Flooding remains one of the most devastating and costly natural disasters. As flooding events grow in frequency and intensity, it has become increasingly important to improve flood monitoring, prediction, and early warning systems. Recent efforts to improve flash flood forecasts using deep learning have shown promise, yet commonly-used techniques such as long short term memory (LSTM) models are unable to extract potentially significant spatial relationships among input datasets. Here we propose a hybrid approach using a Convolutional LSTM (ConvLSTM) network to predict stream stage heights using multi-modal hydrometeorological remote sensing and in-situ inputs. Results suggest the hybrid network can more effectively capture the specific spatiotemporal landscape dynamics of a flash flood-prone catchment relative to the current state-of-the-art, leading to a roughly 26% improvement in model error when predicting elevated stream conditions. Furthermore, the methodology shows promise for improving prediction accuracy and warning times for supporting local decision making.
3 September 2019 in Frontiers in Environmental Science
Information is a critical resource in disaster response scenarios. Data regarding the geographic extent, severity, and socioeconomic impacts of a disaster event can help guide emergency responders and relief operations, particularly when delivered within hours of data acquisition. Information from remote observations provides a valuable tool for assessing conditions “on the ground” more quickly and efficiently. Here, we evaluate the social value of a near real-time flood impact system using a disaster response case study, and quantify the Value of Information (VOI) of satellite-based observations for rapid response using a hypothetical flooding disaster in Bangkok, Thailand. MODIS imagery from NASA's Land, Atmosphere Near real-time Capability for EOS (LANCE) system is used to produce operational estimates of inundation depths and economic damages. These rapid Earth observations are coupled with a decision-analytical model to inform decisions on emergency vehicle routing. Emergency response times from vehicles routed using flood damage data are compared with baseline routes without the benefit of advance information on road conditions. Our results illustrate how the application of near real-time Earth observations can improve the response time and reduce potential encounters with flood hazards when compared with baseline routing strategies. Results indicate a potential significant economic benefit (i.e., millions of dollars) from applying near real-time Earth observations for improved flood disaster response and management.
Flood events pose a severe threat to communities in the Lower Mekong River Basin. The combination of population growth, urbanization, and economic development exacerbate the impacts of these events. Flood damage assessments, critical for understanding the effects of flooding on the local population and informing decision-makers about future risks, are frequently used to quantify the economic losses due to storms. Remote sensing systems provide a valuable tool for monitoring flood conditions and assessing their severity more rapidly than traditional post-event evaluations. The frequency and severity of extreme flood events are projected to increase, further highlighting the need for improved flood monitoring and impact analysis. In this study we integrate a socioeconomic damage assessment model with a near real-time flood remote sensing and decision support tool (NASA’s Project Mekong). Direct damages to populations, infrastructure, and land cover are assessed using the 2011 Southeast Asian flood as a case study. Improved land use/land cover and flood depth assessments result in rapid loss estimates throughout the Mekong River Basin. Results suggest that rapid initial estimates of flood impacts can provide valuable information to governments, international agencies, and disaster responders in the wake of extreme flood events.
Rising sea levels increase the probability of future coastal flooding. Many decision-makers use risk analyses to inform the design of sea-level rise (SLR) adaptation strategies. These analyses are often silent on potentially relevant uncertainties. For example, some previous risk analyses use the expected, best, or large quantile (i.e., 90%) estimate of future SLR. Here, we use a case study to quantify and illustrate how neglecting SLR uncertainties can bias risk projections. Specifically, we focus on the future 100-yr (1% annual exceedance probability) coastal flood height (storm surge including SLR) in the year 2100 in the San Francisco Bay area. We find that accounting for uncertainty in future SLR increases the return level (the height associated with a probability of occurrence) by half a meter from roughly 2.2 to 2.7 m, compared to using the mean sea-level projection. Accounting for this uncertainty also changes the shape of the relationship between the return period (the inverse probability that an event of interest will occur) and the return level. For instance, incorporating uncertainties shortens the return period associated with the 2.2 m return level from a 100-yr to roughly a 7-yr return period (∼15% probability). Additionally, accounting for this uncertainty doubles the area at risk of flooding (the area to be flooded under a certain height; e.g., the 100-yr flood height) in San Francisco. These results indicate that the method of accounting for future SLR can have considerable impacts on the design of flood risk management strategies.