ERCF — Evacuation Risk Classification Framework

Scenario Parameters

Scenario Name

Description

City / Area (optional — auto-fills population)

Population at Risk

Vulnerable Population % (elderly · disabled · children · medical)

Vulnerable population includes: elderly (65+), people with disabilities, children under 5, pregnant and lactating women, and individuals with chronic medical conditions. Typically 20–40% of conflict-affected populations.
Children under 5 specifically require:

1 adult escort per child (reduces effective transport capacity)
Medical monitoring every 4–6 hours (increases paramedic requirements)
Specialized nutrition (RUTF, therapeutic milk) not in standard ration
Psychological support on arrival

Sources: Sphere 2018 Vulnerable Populations chapter; UNICEF Child Protection in Emergencies 2022.

Distance to Safe Destination (km)

Planned Start Date

Affects fuel, transport & shelter costs via ERA5 climate data

Terrain & Infrastructure

Road access · terrain type · seasonal constraints

Conflict Pattern

Population exposure to direct fire · mortality modifier

7-Dimension Risk Indicators (1=Minimal · 5=Critical)

● Validated ● Estimated ● Expert judgment

D1 Kinetic ThreatAirstrikes · artillery · proximity

3.0

Minimal — No active armed conflict in the area. Last incident over 6 months ago.

Low — Sporadic incidents reported, conflict front line more than 100 km (62 mi) away. No impact on planned route.

Moderate — Intermittent combat or airstrikes within 50 km (31 mi). Operations possible but require monitoring.

High — Active airstrikes or artillery in the operational area. Movement requires armed escort or significant route change.

Critical — Direct fire on civilians in the area. No safe window for movement. Immediate suspension of operations required. At D1 ≥ 4, the evacuation window indicator switches to NO SAFE EVACUATION WINDOW — no countdown is shown because movement itself is the hazard.

D2 Mobility ConstraintsWounded · bedridden · no transport · movement barriers

3.0

Minimal — Population is fully mobile. No physical movement barriers identified. Medical vehicle multiplier ×0.8.

Low — Some individuals require assistance but majority can self-evacuate with minimal support. Medical vehicle multiplier ×1.0 (baseline).

Moderate — Significant portion requires physical assistance or adapted transport. Movement is slow. Medical vehicle multiplier ×1.3.

High — Large number of bedridden, critically wounded, or immobile individuals. Standard transport insufficient. Medical vehicle multiplier ×1.8.

Critical — Mass casualty or complete mobility collapse. Self-evacuation impossible. Full medical extraction required. Medical vehicle multiplier ×2.5.

D3 AuthorizationParty consent · ceasefire · IHL compliance

3.0

Minimal — Full consent from all parties. Formal agreement in place. No legal or political obstacles.

Low — Verbal agreement obtained from key parties. Written confirmation pending but not expected to fail.

Moderate — Authorization from one party only. Other parties have not confirmed but have not actively opposed.

High — Consent contested or revoked by at least one armed actor. Significant IHL compliance concerns.

Critical — Active refusal or obstruction by a party. No valid authorization. Operation would violate IHL or expose civilians to retaliation.

D4 LogisticsRoads · transport · fuel · checkpoints

3.0

Minimal — All routes passable. Sufficient vehicles and fuel. No checkpoints or administrative obstacles.

Low — Minor road damage or 1–2 checkpoints expected. Route viable with standard planning.

Moderate — Partial road destruction, fuel scarcity, or multiple hostile checkpoints. Requires contingency route and pre-negotiated passage.

High — Primary route blocked. Only high-risk alternatives available. Fuel access critical. Convoy likely to be stopped or delayed significantly.

Critical — No viable ground route. Complete logistical breakdown. Requires airlift, waterway, or operation must be postponed.

D5 DestinationShelter · food · water · medical capacity

3.0

Minimal — Destination fully equipped. Shelter, food, water, and medical care available for full population load.

