GEEClimT also provides access to our new ERA5-Land-Climatic (ERALClim) dataset, providing data for 19 variables (Table 2) at annual resolution (for 1951–2022), and for World Meteorological Organisation (WMO) climate baseline periods (1951-1980; 1961-1990; 1971-2000; 1981-2010; and 1991–2020). These variables are equivalent to the bioclimatic variables of WorldClim (Fick and Hijmans, 2017), though have potentially much broader applications within the environmental and social sciences beyond ecological and biological data where they are traditionally applied. To maintain consistency with WorldClim, the naming of variables is kept the same (i.e., retaining the bio prefix), so that pre-existing workflows that use WorldClim data can be easily adapted to use ERALClim data.

The ERA5-Land Daily Aggregated and Monthly Aggregated data products (Muñoz-Sabater et al., 2021) were used to calculate ERALClim within GEE, and the baseline data are calculated as the average of the resulting annual ERALClim data (access to the annual and baseline datasets are provided through: GEEClimT; GEE; and can also be downloaded as GeoTIFFs outside of GEEClimT as annual data (https://doi.org/10.5281/zenodo.8120646) and climate baseline data (https://doi.org/10.5281/zenodo.8124385)).

The ERALClim data are provided at a spatial resolution of 0.1 × 0.1° (∼11.1 km²), which is equivalent to the spatial resolution of the ERA5-Land reanalysis model inputs. While it would be possible to provide ERALClim data at a finer scale through simply changing the posting level of the output, this is intentionally not done so as to: 1) preserve the original model resolution that the data have been calculated at; and 2) avoid giving users a potentially false impression of the data product’s spatial resolution. Comparisons of ERALClim to WorldClim data for the WorldClim baseline period (1970–2000) where the latter have been regridded from 1 km² to 0.1 × 0.1° spatial resolution are provided in Supplementary Material S1. GEEClimT does optionally provide functionality to obtain data at sub-grid resolution through bilinear interpolation of data to 1 km², but will otherwise return values from the nearest grid point (see below).

Differences between ERALClim and WorldClim are most likely to occur in regions where there is a low spatial or temporal density of meteorological observations. The reason for this is that WorldClim produces spatially continuous data fields through interpolation of observations, while the ERA5-Land data (on which ERALClim are based) physically models conditions based upon a combination of meteorological observations that have been assimilated into the model and the model’s climate physics (Muñoz-Sabater et al., 2021). This in itself does not suggest that one dataset is better than the other in all cases, but highlights that differences between each dataset will arise due to the different methodologies used in the processing of the underlying data.

The following represents a brief description of how to use GEEClimT and its functionality, though a full walkthrough of how to use the tool is provided at the following https://github.com/jmlea16/GEEClimT. A graphical representation of the tool’s workflow is shown in Figure 1.

Users can access a link to the latest version of GEEClimT at the GitHub link above, and access the tool after signing up for a GEE account. No software download is required, and the tool runs entirely in a user’s internet connected browser. Once the GUI has loaded, users can minimise the code editor section of the page that contains GEEClimT’s underlying code. To use GEEClimT users can take the following steps:

1) Select whether data should be output as a CSV or GeoTIFF.

Where data will be output in CSV format:

5) Select how GEEClimT should handle the data when generating output, either as: a) a single time series representing the mean values of all points that overlap with a user’s region of interest (ROI); or b) multiple time series representing the full record of each grid cell that overlaps with the ROI.

Reanalysis data in CSV format can therefore be extracted for: individual point locations; all grid point observations that fall within a polygon; mean values across multiple points; or the mean of all grid point observations that fall within a polygon.

If data are to be exported in GeoTIFF format:

1. Define a single polygon (at least three vertices) by importing a list of comma separated latitude and longitude coordinates and clicking the “import polygon” button, or by manually defining a polygon using the drawing tools in the top left corner of the map.

Exported CSV or GeoTIFF data can be accessed for download to a user’s local machine via the Google Drive account that is associated with their Google Earth Engine account.

Within glaciology reanalysis and projection data have significant potential for informing investigations into the past, present and future of the Greenland Ice Sheet (GrIS). The low density of meteorological stations across the ice sheet and significant climate gradients from non-glaciated coastal regions to the ice sheet interior mean that observations at one location can differ substantially from conditions experienced a short distance (<10 s km) away. Reanalysis and projection data that are generated by physically based climate models can therefore provide data that are likely to be more representative of a particular locality where weather station data are unavailable. However, reanalysis data can also be subject to spatially varying biases related to the underlying model physics and density of observations (e.g., Cucchi et al., 2020), while the spatial resolution over which both reanalysis and projections are conducted is frequently incapable of resolving local topographic effects and sub-grid resolution weather variability.

In this example, we highlight these effects by comparing ERA5-Land data for Greenland’s capital city Nuuk (64.2000⁰ N 51.6833⁰ W; Jensen et al., 2022), and the on-ice PROMICE weather station KAN_U (67.0007⁰ N 47.0243⁰ W; Fausto et al., 2021; How et al., 2022). These sites are chosen given that Nuuk is a coastal city with significant surrounding topography that is also frequently affected by local conditions off-shore (e.g., sea fog), while KAN_U is located toward the ice sheet interior with little topographic variability and is influenced by general synoptic conditions. Both Nuuk and KAN_U weather stations provide data at hourly resolution, with the analysis below using all data where corresponding ERA5-Land reanalysis data were available (1 November 2000 to 31 December 2021; and 4 April 2009 to 27 March 2023 respectively). It should be noted that while ERA5-Land atmospheric data variables have higher spatial resolution than ERA5 data, these represent lapse-rate corrected regridding of ERA5 data (Muñoz Sabater et al., 2019; https://cds.climate.copernicus.eu/cdsapp#!/dataset/10.24381/cds. e2161bac?tab=overview; accessed: 26/6/2023).

