Probability Distributions for Kaggle Rain Contest I

In the first Kaggle “How Much Did it Rain?” contest, the participants were asked to produce rainfall cumulative distribution functions, with steps from 0 to 69 millimeters in one mm increments. It seems that most of the top finishers, including the winner, used a classification strategy to estimate the 70 individual CDF values for each observation, without directly estimating the amount of rain. An alternative is to estimate the actual rainfall amount, the error standard deviation, and then to generate a suitable CDF based on those estimates. This presentation looks at that approach, focusing on generating the CDF.