Flood Mapping Using Twitter Crowdsourcing Data with Rainfall Data Analysis (Case Study: Jakarta, January 2020)

Jakarta, the capital of Indonesia, experiences frequent flooding every year. In the event of natural disasters such as floods, the government is obliged to implement a series of disaster emergency responses, including disaster modeling. Disaster modeling can be done using various data sources, but field survey data sources, remote sensing, and aerial photographs are considered less efficient. Field surveys take longer, while remote sensing and aerial photography are limited during the rainy season due to high cloud intensity. Currently, social media, particularly Twitter, is receiving significant attention from various groups as a data source for flood modeling. This research aims to construct a spatial database, conduct flood modeling, assess achieved accuracy levels, and examine tweet data in relation to rain events in Jakarta. The Kernel-Based Flood Mapping Model method is used for flood modeling, which utilizes DEMNAS data, river distribution data, administrative boundaries, water depth data at each river floodgate, tweet data, BPBD flood area maps, and surface observation rainfall data provided by BMKG. The accuracy of the flood modeling results was assessed using the overall accuracy method against all the obtained tweet data and BPBD flood area maps. A regression test was conducted to determine the relationship between rainfall and tweet data related to flooding in Jakarta. The results showed that out of 12,345 tweets, 149 could be compiled into a database for modeling. The flood modeling results indicate an accuracy rate of 70% based on the total flood points and 57% based on BPBD flood zones, which are considered low. The regression test indicates that there is no significant relationship between flood points and January rainfall data. The regression test values for rainfall at the flood location and the number of tweets is 0.020822 and for flood depth is 0.049214. This suggests that the rainfall variable has only a 2% effect on the number of tweets and a 4.9% effect on the depth of the flood, with other factors influencing the remainder.