Low — Destination functional but operating near capacity. Minor gaps manageable with pre-positioning.

Moderate — Destination partially functional. Critical gaps in shelter or medical care. Requires parallel humanitarian prepositioning.

High — Destination overwhelmed or insecure. Arriving population at immediate risk of exposure, malnutrition, or violence.

Critical — Destination is unsafe or non-existent. Evacuating population would be placed in equal or greater danger than origin.

D6 UrgencyImminent attack · window closing · timeline

3.0

Minimal — No immediate pressure. Operation can be planned and executed over days or weeks without risk escalation.

Low — Situation is stable but deteriorating. Window of 5–7 days before conditions worsen significantly.

Moderate — Operational window is 48–72 hours. Ceasefire holding but fragile. Action required within days.

High — Window closing within 24 hours. Imminent offensive, flooding, or access denial expected. Rapid deployment required.

Critical — Imminent threat to life. Hours or less to act. Any delay results in mass casualties or complete loss of access.

D7 InformationComms reliability · disinfo · civilian awareness

3.0

Minimal — Reliable communications with all actors and affected population. No significant disinformation environment.

Low — Communications intermittent but functional. Population generally informed. Minor rumors manageable.

Moderate — Significant communications gaps. Active disinformation campaigns. Civilian awareness of evacuation low or distorted.

High — Communications largely down. Population misinformed or unaware. Hostile narratives actively undermining civilian trust in operation.

Critical — Complete communications blackout. No channel to reach population or coordinate with parties. Operation is effectively blind.

Compare on Radar

LEVEL 0

—

Baseline / Monitoring

NATO: Permissive (Stable)

Dimension Profile (Radar)

💰

Evacuation Cost Estimate

— define scenario above

⚖️ Evacuation Decision Analysis — Act Now vs Wait

Evacuate now

—

one-time evac cost

—

ongoing assist/day

No evacuation

—

assist all in zone/day

Break-even

—

define scenario above

Option A: one-time evacuation cost + daily assistance for population remaining in zone. Option B: daily assistance for full population with no evacuation. Break-even: the day at which not evacuating becomes more expensive than having evacuated. Assumes static conflict conditions and costs.

Baseline / Monitoring — Score ≤1.5 · NATO: Permissive (Stable). No evacuation required, maintain preparedness planning. GC IV Art. 49 threshold not yet met.

Low Risk (Advisory) — Score 1.5–2.5 · NATO: Permissive (Degrading). Facilitate voluntary departure for vulnerable groups; issue early warning. Art. 35 GC IV — voluntary departure right applies.

Moderate Risk (Watchful) — Score 2.5–3.5 · NATO: Uncertain. Activate Protection Cluster, negotiate humanitarian corridors, exhaust alternatives before proceeding. AP I Art. 57–58 — precautionary measures required.

High Risk (Contested) — Score 3.5–4.2 · NATO: Hostile (Partial). Mass evacuation — last resort threshold met. Military/CIMIC escort required, notify all parties. GC IV Art. 49: security of the population so demands.

Critical / Emergency — Score >4.2 · NATO: Hostile (Imminent). Emergency extraction; if window closed, shelter-in-place until corridor opens. GC IV Art. 49(1) — forced transfer = grave breach, Rome Statute Art. 8(2)(b)(viii).

Risk Severity

— /5

12345

D1 · D2 · D6

Feasibility

— /5

12345

D3 · D4 · D5 (inverted)

Info Quality

— /5

12345

D7 (inverted)

Minimal — Very low direct threat to civilian life. No active armed conflict or threat is distant/sporadic.

Low — Some elevated threat indicators but not posing immediate danger. Manageable with standard precautions.

Moderate — Elevated threat, direct danger plausible. Combines kinetic activity, vulnerability of the population, and urgency.

High — Direct threat to civilian life. Active combat indicators, high vulnerability, and/or high urgency combine to create serious danger.

