The use of filters based on the Fast Fourier Transform applied to DEMs for the objective mapping of karstic features

Abstract

The suitability of Fast Fourier Transform (FFT) filters to delimit the objective location of geomorphic references is analysed for the case of karstic landscapes, where altitudes feature marked differences. From aerial images and a Light Detection and Ranging (LiDAR) derived high-resolution digital elevation model (DEM), a digital database was created, segregating two geomorphic domains according to scale: domain A (macroforms) and domain B (mesoforms), both subdivided in zones of positive and negative relief polarity. In order to minimise the producer error, the correspondence between the DEM morphology and the features mapped was generalised and certified in a test area, reducing ground truth uncertainty. The efficiency of FFT filters was compared against the most commonly referenced in the literature, such as convolution and openness, in terms of computation cost and geometric position. Two types of FFT filters were created modifying the radius of the high pass mask: short radius filters (appropriate to objectivise macroforms); and middle-large radius filters (appropriate for mesoforms) like convolution and openness. The FFT geomorphic reference obtained offers similar geometric patterns to other filters, but reduced computation times. Moreover, the filters reduce the positional uncertainty of geomorphic contacts, without changing the general altitude trend. The generation of FFT filtered references, combining short and middle-large radius, permits objective mapping of karstic landscapes using a DEM.