5:15 PM - 7:15 PM
[HDS09-P07] Evaluation of Uncertainty in Disaster Prevention Weather Information and Its Potential Use as a Basis for Evacuation Decisions
Keywords:Disaster Prevention Weather Information, Disaster Prevention Literacy, Flood Risk
In recent years, Japan has been experiencing increasingly severe and frequent weather-related disasters, with flood disasters continuing to claim lives. To reduce casualties, the Japan Meteorological Agency and the Cabinet Office have introduced a five-level heavy rain warning system that links disaster prevention weather information—such as evacuation advisories, warnings, and alerts—to specific actions residents should take, encouraging intuitive evacuation behavior.
Ensuring effective evacuation requires accurate weather forecasting and the timely issuance of disaster prevention weather information. While weather forecasting technology has improved over the years, it is not yet flawless. When weather forecasts are inaccurate, the reliability of disaster prevention weather information decreases, reducing the lead time (the time available for evacuation) and increasing the risk of delayed evacuation. To secure sufficient lead time, issuing information at an early stage when the certainty of a disaster is low can lead to more false alarms, potentially diminishing public trust in such information.
By analyzing the uncertainty in disaster prevention weather information issued for each region, we can identify areas where forecast reliability is particularly low. This knowledge allows us to propose region-specific ways to interpret disaster information and provide indicators of how much confidence should be placed in such information when making evacuation decisions.
This study focuses on two key aspects to assess uncertainty: Evaluating the uncertainty of Early Warning Information (which predicts the probability of warning-level events up to five days in advance). Using forecasted JMA Runoff Index data (which indicates the risk of flood disasters) up to six hours ahead to assess uncertainty based on lead time. Additionally, rivers are classified according to uncertainty levels, and the feasibility of using basin rainfall index predictions as a basis for evacuation decisions is examined.
The uncertainty of Early Warning Information, which indicates the likelihood of a warning being issued at "medium" or "high" confidence levels, was analyzed. The data used consisted of Early Warning Information issued from 2017 to 2023, which predicted the likelihood of heavy rain warnings up to five days in advance. The uncertainty assessment was conducted using precision (the proportion of correctly predicted heavy rain warnings) and recall (the proportion of actual heavy rain warnings that were successfully predicted). These metrics were calculated separately for "medium" and "high" confidence levels over a five-day lead time.
The JMA Runoff Index is a measure of relative flood risk and is used as a criterion for flood hazard maps and flood warnings. Data from 8,358 nationwide observation points were collected, comparing forecasted and actual values. Rivers were categorized into three groups based on the uncertainty of their forecast accuracy at different lead times. The usability of basin rainfall index predictions as a basis for evacuation decisions was evaluated for each category. The results showed that forecast accuracy declined as lead time increased for many rivers. In such cases, residents—particularly elderly individuals who require more time to evacuate—must understand the risk of false alarms and evacuate earlier, less certain stages.
Finally, we calculated the probability that all forecasts would be accurate up to Level 5 (disaster occurrence) by combining the uncertainties of early advisory information (Warning Level 1) and JMA Runoff Index forecasts (Warning Levels 2 to 4).
Disaster prevention weather information is an essential element for residents to make swift and appropriate evacuation decisions. This study quantitatively evaluates the uncertainty of disaster prevention weather information and examines strategies for utilizing it to encourage more effective evacuation behavior. The findings will be reported in this presentation.
Ensuring effective evacuation requires accurate weather forecasting and the timely issuance of disaster prevention weather information. While weather forecasting technology has improved over the years, it is not yet flawless. When weather forecasts are inaccurate, the reliability of disaster prevention weather information decreases, reducing the lead time (the time available for evacuation) and increasing the risk of delayed evacuation. To secure sufficient lead time, issuing information at an early stage when the certainty of a disaster is low can lead to more false alarms, potentially diminishing public trust in such information.
By analyzing the uncertainty in disaster prevention weather information issued for each region, we can identify areas where forecast reliability is particularly low. This knowledge allows us to propose region-specific ways to interpret disaster information and provide indicators of how much confidence should be placed in such information when making evacuation decisions.
This study focuses on two key aspects to assess uncertainty: Evaluating the uncertainty of Early Warning Information (which predicts the probability of warning-level events up to five days in advance). Using forecasted JMA Runoff Index data (which indicates the risk of flood disasters) up to six hours ahead to assess uncertainty based on lead time. Additionally, rivers are classified according to uncertainty levels, and the feasibility of using basin rainfall index predictions as a basis for evacuation decisions is examined.
The uncertainty of Early Warning Information, which indicates the likelihood of a warning being issued at "medium" or "high" confidence levels, was analyzed. The data used consisted of Early Warning Information issued from 2017 to 2023, which predicted the likelihood of heavy rain warnings up to five days in advance. The uncertainty assessment was conducted using precision (the proportion of correctly predicted heavy rain warnings) and recall (the proportion of actual heavy rain warnings that were successfully predicted). These metrics were calculated separately for "medium" and "high" confidence levels over a five-day lead time.
The JMA Runoff Index is a measure of relative flood risk and is used as a criterion for flood hazard maps and flood warnings. Data from 8,358 nationwide observation points were collected, comparing forecasted and actual values. Rivers were categorized into three groups based on the uncertainty of their forecast accuracy at different lead times. The usability of basin rainfall index predictions as a basis for evacuation decisions was evaluated for each category. The results showed that forecast accuracy declined as lead time increased for many rivers. In such cases, residents—particularly elderly individuals who require more time to evacuate—must understand the risk of false alarms and evacuate earlier, less certain stages.
Finally, we calculated the probability that all forecasts would be accurate up to Level 5 (disaster occurrence) by combining the uncertainties of early advisory information (Warning Level 1) and JMA Runoff Index forecasts (Warning Levels 2 to 4).
Disaster prevention weather information is an essential element for residents to make swift and appropriate evacuation decisions. This study quantitatively evaluates the uncertainty of disaster prevention weather information and examines strategies for utilizing it to encourage more effective evacuation behavior. The findings will be reported in this presentation.