Critical — Imminent mass-casualty risk. Direct fire, extreme vulnerability, and maximum urgency combine. Derived from D1 (kinetic threat) + D2 (vulnerability) + D6 (urgency).

Not viable — No safe corridor, no authorization, or logistics make movement impossible. Forced evacuation would risk greater harm than staying.

Severely constrained — Partial or contested corridor. Authorization, logistics, or destination conditions are highly unfavorable.

Borderline — Corridor exists but with significant risk or friction across authorization, logistics, or destination readiness.

Viable — Reasonably clear route with manageable friction. Most of authorization, logistics, and destination conditions are favorable.

Fully viable — Clear, authorized route with adequate destination capacity. Derived from D3 (authorization) + D4 (logistics) + D5 (destination), inverted since high raw scores in those dimensions mean worse conditions.

Very poor — Severe intelligence gaps. Decision-makers are operating nearly blind.

Poor — Major information gaps affecting decision quality. Significant uncertainty about conditions on the ground.

Adequate — Some gaps but manageable. Enough information to make a reasonably informed decision.

Good — Reliable information environment with minor gaps. Decision-makers have a clear picture.

Excellent — Comprehensive, reliable situational awareness. Derived from D7 (information environment), inverted since a high raw D7 score means worse information quality.

Most Similar Historical Cases

Loading…

Resources Required for Evacuation

Transport Mode

—

🚌 Vehicles

—

👥 Personnel

—

💰 Total Cost

—

💵 Cost/Person

VEHICLES

PERSONNEL

SUPPLIES (3-day buffer)

Rates reflect national staff and NGO field deployment costs — not UN international professional rates, which are 3–6× higher ($400–580/day fully loaded in Sudan/Yemen context). Sources: MSF international salary $87–97/day + operational overhead (doctorswithoutborders.org, 2022); UN DSA Sudan $270/day, Yemen $225/day (ICSC/CIRC/DSA/574, March 2023). Adjust upward if modelling a UN-led operation.

Current: NGO/national staff rates

Cost estimates marked with are operational assumptions not validated against field data (ICRC, WFP, OCHA, UNDSS). Sphere-validated standards are marked with . Use with caution for academic or operational purposes.
Reference: Sphere Association, The Sphere Handbook, 4th edition, Geneva, 2018.

⚠️ Model Limitation: Conflict Pattern not reflected in resource costs

The Conflict Pattern parameter (Urban Siege, Enclave/Blockade, City Conflict, Regional/Dispersed) currently affects only the mortality and stay-or-evacuate model — it does not adjust evacuation resource costs (security personnel ratios, armoured vehicle requirements, or route length multipliers). In high-intensity conflict contexts (Urban Siege, Enclave/Blockade), actual security and transport costs may be substantially higher than shown. No primary source with validated numerical multipliers by conflict type was identified in the humanitarian operations literature; implementing unvalidated multipliers was judged to risk misleading precision. This limitation should be considered when using cost estimates for Urban Siege or Enclave/Blockade scenarios.

Cost Breakdown (USD)

Sphere 2018 validated Estimated / known basis Unvalidated — requires field data

Cost of Civilians Remaining — Per Day Escalation

What this model is for

Prospective operational planning: estimating the human cost of inaction before a crisis escalates — not retrospective mortality forecasting.
Planning horizon: 0–90 days. Accuracy degrades for very long durations (>120 days) due to linear accumulation.
Decision support: when and whether to evacuate. Financial cost estimates are substantially more reliable than death and injury projections.

Calibration — 16 of 31 cases · v7: 7/16 within 2× (44%) · R²=0.86

The mortality model is calibrated against 16 documented historical cases (1991–2024). 15 cases are excluded as out-of-scope — including 2 'challenge cases' that reveal structural model limitations. Financial cost estimates are substantially more reliable than mortality projections.

