Talkin’ Bout the Weather Some More

One of my current web addictions (1) is the Contra Costa County Flood Control and Water Conservation District  rain gauges page:

First 3 gauges of 29 total.

The Flood Control district maintains 29 automated rain gauges scattered around the 804 square miles of Contra Costa County. This table is automatically updated at the top of the hour, and a quarter after and a quarter til.

I’ve put together a little Google sheet that does a little math, where I can grab the data off this page and paste it in to get some percentages, totals and averages:

Bottom left corner of my rain totals spreadsheet. I’m tracking gauges that meet or exceed their average total rain year inches (24 out of 29 so far), calculating some percentages, and, defying all that is holy, doing some totals and averages across gauges.

Now, if you’re a math guy, and especially a science guy, this little snippet should make your head explode – so, so wrong! Doing totals, percentages and averages by gauge makes complete sense (however limited its use), but doing so *across* gauges?!? Huh?

(Here’s where I expose – to perhaps well-deserved ridicule – how a science-loving non-scientist goes about analysing some data. The key step for me is, as always, philosophical: what am I looking at? What can it tell me in theory? What does it tell me in practice? These are questions that must be answered before you even bother to look at the numbers. Failure to do so is by far the most common technical failure in the Science! news I read: the writer doesn’t know what he’s looking at, doesn’t know the limits of what it can tell him, and then doesn’t understand what it is actually telling him. Stupidity and/or dishonesty is the dominant non-technical problem.)

The sneaky-bad part is that, until you think about it, it sort of makes sense: aren’t I getting an average for rainfall across Contra Costa County? No, I am not – the best I’m getting is the average of a bunch of point samples that are related in a manner that is not clearly understood.

First off, to think that an average of the gauges tells you something about rainfall in general over the area throughout which the gauges are deployed is making some assumptions. These 29 rain gauges represent, at best, a few square feet of the 804 square miles of CCC. Well? Are we supposing that these gauges are representative (whatever that might mean) of the other 803.9999 square miles? Why would we think that? What would we mean by it?

Why are there 29 gauges?  Why not just use one? More obviously, why are the totals at each gauge so different? Season total averages run from 11 inches up to 33 inches, and this year the differences in actual rainfall are at least as pronounced.

Contra Costa County is made up of at least 3 pretty distinct areas: The west-facing slopes of the Richmond/El Cerrito hills and the flats between them and the Bay, extensive hilly areas with a couple of hilly interior valleys punctuated by a big mountain (about 2/3 of the total area), and some flats on the delta to the far east.

Found here. The blue dotted lines do not represent partition based on geographical features. If one wanted to do that, the left hand line would be rotated about 60 degrees clockwise and moved west a bit,  and the right hand one pivoted about 60 degrees counterclockwise from the top point. Then you’d have something like rough climate zones. Very rough, as the south to north differences – farther from water differences – are not captured, and they can make a big difference.

Close to the center of this map is Mount Diablo (DBL 22). This year, Mount Diablo has gotten over 51 inches of rain, which is, according to my fun little spread sheet, 186%  of season average – and we’ve got a couple months more to go.

Immediately to the north of Mount Diablo are two gauges – the Concord Pavilion (CCP 43) and Kregor Peak (KGR 38). These two gauges are among the 5 remaining gauges that have not yet reached their season average total so far. In fact, while Mount Diablo is almost 2 feet of rain over for the season so far, these two are about 3 and a half and 5 and a half inches under. The other three gauges that have not hit their seasonal average total yet are much closer, and might hit them with the storms coming this weekend.

How could this happen? Two gauges within a couple of miles of Mount Diablo are not even getting average rainfall, while the mountain stands to get twice its average.

Consider this current predicted rain map:

Wish I’d have thought to capture yesterday’s, as it was much clearer.

Note Hawaii at the bottom center. That long line of rain from Hawaii to California is pretty much what the weather people call an atmospheric river – a Pinapple Express. This one, which blew through our neighborhood early this morning, was nothing like the size of the last couple. That stuff out to the east looks a bit more exciting. Zoomed in a little:

That thing that looks like a swirl? It is. When it reaches California in the wee hours of Friday, the rain will be pushed from south to north along the stronger, leading edge.

Speculating here: This puts (CCP 43) and (KGR 38) in Mount Diablo’s rain shadow. Gauges just south of Mount Diablo are all above average; the two directly north of it are below.

In a more typical Northern California rain year (2) the storms come down from the Gulf of Alaska, maybe or maybe not picking up some tropical moisture, and hit pretty much directly west to east. (CCP 43) and (KGR 38) would, in such cases, not be in the rain shadow of Mount Diablo, and might therefore get more rain, comparatively, to years like the one we’re having now (3). Thus, the season averages don’t really tell us what to expect. They are useless, really for predictions, as what they tell you is more like what a blended picture of two or more (you can have both Gulf of Alaska storms and some tropical stuff in the same year, for example) mechanisms by which California gets rain and snow.

So, what am I getting if I average Mount Diablo with Concord Pavilion and Kregor Peak? Should I take the average of only 2 out of 3? Add some more gauges? It will make a difference. Fundamentally, there’s nothing magic about these 29 gauges or about the number 29 – we could add or subtract gauges to the mix, or even double count some we think particularly important or ‘representative’. There’s nothing to stop us, it might even make sense, under certain assumptions.

Nope, what my averages across gauges tells me is not that we’re 130% of season average rainfall so far in Contra Costa County. What it tells me is that the average across the gauges is 130% of the average of the total season rainfall for each gauge – and that is all. Which is not all that helpful, and is only interesting in a vaguely cabalistic sort of way.

The point, if any, is that sometimes what may look like reasonable numbers to look at do not, in fact, tell you much. And that I’m a LITTLE bit obsessive on occasion. In a fun way! Really!

1. Other web addictions include: boat building (the 1337 woodworking skillz and empirical engineering fascinate me. Lapstrake for the win!), Sci Fi short films (there are a million of these, some quite good)  and primitive iron smelting (there’s a band out there named Bog Iron Bloom – wish I’da thought of that!). In my fantasy world, I’d dig my own bog iron, smelt it in a clay brick furnace, hammer it into an axe and iron nails, chop down some oak and build a Viking long ship – and make a Sci Fi short film about it! I’d need to find some people who don’t get sea sick to sail it for me, but I’m imagining that’s the least of the problems with this plan.
2. This is when the discussion gets weird: our entire sample size upon which we base our assumption of ‘typical’ is only about 150 years long, and only a fraction of that has anything like the widespread measure-taking we use now. The oldest CCC Water District gauge dates back to only 1937; most are either from the 1970s  – or since 2000. What would be an appropriate timeframe? 10,000 years? 100,000? Why or why not? Certainly, based on physical evidence, (and there are more recent updates that show even more variation I can’t seem to lay my hands on at the moment)  over 10,000 years, the averages would be different – and over 100,000 years, the median prediction would be: much colder, with a chance of more snow.
3. If in fact we have more than one year like this – so far, I’ve only heard things like a 1 in 25 year, but the year isn’t over yet. This seems to me to be a very unusual year, one not captured well by rain gauges such as those discussed above. How many rain and snow gauges are there in the 6,000+ square mile drainage of the Feather River? Because the Oroville Dam is almost 50 years old – and this is the first time the emergency spillway has been used. And there’s more rain on the way, and a massive snowpack to melt. In other words, are we really capturing the full extent of this precipitation year? The physical evidence – reservoirs around the state at or near capacity with a couple months of rain still to go – suggests we’re not.