Large-scale Atmospheric Environment Research Articles

High-resolution downscaling with interpretable deep learning: Rainfall extremes over New Zealand

The gap in resolution between existing global climate model output and that sought by decision-makers drives an ongoing need for climate downscaling. Here we test the extent to which developments in deep learning can out-perform existing statistical approaches for downscaling historical rainfall in the highly complex terrain setting of New Zealand. While deep learning removes the need for manual feature selection when extracting spatial-temporal information from predictor fields, several key considerations need to be addressed. These include: the chosen complexity of the network architecture, suitable loss functions tailored to the problem, as well as input data considerations of domain size and amount of training data required to provide adequate out-of-sample generalization. Sensitivity testing to these considerations reveals that a relatively simple convolutional neural network (CNN) architecture with carefully selected loss functions can considerably outperform existing statistical downscaling models based on multiple linear regression with manual feature selection. When aggregated across the entire region, the fraction of explained variance on wet days increased from 0.35 to 0.52, the root-mean-squared error reduced by over 20% and percentage biases for the 90th percentile of rainfall improved by over 25%. Using interpretable machine learning methods, we demonstrate that the CNN has been capable of self-learning physically plausible relationships between the large-scale atmospheric environment and extreme localized rainfall events. The historical performance and physical interpretability documented here lends support for wider development and application of deep learning in climate downscaling.

Variability and Predictability of Basinwide and Sub-Basin Tropical Cyclone Genesis Frequency in the Northwest Pacific

Abstract The variability and predictability of tropical cyclone genesis frequency (TCGF) during 1973–2010 at both basinwide and sub-basin scales in the northwest Pacific are investigated using a 100-member ensemble of 60-km-resolution atmospheric simulations that are forced with observed sea surface temperatures (SSTs). The sub-basin regions include the South China Sea (SCS) and the four quadrants of the open ocean. The ensemble-mean results well reproduce the observed interannual-to-decadal variability of TCGF in the southeast (SE), northeast (NE), and northwest (NW) quadrants, but show limited skill in the SCS and the southwest (SW) quadrant. The skill in the SE and NE quadrants is responsible for the model’s ability to replicate the observed variability in basinwide TCGF. Above-normal TCGF is tied to enhanced relative SST (i.e., local SST minus tropical-mean SST) either locally or to the southeast of the corresponding regions in both the observations and ensemble mean for the SE, NE, and NW quadrants, but only in the ensemble mean for the SCS and the SW quadrant. These results demonstrate the strong SST control of TCGF in the SE, NE, and NW quadrants; both empirical and theoretical analyses suggest that ensembles of ∼10, 20, 35, and 15 members can capture the SST-forced TCGF variability in these three sub-basin regions and the entire basin, respectively. In the SW quadrant and the SCS, TCGF contains excessive noise, particularly in the observations, and thus shows low predictability. The variability and predictability of the large-scale atmospheric environment and synoptic-scale disturbances and their contributions to those of TCGF are also discussed.

Connection between the South and East Asian Monsoons: Comparing Summer Monsoon Rainfall of Pakistan and South Korea

This study investigates the tele-connection of the southeast Asian monsoon systems by comparing the summer monsoon (June to September) rainfall variability between Pakistan and south Korea. The daily data sets (19812014) of rainfall of Pakistan and south Korea are utilized to explore the possible link. The data products of the National Centers for Environmental Prediction and National Center for Atmospheric Research (NCEP/NCAR) were also used for the understanding of the large-scale atmospheric environments. The patterns of summer monsoon rainfall on a daily basis between Pakistan and south Korea followed to each other throughout the year. Sub-seasonal differences of the summer monsoon revealed that July is the wettest month in both countries. The large-scale atmospheric environment of higher geopotential height revealed that the Tibetan high and the western north Pacific subtropical high are showing positive anomalies during positive phases over south Asia and east Asia, respectively. The anomalies of zonal wind are negative during positive phase and adverse in the negative phase between 20-40oN. The reduced westerly is interpreted as the seasonal variation and moving of jet streams from the east Asian route. The Tibetan high, northwestern Pacific subtropical high and the east Asian jet stream have reliable and sufficient linkage between the Pakistan and south Korea summer monsoon system.

