Flood Disaster Prediction Model Using Long Short-Term Memory (LSTM) in Pekalongan, Central Java.

Abstract

Pekalongan is located in the northern part of Java Island, directly adjacent to the sea in the north. Natural disasters that often occur in Pekalongan are floods, especially in the north of the area, which has a height of 0 meters above sea level. In addition, Pekalongan also has a relatively low land slope of around 0 – 5%, which makes it challenging to distribute water and construct drainage. This study aims to be able to perform predictive modeling of flood-prone areas for the next five years. This study used eight parameters: rainfall, elevation, slope, distance to the river, distance to the sea, groundwater table to surface, soil type, and land use. This research used the Long Short-Term Memory (LSTM) method to predict rainfall parameters using the Python programming language with Jupyter Notebook software. Later, the data will be used as training and test data. Training data testing and tests are conducted to find the minimum failure or error value. The weight scoring method is carried out on each parameter to indicate areas with a high flood vulnerability level. The results showed that Pekalongan has a medium to very high vulnerability level, with a dominant high vulnerability level. The very high level of vulnerability is prevalent in the northern part of the research area, which is directly adjacent to the sea or in the North Pekalongan District. Floods that occur in the northern part of the study area are not only due to high levels of rainfall but can also occur due to the inflow of seawater towards the mainland resulting from high tides and high sea waves. The southern region of the study area has a smaller vulnerability level than the northern region, which has a medium to high vulnerability level.Keywords: Flood ∙ Hazard ∙ Precipitation ∙ LSTM ∙ Rainfall