Water moves in a constant cycle of evaporation, condensation, precipitation, runoff, and re-evaporation. The outflow of water is one of the most important components of the water cycle on earth.
It describes the amount of water that falls as precipitation in a specific area and that reaches our lakes and seas through natural runoff via streams and rivers. While high discharge can cause flooding and stagnation, low discharge can be detrimental to the ecosystems that depend on the river.
In both cases, the consequences can be devastating, leading to severe socioeconomic damage and ecosystem destruction. The outflow of water must therefore be predicted as accurately as possible, since it is of crucial importance for water management planning, ensuring and regulating the drinking water supply and managing droughts.
Numerous variables such as the climate and hydrology, the hydraulic properties of the river (flow rate, flow rate, water level), the altitude and the water withdrawal affect the discharge. Combined with increasing uncertainty and hydroclimatic risks, it is becoming increasingly difficult to accurately predict discharge and address vulnerabilities in river systems.
Ozgur Kisi and Christoph Külls from the TH Lübeck, together with an international team, have conducted a study on the suitability of three machine learning methods (CatBoost (CB), Random Forest (RF) and Extreme Gradient Tree Boosting (XGBoost, XGB)) for monthly runoff forecasts using runoff data from three stations in Turkey (Ducurasu, Sutluce and Kale) and satellite precipitation data from the Tropical Rainfall Measuring Mission (TRMM).
In the first part of the study, the researchers predicted the discharge at each station. In the second part, they predicted it at the downstream station using the data from the upstream stations.
The results showed that the TRMM satellite data is not only very accurate and very useful for the monthly runoff forecasts, but also that it significantly improves the ability of the machine learning methods (e.g. CB, RF and XGB).
The results from the study can provide useful information for decision-makers, especially in developing countries where precipitation data are missing or not available for technical reasons. On the original publication by Mojtaba Mehraein, Aadhityaa Mohanavelu, Sujay Raghavendra Naganna, Christoph Külls and Ozgur Kisi at MDPI (Multidisciplinary Digital Publishing Institute): https://www.mdpi.com/2073-4441/14/22/3636
About dr Ozgur Kisi
Ozgur Kisi has been a visiting professor at TH Lübeck since October 2021. The professor came to Lübeck with a grant from the Alexander von Humboldt Foundation. The engineer for hydraulic engineering, hydrology and water management is one of the ten most cited scientists worldwide in his field.
Further information
The department construction