Modeling Tropical Cyclone Induced Stream Flow in Tar Pamlico River of North Carolina

ABSTRACT In order to better understand and simulate large coastal watershed hydrology and hydro-meteorological processes associated with tropical cyclone (TC) induced inland flooding, the Annualized Agricultural Nonpoint source pollution Model (A nn AGNPS) with Muskingum channel routing has been applied to the Tar-pamlico river basin in North Carolina, USA. The study focuses on three major hydro-meteorological processes: 1) large-scale atmospheric environment, 2) watershed hydrological processes and 3) groundwater response. The modeling results indicate that although the AnnAGNPS model can perform well in predicting the total amount of watershed runoff, channel routing is needed to improve the hydrographs of flow discharge during hurricane events. Sensitivity analysis of soil saturated hydrological conductivity ( K s ) indicates that both base flow and event total runoff are sensitive to K s . Base flow increases as K s increases when 0 K s ≤ 15 m/day, but decreases slightly when K s > 15 m/day. Peak runoff exponentially decreases as K s increases. The results also show that without the preceding Hurricane Dennis, the outlet discharge as a result of Hurricane Floyd would have been as much as 37% lower than that caused by the combined Dennis-Floyd effect.

Characteristics of Explosive Cyclones over the Northern Pacific

AbstractExplosive cyclones (ECs) over the northern Pacific Ocean during the cold season (October–April) over a 15-yr (2000–15) period are analyzed by using the Final (FNL) Analysis data provided by the National Centers for Environmental Prediction. These ECs are stratified into four categories according to their intensity: weak, moderate, strong, and super ECs. In addition, according to the spatial distribution of their maximum-deepening-rate positions, ECs are further classified into five regions: the Japan–Okhotsk Sea (JOS), the northwestern Pacific (NWP), the west-central Pacific (WCP), the east-central Pacific (ECP), and the northeastern Pacific (NEP). The occurrence frequency of ECs shows evident seasonal variations for the various regions over the northern Pacific. NWP ECs frequently occur in winter and early spring, WCP and ECP ECs frequently occur in winter, and JOS and NEP ECs mainly occur in autumn and early spring. The occurrence frequency, averaged maximum deepening rate, and developing and explosive-developing lifetimes of ECs decrease eastward over the northern Pacific, excluding JOS ECs, consistent with the climatological intensity distributions of the upper-level jet stream, midlevel positive vorticity, and low-level baroclinicity. On the seasonal scale, the occurrence frequency and spatial distribution of ECs are highly correlated with the intensity and position of the upper-level jet stream, respectively, and also with those of midlevel positive vorticity and low-level baroclinicity. Over the northwestern Pacific, the warm ocean surface also contributes to the rapid development of ECs. The composite analysis indicates that the large-scale atmospheric environment for NWP and NEP ECs shows significant differences from that for the 15-yr cold-season average. The southwesterly anomalies of the upper-level jet stream and positive anomalies of midlevel vorticity favor the prevalence of NWP and NEP ECs.

Statistical characteristics of environmental parameters for warm season short-duration heavy rainfall over central and eastern China

Water vapor content, instability, and convergence conditions are the key to short-duration heavy rainfall forecasting. It is necessary to understand the large-scale atmospheric environment characteristics of short-duration heavy rainfall by investigating the distribution of physical parameters for different hourly rainfall intensities. The observed hourly rainfall data in China and the NCEP final analysis (FNL) data during 1 May and 30 September from 2002 to 2009 are used. NCEP FNL data are 6-hourly, resulting in sample sizes of 1573370, 355346, and 11401 for three categories of hourly rainfall (P) of no precipitation (P < 0.1 mm h−1), ordinary precipitation (0.1⩽ P < 20 mm h−1), and short-duration heavy rainfall (P ⩾ 20.0 mm h−1), respectively, by adopting a temporal matching method. The results show that the total precipitable water (PWAT) is the best parameter indicating the hourly rainfall intensity. A PWAT of 28 mm is necessary for any short-duration heavy rainfall. The possibility of short-duration heavy rainfall occurrence increases with PWAT, and a PWAT of 59 mm is nearly sufficient. The specific humidity is a better indicator than relative humidity. Both 700- and 850-hPa relative humidity greater than 80% could be used to determine whether or not it is going to rain, but could not be used to estimate the rainfall intensity. Temperature and potential pseudo-equivalent temperature are also reasonable indicators of short-duration heavy rainfall. Among the atmospheric instability parameters, the best lifted index (BLI) performs best on the short-duration rainfall discrimination; the next best is the K index (KI). The three rainfall categories are not well recognized by total totals (TT) or the temperature difference between 850 and 500 hPa (DT85). Three-quarters of short-duration heavy rainfall occurred with BLI less than -0.9, while no short-duration heavy rainfall occurred when BLI was greater than 2.6. The minimum threshold of KI was 28.1 for short-duration heavy rainfall. The importance of dynamic conditions was well demonstrated by the 925- and 850-hPa divergence. The representativeness of 925-hPa divergence is stronger than that of 850 hPa. Three-quarters of short-duration heavy rainfall occurred under a negative divergence environment. However, both the best convective potential energy (BCAPE) and vertical wind shear were unable to discriminate the hourly rainfall intensities.

