Saturday, October 27, 2012

Frankenstorm’s River Forecasts

This coming week, Hurricane Sandy could merge with a snow-bearing cold front to cause major losses along the East Coast of the US. Occurring at high tide, the financial district of Manhattan could be flooded. This trifecta of two storms colliding at high tide has been dubbed the “Frankenstorm”. Roger Pielke Jr discusses the projected losses as being somewhere in the $1-$10 billion range, depending on where the storm hits and how strong it is.


NASA image of Sandy

Bill Hooke has an excellent summary of where to go for information and he quotes the Capitol Weather Gang as saying the scenarios for Washington DC are

(A) worst case, – direct hit (on Washington), severe impacts (30% chance);

(B) indirect hit, major impacts (N.B. this scenario also implies a direct hit, only north of here, say around New York or New Jersey) (45% chance);

(C) glancing blow, minor impacts (but Massachusetts or Maine might still be hit) (20% chance);

(D) out to sea, few impacts (5% chance).

The best source of information on flooding is the National Weather Service Advanced Hydrologic Prediction Page. These are the official “single valued” forecasts that go a relatively short time into the future.


The forecast for the Hudson River upstream of New York City. The left side is what’s happened in the past, the right side is projected into the future. The oscillations are the ocean tides.


Here’s another station northeast of Philadelphia, southwest of Trenton

The official suite of products does have some information about the chance of flooding reaching different levels. These are the Ensemble Streamflow Prediction (ESP) products like so


Chance of flooding at Langhorne, PA.

The problem with that ESP forecast is that it doesn’t include anything specific about a rainfall forecast, it just assumes a wide range of future rainfall scenarios are possible.

That’s obviously not the case now because a hurricane is bearing down on the region. Now the National Weather Service is experimenting with linking the river forecasting models up to various weather forecasting models (in a process called MMEFS). Three weather models give a range of possibilities of what could happen. Here are the expected areas of flooding from one model (GEFS):


Purplish dots mean more severe flooding. Square dots mean higher chance of that happening.

Here are some more charts of Langhorne:


The possibilities of flows 13 days into the future. Each line is a possible future.


Like the ESP plot, except when you consider the specific weather forecast for the next 13 days. Earlier there was about a 5% chance of a big flood, this now says there’s an over 85% chance. Some of the worst flood scenarios are eye-poppingly high (homes under ~15 feet of water)

Naturally, these are experimental forecasts and are dependent very much on where the hurricane actually goes. These experimental products are also not reviewed and approved by a human hydrologist, they are straight from the model, hands-off. In contrast, the official products are inspected and adjusted as necessary by a professional hydrologist who has local knowledge of ongoing conditions.

The amazing thing, however, is that every piece of information in this post was freely available on the Internet (from my room in England). Nearly everywhere else that I know of, the river forecasts are only accessible to a select group of people, such as emergency services personnel. They control access through special logins and such. If you have some thoughts on public access to this kind of information, leave a note in the comments section!


  1. Another striking aspect: As of last night (Fri 26th), just days before the hurricane is likely to hit, the AHPS map of flood outcomes (single value) showed all green lights, meaning zero possibility of high water. This is because the Sandy's main impacts are beyond the length of the forecasts (5 days or so), and also because NWS forecasters are cautious about crossing thresholds and are probably uncertain about when/where the rain will fall. The ESP weekly risk of flooding products also showed zero risk of flooding in the next 7 days. At the same time, in contrast, the longer lead automated MMEFS forecasts showed significant risk of flooding everywhere. Though MMEFS is technically still a naive approach, with various scientific flaws, one can also easily argue for the qualitative value of advance warning that it nonetheless provides, relative to the less ambitious traditional NWS single-value forecasts.

  2. Andy, great comments. I read somewhere that the National Weather Service's false alarm rate for tornadoes has been about 75% for many years. (you also see it at A researcher recently asked me if I knew if hydrologists have more or fewer false alarms compared to meteorologists. My sense was that hydrologists issue far fewer warnings because they don't want to over-warn. Typically you'll see the weather warning maps light up long before you see the flood warning maps change color. Is there any was to put some numbers behind this?