Postingan ini mestinya dari dulu-dulu yaa.. Tapi tidak apa-apalah, for you to know, atau barangkali ada yang lagi cari-cari ide tentang topik thesis di seputar bidang teknik sumber daya air, ini bisa jadi salah satu alternatif (walau di Indonesia masih jauh kali ya kalau mau diterapkan).
Idenya tentu saja dari si promoter, dan supervisor sebagai pelaksana dan yang turut mengembangkan ide. Tulisan abstract ini tentunya setelah melalui editing dan koreksi dari promoter dan supervisor ya, mengingat tata bahasa Inggris saya yang tidak bagus-bagus amat (tapi bagus saja :D).
So, here I proudly present my thesis abstract, taken from my master dissertation in partial fulfilment of the requirements for the Degree of Master of Science in Water Resources Engineering (in Katholieke Universiteit Leuven and Vrije Universiteit Brussel). Fyi, the defense was in September 2012.
Uncertainty Estimation and Communication for Probabilistic River Flood Forecasting – Case Study of Demer River, Belgium
(by: Erin Priandini)
Promoter: Prof. P. Willems
Supervisor: Ir. N. Van Steenbergen
To minimize negative impact of floods, a real-time flood-forecasting system in combination with early warning system is used to predict water levels and discharges along the rivers in Flanders, a region of Belgium. It requires prior and posterior information as input data. The current output is deterministic, which could have strong implications if the real observation highly deviates from the forecast. Considering that there is always uncertainty in the forecast, a probabilistic approach is applied in order to complement the forecast output with an uncertainty estimate. A statistical analysis is used to calculate the water level uncertainty by analyzing the historical forecast errors using a non-parametric approach. The forecast residuals are divided into a number of classes based on combination of the different factors, e.g. forecasted water level, time horizon, measured rainfall and the slope of the water level. The percentile values of residuals are calculated for each combination of classes, where the distribution of the forecast residuals is assumed constant over the class. The median value of percentile is used to correct the systematic deviation. The residuals are placed in an error matrix. To investigate the contribution of the different factors to the water level forecast uncertainty, the methods are analyzed in three different scenarios of error matrices. The water level forecast uncertainty is then calculated based on interpolation in the error matrix, and is presented in the form of confidence intervals through time. The exceedance probabilities of the alert and alarm levels are also calculated based on confidence intervals estimation.
Probabilistic information may confuse the users and could lead to misinterpretation. Considering that the information is very useful for decision-making process, hence presenting the result of river stage uncertainty in a useful form is as important as the output reliability itself. Different alternatives for the presentation of the uncertainty results are therefore worked out. The preference of an appropriate uncertainty communication form depends on the needs and to whom this information is addressed.
Keywords: uncertainty, probabilistic forecast, non-parametric method, uncertainty communication