Machine Learning for the Development of Tools for Planning of Climate Adaptation

19:00 Tuesday 28 May




Henrik Vest Soerensen (Denmark) 1; Rikke Nan Valdemarsen (Denmark) 1

1 - Coast to Coast Climate Challenge

Planning climate adaptation initiatives require valid data and tools for municipalities and utilities in order to form an overview of which climate challenges are present currently and in future. In the Danish climate adaptation project Coast to Coast Climate Challenge (C2C CC) with 31 partners, we develop tools precisely for the partnership that the partners need in their planning of climate initiatives. Central Denmark Region is the facilitator in the joint project and has among others the task to ensure adequate knowledge in the collaboration. Thus the partners can create good and innovative climate adaptation solutions with an added value for the citizens.

Municipalities and utilities are challenged by the variation of precipitation with more and intensive occurrences, and more frequent periods of persistent rain. It leads to floodings on land and in the cities from creeks and larger developed areas in the cities. The increased precipitation, also influence the ground water table which in periods are considerably above the average level.

In C2C CC, we develop a combined tool, consisting of surface and groundwater models and a map of torrential rain storms, showing where the water compiles in dips at larger cloudburst incidents. Especially, as regards the groundwater, the tool must show the water table close to the ground. Here, we are challenged by the fact that the data density is very small at several places in the geography.

Therefore, we have chosen to base the development of the groundwater part of the tool on data driven models (“machine learning”) where a lot of data types about geology, water tables etc. are elaborated in decision trees, and a guesstimate on the groundwater table in a given area is provided as output. The advantage of this approach is that the tool can bring knowledge from data intensive areas to areas with less data (e.g. the open farmland). Also, the tool is believed to be very adaptive to new information and data collected on a local scale. The performance of the tool will be tested against detailed models for local areas established in other projects like TOPSOIL, KIMONO, FODS 4.1 or CLIWAT.

The first results from the method show very impressive outputs. The water table, even in data scarce areas, can be identified for certain. The development of the tool will be finalised in the course of 2018, and will be implemented at the partners in the beginning of 2019.