Introduction to CORDEX-ML-Bench: A Benchmarking Dataset for Data-driven Regional Climate DownscalingCORDEXBench is a standardized benchmarking dataset designed to evaluate empirical-statistical downscaling (ESD) and deep learning-based regional climate emulators. It supports rigorous model evaluation across multiple climate zones and experimental setups. This dataset spans three distinct geographic domains:New Zealand (NZ) - 0.11° resolutionEurope (ALPS) - 0.11° resolutionSouth Africa - 0.10° resolutionEach region includes carefully structured training and testing data derived from dynamically downscaled Global Climate Models (GCMs), enabling systematic analysis of emulator performance in both historical and future climates.The dataset provides two core training experiments:ESD Pseudo-Reality (1961–1980)A 20-year historical training period using a single GCM (e.g., ACCESS-CM2 for NZ), designed to mimic ESD training. Emulator Hist+Future (1961–1980 + 2081–2100)A more comprehensive 40-year training period combining historical and future climates. This experiment supports evaluation of extrapolative skill, including transferability across GCMs.Both setups will be tested with and without topography as a predictor. For each training setup, the dataset enables evaluation across multiple test periods and inference conditions:Historical (1981–2000): For both perfect and imperfect inference.Mid-century (2041–2060) and End-century (2081–2100): To assess extrapolation to future climates, including hard transferability scenarios using unseen GCMs.Evaluation TypesThe dataset supports several benchmarking configurations:PP cross-validation: Same GCM used in training and testing.Imperfect inference: Same GCM but different realizations or noise.Transferability testing: Inference using a different GCM than the training set.Change signal evaluation: Assessment of climate change response in future periods.Data Structure Example: New Zealand DomainEach domain follows a consistent file structure, with subdirectories for training and testing data, and further divisions by period, GCM, and evaluation type. Predictors include both dynamic variables (e.g., temperature, precipitation) and optional static fields (e.g., topography).NZ_Domain/├── train/│ ├── ESD_pseudo-reality/│ │ ├── predictors/│ │ └── target/│ ├── Emulator_hist_future/│ │ ├── predictors/│ │ └── target/├── test/│ ├── historical/│ ├── mid_century/│ └── end_century/A More Detailied Overview of the Experiments📘 Table 1: Evaluation using ESD “pseudo-reality” (T1)Training SetupInference SetEvaluation TypeNotesEvalRequiredESD “pseudo-reality”Period: 1961–1980Static fields: Yes/Nohistorical (1981–2000) PP cross-validationSame GCM used in training, perfectlyError, ClimX historical 1981–2000Imperfect cross validationSame GCM, but imperfectlyError, Clim  2041-2060 + 2081-2100ExtrapolationSame GCM, but perfectlychange signalX 2041-2060 + 2081-2100ExtrapolationSame GCM but imperfectlychange signal  📗 Table 2: Evaluation using Emulator (T2)Training SetupInference SetEvaluation TypeNotesEvalRequiredEmulator hist + future period: 1961-1980 + 2081-2100 Static fields: Yes/Nohistorical (1981–2000) PP cross-validationSame GCM used in training, perfectlyError, ClimX historical 1981–2000Imperfect cross validationSame GCM, but imperfectlyError, ClimX 2041-2060 + 2081-2100ExtrapolationSame GCM, but perfectlychange signalX 2041-2060 + 2081-2100Extrapolation / Hard TransferibilityDifferent GCM, but perfectlychange signalX 2041-2060 + 2081-2100Extrapolation / Hard TransferibilityDifferent GCM, but imperfectlychange signalX            # NZ Domain/├── train/│   ├── ESD_pseudo-reality/│   │   ├── predictors/│   │   │   ├── ACCESS-CM2_1961-1980.nc│   │   │   └── static.nc│   │   └── target/│   │       └── pr_tasmax_ACCESS-CM2_1961-1980.nc││   ├── Emulator_hist_future/│   │   ├── predictors/│   │   │   ├── ACCESS-CM2_1961-1980_2080-2099.nc│   │   │   └── static.nc│   │   └── target/│   │       └── pr_tasmax_ACCESS-CM2_1961-1980_2080-2099.nc│├── test/│   ├── historical/│   │   ├── predictors/│   │   │   ├── perfect/│   │   │   │   ├── ACCESS-CM2_1981-2000.nc│   │   │   │   └── EC-Earth3_1981-2000.nc│   │   │   └── imperfect/│   │   │       ├── ACCESS-CM2_1981-2000.nc│   │   │       └── EC-Earth3_1981-2000.nc│   │   └── target/│   │       ├── pr_tasmax_ACCESS-CM2_1981-2000.nc│   │       └── pr_tasmax_EC-Earth3_1981-2000.nc││   ├── mid_century/│   │   ├── predictors/│   │   │   ├── perfect/│   │   │   │   ├── ACCESS-CM2_2040-2059.nc│   │   │   │   └── ...│   │   │   └── imperfect/│   │   │       ├── ACCESS-CM2_2040-2059.nc│   │   │       └── ...│   │   └── target/│   │       ├── pr_tasmax_ACCESS-CM2_2040-2059.nc│   │       └── ...││   └── end_century/│       ├── predictors/│       │   ├── perfect/│       │   │   ├── ACCESS-CM2_2080-2099.nc│       │   │   └── ...│       │   └── imperfect/│       │       ├── ACCESS-CM2_2080-2099.nc│       │       └── ...│       └── target/│           ├── pr_tasmax_ACCESS-CM2_2080-2099.nc│           └── ...# ALPS Domain/├── train/│   ├── ESD_pseudo-reality/│   │   ├── predictors/│   │   │   ├── CNRM-CM5_1961-1980.nc│   │   │   └── static.nc│   │   └── target/│   │       └── pr_tasmax_CNRM-CM5_1961-1980.nc││   ├── Emulator_hist_future/│   │   ├── predictors/│   │   │   ├── CNRM-CM5_1961-1980_2080-2099.nc│   │   │   └── static.nc│   │   └── target/│   │       └── pr_tasmax_CNRM-CM5_1961-1980_2080-2099.nc│├── test/│   ├── historical/│   │   ├── predictors/│   │   │   ├── perfect/│   │   │   │   ├── CNRM-CM5_1981-2000.nc│   │   │   │   └── MPI-ESM-LR_1981-2000.nc│   │   │   └── imperfect/│   │   │       ├── CNRM-CM5_1981-2000.nc│   │   │       └── MPI-ESM-LR_1981-2000.nc│   │   └── target/│   │       ├── pr_tasmax_CNRM-CM5_1981-2000.nc│   │       └── pr_tasmax_MPI-ESM-LR_1981-2000.nc││   ├── mid_century/│   │   ├── predictors/│   │   │   ├── perfect/│   │   │   │   ├── CNRM-CM5_2040-2059.nc│   │   │   │   └── MPI-ESM-LR_2040-2059.nc│   │   │   └── imperfect/│   │   │       ├── CNRM-CM5_2040-2059.nc│   │   │       └── MPI-ESM-LR_2040-2059.nc│   │   └── target/│   │       ├── pr_tasmax_CNRM-CM5_2040-2059.nc│   │       └── pr_tasmax_MPI-ESM-LR_2040-2059.nc││   └── end_century/│       ├── predictors/│       │   ├── perfect/│       │   │   ├── CNRM-CM5_2080-2099.nc│       │   │   └── MPI-ESM-LR_2080-2099.nc│       │   └── imperfect/│       │       ├── CNRM-CM5_2080-2099.nc│       │       └── MPI-ESM-LR_2080-2099.nc│       └── target/│           ├── pr_tasmax_CNRM-CM5_2080-2099.nc│           └── pr_tasmax_MPI-ESM-LR_2080-2099.ncData PreprocessingFor information (roughly) on how the data was preprocessed, please see the following repositories for the different regions: NZ: https://github.com/nram812/CORDEXBench-nzdomain-preprocessingALPS: https://github.com/jgonzalezab/CORDEXBench-alpsdomain-preprocessing