Case	Recorded	v7 model	Ratio	Status
Mariupol, Ukraine [ID]	8,000	21,656	2.71×	✗ overcount
Gaza, Palestine	45,000	78,752	1.75× ✓	✓ within 2×
Aleppo, Syria [ID]	31,000	8,266	0.27×	✗ undercount
Kherson, Ukraine	200	71	0.36×	✗ undercount
Mosul, Iraq	9,000	7,216	0.80× ✓	✓ within 2×
DRC-Goma (M23)	3,000	6,082	2.03×	✗ overcount
Grozny II, Chechnya	6,500	16,128	2.48×	✗ overcount
Fallujah I, Iraq	700	666	0.95× ✓	✓ within 2×
Jenin Camp, W. Bank	37	23	0.63× ✓	✓ within 2×
Vukovar, Croatia [ID]	2,000	1,398	0.70× ✓	✓ within 2× (ID)
Fallujah II, Iraq	700	1,576	2.25×	✗ overcount
Gaza — Cast Lead (2008)	965	4,290	4.45×	✗ large-enclave
Gaza — Protective Edge (2014)	1,540	9,734	6.32×	✗ large-enclave
West Beirut, Lebanon	4,400	9,181	2.09×	✗ overcount
Nagorno-Karabakh 2020	115	91	0.80× ✓	✓ within 2×
Huambo, Angola [ID]	7,000	10,547	1.51× ✓	✓ within 2× (ID)
Srebrenica, Bosnia (1995†)	8,000	—	—	OOS — genocide†
Kosovo — Albanian disp. (1999‡)	10,317	—	—	OOS — open corridor‡
Central African Republic (2014‡)	5,000	—	—	OOS — regional dispersed‡
Sudan — Khartoum (2023‡)	15,000	—	—	OOS — duration+scale‡
Sarajevo, Bosnia (1992–95§)	5,434	—	—	OOS — duration§
Battle of Raqqa, Syria (2017‡)	1,600	—	—	OOS — air campaign‡
Hue, Vietnam (1968¶)	6,900	—	—	OOS — massacre¶
Manila, Philippines (1945¶)	100,000	—	—	OOS — massacre¶
Bucha, Ukraine (2022¶)	480	—	—	OOS — executions¶
Kuito/Bié, Angola (1993§)	25,000	—	—	OOS — duration§
Misrata, Libya (2011**)	~400	—	—	OOS — data uncertain**
Marawi, Philippines (2017§)	400	—	—	OOS — duration§
Sri Lanka — Vanni (2009§)	7,000	—	—	OOS — duration§
E. Ghouta, Syria (2018††)	1,768	15,105	8.54×	Challenge case††
Gaza — Pillar of Def. (2012††)	103	1,768	17.2×	Challenge case††

v7: 7/16 within 2× (44%) · R²=0.855 · LOOCV R²=0.807 · L3 R²=0.958 · L4 R²=0.624 · Calibration script: calibration/calibrate.py
[ID] = infra-denial multiplier applied (Mariupol, Aleppo, Vukovar, Huambo). Gaza Cast Lead & Protective Edge: large-enclave pattern — L4 rate calibrated for siege attrition, not precision air campaigns on pop >500k.

