Simulation of the impacts of a sensor’s PSF on mixed-pixel unmixing: adjacency and overall effect

The point spread function (PSF) of optical sensors is a critical source of error in mixed-pixel unmixing, and the statistical characteristics of PSF-induced unmixing errors remain poorly understood. Characterizing these statistical behaviours can clarify the underlying mechanisms through which the PSF affects unmixing procedures and provide a theoretical foundation for developing high-precision models for correcting PSF-induced errors. Based on their distinct underlying mechanisms, PSF-induced artefacts in mixed-pixel unmixing can be classified into two types: non-uniformity effects and adjacency effects. Using Monte Carlo simulations and randomly configured landscape scenarios, this study systematically investigated the independent impact of the adjacency effect and the overall impact of the PSF on mixed-pixel unmixing accuracy. The key findings are summarized as follows: (1) Systematic errors induced by the adjacent response range from −0.32% to 0.32% and exhibit a linear decreasing trend. In comparison, the standard deviation ranges from 0.33‒2.25%, indicating a symmetric parabolic distribution centred at a true abundance of 50%. (2) Both systematic and random error trends caused by the PSF are highly consistent with those generated by the pure adjacency response, whereas the overall error magnitude is lower under the actual PSF response; the systematic error ranges from −0.31% to 0.31%, and the standard deviation of the random error ranges from 0.35‒2.41%. (3) Under the parameter settings considered in this study, PSF-induced systematic unmixing errors were predominantly governed by the adjacency effect, whereas random errors resulted from the combined influence of the adjacency and non-uniformity effects. Furthermore, the magnitude of PSF-related errors depends on both landscape features and sensor response characteristics.