Orographic precipitation is a critical freshwater source and major flooding hazard, but its distribution and behavior over complex terrain are often uncertain due to sparse observations. We examine precipitation and its drivers in one of the wettest regions in the world, the Southern Alps of New Zealand (NZ), using the first multi-decadal simulation by a convection-permitting regional climate model across all mainland NZ at 2.2 km grid-scale. Model skill is primarily assessed against direct measurements by more than 170 rain gauges to avoid uncertainty commonly introduced by gridded observations in remote regions. Peak intensity and duration of sub-daily rainfall over mountains appear markedly improved in the 2.2 km model relative to the 12 km driving model. The orientation of water vapor flux relative to the mountain barrier strongly affects both climatological and daily extreme precipitation. Transects illustrate the influence of steep local topography on strong landfalling atmospheric rivers to produce high vertical velocities and extremely high accumulations of rainfall over windward upper mountain flanks, which do not appear unreasonable against available gauge observations. These transects also reveal the finer spatial structure of mountain waves in the 2.2 km model, which may contribute to its more realistic windward enhancement of orographic precipitation, but with excessive leeward precipitation and an annual mean dry bias over mountains. Despite the computational burden, these results support further targeted dynamical modeling at kilometer scales to improve physical understanding of precipitation in the current climate and its potential future change in NZ and other mountainous regions of the world.

Three similar to 12-km reanalysis-driven regional climate models (RCMs) are evaluated in terms of capturing climatologies and extremes of precipitation, temperature and surface wind over Aotearoa/New Zealand (NZ). NZ provides an excellent case study for evaluating high-resolution RCMs due to its coastal and complex terrain and isolated geographical position in the midlatitudes, exposed to both tropical and polar influences. Overall, we find that the RCMs faithfully reproduce the observed climate, with precipitation and temperature climatologies particularly well-captured. However, the RCMs display significant differences from observations in capturing the surface wind climatology, highlighting a remaining key challenge. The excess "drizzle problem" is apparent to varying degrees, leading to a weaker representation in the length of meteorological drought in some regions. The RCMs also diverge in reproducing diurnal temperature range which appears partly related to cloud cover. Finally, we discuss the important role of observational uncertainty in the context of model evaluation.

New Zealand's National Institute of Water and Atmospheric Research publishes climate normals for New Zealand that are used for reporting the regional state of the climate, climate extremes and variability. Temperature and precipitation patterns are affected by both anthropogenic climate change and natural climate variability which in turn affect the climatological normal values calculated every decade. This study investigates how New Zealand's normals for temperature and precipitation have shifted over time at the national, regional and seasonal scales from the 1941–1970 period to the 1991–2020 period. Contrary to WMO recommendations, but aligned with many other countries, New Zealand's climate normals have traditionally not undergone homogenisation. The impact of introducing some homogenisation in the latest 1991–2020 normals, in contrast to historical non‐homogenized station normals, has been assessed using a new homogenized “Seventeen‐Station” temperature series from 1941 to 2020. We demonstrate that interpolating sparse non‐homogenized normals spatially to a grid can produce significant erroneous patterns and therefore undermine the accuracy of the conclusions drawn when using such normals. We find that the historical non‐homogenized temperature normals have a consistent negative bias in the long‐term trend at national, regional and seasonal scales. Our analysis of homogeneity tested precipitation showed consistent decreases at a national scale across all normal periods relative to the 1951–1980 precipitation normal. We also highlight how fixed period temperature and precipitation normals do not fully reflect the current state of a climate that is influenced by decadal variability and global warming. To derive normals fit for use in a changing climate it is suggested that automated methods for broad data homogenisation be developed along with alternative methods to derive normals that account for a non‐stationary climate.

Dynamical downscaling provides physics-based high-resolution climate change projections across regional and local scales. This is particularly important for island nations characterized by complex terrain, where the coarse resolution of global climate model (GCM) output often prohibits direct use. One of the main motivations for dynamical downscaling is to reduce biases relative to the host GCM at the local scale, which can be quantified through assessing ‘added value’. However, added value from downscaling is not guaranteed; quantifying this can help users make informed decisions about how best to use available climate projection data. Here we describe the experiment design of the updated national climate projections for New Zealand based on dynamical downscaling. The global non-hydrostatic Conformal Cubic Atmospheric Model (CCAM) is primarily used for downscaling, with a global stretched grid targeting high resolution over New Zealand (12-km) and the wider South Pacific region (12–35-km). Focusing on the historical simulations, we assess added value for a range of metrics, climatological fields, extreme indices, and tropical cyclones. The main strengths of the downscaling include generally large improvements relative to the host GCM for temperature and orographic precipitation. Inter-annual variability in temperature is well captured across New Zealand, and several temperature and precipitation-based extreme indices show large improvements. The representation of tropical cyclones reaching at least category 2 intensity is generally improved relative to the large consistent under-representation in the host GCMs. The remaining biases are explored and discussed forming the basis for ongoing bias-correction work.

