LBLeroy Bird

Leroy Bird

*This is an algorithmically generated author record based on publicly available information.

Current Organizations / Institutes

The organizations/institutes from the scholar's latest article


Leroy Bird has a total of 51 co-authors. Here are the top co-authors by number of shared publications:


Latest Publications

  • Open Access

    Sensitivity of extreme precipitation to climate change inferred using artificial intelligence shows high spatial variability

    by Leroy J. Bird, Gregory E. Bodeker, Kyle R. Clem

    Evaluating how extreme precipitation changes with climate is challenged by the paucity, brevity and inhomogeneity of observational records. Even when aggregating precipitation observations over large regions (obscuring potentially important spatial heterogeneity) the statistics describing extreme precipitation are often too uncertain to be of any practical value. Here we present an approach where a convolutional neural network (an artificial intelligence model) is trained on precipitation measurements from 10,000 stations to learn the spatial structure of the parameters of a generalised extreme value model, and the sensitivity of those parameters to the annual mean, global mean, surface temperature. The method is robust against the limitations of the observational record and avoids the short-comings of regional and global climate models in simulating the sensitivity of extreme precipitation to climate change. The maps of the sensitivity of extreme precipitation to climate change, on ~1.5 km × 1.5 km grids over North America, Europe, Australia and New Zealand, derived using the successfully trained convolutional neural network, show high spatial variability.


  • Open Access

    A forecast-model-based extreme weather event attribution system developed for Aotearoa New Zealand

    by Jordis Sonnhild Tradowsky, Greg E Bodeker, Christopher John Noble, Dáithí A Stone, Graham David Rye, Leroy James Bird, William I Herewini, Sapna Rana, Johannes Rausch, Iman Soltanzadeh

    A largely automated extreme weather event (EWE) attribution system has been developed that uses the Weather Research and Forecast numerical weather prediction model to simulate EWEs under current and pre-industrial climate conditions. The system has been applied to two extreme precipitation events in Aotearoa New Zealand with the goal of quantifying the effect of anthropogenic climate change on the severity of these events. The forecast simulation of the target event under current climate conditions constitutes the first scenario (ALL). We then apply a climate change signal in the form of delta fields in sea-surface temperature, atmospheric temperature and specific humidity, creating a second ‘naturalised’ scenario (NAT) which is designed to represent the weather system in the absence of human interference with the climate system. A third scenario, designed to test for coherence, is generated by applying deltas of opposite sign compared to the naturalised scenario (ALL+). Each scenario comprises a 22-member ensemble which includes one simulation that was not subject to stochastic perturbation. Comparison of the three ensembles shows that: (1) the NAT ensemble develops an extreme event which resembles the observed event, (2) the severity, i.e. maximum intensity and/or the size of area affected by heavy precipitation, changes when naturalising the boundary conditions, (3) the change in severity is consistently represented within the three scenarios and the signal is robust across the different ensemble members, i.e. it is typically shown in most of the 22 ensemble members. Thus, the attribution system presented here can be used to provide information about the influence of anthropogenic climate change on the severity of specific extreme events.


  • Open Access

    Deep learning for stochastic precipitation generation – deep SPG v1.0

    by Leroy J. Bird, Matthew G. W. Walker, Greg E. Bodeker, Isaac H. Campbell, Guangzhong Liu, Swapna Josmi Sam, Jared Lewis, Suzanne M. Rosier

    We present a deep-neural-network-based single-site stochastic precipitation generator (SPG), capable of producing realistic time series of daily and hourly precipitation. The neural network outputs a wet-day probability and precipitation distributions in the form of a mixture model. The SPG was tested in four different locations in New Zealand, and we found it accurately reproduced the precipitation depth, the autocorrelations seen in the original data, the observed dry-spell lengths, and the seasonality in precipitation. We present two versions of the hourly and daily SPGs: (i) a stationary version of the SPG that assumes that the statistics of the precipitation are time independent and (ii) a non-stationary version that captures the secular drift in precipitation statistics resulting from climate change. The latter was developed to be applicable to climate change impact studies, especially studies reliant on SPG projections of future precipitation. We highlight many of the pitfalls associated with the training of a non-stationary SPG on observations alone and offer an alternative method that replicates the secular drift in precipitation seen in a large-ensemble regional climate model. The SPG runs several orders of magnitude faster than a typical regional climate model and permits the generation of very large ensembles of realistic precipitation time series under many climate change scenarios. These ensembles will also contain many extreme events not seen in the historical record.


