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

Area of Science:

• Environmental science
• Data science
• Geosciences

Background:

• Accurate spatiotemporal modeling of precipitation is vital for water resource management and hazard mitigation.
• Existing global precipitation models are limited by the intermittent, non-Gaussian, and geographically dependent nature of precipitation.
• There is a need for advanced modeling techniques to address these complexities.

Purpose of the Study:

• To propose and investigate a novel data-driven model for precipitation amount forecasting.
• To implement a non-parametric, warped Gaussian process regression (wGPR) for enhanced spatiotemporal precipitation modeling.
• To evaluate the performance of the proposed wGPR model using synthetic and real-world precipitation data.

Main Methods:

• Development of a data-driven, non-parametric warped Gaussian process regression (wGPR) model.
• Application of the wGPR model to a synthetic test function with non-Gaussian noise.
• Validation using a reanalysis dataset of monthly precipitation from Crete.
• Cross-validation analysis to assess interpolation accuracy for incomplete data.

Main Results:

• The proposed wGPR model demonstrated enhanced flexibility in handling non-Gaussian data.
• Cross-validation confirmed the advantages of non-parametric warping for interpolating incomplete precipitation data.
• The wGPR model achieved improved predictive accuracy compared to traditional methods for the studied cases.

Conclusions:

• The data-driven warped Gaussian process regression (wGPR) offers a powerful approach for spatiotemporal precipitation modeling.
• Non-parametric warping significantly improves the model's ability to handle the complexities of precipitation data.
• The proposed wGPR method provides a promising tool for more accurate precipitation forecasting and water resource management.