## Erosion prediction using Graphical Processing Units in
RUSLE-model

** **

**Harri Lilja**

**MTT**

Erosion is a global
problem which each year reduces the amount of arable land left on Earth. Using
erosion prediction models, areas susceptible to erosion can be identified and
reinforced but the prediction process can be slow with detailed data source
like LidarDEM. Using highly parallel graphical processing units to compute all
steps required to predict erosion the calculation efficiency can be
significantly improved.

In the RUSLE-model the
most time-demanding task is the calculation of a flow accumulation (LS-factor).
The LSCL program has been programmed with C++ in OpenCL-framework. The program
includes several algorithms to calculate flow accumulation D8, FD8, FFD8 and
the new anisotropic algorithm developed in MTT. The new algorithm can
topologically sort the elevation data to perform almost ten times faster than
previous algorithms. All other parts of the RUSLE computation map perfectly to
the GPU architecture and are able to complete in less than a second each for a
grids containing 72 million cells.