Accelerating Geomorphometric Derivative Computation Using GPUs
Identifying underwater features, ranging from natural seabed habitats to anthropogenic structures like shipwrecks, is important for ocean research and mapping. Scientists rely on geomorphometric derivatives, such as slope, aspect, openness, the topographic position index (TPI), and the vector ruggedness measure (VRM) to interpret digital elevation models of multibeam echosounder data. However, isolating specific features often requires fine-tuning derivative parameters, a process that involves repeated computations. On high-resolution datasets, this iterative workflow can be time-consuming, which may hinder the efficient interpretation of seafloor data. In this work, we present a derivative engine implemented in Rust and integrated into the ValidITy software. The system uses a tiling scheme and GPU-accelerated OpenCL kernels to manage memory and compute requirements. This architecture allows for the processing of datasets that exceed available GPU memory and supports interactive parameter adjustment through real-time visualisation. Benchmarking against SAGA GIS shows that the engine achieves speedups of several orders of magnitude for TPI and more than 100 times for VRM and negative openness, supporting a more efficient workflow for seafloor annotation. (Deep Sea Monitoring Reports, DSMR-0004)