Precipitation and its extremes: a study into the hydro-climatological controls of extreme precipitation

Abstract

Extreme precipitation can be characterized by the tail behavior of precipitation probability distributions.The tail contains the most extreme precipitation events and tells something about both the magnitude and frequency of these events. Here the heaviness amplification factor is used to represent the tail behavior. This research determines the most important hydro-climatological controls of the heaviness amplification factor. The study area is the entire world, where Antarctica and the oceans are not taken into account. A pre-defined set of 17 controls divided into four groups is used. The four groups are: general climate condition, geography, land cover and climate variability. The relation between the heaviness and the controls is determined with multiple linear regression, where the standardized controls are used. Since multiple linear regression is applied, multicollinearity might be a problem. This term refers to the occasion where independent variables are correlated to another independent variable or a linear combination of variables. This could lead to erroneous and unreliable results. To deal with this problem, three elimination methods are applied. This first method consists of a correlation analysis, where two controls are highly correlated if their correlation coefficient is higher than 0.9. The control with the lowest regression coefficient is eliminated. This is followed by another regression analysis with all the remaining controls, where controls with a low regression coefficient are also eliminated. The second method applies an iterative regression procedure, where in each iteration the control with the lowest coefficient is eliminated until three controls remain. The last method calculates the variance inflation factor, which is a measure for multicollinearity. It removes the controls with the highest variance inflation factor if higher than 10, until none of the controls has a score above this limit. The described analysis is done for the entire world, but also for 44 IPCC climate reference regions. The conclusion of the analysis is that the most important controls for the heaviness belong to the general climate condition and the climate variability. In total, 13 of the 17 controls occur in a top three for at least one region in one or more of the elimination methods. The most important control is precipitation variability which is in 57% of the regions in the top three based on the results of methods 1, 2 and 3. Each region gives a different top three or at least a different order of controls. The heaviness is calculated based on the outcomes of the multiple linear regression. The R-squared ranges from 0.2 to 0.8, with an average value around 0.3. This suggests that the controls explain about 30% of the variance. The average error for the world between the data and the calculation of the heaviness is 0.11. When the heaviness of a single region is calculated, the absolute error between the data and the calculation of the heaviness reduces from 0.12 to 0.10 when using regional coefficients. This result leads to the conclusion that regional processes do play a role in the behavior of extreme precipitation. This makes it less appropriate to define a global set of most important controls. None of the three elimination methods is able to solve the problem of multicollinearity completely. On average, 35% of the controls have a opposite sign between the correlation and regression coefficient.