Wet area maps

The work with developing wet area maps within WAMBAF is advancing. All of our demo areas now have high resolution wet area maps. The project has developed and tested a new map of Swedish stream channels.

The new map over Swedish stream channels is much more accurate than anything current and forms an integrated drainage network. This means that it's now possible to follow small stream channels from anywhere in the country all the way to the Baltic Sea.

Tests are done in Sweden

The map also follows the inundated channel that is present in the landscape instead of a straight line somewhere nearby the actual stream. We are using Sweden as a test bench but this approach will likely work in all countries surrounding the Baltic Sea. There is still some room for improvement but this new map will be a powerful tool in managing small streams in the Baltic region.

We also have exciting results from the development of a new type of wet area map. Previously we have worked with topographical indices, such as the depth to water maps we created for all demo areas. These maps have proven to be useful in planning forestry operations in Canada and Sweden.

These methods are easy to implement on large scales but are also static and do not take differences in spatial runoff patterns or soil textures into account. Therefore we have been working with a different approach involving Machine learning. Machine learning is a datamining technique that finds patterns in datasets and use these patterns to predict new data. It works by using field data and environmental data such as maps of runoff patterns and quaternary deposits or local topography to train a "machine".

Accurate and cost effective

In our case we used the Swedish national forest inventory as training data and multiple other data sources for environmental variables. For example, we used all maps of quaternary deposits from the Swedish geological survey and 30 years of runoff data from the Swedish meteorological and hydrological institute together with multiple wetness index maps derived from airborne laser scans. The result is a trained "machine" that can be used to map wet areas with unparalleled accuracy.

This is a new cost effective way to map wet soils on high resolutions across large areas. However, the most fascinating result from this research is that there is sufficient information in a digital elevation model alone to produce accurate maps of wet areas. Including additional factors such as quaternary deposits and runoff only marginally improved the accuracy of the maps.

Will be published in scientific publications

This is the first time machine learning has been used to map wet areas on this high resolution and large scale. Both these studies will result in scientific publications over the coming months.

  • Last Updated: 6/15/2018