Social-media and newspaper reports reveal large-scale meteorological drivers of floods on Sumatra

Flooding is a significant global hazard, causing substantial societal and economic damage, particularly in developing nations and coastal tropical regions. Climate change is expected to exacerbate this risk. The Indonesian island of Sumatra, part of the Maritime Continent (a region of high average precipitation), experiences frequent and devastating floods, impacting millions. While factors like land use and human settlement patterns contribute to flood severity, precipitation intensity and accumulation are paramount. Understanding atmospheric circulation patterns that lead to increased rainfall is crucial for mitigating the adverse effects. The Maritime Continent's precipitation is influenced by a strong diurnal cycle of local convection and large-scale weather patterns such as the Madden-Julian Oscillation (MJO) and CCKWs. The MJO is a large-scale phenomenon propagating eastward at ~5 m/s, while CCKWs are faster (~12 m/s) and can occur within or independently of MJO events. Both can significantly affect rainfall in the region. Sumatra's equatorial location makes it particularly susceptible to these eastward-propagating waves, which drive atmospheric convection and precipitation, leading to flooding. However, a direct link between these large-scale disturbances and specific flooding events in the region has not been definitively established. This study aims to address this gap by analyzing flood data from multiple sources and meteorological data to determine the contribution of these large-scale atmospheric phenomena to flooding events in Sumatra. Accurate prediction of these events is critical for reducing societal impacts, improving upon current early warning systems which rely on limited real-time data and lack long-range forecasting capabilities.

Existing literature highlights the significant impact of floods globally, particularly in vulnerable regions. Studies have shown the increasing vulnerability of coastal cities in tropical regions to flood damage due to a combination of socioeconomic and natural factors, exacerbated by climate change. The Maritime Continent, with its high average precipitation, experiences frequent flood events. Previous research has investigated the influence of various atmospheric phenomena on precipitation in this region, including the MJO and CCKWs. The MJO, a large-scale intraseasonal oscillation, has been linked to enhanced precipitation in Indonesia. CCKWs, another significant mode of variability, propagate much faster than the MJO and have been shown to directly impact precipitation and diurnal convection over the Maritime Continent. While studies have highlighted the individual roles of these weather patterns, their direct link to specific flooding events in areas like Sumatra remained unclear, necessitating the present study's focus on establishing this crucial connection. This study leverages past research on MJO and CCKW influences on precipitation to create a predictive model. Studies on social media for detecting floods show promise in supplementing existing datasets to provide a more comprehensive picture.

This study uses a synergistic approach combining flood data from multiple sources with meteorological data analysis. Three independent flood datasets were used: (1) a database developed from Twitter messages about floods in Sumatra between 2014 and 2018, geoparse using TAGGS algorithm to identify Sumatran messages which were then binned within Sumatra's sub-regions; (2) a database compiled from local newspaper reports, also binned in the same way for consistency; and (3) a database from Indonesia's National Board for Disaster Management (BNPB). Flood onset and termination times were determined using a surge in tweets or newspaper reports, considering floods within 3 days as a single event. Meteorological data included precipitation estimates from the TRMM and GPM satellite missions (3B42v7 product, providing 3-hourly maps on a 0.25° x 0.25° grid), rain gauge data from Sumatra (three stations near Padang), 850 mb zonal winds from ECMWF's ERA5 reanalysis, and the Realtime Multivariate MJO index (RMM) to assess MJO strength and location. Precipitation anomalies associated with CCKWs were determined using Fourier space wavenumber-frequency filtering, and CCKW trajectories were calculated based on filtered precipitation anomalies within the equatorial band. Seasonal precipitation was analyzed using a 90-day running average low-pass filter of area-averaged 5-day precipitation accumulation estimates. Consistency between flood databases and meteorological data was maintained through regional averaging of meteorological data. CCKW trajectory analysis used base points at 90°E, tracking robust trajectories continuous between 80°E and 110°E (indicating initiation at least two days before reaching Sumatra).

The study found substantial agreement among the three independent flood databases, despite some regional variations likely due to factors like internet access and media preferences. Analysis revealed that most floods in West Sumatra (a region experiencing the highest number of floods), coincided with spikes in precipitation and increased CCKW activity. Eight out of nine major flood events (present in all three databases) were associated with 5-day precipitation accumulation exceeding 50 mm, with seven showing rapid increases in accumulation before or during the flood. Many floods occurred during or immediately after robust CCKW passages, even when the MJO was not in its active phase over Indonesia, indicating CCKWs' independent importance in generating flood conditions. A case study of a major flood near Padang demonstrated that even when rainfall was not exceptionally high at a local scale, the larger-scale precipitation anomaly associated with the CCKW was the main contributor to the flood. Analysis of all floods across Sumatra during 2014-2018 showed that while a substantial number (34-46%) occurred outside of active MJO phases, over 91% were associated with some CCKW activity, and over 63% began during strong CCKWs. Approximately 45% of robust CCKW trajectories (between 80°E and 110°E) were associated with flooding periods in Sumatra. This highlights that CCKWs serve as a significant and frequently independent predictor of flood events across Sumatra. Seasonal flood patterns largely align with the annual movement of the Intertropical Convergence Zone. While MJO activity can increase precipitation, only about 28% of Sumatra floods were directly preceded by favorable MJO conditions, whereas nearly half of the flood periods had a well-defined CCKW precursor at least two days in advance.

The study's findings directly address the research question by demonstrating a strong link between large-scale equatorial disturbances (primarily CCKWs) and flooding events in Sumatra. The significant number of floods linked to CCKWs (over 90%) and the frequent occurrence of CCKWs without active MJO phases highlight the CCKWs' critical role independent of other factors. The use of multiple flood databases enhances the reliability of the results. The study’s findings have significant implications for improving flood prediction in Sumatra. The predictability offered by CCKWs (lead time of 3-5 days) can be leveraged alongside MJO predictions to develop more accurate and timely flood warning systems, which is particularly crucial given the vulnerability of Sumatra's population to flood risks. The conservative nature of the CCKW estimates suggests the true impact of CCKWs on flooding could be even higher.

This study establishes a strong correlation between CCKWs and flood events on Sumatra. The use of social media and newspaper data significantly enhances the accuracy and completeness of flood records. The high percentage of floods associated with CCKWs (over 90%), independent of the MJO's active phase in many cases, suggests significant potential for improving flood prediction. Future research should investigate the interaction between CCKWs, off-equatorial vortices, and equatorial Rossby waves. Improved models that consider the interactions of various equatorial modes and better CCKW prediction methods are needed to create even more effective flood warning systems in Sumatra and similar regions.

The study's time frame (2014-2018) is relatively short. The geoparsing of Twitter data, while utilizing a robust algorithm, remains an approximation and might introduce some error. The underestimation of precipitation rates in the TRMM3B42v7 data, especially over complex topography, might affect the quantification of CCKW impacts. Future studies incorporating a longer dataset, improved geoparsing techniques, and higher-resolution precipitation estimates would further refine the findings and improve the robustness of the CCKW-flood correlation. This study focused primarily on flood initiation; future research should examine CCKW's influence on precipitation enhancement after a flood has started.