A study of polarimetric error induced by satellite motion: application to the 3MI and similar sensors

Abstract

Abstract. This study investigates the magnitude of the error\nintroduced by the co-registration and interpolation in computing Stokes\nvector elements from observations by the Multi-viewing, Multi-channel,\nMulti-polarisation Imager (3MI). The Stokes parameter derivation from the\n3MI measurements requires the syntheses of three wide-field-of-view images\ntaken by the instrument at 0.25\u2009s interval with polarizers at different\nangles. Even though the synthesis of spatially or temporally inhomogeneous\ndata is inevitable for a number of polarimetric instruments, it is\nparticularly challenging for 3MI because of the instrument design, which prioritizes\nthe stability during a long life cycle and enables the\nwide-field-of-view and multiwavelength capabilities. This study therefore\nfocuses on 3MI s motion-induced error brought in by the co-registration\nand interpolation that are necessary for the synthesis of three images. The\n2-D polarimetric measurements from the Second-generation Global Imager\n(SGLI) are weighted and averaged to produce two proxy datasets of the 3MI\nmeasurements, with and without considering the effect of the satellite\nmotion along the orbit. The comparison of these two datasets shows that the\nmotion-induced error is not symmetric about zero and not negligible when the\nintensity variability of the observed scene is large. The results are\nanalyzed in five categories of pixels: (1) cloud over water, (2) clear sky over water, (3) coastlines, (4) cloud over\nland, and (5) clear sky over land. The\nmost spread distribution of normalized polarized radiance ( Lp )\ndifference is in the cloud-over-water class, and the most spread\ndistribution of degree of linear polarization (DOLP) difference is in the\nclear-sky-over-water class. The 5th to 95th percentile ranges of Lp difference for\neach class are (1) [ - 0.0051 , 0.012 ], (2) [ - 0.0040 , 0.0088 ], (3)\n[ - 0.0033 , 0.012 ], (4) [ - 0.0033 , 0.0062 ], and (5) [ - 0.0023 , 0.0032 ]. The same\npercentile range of DOLP difference for each class are (1) [ - 0.023 , 0.060 ],\n(2) [ - 0.043 , 0.093 ], (3) [ - 0.019 , 0.082 ], (4) [ - 0.0075 , 0.014 ], and (5)\n[ - 0.011 , 0.016 ]. The medians of the Lp difference are (1) 0.00035, (2)\n0.000049, (3) 0.00031, (4), 0.000089, and (5) 0.000037, whereas the medians\nof the DOLP difference are (1) 0.0014, (2) 0.0015, (3) 0.0025, (4) 0.00027,\nand (5) 0.00014. A model using Monte Carlo simulation confirms that the\nmagnitude of these errors over clouds are closely related to the spatial\ncorrelation in the horizontal cloud structure. For the cloud-over-water\ncategory, it is shown that the error model developed in this study can\nstatistically simulate the magnitude and trends of the 3MI s motion-induced\nerror estimated from SGLI data. The obtained statistics and the simulation\ntechnique can be utilized to provide pixel-level quality information for 3MI\nLevel 1B products. In addition, the simulation method can be applied to the\npast, current, and future spaceborne instruments with a similar design.