When does Metro Manila flood?

A natural laboratory for threshold dynamics

Metro Manila sits between Manila Bay and Laguna de Bay, with the Marikina–Pasig river system threading through the metro and draining toward the bay under wet-season conditions. Every wet season, the Habagat brings warm, moisture-laden southwest monsoon flows that can deliver sustained rainfall over the western side of Luzon.

What engineers and residents repeatedly observe is not merely that more rain produces more water. It is that the city can appear to hold through one storm, then cross into widespread flooding under the next. The research question is this threshold: when does local wetting stop dying out and begin spreading as a persistent, system-level flood state?

Schematic map capturing the essential topology: bay to the west, lake to the southeast, Marikina and Pasig corridors threading through. Not survey-grade — the point is the drainage geometry, not the cadastral lines.

The minimal faithful model

Reduce the landscape to a binary grid: each cell is wet (1) or dry (0). At each discrete time step, a cell updates using only its lateral neighbours. If at least one lateral neighbour is wet, the cell becomes wet with probability p. If neither lateral neighbour is wet, it remains dry. That is the complete rule.

At p = 1, the model becomes Wolfram Rule 250 exactly: deterministic spreading by the OR of the left and right neighbours. Below p = 1, the same rule becomes a Probabilistic Cellular Automaton governed by a single effective parameter p, representing the combined pressure of rainfall intensity, soil saturation, surface runoff, and drainage limitation.

In the theoretical model, p is spatially uniform. This preserves the clean Domany–Kinzel / directed-percolation interpretation and allows the flood threshold to be measured as a true phase transition.

Metro Manila enters as the visualization and motivation layer. Low-lying corridors such as Marikina and Pasig can be represented through the initial wet cells, basin geometry, or a separate topographic weighting extension. But the core rule remains local, binary, and uniform: neighbour influence plus probability p.

YELLOW the absorbing state holds

PAGASA's yellow warning — 7.5 to 15 mm per hour — sits below the critical threshold. Wet patches nucleate along the river corridors, then die out. With p < p_c, the dynamics converge to the all-dry absorbing state: once every cell is dry, no update rule can reactivate them. This is the regime that Rule 0 represents in its deterministic limit — all outputs zero, regardless of neighbourhood. The city is stable, not because it cannot get wet, but because the absorbing state dominates.

ORANGE the critical filament

PAGASA's orange rainfall level corresponds to intense rain, around 15 to 30 mm per hour. In this model, orange is interpreted as the critical-warning regime: the system is near the threshold where local wetting stops dying out quickly but has not yet become a persistent metro-wide flood.

At p ≈ p_c, wet activity survives as thin, branching ribbons along vulnerable corridors such as the Marikina valley and Pasig system. These clusters are neither safely extinguished nor fully spreading. They persist because the birth of new wet cells and the death of existing wet cells are nearly balanced.

This branching survivor pattern is the stochastic analogue of Wolfram's edge-of-chaos idea: a state between complete order and complete disorder. Rule 110 is the correct Class IV reference point, famous for Turing completeness and complex localized structures. Rule 22 can remain as a chaotic/fractal comparator because its filaments visually resemble critical spreading, but it should not be treated as the main Class IV anchor.

This is the most structurally rich and information-dense regime in the model. It is the warning zone where the city may still hold — but small changes in rainfall, saturation, drainage, or initial wetting can push the system into persistent flooding.

RED the active phase

PAGASA's red rainfall level corresponds to torrential rain, above 30 mm per hour. In this model, red is interpreted as the supercritical regime: the system has crossed p_c, moving from the absorbing phase into the active directed-percolation phase.

In CA terms, wet activity no longer dies out quickly. A macroscopic fraction of cells remains active over time, and the system sustains persistent spreading rather than returning easily to the all-dry state. This is the stochastic analogue of Wolfram's Class III behaviour: active, irregular, and disordered.

The deterministic endpoint of this model remains Rule 250 at p = 1. Rule 255 — where every cell activates even with no wet neighbours — is a separate saturation corner of the rule cube, not the limit of the Rule 250 probabilistic path.

The transition is sharp but continuous. The same marginal increase in p that produced only short-lived wet patches below the threshold can produce persistent spreading above it. That asymmetry — extinction below p_c, survival above p_c — is the signature of a continuous directed-percolation phase transition. It gives a mathematical reason why an orange warning can escalate into red under the same weather system: once the effective wetting probability crosses the threshold, local flooding can become system-level persistence.

The order parameter reveals it

Measure the long-time flooded fraction across the full range of p, and the result is unambiguous. Below the threshold, wet activity dies out and the flooded fraction approaches zero. Near p_c, the system becomes critical. Above p_c, the flooded fraction rises sharply but continuously, following the directed-percolation order-parameter relation:

ρ ~ (p − p_c)^β

For 1+1D directed percolation, β ≈ 0.276. The transition is continuous — not a true jump — but on a finite lattice it can look abrupt because small changes in p produce large differences in survival time and flooded extent.

The measured critical point p_c ≈ 0.70 is consistent with the published directed-site-percolation threshold (p_c ≈ 0.705485) for the 1+1D square/diagonal lattice, within rounding and finite-size effects. The order-parameter curve is therefore the quantitative signature of the absorbing-to-active transition.

Flooding is emergent — and so is the solution

Not one cell knows what a flood is. Each cell executes the same local rule: check its neighbours, then activate with probability p. The flood is emergent — a global pattern produced by many local interactions, with no central coordinator and no cell-level awareness of the whole system.

This is the central lesson of cellular automata. Rule 110, which Matthew Cook proved Turing complete, shows the outer edge of what local CA rules can generate when information is encoded in their patterns. This investigation makes a simpler claim: if local rules can support computation, then they can certainly support spreading, clustering, extinction, and persistence. A flood threshold emerging from a probabilistic Rule 250 update is therefore theoretically natural.

The policy implication is also local. In the clean model, p_c is a system-level threshold. But in a real city, the effective wetting probability is not evenly distributed. Low-lying, poorly drained corridors such as Marikina and Pasig are the places where wet activity is most likely to survive, branch, and carry the transition from local flooding to system-wide persistence.

Intervening there — restoring floodplains, clearing drainage, improving outflow, raising embankments, and protecting retention areas — reduces the local effective p or raises the local capacity required for spreading. The CA framework makes the intervention logic clear: flood control does not only mean treating the whole city uniformly. It means identifying the cells and corridors that carry the transition, then weakening the paths through which local wetting becomes persistent flood.