## Thursday, May 12, 2016

### Rain is constant, videos are van Goghized

A comment about two interesting enough papers.

The changes in global mean temperature are discussed all the time but what has happened with the rain since 1850? Has it been increasing or decreasing? The Journal of Hydrology has published a paper giving a clear answer:
Changes in annual precipitation over the Earth’s land mass excluding Antarctica from the 18th century to 2013
Hat tip: Not a lot of people know that and Climate Depot.

Physicists W.A. van Wijngaarden and A. Syed from Toronto took 1,000 stations into account and the trends (changes of precipitation expressed in percents per century) are the following:

1850-2000: -1.2 ± 1.7% per century
1900-2000: +2.6 ± 2.5% per century
1950-2000: -5.4 ± 8.1% per century

The trends seem to be zero: all of the deviations are safely below two sigma. There's no significant signal. Note that the longer periods, especially from 1850, produce more precise values for the "basically zero" trend, despite the fact that they depend on some "ancient" data, because the noise from a larger number of years averages out more accurately.

The total amount of precipitation doesn't detectably increase. It doesn't significantly decrease, either. The maximum accuracy one can extract is "plus minus a few percent per century". This is much poorer precision than the estimated trends of the temperature. If you naturally talk about the absolute temperature (in kelvins), the centennial trend is being quantified with the precision of 0.1 °C which is 0.03% of the temperature around 300 kelvins.

The poorer precision of the precipitation data is basically because the amount of precipitation every year behaves like the white noise. A new year is basically independent from the previous one. A new random number emerges every year. There's not much autocorrelation. On the other hand, the temperature is closer to the red noise – the function $$T(t)$$, the temperature as a function of time, is continuous (although not necessarily nicely differentiable) and it's basically the year-on-year temperature change that is an independent random number every year (up to some pinkness of the noise, and some regulation when the accumulated deviation becomes too high).

This red-noise or random-walk character of the temperatures implies that it doesn't matter much whether you include another year – the function looks smoother (at least continuous) than the white-noise precipitation data.

Some people's panic about the detectably nonzero temperature trend is a self-inflicted injury, however. As I just explained, it's much easier to show that the trend is different from zero, due to the relative smoothness of the data. On the other hand, the precipitation data which behave as white noise admit changes of the precipitation by 5% per century. If such a trend existed (with either sign), it would arguably have a greater effect on the ecosystems and human societies than the observed change of the temperature by 1 °C per century or so.

But we can't reliably prove, using the empirical data, that the trend is 0% rather than 5%, or 5% rather than 0%. So the precipitation alarmism isn't as popular as the temperature alarmism. But the main reason why the precipitation alarmists haven't proliferated as much as the temperature alarmists is the fact that the white-noise-like quantities such as precipitation inevitably produce greater error margins of the trend, as I just argued. That's why the trend for the temperature could have been distinguished from zero but the precipitation trend couldn't have. The actual magnitude of the trend was pretty much irrelevant for the question whether a type of the hysteria emerges or not.

Van Gogh effects for your movie maker

MIT Technology Review has mentioned an interesting preprint
Artistic style transfer for videos
by Manuel Ruder, Alexey Dosovitskiy, Thomas Brox who work in Freiburg, Germany. First, remind yourself what the paintings by van Gogh looked like. One of these messy papers may be sold for millions.

Now, the folks may produce whole movies – with thousands of frames – in the van Gogh style. And it costs almost nothing.

A longer video is available.

They may start with animated or human movies and "transfer the style" of van Gogh to each frame of the video, according to a recently introduced algorithm. The resulting videos look cute. Maybe I would prefer to watch many things in the van Gogh (or another) style.

I am surprised they are posting it as a preprint. They should have sold it to companies producing movie makers first. I would like to have the "Effects / Van Gogh" and perhaps similar entries in the menus! The authors are recommended to monetize their algorithm and pay me 5% of the profit for their gratitude for my excellent idea.