.

• Superpixel Segmentation & Empirical Refinement: piSOFIT segments images into spectrally similar superpixels, solves atmospheric parameters at the superpixel level, and applies an empirical line adjustment to refine surface reflectance for each pixel.

For more technical details, visit the Isofit documentation.