Number of flood disaster estimation in Indonesia using local linear and geographically weighted regression approach

• Main Authors: Mardianto, M. Fariz Fadillah, Sediono, Aprilianti, Novia Anggita, Ardhani, Belindha Ayu, Rahmadina, Rizka Firdaus, Ulyah, Siti Maghfirotul
• Format: Conference Proceeding
• Language: English
• Subjects: Disaster management; Ecological effects; Flood management; Regression analysis; River banks
• ISSN: 0094-243X; 1551-7616
• DOI: 10.1063/5.0042118

Flood is an annual problem in Indonesia. Based on data from Badan Nasional Penanggulangan Bencana (BNPB) or National Disaster Management Agency in 2020, flood had an average incidence of 649 times in all regions of Indonesia from 2011 to 2019. One of the main factors causing this disaster is ecological damage, which is triggered by settlements along riverbanks. Refer to this problem, it is necessary to estimate the number of flood disaster based on the influence of this settlement factor. This is important to prepare appropriate mitigation efforts to reduce the number of disaster and the impact of losses. The estimation was carried out using two types of approaches that are Geographically Weigthed Regression (GWR) and local linear nonparametric regression. This study used secondary data from Badan Pusat Statistik (BPS) or Central Bureau of Statistics that are the number of flood disaster and the number of settlements along riverbanks in 34 provinces in 2018. The results of this study showed that the nonparametric regression method with local linear approach produces better estimation than GWR method in analyzing flood cases. It was based on the R2 value of the nonparametric regression with local linear approach of 51.48%, which is greater than the GWR value of 46.52%; and the MSE value of the nonparametric regression with local linear approach of 24.26, which is less than the GWR value of 32.16. Although the local linear nonparametric regression method had better estimation ability, the use of this method is not enough to estimate the number of flood disaster. This relates to spatial effects that cannot be separated from the phenomenon of flood where disaster at a particular location can cause flood around the nearest location. The results can be used to develop alternative methods by combining GWR and nonparametric regression with local linear approach.