Geometric Modeling and Validation of ALOS/PRISM Imagery and Products

Abstract

PRISM is a panchromatic radiometer carried onboard of the new generation Japanese remote sensing satellite ALOS (Advanced Land Observing Satellite). It has three optical systems for forward, nadir and backward views with 2.5 meter spatial resolution. Multiple Linear Array CCD chips are located on the focal plane of each camera, along one across-track line. Three PRISM images per scene are acquired almost simultaneously in forward, nadir and backward viewing angles in along-track direction.The photogrammetric processing of PRISM imagery has special requirements due to the Linear Array CCD sensor structure. As a Member of the ALOS Calibration/Validation Team, we have implemented new algorithms for the geometric processing of the PRISM images, in particular for the interior orientation and self-calibration. In addition, we have refined our physical sensor model according to the multiple optical camera heads of the sensor. Our rigorous model for the PRISM sensor is based on a modified bundle adjustment algorithm with the possibility to use two different trajectory models: the Direct Georeferencing Model with Stochastic Exterior Orientation Elements (DGR) and the Piecewise Polynomial Model (PPM). The given trajectory values are used as stochastic unknowns (observed values) in both approaches in the adjustment. For the self-calibration of the PRISM imagery, we have initially defined 30 additional parameters for the 3 cameras. The parameters are described in accordance with the physical structure of the PRISM imaging sensors. In this paper, PRISM images acquired over three recently generated testfields are used for calibration and geometric validation purposes. We have tested our rigorous sensor model, both with the DGR model and the PPM, using self-calibration, in all testfields. In addition, we have evaluated the accuracies of the RPCs (Rational Polynomial Coefficients) provided by JAXA/RESTEC in two of the testfields, using three methods: the direct georeferencing (with forward intersection), 2D affine transformation with 6 parameters, and translational correction with 2 shift parameters.PRISM level 1B1 images are used in all tests. The rigorous model produces RMSE values, computed from check points, of about 1/2 pixel in planimetry and 1/3-1/2 pixel in height. Both trajectory models provide sub-pixel accuracy for georeferencing and point positioning in an optimal GCP configuration. However, the PPM requires a higher number of ground control points to obtain stable adjustment results. The direct georeferencing accuracy obtained from the given RPCs is fairly good, and results in 2.5 pixels RMSE in the worst case. The RPC triangulation results show differences between the testfields, requiring different bias-correction terms and GCP distributions for optimal results.