In Aotearoa-New Zealand, limited research has focused on the swimming performance of native fish species, particularly across life stages and varying environmental conditions. This study focussed on quantifying the swimming capabilities of longfin eel elvers (Anguilla dieffenbachii) by 1) determining the critical swimming speed (Ucrit), 2) testing the time to fatigue across a range of water speeds (swimming endurance) and, 3) comparing the swimming capabilities of elvers as individuals and within groups. Individual trials showed Ucrit ranged from 0.13 to 0.67 m s-1 (mean Ucrit = 0.34), while group trials ranged from 0.13 to 0.37 m -1s (mean Ucrit = 0.25). After controlling for differences in water temperatures and oxygen levels, Ucrit had a clear positive relationship with elver length, while no statistical difference was observed in mean Ucrit between individuals and group swim trials. As water speeds increased, elver swimming endurance decreased, with elvers swimming up to 210 s at the lowest water speed (0.1 m s-1) and only swimming an average of 22.1 s at water speed of 1 m s-1.

In Aotearoa-New Zealand, limited research has focused on the swimming performance of native fish species, particularly across life stages and varying environmental conditions. This study focussed on quantifying the swimming capabilities of longfin eel elvers (Anguilla dieffenbachii) by 1) determining the critical swimming speed (Ucrit), 2) testing the time to fatigue across a range of water speeds (swimming endurance) and, 3) comparing the swimming capabilities of elvers as individuals and within groups. Individual trials showed Ucrit ranged from 0.13 to 0.67 m s-1 (mean Ucrit = 0.34), while group trials ranged from 0.13 to 0.37 m -1s (mean Ucrit = 0.25). After controlling for differences in water temperatures and oxygen levels, Ucrit had a clear positive relationship with elver length, while no statistical difference was observed in mean Ucrit between individuals and group swim trials. As water speeds increased, elver swimming endurance decreased, with elvers swimming up to 210 s at the lowest water speed (0.1 m s-1) and only swimming an average of 22.1 s at water speed of 1 m s-1.

Datasets and outputs from the research article "Achieving full representativity of seafloor biodiversity with minimal expansion: exploring a strategy for New Zealand marine conservation"

Globally, many large rivers are modified to meet human needs, often with adverse impacts on fish populations. In New Zealand, such rivers support important recreational brown trout (Salmo trutta) fisheries, but understanding of river alterations on trout is limited. This study utilized New Zealand’s most extensive brown trout dataset from Glenariffe Stream (1965–1993). Annual brown trout spawning numbers varied eight-fold; larger runs had more small, first-time spawners, while smaller runs were sustained by return spawners. Spawning timing differed by sex, with larger fish arriving two months later than initial smaller spawners. Juvenile outmigration was driven by water level, time of year, and lunar phase. Tagged fish data highlighted the significance of longitudinal connectivity for post-spawning adults, particularly females, which travelled over 100 km downstream to estuarine habitats to rapidly regain condition. Collectively, our findings quantify the inherent annual variability in brown trout spawning runs and the catchment-wide scale that brown trout population dynamics occur over. Managers need to understand these upstream-downstream linkages when making river-modification decisions.

This dataset includes annotated imagry of Sabella spallanzanii, mediterranean fan worm, and other generic aquatic imagry used to train computer vision models for the detection of S. spallanzanii. These are stored in KITTI format within the attached Annotated_Sabella_and_Aquatic_Environment_Imagery.tar. The S. spallanzanii imagery is all collected from Auckland Marina. This dataset also includes eight videos of four different object detection computer vision models of different architectures (RenNet50 and GoogLeNet) and resolutions (Full HD or 1920x1080, and PAL or 1024x576) each applied to video containing S. spallanzanii from two different sites in Aotearoa New Zealand (Auckland Marina, and Whangarei Harbour). These are stored in mp4 format within the attached Inference_Videos_of_Sabella.tar.Finally an configuration file, Example training configuration file.json, is attached. This showing the data contributing to the models used to inference the attached videos as well as some of the dataset plitting, architectural, backbone and data augmentation settings.

This dataset includes annotated imagry of Sabella spallanzanii, mediterranean fan worm, and other generic aquatic imagry used to train computer vision models for the detection of S. spallanzanii. These are stored in KITTI format within the attached Annotated_Sabella_and_Aquatic_Environment_Imagery.tar. The S. spallanzanii imagery is all collected from Auckland Marina. This dataset also includes eight videos of four different object detection computer vision models of different architectures (RenNet50 and GoogLeNet) and resolutions (Full HD or 1920x1080, and PAL or 1024x576) each applied to video containing S. spallanzanii from two different sites in Aotearoa New Zealand (Auckland Marina, and Whangarei Harbour). These are stored in mp4 format within the attached Inference_Videos_of_Sabella.tar.Finally an configuration file, Example training configuration file.json, is attached. This showing the data contributing to the models used to inference the attached videos as well as some of the dataset plitting, architectural, backbone and data augmentation settings.

