About this map

This map was made by Rob Ennals, a software engineer based in Sunnyvale, California. It is a personal project, not a product of any company or government agency. If you spot a mistake or have a question, email me at rob.ennals@gmail.com.

The four levels

How the levels are worked out

1. Start from the studied flood zones

FEMA and California's Department of Water Resources (DWR) have run detailed studies of how specific creeks flood. Those studies produce "1% annual chance" (100-year) flood zones. We treat these as the most reliable official information, and every spot inside one is Higher. They matched real floods well: 86–100% of the land that actually flooded in the events below was inside them.

2. Measure how high each other spot sits above those zones

Flood water spreads onto nearby land that is not much higher than the flood itself. So for every spot outside a flood zone, we find the nearest large flood zone and compare ground heights:

• Height above the flood level: the spot's ground height minus the ground height at the nearest edge of the flood zone (used as a stand-in for the flood water's level). Spots lower than that edge count as level with it.
• Distance: we add 1 m of height for every 400 m of distance from the zone. On flat ground this makes the risk fade gradually with distance instead of stopping at a line.
• Small zones don't spread risk: only flood zones of at least 5 hectares (about 220 m × 220 m) are used here. Hundreds of tiny zones (ponds, basins, slivers) would otherwise each spread a circle of risk across flat land. The small zones themselves are still Higher.

3. Measure how high each spot sits above the nearest creek

Many small creeks were never studied by anyone. To cover them, we use a national dataset that gives, for every 10 m of ground, its height above the nearest creek or drainage channel ("height above nearest drainage"). Land only a metre or two above a creek is where creeks overflow first.

Only real stream channels count. A channel's flood potential comes from the water arriving from upstream (rain and snowmelt over its whole drainage area), not from the rain where it happens to be, so:

• Canals, ditches and pipelines never count. They carry managed water, not floods.
• In wet regions, every natural stream counts.
• In dry regions (average rainfall under 250 mm, or 10 inches, a year: most of the desert and the southern San Joaquin Valley floor) only real rivers count: those draining at least 100 km² (like the Mojave River) or carrying an average of at least 10 cubic feet a second, which includes rivers fed by mountain snowmelt (like the Kern and Owens). Small desert washes rarely run, and on flat desert ground they made huge areas look risky.
• Flat basins that drain to no stream (such as the Imperial Valley floor) are left to the official flood zones, since there is no channel to measure height above.

4. Add mud and debris flows below recent burn areas (Los Angeles County)

For several years after a wildfire, heavy rain on the bare slopes can send fast-moving mud, rocks and burned trees down the canyons below, and these are often what destroys homes. Flood insurance covers this "mudflow", but flood zone maps don't show it. Los Angeles County Public Works keeps a list of the properties that mud and debris flows from burn areas of the last five years could reach, and notes the storm size at which each would be reached (its mud and debris flow forecast uses the same storm sizes):

• Higher: the property itself could be endangered in a small or moderate storm (phase 1 or 2).
• Elevated: the property itself could be endangered in a big storm (phase 3).
• Moderate: the road to it could be blocked, but the property itself isn't expected to be hit.

These show up as individual properties when you zoom in. Other counties don't publish a list like this yet. If you live below a recent burn area anywhere in California, check your county's warnings whenever heavy rain is forecast.

5. Add UC Irvine's flood model of the Los Angeles basin

Most of Los Angeles's flood insurance claims don't come from the creek and river zones above. Much of the basin's flooding happens when heavy rain overwhelms the storm drains, or water runs down the streets from the hills. A team at UC Irvine led by Prof. Brett Sanders built a detailed computer model of the Los Angeles basin, PRIMo-Drain. Working on a 3 m (10 ft) grid, it rains on the whole city, sends the water through the storm drains, rivers and channels, and tracks where it collects, together with high tides and storm surge at the coast. They published the flood depth it predicts for every land parcel (about 1.8 million) in a 100-year storm and a storm expected about once in 20 years. We use those depths for each parcel:

• Elevated: more than 30 cm (1 ft) deep in the storm expected about once in 20 years.
• Moderate: more than 10 cm (4 in) deep in the 100-year storm.

These cut-offs and levels come from checking the model against flood insurance claims. Neighbourhoods with a parcel flooded more than 30 cm deep in the 20-year storm averaged 1.23 claims since 1995, eight times the 0.15 of neighbourhoods the model keeps dry; shallower flooding in the 100-year storm was a weaker but real signal (0.44). Even so, these neighbourhoods have far fewer claims than those in FEMA's 100-year zones, about as many as our other Elevated and Moderate land, so that's how we rate them. Claims are only used for this calibration; everything else in this method is set without them. The model covers the urban basin, from the coast to the foot of the San Gabriel Mountains (84 of the county's 88 cities), but not the Antelope Valley, Malibu or Catalina Island. Like the mud-flow list, it shows up as individual properties when you zoom in.

