Search

Over the past decade the use of machine learning in meteorology has grown rapidly. Specifically neural networks and deep learning have been used at an unprecedented rate. To fill the dearth of resources covering neural networks with a meteorological lens, this paper discusses machine learning methods in a plain language format that is targeted to the operational meteorological community. This is the second paper in a pair that aim to serve as a machine learning resource for meteorologists. While the first paper focused on traditional machine learning methods (e.g., random forest), here a broad spectrum of neural networks and deep learning methods is discussed. Specifically, this paper covers perceptrons, artificial neural networks, convolutional neural networks, and U-networks. Like the Part I paper, this manuscript discusses the terms associated with neural networks and their training. Then the manuscript provides some intuition behind every method and concludes by showing each method used in a meteorological example of diagnosing thunderstorms from satellite images (e.g., lightning flashes). This paper is accompanied with an open-source code repository to allow readers to explore neural networks using either the dataset provided (which is used in the paper) or as a template for alternate datasets. [ABSTRACT FROM AUTHOR]

The Cyclone Global Navigation Satellite System (CYGNSS) mission has generated several new ocean surface data products. Because of its high overpass rate and ability to measure surface wind speeds through weather, these data products are typically higher temporal resolution than traditional satellite tropical cyclone (TC) data products. However, the nature of the Global Navigation Satellite Systems reflectometry (GNSS-R) signal, which is essentially a single pixelwide measurement along the CYGNSS specular point (SP) track on the ground, necessitates aggregating the wind speed data over a period of time to accurately characterize TCs in terms of intensity and structure. The standard CYGNSS level 3 (L3) wind products are averaged over 1 h, which typically is not sufficient to produce an accurate characterization. A new L3 storm-centric gridded (SCG) wind speed product is presented here. It has been developed to improve upon the standard L3 algorithm for the purpose of providing better characterization of TC wind fields over their storm life cycle. This paper describes the L3 SCG algorithm, provides examples of L3 SCG wind fields, and discusses the potential uses and limitations of the new data product. [ABSTRACT FROM AUTHOR]

There are many sources of uncertainty in satellite precipitation retrievals of warm rain. In this paper, the second of a two-part study, we focus on uncertainties related to spatial heterogeneity and surface clutter. A cloud-resolving model simulation of warm, shallow clouds is used to simulate satellite observations from three theoretical satellite architectures—one similar to the Global Precipitation Measurement Core Observatory, one similar to CloudSat, and one similar to the planned Atmosphere Observing System (AOS). Rain rates are then retrieved using a common optimal estimation framework. For this case, retrieval biases due to nonuniform beamfilling are very large, with retrieved rain rates negatively (low) biased by as much as 40%–50% (depending on satellite architecture) at 5 km horizontal resolution. Surface clutter also acts to negatively bias retrieved rain rates. Combining all sources of uncertainty, the theoretical AOS satellite is found to outperform CloudSat in terms of retrieved surface rain rate, with a bias of −19% as compared with −28%, a reduced spread of retrieval errors, and an additional 17.5% of cases falling within desired uncertainty limits. The results speak to the need for additional high-resolution modeling simulations of warm rain so as to better characterize the uncertainties in satellite precipitation retrievals. [ABSTRACT FROM AUTHOR]

Sea surface temperature (SST) fronts are important for fisheries and marine ecology, as well as upper-ocean dynamics, weather forecasting, and climate monitoring. In this paper, we propose a new approach to detect SST fronts from RADARSAT-2 ScanSAR images, based on the correlation of SAR-derived wind speeds using the gray level cooccurrence matrix (GLCM) approach. Due to the large differences between the correlation of wind speeds for SST fronts compared to other areas, SST fronts can be detected by the threshold method. To eliminate small-scale features (or noise), the 30 km scale is used as the length threshold for the detection of the SST fronts. The proposed method is effective when wind speeds are between 3 and 13 m s−1. The overall accuracy of our method is about 93.6%, which is sufficient for operational applications. [ABSTRACT FROM AUTHOR]

Two spaceborne radars currently in orbit enable the sampling of snowfall near the surface and throughout the atmospheric column, namely, CloudSat's Cloud Profiling Radar (CPR) and the Global Precipitation Measurement mission's Dual-Frequency Precipitation Radar (GPM-DPR). In this paper, a direct comparison of the CPR's 2C-SNOW-PROFILE (2CSP), the operational GPM-DPR algorithm (2ADPR) and a neural network (NN) retrieval applied to the GPM-DPR data is performed using coincident observations between both radars. Examination of over 3500 profiles within moderate to strong precipitation (Ka band ≥ 18 dBZ) show that the NN retrieval provides the closest retrieval of liquid equivalent precipitation rate R immediately above the melting level to the R retrieved just below the melting layer, agreeing within 5%. Meanwhile, 2CSP retrieves a maximum value of R at −15°C, decreases by 35% just above the melting layer, and is about 50% smaller than the GPM-DPR retrieved R below the melting layer. CPR-measured reflectivity shows median reduction of 2–3 dB from −15° to −2.5°C, likely the reason for the 2CSP retrieval reduction of R. Two case studies from NASA field campaigns [i.e., Olympic Mountains Experiment (OLYMPEX) and Investigation of Microphysics and Precipitation for Atlantic Coast-Threatening Snowstorms (IMPACTS)] provide analogs to the type of precipitating systems found in the comparison between retrieval products. For the snowfall events that GPM-DPR can observe, this work suggests that the 2CSP retrieval is likely underestimating the unattenuated reflectivity, resulting in a potential negative, or low, bias in R. Future work should investigate how frequently the underestimated reflectivity profiles occur within the CPR record and quantify its potential effects on global snowfall accumulation estimation. [ABSTRACT FROM AUTHOR]