  • Open Access

    Sensitivity of extreme precipitation to climate change inferred using artificial intelligence shows high spatial variability

    by Leroy Bird, Gregory Bodeker, Kyle Clem

    Evaluating changes in extreme precipitation, and how those extremes will change with climate, is challenged by the paucity, brevity and inhomogeneity of observational time series. Using global precipitation measurements from 10,000 stations, we successfully demonstrate an artificial intelligence-based method for quantifying the sensitivity of extreme precipitation to climate change that is robust against the limitations of the observational record; a convolutional neural network (CNN) is trained to learn the spatial structure of Generalized Extreme Value (GEV) parameters describing the statistics of precipitation extremes. Given an annual mean, global mean surface temperature time series as a climate covariate, the CNN also learns the non-stationarity of the GEV fit parameters. The trained CNN is used to generate maps of the sensitivity of extreme precipitation to global temperature change on ∼1.5km×1.5km grids over North America, Europe, Australia and New Zealand. This observations based method avoids the short-comings of regional and global climate models in simulating expected changes in extreme precipitation.

  • Open Access

    Deep learning for stochastic precipitation generation – Deep SPG v1.0

    by Leroy Bird, Matthew Walker, Greg Bodeker, Isaac Campbell, Guangzhong Liu, Swapna Josmi Sam, Jared Lewis, Suzanne Rosier

    We present a deep neural network based single site stochastic precipitation generator (SPG), capable of producing realistic time series of daily and hourly precipitation. The neural network outputs a wet day probability and precipitation distributions in the form of a mixture model. The SPG was tested in four different locations in New Zealand, and we found it accurately reproduced the precipitation depth, the autocorrelations seen in the original data, the observed dry-spell lengths and the seasonality in precipitation. We present two versions of the hourly and daily SPGs: (i) a stationary version of the SPG that assumes that the statistics of the precipitation are time independent (ii) a non-stationary version that captures the secular drift in precipitation statistics resulting from climate change. The latter was developed to be applicable to climate change impact studies, especially, studies reliant on SPG projections of future precipitation. We highlight many of the pitfalls associated with the training of a non-stationary SPG on observations alone, and offer an alternative method that replicates the secular drift in precipitation seen in a large-ensemble regional climate model. The SPG runs several orders of magnitude faster than a typical regional climate model, and permits the generation of very large ensembles of realistic precipitation time series under many climate change scenarios, these ensembles will also contain many extreme events not seen in the historical record.

  • Open Access

    The effect of experiment conditioning on estimates of human influence on extreme weather

    by Dáithí A. Stone, Suzanne M. Rosier, Leroy Bird, Luke J. Harrington, Sapna Rana, Stephen Stuart, Sam M. Dean


  • Open Access

    The MAPM (Mapping Air Pollution eMissions) method for inferring particulate matter emissions maps at city scale from in situ concentration measurements: description and demonstration of capability

    by Brian Nathan, Stefanie Kremser, Sara Mikaloff-Fletcher, Greg Bodeker, Leroy Bird, Ethan Dale, Dongqi Lin, Gustavo Olivares, Elizabeth Somervell

    Mapping Air Pollution eMissions (MAPM) is a 2-year project whose goal is to develop a method to infer particulate matter (PM) emissions maps from in situ PM concentration measurements. Central to the functionality of MAPM is an inverse model. The input of the inverse model includes a spatially distributed prior emissions estimate and PM measurement time series from instruments distributed across the desired domain. In this proof-of-concept study, we describe the construction of this inverse model, the mathematics underlying the retrieval of the resultant posterior PM emissions maps, the way in which uncertainties are traced through the MAPM processing chain, and plans for future developments. To demonstrate the capability of the inverse model developed for MAPM, we use the PM2.5 measurements obtained during a dedicated winter field campaign in Christchurch, New Zealand, in 2019 to infer PM2.5 emissions maps on a city scale. The results indicate a systematic overestimation in the prior emissions for Christchurch of at least 40 %–60 %, which is consistent with some of the underlying assumptions used in the composition of the bottom-up emissions map used as the prior, highlighting the uncertainties in bottom-up approaches for estimating PM2.5 emissions maps.


  • Open Access

    The winter 2019 air pollution (PM2.5) measurement campaign in Christchurch, New Zealand

    by Ethan R. Dale, Stefanie Kremser, Jordis S. Tradowsky, Greg E. Bodeker, Leroy J. Bird, Gustavo Olivares, Guy Coulson, Elizabeth Somervell, Woodrow Pattinson, Jonathan Barte, et al.

