Implications of the point spread function for downscaling and data fusion in remote sensing
The "support" is a geostatistical concept which has been studied over many years. It means the space on which an observation is made, or the space on which a datum is represented. In remote sensing, the support is often equated to the pixel. In remote sensing, and for images in general, the support has a special significance because, roughly speaking, it characterises the whole sampling framework. The sampling framework is often neglected in research on spatial data, and it is commonly a missing element in the modern world of data science and spatial data science. This is a critical omission because data are a function of two things: reality and the sampling framework. They are not reality itself although much data analysis treats them as if there were. In geostatistics, downscaling methods have been developed based on the support - in remote sensing treated as the pixel. This is an advance on much data science taking place currently which ignores the support. However, interestingly the support in remote sensing isn't actually the pixel at all - it is a centre weighted function called the point spread function or "PSF". Thus, the PSF represents an opportunity to improve the way that we integrate spatial datasets and handle scale in spatial data. Using examples from research undertaken with Prof. Qunming Wang and others, I will explore the effect of properly representing the PSF on geostatistical downscaling and spatial data fusion. The results have implications potentially for all remote sensing images ever produced!
- When: 12 November 2020, 15:00
- Who: IAMG Distinguished Prof. Peter M. Atkinson (Lancaster University)
- Where: Online, BigBlueButton