September CSA “Code Red” – Public Warning in Operation “Cast Lead” ISMOR 2009 Lt.Col. Ami Mizrahi, M.Sc. Center for System Analysis Planning Division IDF
September CSA Background In operation “Cast Lead” Hamas fired rockets on civilians Protection based on “Most Protected Room” Concrete room / Inner room / Lower floors / Staircases Alerting the population Sensor constellation / Sirens
September CSA The Dilemma Increase alert probability Maximize P alert Minimal distraction of civilian life under ongoing rocket attacks Minimize P unnecessary alert Analogous to P fa and P d
September CSA P unnecessary alert (P d ) (P fa ) P alert Vector of launch Logic Change Hit area prediction Schematic Tradeoffs
September CSA Protection Policy Partial Alert “Stay in Shelters” Halt Civilian Routine Psychological Impact Alert Everyone The “Blitz” on London Yom Kippur War Focused Alert Desert Storm2 nd Lebanon War Cast Lead
September CSA Desert Storm 1991 39 Al-Hussein Missiles (SCUD) were launched at Israel Israel was divided into 6 alert zones “Sealed Room” Gas Masks Every missile caused ~1/3 of Israel’s population to be alerted (~2M)
September CSA 2 nd Lebanon War 2006 ~4,000 rockets were fired at Israel Improved hit predictions New sensors More Public Warning Zones On the average ~100K people were alerted
September CSA Prior to Operation “ Cast Lead ” Experience gathered for 8 years A few rockets per day rockets over the years Order of magnitude increase in # of zones Order of magnitude decrease in # of people in a zone Improved hit prediction Implement relevant alerting logic
September CSA Logic for Improvised Rockets
September CSA Operation “ Cast Lead ” Dec. 27 th, Jan. 17 th, 2009 During “Cast Lead” ~1000 Rockets and Mortars Fired New threats - longer range rockets Alerting logic problematic for longer range rockets
September CSA Longer Threats -> more Unnecessary Alerts
September CSA Longer Threats -> more Unnecessary Alerts
September CSA Logic Change Vector of launch Hit Area Prediction
September CSA What was done? Empirical estimation of hit prediction accuracy Why empirical? Non-standard rockets Adaptation to theater (sensor combination, specific locations…) Define MOEs P alert Number of people affected by the Unnecessary Alerts Define new logic Calibrate parameters Test new logic on new cases (Validation) Test stability of new logic “Go to the decision makers”
September CSA Research Timeline Identify Problem Collect Data ~ 3 Days~ 1 Week Change Code ~ 3 Days Test Stability & Validate ~ 1 Week Approval End of Operation
September CSA Research Difficulties Data collection Vs. Real time crisis management Hard to get data from rescue personnel Rocket location Low priority to locate rockets falling outside residential areas Possible sample skewing Numerous authorities Military / Civilian - Police, Intel, Home Front Command, Our teams Very Noisy Data Limited accuracy of data Cross-check the data. Go to the field. Limited High Level Attention
September CSA Summary Basic dilemma remains: P alert <> P unnecessary alerts Requirement for flexibility for the alert system Alert Time <> Accuracy Local optimizations based on scenario Real time Alerting Zone Control Active Defense (intercept) poses more questions: Danger from Debris OR During Hostilities Improved Data Collection During Hostilities Relevant research time