    MAPM (Mapping Air Pollution eMissions) is a project whose goal is to develop a method to infer airborne particulate matter (PM) emissions maps from in situ PM concentration measurements. In support of MAPM, a winter field campaign was conducted in New Zealand in 2019 (June to September) to obtain the measurements required to test and validate the MAPM methodology. Two different types of instruments measuring PM were deployed: ES-642 remote dust monitors (17 instruments) and Outdoor Dust Information Nodes (ODINs; 50 instruments). The measurement campaign was bracketed by two intercomparisons where all instruments were co-located, with a permanently installed tapered element oscillating membrane (TEOM) instrument, to determine any instrument biases. Changes in biases between the pre- and post-campaign intercomparisons were used to determine instrument drift over the campaign period. Once deployed, each ES-642 was co-located with an ODIN. In addition to the PM measurements, meteorological variables (temperature, pressure, wind speed, and wind direction) were measured at three automatic weather station (AWS) sites established as part of the campaign, with additional data being sourced from 27 further AWSs operated by other agencies. Vertical profile measurements were made with 12 radiosondes during two 24 h periods and complimented measurements made with a mini micropulse lidar and ceilometer. Here we present the data collected during the campaign and discuss the correction of the measurements made by various PM instruments. We find that when compared to measurements made with a simple linear correction, a correction based on environmental conditions improves the quality of measurements retrieved from ODINs but results in over-fitting and increases the uncertainties when applied to the more sophisticated ES-642 instruments. We also compare PM2.5 and PM10 measured by ODINs which, in some cases, allows us to identify PM from natural and anthropogenic sources. The PM data collected during the campaign are publicly available from https://doi.org/10.5281/zenodo.4542559 (Dale et al., 2020 b), and the data from other instruments are available from https://doi.org/10.5281/zenodo.4536640 (Dale et al., 2020 a).


  • Open Access

    WRF4PALM v1.0: a mesoscale dynamical driver for the microscale PALM model system 6.0

    by Dongqi Lin, Basit Khan, Marwan Katurji, Leroy Bird, Ricardo Faria, Laura E. Revell

    A set of Python-based tools, WRF4PALM, has been developed for offline nesting of the PALM model system 6.0 into the Weather Research and Forecasting (WRF) modelling system. Time-dependent boundary conditions of the atmosphere are critical for accurate representation of microscale meteorological dynamics in high-resolution real-data simulations. WRF4PALM generates initial and boundary conditions from WRF outputs to provide time-varying meteorological forcing for PALM. The WRF model has been used across the atmospheric science community for a broad range of multidisciplinary applications. The PALM model system 6.0 is a turbulence-resolving large-eddy simulation model with an additional Reynolds-averaged Navier–Stokes (RANS) mode for atmospheric and oceanic boundary layer studies at microscale (Maronga et al., 2020). Currently PALM has the capability to ingest output from the regional scale Consortium for Small-scale Modelling (COSMO) atmospheric prediction model. However, COSMO is not an open source model and requires a licence agreement for operational use or academic research (http://www.cosmo-model.org/, last access: 23 April 2021). This paper describes and validates the new free and open-source WRF4PALM tools (available at https://github.com/dongqi-DQ/WRF4PALM, last access: 23 April 2021). Two case studies using WRF4PALM are presented for Christchurch, New Zealand, which demonstrate successful PALM simulations driven by meteorological forcing from WRF outputs. The WRF4PALM tools presented here can potentially be used for micro- and mesoscale studies worldwide, for example in boundary layer studies, air pollution dispersion modelling, wildfire emissions and spread, urban weather forecasting, and agricultural meteorology.


  • Open Access

    The MAPM (Mapping Air Pollution eMissions) method for inferring particulate matter emissions maps at city-scale from in situ concentration measurements: description and demonstration of capability

    by Brian Nathan, Stefanie Kremser, Sara Mikaloff-Fletcher, Greg Bodeker, Leroy Bird, Ethan Dale, Dongqi Lin, Gustavo Olivares, Elizabeth Somervell

    MAPM (Mapping Air Pollution eMissions) is a two-year project whose goal is to develop a method to infer particulate matter (PM) emissions maps from in situ PM concentration measurements. Central to the functionality of MAPM is an inverse model. The input of the inverse model includes a spatially-distributed prior emissions estimate and PM measurement time series from instruments distributed across the desired domain. Here we describe the construction of this inverse model, the mathematics underlying the retrieval of the resultant posterior PM emissions maps, the way in which uncertainties are traced through the MAPM processing chain, and plans for future development of the processing chain. To demonstrate the capability of the inverse model developed for MAPM, we use the PM2.5 measurements obtained during a dedicated winter field campaign in Christchurch, New Zealand, in 2019 to infer PM2.5 emissions maps on city scale. The results indicate a systematic overestimation in the prior emissions for Christchurch of at least 40–60 %, which is consistent with some of the underlying assumptions used in the composition of the bottom-up emissions map used as the prior, highlighting the uncertainties in bottom-up approaches for estimating PM2.5 emissions maps. The paper also presents the results of two sets of observing system simulation experiments (OSSEs) that explore how measurement uncertainties affect the computation of the derived emissions maps, and the extent to which using emissions maps from one day as the prior for the next day improves the ability of the inversion system to characterize the emissions sources. We find in the first case that a smaller number of high-accuracy instruments performs significantly better than a higher number of low-accuracy instruments. In the second case, the results are ultimately inconclusive, showing the need for further investigations that are beyond the scope of this study.