The application of Machine Learning (ML) is an emerging field of research in hydrology and far beyond. ML allows to identify complex relations in large data sets and may therefore allow to address complex problems in hydrology, as well (Shen, 2018). One of these problems comprises the assessment of weathering strength, often referred to as the degree of karstification (Worthington et al., 2017), at carbonate rock landscapes by remote sensing data. In strongly weathered carbonate rock areas, distinct hydrological behaviour (Ford and Williams, 2007) and elevated vulnerability to contamination (Vías et al., 2006) can be expected. The precise definition of their extent is therefore relevant for water management. Despite recent efforts to map carbonate rock regions globally (Chen et al., 2017), map resolutions remain coarse and imprecise at regions with limited data availability and yet no information on the strength of weathering is available.

The aim of this MSc thesis is the first known exploration of ML approaches (1) to improve the mapping of carbonate rock regions and (2) to quantify the strength of weathering using readily available global topographic data products.

The student will extract highly resolved topographic information for all carbonate rock regions obtained from the World Karst Map (Chen et al., 2017). Continental and global topographic information is available through products like the Shuttle Radar Topography Mission (USGS, 2006). Topographical metrics like average elevation, slope, variability of elevation and slope, etc., will be derived for regions where non-carbonate rock regions and carbonate rock regions are close and available at a high precision (Central Europe or Northern America). Within the same regions, river network density maps (Lehner et al., 2008) will be downloaded as proxies for the degree of weathering. The student will use this data and DL approaches to (1) analyse whether topographic variability can be used as proxy to distinguish carbonate rock regions from non-carbonate rock regions and (2) to explore the predictive power of topographic variability to estimate the strength of weathering within those carbonate rock regions. Additional information like the locations of karst springs or discharge variability will be incorporated if necessary. If successful, the trained DL algorithms will be applied at regions where carbonate rock estimates are coarse and unreliable (Northern Asia or Southern America).

The primary challenge of this MSc thesis is the handling and analysis of very large spatial data set. To succeed, programming and data analysis skills are required (Python for the DL application, R or Matlab for data analysis and visualisation). The student id expected to get familiar with the Convolutional Neural Network (CNN, Ding et al., 2017)) method, for instance using TensorFlow (https://cv-tricks.com/tensorflow-tutorial/training-convolutional-neural-network-for-image-classification/).

Junior-Prof. Andreas Hartmann, Dr. Benjamin Mewes (Ruhr-Universität Bochum).

Andreas Hartmann andreas.hartmann@hydrology.uni-freiburg.de Tel. +49 (0)761 / 203-3520

The TensorFlow codes are available at https://cv-tricks.com/tensorflow-tutorial/training-convolutional-neural-network-for-image-classification/ and support when applying them will be provided by Dr Mewes.

English or German

Chen, Z., Auler, A. S., Bakalowicz, M., Drew, D., Griger, F., Hartmann, J., Jiang, G., Moosdorf, N., Richts, A., Stevanovic, Z., Veni, G. and Goldscheider, N.: The World Karst Aquifer Mapping project: concept, mapping procedure and map of Europe, Hydrogeol. J., doi:10.1007/s10040-016-1519-3, 2017.

Ding, A., Zhang, Q., Zhou, X. and Dai, B.: Automatic recognition of landslide based on CNN and texture change detection, Proc. - 2016 31st Youth Acad. Annu. Conf. Chinese Assoc. Autom. YAC 2016, (Subsection C), 444–448, doi:10.1109/YAC.2016.7804935, 2017.

Ford, D. C. and Williams, P. W. D. W.: Karst Hydrogeology and Geomorphology, Wiley, Chichester., 2007.

Lehner, B., Verdin, K. and Jarvis, A.: New global hydrography derived from spaceborne elevation data, Eos (Washington. DC)., 89(10), 93–94, doi:10.1029/2008EO100001, 2008.

Shen, C.: A Transdisciplinary Review of Deep Learning Research and Its Relevance for Water Resources Scientists, Water Resour. Res., 54(11), 8558–8593, doi:10.1029/2018WR022643, 2018.

USGS: Shuttle Radar Topography Mission, 3 Arc Second scene SRTM V2.1, edited by U. of M. Global Land Cover Facility, 2006.

Vías, J. M., Andreo, B., Perles, M. J., Carrasco, F., Vadillo, I. and Jiménez, P.: Proposed method for groundwater vulnerability mapping in carbonate (karstic) aquifers: The COP method, Hydrogeol. J., 14(6), 912–925, doi:10.1007/s10040-006-0023-6, 2006.

Worthington, S. R. H., Jeannin, P.-Y., Alexander, E. C., Davies, G. J. and Schindel, G. M.: Contrasting definitions for the term ‘karst aquifer,’ Hydrogeol. J., 25(5), 1237–1240, doi:10.1007/s10040-017-1628-7, 2017.