An Analysis of the Environmental Moisture Impacts of Western North Pacific Tropical Cyclones

Abstract The present study examines the environmental moisture anomalies present during western North Pacific tropical cyclone (TC) passage using storm-relative composites. Composited precipitable water anomalies reveal asymmetric anomalies with dry anomalies to the northwest and southwest of the TC and moist anomalies to the east of the TC. Precipitable water anomalies filtered in space and time suggest that the moisture anomalies in the northwest, southwest, and east regions (NWR, SWR, and ER, respectively) are partially due to the TC, while the anomalies in the SWR are also forced by a convectively suppressed Madden–Julian oscillation (MJO) and equatorial Rossby wave (ERW). Composited vertically integrated moisture budgets and backward parcel trajectories reveal that the moisture anomalies in the NWR, SWR, and ER are primarily due to the convergence of climatological mean moisture by the anomalous meridional wind. This convergence is induced by the secondary circulation of the TC in the NWR and ER and by inertial instability induced by the TC, MJO, and ERW in the SWR and ER as also suggested by prior work. Dry anomalies in the NWR are also forced by the advection of moisture by lower-tropospheric northerly wind anomalies associated with the primary circulation of the TC. Together with prior work, these results suggest that TCs can have significant impacts on their large-scale atmospheric environment extending well beyond the spatiotemporal scales of the lower-tropospheric cyclonic circulation of the TC.

Cloud fraction at the ARM SGP site: reducing uncertainty with self-organizing maps.

Instrument downtime leads to uncertainty in the monthly and annual record of cloud fraction (CF), making it difficult to perform time series analyses of cloud properties and perform detailed evaluations of model simulations. As cloud occurrence is partially controlled by the large-scale atmospheric environment, this knowledge is used to reduce uncertainties in the instrument record. Synoptic patterns diagnosed from the North American Regional Reanalysis (NARR) during the period 1997–2010 are classified using a competitive neural network known as the self-organizing map (SOM). The classified synoptic states are then compared to the Atmospheric Radiation Measurement (ARM) Southern Great Plains (SGP) instrument record to determine the expected CF. A number of SOMs are tested to understand how the number of classes and the period of classifications impact the relationship between classified states and CFs. Bootstrapping is utilized to quantify the uncertainty of the instrument record when statistical information from the SOM is included. Although all SOMs significantly reduce the uncertainty of the CF record calculated in Kennedy et al. (Theor Appl Climatol 115:91–105, 2014), SOMs with a large number of classes and separated by month are required to produce the lowest uncertainty and best agreement with the annual cycle of CF. This result may be due to a manifestation of seasonally dependent biases in NARR. With use of the SOMs, the average uncertainty in monthly CF is reduced in half from the values calculated in Kennedy et al. (Theor Appl Climatol 115:91–105, 2014).

Modeling the Impacts of the Large-Scale Atmospheric Environment on Inland Flooding during the Landfall of Hurricane Floyd (1999)

The contribution of the large-scale atmospheric environment to precipitation and flooding during Hurricane Floyd was investigated in this study. Through the vortex removal technique in the Weather Research and Forecasting (WRF) model, the vortex associated with Hurricane Floyd (1999) was mostly removed in the model initial conditions and subsequent integration. Results show that the environment-induced precipitation can account for as much as 22% of total precipitation in the innermost model domain covering North Carolina coastal area and 7% in the focused hydrological study area. The high-resolution precipitation data from the WRF model was then used for input in a hydrological model to simulate river runoff. Hydrological simulation results demonstrate that without the tropical systems and their interactions with the large-scale synoptic environment the synoptic environment would only contribute 10% to the total discharge at the Tarboro gauge station. This suggests that Hurricane Floyd and Hurricane Dennis preceding it, along with the interactions between these tropical systems and the large-scale environment, have contributed to the bulk (90%) of the record amount of flood water in the Tar-Pamlico River Basin.

Forecasting in the Vertical with a Local Dynamical Interpretation Method

Abstract A one-dimensional (I-D) planetary boundary-layer model, including a complete set of simple physical parameterizations, has been used since July 1986 to predict daily soundings at Trappes in the suburbs of Paris. This model is coupled to the French operational spectral hemispheric model, representing the large-scale atmospheric environment by means of horizontal gradients and vertical velocity in the advective terms, and geostrophic wind. The study of the statistical scores (mean absolute error and mean error) for 220 cases during the year 1986–87 provides good accuracy for the 12-h forecast (as compared to fine mesh modeling and statistical interpretation methods) but exhibits a loss of accuracy for periods longer than 12 h. Improvement in the results through adjustment of some of the model parameters with the help of a file containing 22 test situations appreciably reduces the mean absolute error, this method would thus be useful for local forecasts of sensitive weather elements. Further improve...