The 2015-2017 Cape Town Drought: Attribution and Prediction Using Machine Learning

• Published in: Procedia computer science 2018, Vol.140, p.248-257
• Main Authors: Richman, Michael B., Leslie, Lance M.
• Format: Article
• Language: English
• Subjects: Attribution; Drought; Global Climate Change; Machine Learning; Support Vector Machines; Variable Selection
• ISSN: 1877-0509; 1877-0509
• DOI: 10.1016/j.procs.2018.10.323

Cape Town was declared a disaster area after the worst drought in almost a century, following its driest three consecutive wet seasons (April 1-October 31), in 2015 -2017. Cape Town’s drought was extreme, with “zero day” water storage months away, causing severe water rationing to Cape Town’s ~3.8 million population. The crisis extended into surrounding farmlands, as agriculture is vital for the region’s economy. Possible drought causes are numerous and, aside from the decreasing wet season precipitation, the effects are exacerbated by the increasing population with associated water demand, greater agricultural acreage and land surface changes. As rainfall decreases, water management becomes critical, requiring predictions for future rainfall. Possible climate drivers associated with available Cape Town precipitation and temperature include: The Southern Annular Mode, Atlantic Meridional Mode, Indian Ocean Dipole, an Integrated Southern Hemisphere temperature index and several El Niño indices. Several variable selection techniques suggest signals in both the Atlantic and Indian Oceans contribute to Cape Town droughts. Machine learning techniques are applied to these drivers for the first time and provide encouraging predictive skill levels. Results suggest that machine learning holds promise for adapting to drought by managing water resources in Cape Town and, more generally for global locations depending solely on rainfall under a warming climate.