Known limitations

Sudan and Gaza (2023) exceed model scope: Sudan is a conflict dispersed across 1.8M km² (continental scale); Gaza mortality significantly driven by famine/healthcare collapse. Both require purpose-built models.
Srebrenica (genocide): targeted mass killing in 3 days. Model assumes conflict attrition over time. Genocide dynamics are categorically different and not modeled.
Challenge cases — structural model limits: Eastern Ghouta (2018) and Operation Pillar of Defense (2012) are retained in the corpus but excluded from calibration. E. Ghouta: L4 high-mortality siege undercounted (8.54×, same limitation as Aleppo 2016 — conf_mult × infra_denial insufficient for sieges with >1,500 deaths in <60 days). Pillar of Defense: L4 large-enclave 8-day precision air campaign overcounted by 17× (same structural pattern as Cast Lead and Protective Edge).
Famine / disease mortality not included: blockade-induced starvation and healthcare collapse are absent. This is the primary source of underestimation in prolonged enclave scenarios.
Within-2× rate of 44% (7/16 cases): lower than v6 (80%) due to honest recalibration — v6 metrics were not reproducible from the codebase. v7 metrics are fully reproducible via calibration/calibrate.py. L3 R²=0.958 (strong); L4 R²=0.624 (moderate — heterogeneous conflict types).
L4 structural limitation: the model cannot simultaneously capture high-mortality urban sieges (Aleppo: 31,000 deaths, 300k pop) and large-enclave precision operations (Gaza Cast Lead: 965 deaths, 1.5M pop). Two 'challenge cases' (Eastern Ghouta 2018, Pillar of Defense 2012) document this boundary explicitly.
Sample size n=16 (calibration set). 15 additional cases excluded as out-of-scope (mass atrocities, duration >90 days, open-corridor displacement, regional dispersed, or structural model limitations). No confidence intervals. All parameters are modelled estimates, not statistically derived.
Siege cap (pop ≤ 500k): heuristic threshold. Gaza (2.3M) is excluded from siege cap even though movement was under fire, resulting in some underprotection correction.

How to use this responsibly

Mortality estimates should be treated as order-of-magnitude indicators with wide uncertainty bands — a ratio of 2× to 5× vs the historical record is typical even for the best-calibrated cases. Financial cost estimates (evacuation resources, in-situ assistance) are the primary reliable output of the ERCF and should be prioritised for operational planning. Always validate mortality projections against country-specific intelligence and field assessment before any operational application.

† Srebrenica: targeted genocide — outside model domain regardless of parameter settings. ‡ Sudan 2023: dispersed conflict across 1.8M km² — continental scale exceeds design assumptions. § Duration exceeds ERCF 0–90 day planning window. ¶ Deliberate systematic massacre — outside armed conflict attrition model domain. ** Death count too uncertain for calibration. †† Challenge case: retained in corpus as documented boundary case but excluded from calibration set. E. Ghouta: L4 high-mortality siege, model undercounts (same structural limitation as Aleppo). Pillar of Defense: L4 large-enclave short-duration precision air campaign, model overcounts by ~17× (same pattern as Cast Lead and Protective Edge). Source: ERCF Mortality Model v7, ERCF Historical Case Database (31 cases, 1991–2024).

Population remaining in zone:

% mode

0% 100% 100%

Days remaining:

90d

30 days

—

Financial (USD)

Survival cost only — excludes logistics overhead. Risk-scaled estimate ($1–$12/person/day). · see full delivery cost ↓

—

EST. DEATHS MODERATE

—

INJURIES MODERATE

Model structure: R²=0.855, p<0.00001 (statistically significant). Calibrated on 16 of 31 historical cases (15 excluded: out-of-scope). Infrastructure-denial multiplier (α=0.4251) applied to 4 documented cases (Mariupol, Aleppo, Vukovar, Huambo). L3: Strong (R²=0.958, 56% within 2×). L4: Moderate (R²=0.624, 29% within 2×). See methodology

CUMULATIVE FINANCIAL COST (USD)

CUMULATIVE HUMAN IMPACT

Assistance Cost — People Remaining in Zone

Supply delivery + emergency extraction + field medical · all dimension modifiers applied · distinct from the "Financial" survival estimate above

Cost configuration

—

Total Assistance Cost

—

Crisis Cost (incl. extraction)

—

Extraction Probability

—

Cost/Person Remaining

Historical Conflict Cases

Level	Case	Year	Pop. at Risk	Deaths	Duration	Type

Dimension Comparison — Current vs Historical

Red = current scenario · Blue = selected historical case