Detection and attribution experiments are designed for the causal diagnosis of features in the climate system, including trends in mean climate and extreme events. While several detection and attribution data sets now exist, the coarse resolution of the climate models used (∼100‐km) often hinders their application to topographically complex regions like Aotearoa New Zealand and small island nations. The coarse atmospheric resolution may also be detrimental for simulating certain features of the atmospheric circulation, including the jets, blocking and cyclones. To address this, here we introduce a new set of climate model runs consisting of high‐resolution atmospheric simulations from the Conformal Cubic Atmospheric Model (CCAM) non‐hydrostatic global model. The variable‐resolution grid employed by CCAM enables targeted high‐resolution simulations over New Zealand (12‐km) and intermediate resolution over the wider South Pacific region (12–35‐km). Simulations from the historical experiment (years 1982–2021), consisting of ten initial condition ensemble members, are presented and evaluated here. The evaluation focuses on the representation of the large‐scale atmospheric circulation over the Southern Hemisphere including the jet streams, storm tracks, cyclones, blocking and teleconnections, as well as more localized temperature and precipitation variability and extremes specifically over New Zealand. While certain biases are highlighted and discussed for the large‐scale atmospheric circulation, CCAM is found to perform especially well for various precipitation and temperature‐based extreme indices at smaller scales across New Zealand, generally outperforming state‐of‐the‐art reanalysis and coarser resolution global atmospheric models. These results support further application of the CCAM ensemble for studying weather and climate extremes in attribution studies.

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.

Many methods and climate/weather modelling tools have been used over the past decade for assessment of the role of anthropogenic emissions in recent specific weather events (“event attribution”). Differences in the methods and models often correspond to differences in the characterisation, or conditioning, of an observed extreme event within a model, and this might be expected to affect any attribution statement. In practice, however, it may not always be feasible or practical to use the most appropriate method for the question at hand, or to use multiple methods so as to arrive at a generic conclusion. This is especially true given the growing interest in making rapid assessments of extreme events within operational forecast centres. How transferable are conclusions across methods, hence allowing the substitution of one method or modelling tool for another? In this paper we investigate differences in event attribution conclusions across a wide range of experiment designs, running from free-running simulations of atmosphere–ocean climate models, through to weather forecasts constrained to reproduce the nature of the event quite closely. Across a number of recent extreme weather events over Aotearoa New Zealand, we find no systematic differences in conclusions across the various experiment setups. This surprising result offers hope that attribution statements may be transferable across methods, because errors that arise when transferring results across methods are overshadowed by other errors and uncertainties given current technology.

An assessment has been made of the ability of the UK Met Office Unified Model (UM) to simulate the Antarctic stratospheric circumpolar vortex and, in particular, the extent to which the vortex acts as a barrier to meridional transport. It is important that models simulate this barrier well as it determines spatial gradients in radiatively active gases, such as ozone, which then determine the spatial morphology of the radiative forcing field. The assessment was made by comparing metrics of meridional impermeability calculated from dynamical fields extracted from UM simulations and from analogous fields obtained from NCEP-CFSR reanalysis. Two different UM configurations were assessed: global atmosphere 3.0 (GA3.0) using the New Dynamics dynamical core, and GA7.0 using the newer ENDGame dynamical core, with both versions run at N96 resolution (1.25\(^{\circ }\) latitude by 1.875\(^{\circ }\) longitude). The GA7.0 configuration appears to better simulate the dynamical isolation of the Antarctic stratospheric vortex in the lower stratosphere up to about 600 K, while GA3.0 provides a better simulation in the upper stratosphere. However, neither UM configuration simulates the same degree of dynamical isolation suggested by the reanalysis. In particular the UM configurations produce a wider and more poleward meridional band of high wind-speed and steep PV gradients when compared with the NCEP-CFSR reanalysis, leading to a stronger barrier in GA7.0 and a weaker barrier in GA3.0. Possible causes of discrepancies between model simulations and reanalysis and between the two model configurations are discussed. It is pointed out that further work is needed to identify ways of resolving these discrepancies in model simulations.

1 Landcare Research Manaaki Whenua Ltd, Hamilton, New Zealand, 2 Landcare Research Manaaki Whenua Ltd, Wellington, New Zealand 3 GNS Science, Lower Hutt, New Zealand, 4 Bodeker Scientific, Alexandra, New Zealand, 5 National Institute of Water & Atmospheric Research Ltd, Chrischurch, New Zealand, 6 University of Waikato, Hamilton, New Zealand, 7 Landcare Research Manaaki Whenua, Auckland, New Zealand, 8 Landcare Research Manaaki Whenua, Lincoln, New Zealand, 9 Landcare Research Manaaki Whenua, Palmerston North, New Zealand, 10 National Institute of Water & Atmospheric Research Ltd, Wellington, New Zealand, 11 AgResearch Ltd, Wellington, New Zealand 12 Plant and Food Research, Wellington, New Zealand, 13 Motu Economic and Public Policy Research, Wellington, New Zealand

“Ex-treme There an Anthropogenic This study investigates whether there was an anthropogenic influence on the extreme five-day rainfall observed in Northland, New Zealand, in early July 2014. The study is limited to an investigation of possible anthropogenic influence only on July rainfall amounts of the order of those observed. It does not make any definitive statement about anthropogenic influence on the drivers of such extreme rainfall, although this is the subject of ongoing research using the same datasets.

For the 2013 New Zealand drought, evidence from a number of models suggests that the meteorological drivers were more favorable for drought as a result of anthropogenic climate change.