| dc.description.abstract | Climate change has impacted the global water cycle and has changed how streamflow behaves in Norway. Knowledge about high and low flows are important to be prepared for future changes, such as changes in water availability for humans, electricity production, agriculture and more. This thesis investigates historical changes (01.01.1991-31.12.2019) in seasonal high and low flow in Norway and their co-occurrence, and uses machine learning to identify which catchment characteristics, climate indices or other trends in seasonal high and low flow are important predictors to explain these trends. Discharge data from the Norwegian Water Resources and Energy Directorate (NVE) is used to calculate trends in high and low flow. Low flow was divided into two periods: summer low flow (June-September) and winter low flow (October-May), and high flow was divided into two periods: spring high flow (March-August) and autumn high flow (September-February). The trends were calculated over smoothing intervals of minimum/maximum discharge over 7 and 30 days. Precipitation, temperature, evaporation and snow data from seNorge is used to create climate indices while catchment characteristics are from NVE. The machine learning methods decision tree and random forest were applied to find the most important predictors for each seasonal trend. The trend results showed a clear divide in trend direction between southern and northern Norway across all seasonal trends. Southern Norway displayed only significantly increasing trends for summer low flow and autumn high flow, where as some decreasing trends showed up in the western part of southern Norway for winter low flow and spring high flow. Northern Norway displayed mostly significantly decreasing trends across all seasons, except for Troms and Finnmark which had increasing trends in winter low flow. The increasing trends in southern Norway may be because southern Norway experienced increased precipitation across all seasons during the research period, while the decreasing trends in northern Norway may be due to increasing winter temperatures and less snow melting in spring and summer. Increasing winter temperature may lead to more frequent melting during the winter period in Troms and Finnmark, explaining the increasing trends in winter low flow. The machine learning highlighted latitude and longitude as important predictors, within top tree for at least one machine learning model for all trends except AM(7) winter low flow. Precipitation, temperature and streamflow often showed up as important as well. Other seasonal trends were not often used as predictors, probably because a catchment displaying a significant trend for one season did not always show trends in the other seasons. This thesis provides insight into trends in high and low flow their co-occurrence during the current climate period, and main predictors explaining the observed changes. | eng |