This research is co-financed by Greece and the European Union (European Social Fund- ESF) through the Operational Programme «Human Resources Development, Education and Lifelong Learning 2014-2020»in the context of the project “Gaussian Anamorphosis with Kernel Estimators for Spatially Distributed Data and Time Series and Applications in the Analysis of Precipitation ” (MIS 5052133).

EGU 2021 Presentation by the Research Team

Agou, V. D., Pavlides, A., and Hristopulos, D. T., “Space-Time Analysis of Precipitation Reanalysis Data for the Island of Crete using Gaussian Anamorphosis with Hermite Polynomials”, 2021.

Preprint of Research Paper

Andrew Pavlides, Vasiliki Agou, Dionissios T. Hristopulos, "Non-parametric Kernel-Based Estimation of Probability Distributions for Precipitation Modeling," September 2021.  Manuscript submitted to the journal Advances in Water Resources.


Preprint available on the Arxiv e-print open-access repository. 

Read our publication in ENTROPY journal

Spatial Modeling of Precipitation Based on Data-Driven Warping of Gaussian Processes

Gaussian Anamorphosis with Kernel Function Estimators for Spatially Distributed Data and Time Series with Applications in Precipitation Analysis

Using Kernel Functions to Estimate the Probability Distribution of Skewed Data

The probability distribution of precipitation amount strongly depends on geography, climate zone, and time scale considered. Closed-form parametric probability distributions are not sufficiently flexible to provide accurate and universal models for precipitation amount over different time scales. In this paper we derive non-parametric estimates of the cumulative distribution function (CDF) of precipitation amount for wet time intervals. The CDF estimates are obtained by integrating the kernel density estimator leading to semi-explicit CDF expressions for different kernel functions. We investigate kernel-based CDF estimation with an adaptive plug-in bandwidth (KCDE), using both synthetic data sets and reanalysis precipitation data from the island of Crete (Greece). We show that KCDE provides better estimates of the probability distribution than the standard empirical (staircase) estimate and kernel-based estimates that use the normal reference bandwidth. We also demonstrate that KCDE enables the simulation of non-parametric precipitation amount distributions by means of the inverse transform sampling method.


Space-Time Analysis of Precipitation Reanalysis Data for the Island of Crete using Gaussian Anamorphosis with Hermite Polynomials

  • Climate change poses a serious threat to environmental and economic sustainability
  • Lack of a universal probability distribution that can fit monthly precipitation data
  • Spatial variability of rainfall and dependence on complex topography
  • We investigate flexible normalizing transformations for the geostatistical analysis of precipitation
  • We generate new, improved precipitation maps and uncertainty estimates based on reanalysis data for the island of Crete


Analyzing Precipitation Data from Crete

We used observational, synthetic (simulated) and reanalysis precipitation data.  The observation record is very sparse for Crete. Hence, we relied heavily on ERA5 precipitation data. The ERA5 spatial grid around Crete is shown below.  

Precipitation Maps: Year 2008

We constructed precipitation maps for different years based on ERA5 reanalysis data for the island of Crete. The generation of the maps employed geostatistical methods and Gaussian anamorphosis by means of Hermite polynomials. 

Cross-validation shows improved performance of Gaussian anamorphosis with Hermite polynomials

Using cross-validation it was shown that the maps obtained using Gaussian anamorphosis with Hermite polynomials are more accurate than those generated by means of the standard Ordinary Kriging approach.