The met department and randomness

Ok. Nothing unusual about the title of this post. There is intuitively a lot of randomness where the met department is concerned. This post is about an editorial in the Business Standard.

Now, the Indian Met department used a new process for forecasting the monsoons last year. Now, this process yielded good results in the north-east and north-west in terms of forecast accuracy. In the center and south, however, it was a disaster. The process had predicted a small shortfall in rain in these two regions, and it turned out the rains here were more than a quarter more than normal!

So what does the met do now? They decide to discard the process for the center and south. And will continue to use it for the north-west and north-east. Even if you know a little bit about randomness and testing, and I assume that the people at the met department should definitely be well-versed in this, you will know that they have done is ridiculous. How can you form an opinion about something by looking at just one data point? Wouldn’t there have been a good chance that this an anomalous result? Now, what will the met do if the method fails for the north-east and north-west also? Will they completely abandon this new method?

I find the system that the met department is using no more intelligent than the one that I use to classify my shirts as “lucky” or “unlucky” (and trust me that isn’t very intelligent; I just use 1/2 data points and quickly derive an opinion).

God help us, if the met department is like this. The sooner weather derivatives (rainfall, temperature, etc.) get launched (or have they already been launched? I know they are now legal in india) the better for us. At least in that case we will get the wisdom of crowds to forecast the monsoon.

Put Comment