6. Pick the level

A spot gets the highest level that any of the rules above gives it. The outlines are traced like contour lines on a hiking map, so they follow the land rather than a grid of squares.

What we deliberately don't use to set the level

FEMA also labels land as "500-year flood zone", "minimal hazard" or "undetermined". Outside the studied creek corridors, these labels follow city lines rather than the land. In Santa Clara County, for example:

• FEMA labels about 92% of Saratoga as a 500-year zone, but 96% of neighbouring Monte Sereno, on similar ground, as minimal hazard.
• About 70% of San José, including built-up neighbourhoods, is labelled "undetermined", meaning never studied.

Using those labels made hillsides look risky and low ground look safe, so we show them as notes in the "Why" list but let the terrain decide the level.

Checking it against real floods

None of the historical data below is used to set the levels, so it is a fair test. The one exception is choosing which level UC Irvine's Los Angeles model maps to (step 5), which used the Los Angeles claims.

Real flood outlines (Santa Clara Valley Water District)

For comparison, Higher covers 7% of Santa Clara County's land and Elevated 8%.

Flood insurance claims, 1995–2026 (FEMA)

FEMA publishes every flood insurance claim, tagged with its census neighbourhood (a "block group" of a few hundred homes), not its address. On the map each neighbourhood with claims is a circle; clicking it outlines the neighbourhood, since the claims could have come from anywhere inside it. Across California, 33,094 claims could be placed in a neighbourhood, and the claims rate rises steadily with the share of a neighbourhood the map flags:

In 36 of the 38 counties with enough claims to judge, neighbourhoods with flagged land have more than twice the claims of those without. Where the fit is weaker:

• Los Angeles: before we added UC Irvine's model, a third of its claims (1,472 of 4,416) came from neighbourhoods the map didn't flag; now 13% (555) do, mostly in hillside canyons. Its most flagged neighbourhoods are largely on the levee-protected plain, where the levees have kept claims low: neighbourhoods with levee-protected land average 0.29 claims, against 2.4 in those with FEMA's 100-year zones. Below recent burn areas the county's mud-flow list fits the claims well: the 34 neighbourhoods with a property rated Higher on it averaged 13.7 claims each, 20 times the county average, and 428 of their 466 claims came before the fires behind the list. Those canyons flooded long before they burned.
• Alameda has few claims (349) and only a weak pattern, which suggests much of its flooding comes from overwhelmed storm drains rather than creeks.
• San Francisco has only 19 claims, 16 of them in neighbourhoods the map doesn't flag. This fits its known sewer back-ups, which no source used here models.

The full numbers for every county are produced by make check in the source code.

Known limitations

• Small hillside canyons: water and mud rushing down small canyons in heavy storms (for example in Los Angeles's hillside neighbourhoods in 1998 and 2005) is only shown below burn areas of the last five years in Los Angeles County. Elsewhere, the stream data used here leaves out the smallest canyons, and no flood map covers them. Insurance claims there show up on the map as circles in uncoloured areas.
• Storm drains: street flooding from overwhelmed storm drains is only modelled in the Los Angeles basin, by UC Irvine's model. Elsewhere it can happen on any street.
• Desert flash floods: in dry regions, small washes are not counted, so flash floods down desert washes and across alluvial fans (the fan-shaped slopes below desert mountains) are mostly not shown unless an official study mapped them.
• Levees and channels: we rely on FEMA's and the Army Corps' records of levees. Recent creek improvements may make some areas safer than shown; failing ones may make them riskier.
• Detail: the terrain levels are worked out on a 30 m grid, and creek heights come from 10 m data averaged to that grid. Use this map for the neighbourhood picture, not to compare one house with the next.
• Coastal and Bay flooding: this version relies on FEMA's Bay shoreline zones. Storm-surge modelling (USGS CoSMoS) is not yet included.
• Insurance claims: only insured homes file claims, so more homes may have flooded. The census neighbourhood codes on some claims may not match the neighbourhood outlines we used, so a few may be misplaced.
• Coverage: all 58 California counties. Over the ocean, the Bay and large lakes the map shows no data.

Data sources

All data downloaded November 12, 2026.

← Back to the map