Replicating the micro structure of disdrometer data leads to a rainfall generator that correctly reproduces the large scale structure of the data set

Analyzing the transition probability of disdrometer data revealed a sigmoid relation between precipitation intensity of the current and next minute. The sigmoid changes in it's parameters slope, location and asymmetry based on the intensity of the current value. In particular the evolution of the parameters shows some distinct bends that mark transition points. Replicating how the sigmoid morphs with intensity we build a Markov chain model that generates realistic precipitation data. In particular it can generate the power law relation in the high intensity range of the distribution and also correctly includes a transition to exponential distribution at low intensities. To complete the algorithm we included a threshold based transition to dry periods. This introduces realistic intermittency into the data. What makes our findings compelling is that we strictly replicated the micro structures we found in the data and ended up with a random walk that generates the large scale structure of the data set. No optimizing was involved. We still have to fully validate the performance of our algorithm and understand the essential components that generate key characteristics as for example the transition between exponential and power law. With that we hope to find a universal mechanism that is able to generate very different precipitation distributions based on how we shape the morphing of the sigmoid function.