Search Results
You are looking at 1 - 3 of 3 items for
- Author or Editor: Marko Orescanin x
- Refine by Access: All Content x
Enhancing PMW Satellite Precipitation Estimation: Detecting Convective ClassAbstract
A decades-long effort in observing precipitation from space has led to continuous improvements of satellite-derived passive microwave (PMW) large-scale precipitation products. However, due to a limited ability to relate observed radiometric signatures to precipitation type (convective and stratiform) and associated precipitation rate variability, PMW retrievals are prone to large systematic errors at instantaneous scales. The present study explores the use of deep learning approach in extracting the information content from PMW observation vectors to help identify precipitation types. A deep learning neural network model (DNN) is developed to retrieve the convective type in precipitating systems from PMW observations. A 12-month period of Global Precipitation Measurement mission Microwave Imager (GMI) observations is used as a dataset for model development and verification. The proposed DNN model is shown to accurately predict precipitation types for 85% of total precipitation volume. The model reduces precipitation rate bias associated with convective and stratiform precipitation in the GPM operational algorithm by a factor of 2 while preserving the correlation with reference precipitation rates, and is insensitive to surface type variability. Based on comparisons against currently used convective schemes, it is concluded that the neural network approach has the potential to address regime-specific PMW satellite precipitation biases affecting GPM operations.
Abstract
A decades-long effort in observing precipitation from space has led to continuous improvements of satellite-derived passive microwave (PMW) large-scale precipitation products. However, due to a limited ability to relate observed radiometric signatures to precipitation type (convective and stratiform) and associated precipitation rate variability, PMW retrievals are prone to large systematic errors at instantaneous scales. The present study explores the use of deep learning approach in extracting the information content from PMW observation vectors to help identify precipitation types. A deep learning neural network model (DNN) is developed to retrieve the convective type in precipitating systems from PMW observations. A 12-month period of Global Precipitation Measurement mission Microwave Imager (GMI) observations is used as a dataset for model development and verification. The proposed DNN model is shown to accurately predict precipitation types for 85% of total precipitation volume. The model reduces precipitation rate bias associated with convective and stratiform precipitation in the GPM operational algorithm by a factor of 2 while preserving the correlation with reference precipitation rates, and is insensitive to surface type variability. Based on comparisons against currently used convective schemes, it is concluded that the neural network approach has the potential to address regime-specific PMW satellite precipitation biases affecting GPM operations.
Abstract
The recently installed S-band phased-array radar (PAR) at the National Weather Radar Testbed (NWRT) offers fast and flexible beam steering through electronic beam forming. This capability allows the implementation of a novel scanning strategy termed beam multiplexing (BMX), with the goal of providing fast updates of weather information with high statistical accuracy. For conventional weather radar the data acquisition time for a sector scan or a volume coverage pattern (VCP) can be reduced by increasing the antenna’s rotation rate to the extent that the pedestal allows. However, statistical errors of the spectral moment estimates will increase due to the fewer samples that are available for the estimation. BMX is developed to exploit the idea of collecting independent samples and maximizing the usage of radar resources. An improvement factor is introduced to quantify the BMX performance, which is defined by the reduction in data acquisition time using BMX when the same data accuracy obtained by a conventional scanning strategy is maintained. It is shown theoretically that a fast update without compromising data quality can be achieved using BMX at small spectrum widths and a high signal-to-noise ratio (SNR). Applications of BMX to weather observations are demonstrated using the PAR, and the results indicate that an average improvement factor of 2–4 can be obtained for SNR higher than 10 dB.
Abstract
The recently installed S-band phased-array radar (PAR) at the National Weather Radar Testbed (NWRT) offers fast and flexible beam steering through electronic beam forming. This capability allows the implementation of a novel scanning strategy termed beam multiplexing (BMX), with the goal of providing fast updates of weather information with high statistical accuracy. For conventional weather radar the data acquisition time for a sector scan or a volume coverage pattern (VCP) can be reduced by increasing the antenna’s rotation rate to the extent that the pedestal allows. However, statistical errors of the spectral moment estimates will increase due to the fewer samples that are available for the estimation. BMX is developed to exploit the idea of collecting independent samples and maximizing the usage of radar resources. An improvement factor is introduced to quantify the BMX performance, which is defined by the reduction in data acquisition time using BMX when the same data accuracy obtained by a conventional scanning strategy is maintained. It is shown theoretically that a fast update without compromising data quality can be achieved using BMX at small spectrum widths and a high signal-to-noise ratio (SNR). Applications of BMX to weather observations are demonstrated using the PAR, and the results indicate that an average improvement factor of 2–4 can be obtained for SNR higher than 10 dB.
Abstract
Visible and infrared radiance products of geostationary orbiting platforms provide virtually continuous observations of Earth. In contrast, low Earth orbiters observe passive microwave (PMW) radiances at any location much less frequently. Prior literature demonstrates the ability of a Machine Learning (ML) approach to build a link between these two complementary radiance spectra by predicting PMW observations using infrared and visible products collected from geostationary instruments, which could potentially deliver a highly-desirable synthetic PMW product with nearly continuous spatio-temporal coverage. However, current ML models lack the ability to provide a measure of uncertainty of such a product, significantly limiting its applications. In this work, Bayesian Deep Learning is employed to generate synthetic Global Precipitation Measurement (GPM) mission Microwave Imager (GMI) data from Advanced Baseline Imager (ABI) observations with attached uncertainties over the ocean. The study first uses deterministic Residual Networks (ResNets) to generate synthetic GMI brightness temperatures with as little mean absolute error as 1.72 K at the ABI spatio-temporal resolution. Then, for the same task, we use three Bayesian ResNet models to produce a comparable amount of error while providing previously unavailable predictive variance (i.e. uncertainty) for each synthetic data point. We find that the Flipout configuration provides the most robust calibration between uncertainty and error across GMI frequencies, and then demonstrate how this additional information is useful for discarding high-error synthetic data points prior to use by downstream applications.
Abstract
Visible and infrared radiance products of geostationary orbiting platforms provide virtually continuous observations of Earth. In contrast, low Earth orbiters observe passive microwave (PMW) radiances at any location much less frequently. Prior literature demonstrates the ability of a Machine Learning (ML) approach to build a link between these two complementary radiance spectra by predicting PMW observations using infrared and visible products collected from geostationary instruments, which could potentially deliver a highly-desirable synthetic PMW product with nearly continuous spatio-temporal coverage. However, current ML models lack the ability to provide a measure of uncertainty of such a product, significantly limiting its applications. In this work, Bayesian Deep Learning is employed to generate synthetic Global Precipitation Measurement (GPM) mission Microwave Imager (GMI) data from Advanced Baseline Imager (ABI) observations with attached uncertainties over the ocean. The study first uses deterministic Residual Networks (ResNets) to generate synthetic GMI brightness temperatures with as little mean absolute error as 1.72 K at the ABI spatio-temporal resolution. Then, for the same task, we use three Bayesian ResNet models to produce a comparable amount of error while providing previously unavailable predictive variance (i.e. uncertainty) for each synthetic data point. We find that the Flipout configuration provides the most robust calibration between uncertainty and error across GMI frequencies, and then demonstrate how this additional information is useful for discarding high-error synthetic data points prior to use by downstream applications.