This is an 8m hydraulically conditioned DEM and hydraulic roughness dataset covering the three main islands of New Zealand (North, South and Rakiura Islands) to a minimum distance of 20km offshore. The dataset is separated into 526 30km by 30km tiles included in geofabrics-tiles-2025-04-07-8m.tar. Each tile is stored as netCDF file containing four layers: hydrologically conditioned DEM, hydraulic roughness, data source and LiDAR source. The data source and LiDAR source layers provide pixel-wise metadata about the processes and data contributing to each pixel. The location of these tiles is defined in the tile.gpkg file.  The dataset was developed for freshwater flood modelling. Land elevations are primarily derived from LiDAR surveys, and offshore elevations from a curated elevation dataset developed for tidal modelling, integrated on an unstructured mesh with spatial resolution down to 5 m in estuaries to 15 km offshore. Hydrological conditioning was achieved using the GeoFabrics software package (Pearson, R et al, 2023). This included the direct inclusion of measured bed elevations for 97 estuaries and rivers, and 57 lakes; and the estimation and inclusion of bed elevations for a further 98 rivers and river mouths. Additionally, all culverts and waterways (as defined on Open Street Map) were daylighted (obstructions removed and opened to the sky). Finally, the crest elevation of 5368 stop banks were preserved.This data has been prepared by the Earth Sciences New Zealand (ESNZ) for its own internal purposes. The information contained in this data is derived from multiple data sources, including 3rd party data sources. As there is always uncertainty associated with such data, ESNZ gives no warranties of any kind concerning its assessment and estimates, including accuracy, completeness, timelines or fitness for purpose and accepts no responsibility for any actions taken based on, or reliance placed on them by any person or organisation. ESNZ excludes to the full extent permitted by law any liability to any person or organisation for any loss, damage or expense, direct or indirect, and however caused, whether through negligence  or otherwise, resulting from any person or organisations use of, or reliance on the information contained in this data.Pearson, R et al., 2023, Geofabrics 1.0.0: An Open-Source Python Package for Automatic Hydrological Conditioning of Digital Elevation Models for Flood Modelling. Environmental Modelling and Software. http://dx.doi.org/10.2139/ssrn.4463610

This is an 8m hydraulically conditioned DEM and hydraulic roughness dataset covering the three main islands of New Zealand (North, South and Rakiura Islands) to a minimum distance of 20km offshore. The dataset is separated into 526 30km by 30km tiles included in geofabrics-tiles-2025-04-07-8m.tar. Each tile is stored as netCDF file containing four layers: hydrologically conditioned DEM, hydraulic roughness, data source and LiDAR source. The data source and LiDAR source layers provide pixel-wise metadata about the processes and data contributing to each pixel. The location of these tiles is defined in the tile.gpkg file.  The dataset was developed for freshwater flood modelling. Land elevations are primarily derived from LiDAR surveys, and offshore elevations from a curated elevation dataset developed for tidal modelling, integrated on an unstructured mesh with spatial resolution down to 5 m in estuaries to 15 km offshore. Hydrological conditioning was achieved using the GeoFabrics software package (Pearson, R et al, 2023). This included the direct inclusion of measured bed elevations for 97 estuaries and rivers, and 57 lakes; and the estimation and inclusion of bed elevations for a further 98 rivers and river mouths. Additionally, all culverts and waterways (as defined on Open Street Map) were daylighted (obstructions removed and opened to the sky). Finally, the crest elevation of 5368 stop banks were preserved.This data has been prepared by the Earth Sciences New Zealand (ESNZ) for its own internal purposes. The information contained in this data is derived from multiple data sources, including 3rd party data sources. As there is always uncertainty associated with such data, ESNZ gives no warranties of any kind concerning its assessment and estimates, including accuracy, completeness, timelines or fitness for purpose and accepts no responsibility for any actions taken based on, or reliance placed on them by any person or organisation. ESNZ excludes to the full extent permitted by law any liability to any person or organisation for any loss, damage or expense, direct or indirect, and however caused, whether through negligence  or otherwise, resulting from any person or organisations use of, or reliance on the information contained in this data.Pearson, R et al., 2023, Geofabrics 1.0.0: An Open-Source Python Package for Automatic Hydrological Conditioning of Digital Elevation Models for Flood Modelling. Environmental Modelling and Software. http://dx.doi.org/10.2139/ssrn.4463610

The microbial metagenome assembled genomes, metabarcoding data, and original and processed images used for AI-based imagery recognition are available on Science Data Bank.