Monthly Rainfall Prediction Using Statistical Downscaling with Combination of Grid Boxes and Adaptive Neuro Fuzzy and Inference System in Lombok

• Published in: IOP conference series. Materials Science and Engineering 2019-11, Vol.676 (1), p.12009
• Main Authors: Safril, A, Ulfiana, A
• Format: Article
• Language: English
• Subjects: Air temperature; Boxes; Early warning systems; El Nino; Emergency communications systems; Geopotential; Geopotential height; Inference; La Nina; Lead time; Natural disasters; Rainfall; Risk management; Root-mean-square errors; Sea level; Singular value decomposition; Statistical prediction; Variability
• ISSN: 1757-8981; 1757-899X
• DOI: 10.1088/1757-899X/676/1/012009

Lombok is an island in Indonesia that has more people live in this area. Flood is a natural disaster that cause severe impact on society. To manage the risk of natural disaster, seasonal prediction of rainfall for early warning system are needed. However, prediction is difficult to predict due to El Niño and La Niña. Seasonal prediction of rainfall using Global Circulation Model (GCM) is useful to capture rainfall variability, but has a lack information because it has coarse of grid resolution (more than 200 kilometers). This model cannot provide in the local area so statistical downscaling model is used to get detail information. To optimize the result of prediction from global model, a method to capture the extreme condition is needed. Taking a best predictor that will be used to predict the rainfall is done to get the high performance of model used Singular Value Decomposition (SVD). ANFIS (Adaptive Neuro Fuzzy and Inference System) is a model prediction to capture the rainfall variability. Predictor is chosen based on the physical condition of atmosphere and ocean that have impact on Lombok Region. Selected area of predictor (atmospheric variable) are grid boxes in the Maritime Continent (80°E - 150°E and 12.5°N - 12.5°S) and Nusa Tenggara (105°E - 120°E, 10°S - 0). Eight variable predictors are used for prediction, they are air temperature at 2 m (T2M), geopotential height at 500 mb (Z500), zonal and meridional wind at 850 mb (U850 and V850) and at 200 mb (U200 and V200), sea level pressure (SLP), and air temperature at 850 millibar (T850). Both predictor variables (grid box combination) are used to predict for lead time one month. The result of research shows that rainfall prediction can capture the rainfall variability (correlation 0.70) with RMSE (Root Mean Square Error) is 69.