Extreme Precipitation Modelling Using Geostatistics and Machine Learning Algorithms

The paper presents an approach for mapping of precipitation data. The main goal is to perform spatial predictions and simulations of precipitation fields using geostatistical methods (ordinary kriging, kriging with external drift) as well as machine learning algorithms (neural networks). More practically, the objective is to reproduce simultaneously both the spatial patterns and the extreme values. This objective is best reached by models integrating geostatistics and machine learning algorithms. To demonstrate how such models work, two case studies have been considered: first, a 2-day accumulation of heavy precipitation and second, a 6-day accumulation of extreme orographic precipitation. The first example is used to compare the performance of two optimization algorithms (conjugate gradients and Levenberg-Marquardt) of a neural network for the reproduction of extreme values. Hybrid models, which combine geostatistical and machine learning algorithms, are also treated in this context. The second dataset is used to analyze the contribution of radar Doppler imagery when used as external drift or as input in the models (kriging with external drift and neural networks). Model assessment is carried out by comparing independent validation errors as well as analyzing data patterns.

Simultaneous B(0)- and B(1)+-map acquisition for fast localized shim, frequency, and RF power determination in the heart at 3 T.

High-field (>or=3 T) cardiac MRI is challenged by inhomogeneities of both the static magnetic field (B(0)) and the transmit radiofrequency field (B(1)+). The inhomogeneous B fields not only demand improved shimming methods but also impede the correct determination of the zero-order terms, i.e., the local resonance frequency f(0) and the radiofrequency power to generate the intended local B(1)+ field. In this work, dual echo time B(0)-map and dual flip angle B(1)+-map acquisition methods are combined to acquire multislice B(0)- and B(1)+-maps simultaneously covering the entire heart in a single breath hold of 18 heartbeats. A previously proposed excitation pulse shape dependent slice profile correction is tested and applied to reduce systematic errors of the multislice B(1)+-map. Localized higher-order shim correction values including the zero-order terms for frequency f(0) and radiofrequency power can be determined based on the acquired B(0)- and B(1)+-maps. This method has been tested in 7 healthy adult human subjects at 3 T and improved the B(0) field homogeneity (standard deviation) from 60 Hz to 35 Hz and the average B(1)+ field from 77% to 100% of the desired B(1)+ field when compared to more commonly used preparation methods.