Comparison of observations to reanalysis data were performed at hourly, daily and monthly temporal resolution, with an ensemble of 24 CMIP6 models used to obtain data for historical (1950–2015), and SSP2-4.5 and SSP5-8.5 projection scenarios (2015–2,100) at annual resolution (Figure 3; Bentsen et al., 2019; Byun et al., 2019; Dix et al., 2019; EC-Earth Consortium, 2019a; EC-Earth Consortium, 2019a; Guo et al., 2018; Hajima et al., 2019; Jungclaus et al., 2019; Krasting et al., 2018; Li, 2019; Lovato et al., 2021; Lovato and Peano, 2020; NASA Goddard Institute for Space Studies, 2018; Ridley et al., 2018; Ridley et al., 2019; Volodin et al., 2019a; 2019b; Wieners et al., 2019; Yukimoto et al., 2019; Ziehn et al., 2019). The CMIP6 ensemble excludes simulations available in GEECE where there are any missing data (Boucher et al., 2018; Tatebe and Wanatabe, 2018; Xin et al., 2018; Danabasoglu, 2019a; Danabasoglu, 2019b; Cao and Wang, 2019; Kim et al., 2019; Lee and Liang, 2019; Panickal et al., 2019; Seland et al., 2019).

To enable comparison with observations, ERA5-Land 2 m air temperature data (Muñoz-Sabater et al., 2021) were extracted using GEEClimT from Hourly, Daily Aggregated and Monthly Aggregated data products, while GEECE was used to define the CMIP6 model ensemble, process input data from daily to annual resolution, and calculate respective ensemble means and standard deviations for the time series. Within the tools we take advantage of GEEClimT and GEECE’s functionality for extracting to a CSV file data from given point locations where values are interpolated from the nearest four grid cells, though users should note the different spatial resolutions of the climate projection data (0.25 × 0.25°) and ERA5-Land reanalysis data (0.1 × 0.1°). Through using the point interpolation option in the tools we aim to mitigate the effects of the topography around Nuuk, though the differing spatial resolutions of the underlying data place limits on direct comparison. Observational data for each site were resampled to the corresponding ERA5-Land data product time interval by taking the mean of available observations. Time periods where observation data were missing within each timestep were discarded from the analysis.

Results show strong correlations exist at all temporal intervals between observations and ERA5-Land reanalysis data products for both Nuuk and KAN_U locations. However, ERA5-Land data for Nuuk show evidence for a cold bias that increases with more negative temperatures. This bias is replicated at hourly, daily and monthly temporal intervals, though the correlation between observations and reanalysis is observed to strengthen with longer temporal intervals (Figures 3A–C). At KAN_U, less bias is observed between reanalysis and observational data compared to Nuuk, though ERA5-Land data do exhibit a warm bias at the coldest temperatures (Figures 3E–G). Stronger correlations between observations and ERA5-Land data are found at KAN_U than at Nuuk for all temporal intervals analysed. Results from the CMIP6 ensemble show that Nuuk and KAN_U are projected to experience similar magnitudes of warming for SSP2-4.5 and SSP5-8.5 by 2,100, with divergence between the two projection scenarios beginning in approximately 2040 for both locations (Figures 3D, H).

Differences in the strength and bias of relationships observed at Nuuk and KAN_U likely reflect a combination of inherent model bias, regridding effects associated with ERA5-Land using downscaled data from ERA5 output, the inability of the reanalysis model to resolve the effects of sub-grid topographic variability (i.e., high variability at Nuuk resulting in weaker correlations compared to low variability at KAN_U), and sub-grid scale variability in weather conditions (also higher at Nuuk than at KAN_U). The warm bias of reanalysis data at KAN_U at lower temperatures may also be partly explained by winter snow accumulation partially burying the weather station, effectively raising the land surface meaning that the temperature sensor height may be < 2 m.

The above example aims to highlight that while reanalysis data will likely be strongly correlated to reality, biases may exist; outliers become more likely at shorter time intervals; sub-grid scale topography and local weather variability will impact the accuracy of reanalysis data; and how the observational data have been acquired all need to be considered before taking results of reanalysis data at face value.

A key approach within biology, ecology and evolution is to correlate biological data with the abiotic environment (e.g., climate, topography) to establish mechanistic links between the environment and species distributions, with this approach broadly being referred to as biogeography (Lomolino et al., 2017). Such biogeography studies can be used to inform on population characteristics, demography, species distributions, biodiversity and responses to climate change (Fitt and Lancaster, 2017; Lomolino et al., 2017). The ability to access high quality, geolocated abiotic environmental data therefore holds huge importance. Here, we demonstrate the utility of GEEClimT for such studies, demonstrating its application in species distribution modelling (SDM) of a native British damselfly, Erythromma najas.

Within this case study, results from SDM’s will be compared between models where climate variables were obtained from ERALClim dataset within GEEClimT, and models where WorldClim V2 (Fick and Hijmans, 2017) environmental variables were used. In this case study GEEClimT is used to extract a raster grid of ERALClim data of the British Isles from a user drawn polygon and download equivalent WorldClim V2 data that are both clipped to the United Kingdom borders in post processing. WorldClim V2 represents the environmental dataset most widely used for this application (Title and Bemmels, 2018), while ERALClim represents a new dataset developed in this study that is comparable to WorldClim, derived from ERA5-Land data (Munoz-Sabater et al., 2021).

Bioclimatic variables were assessed for co-linearity, and variables with a correlation score greater than 0.8 were removed, leaving Bio1, 2, 3, 4, 5, 8, 9, 10, 15, and 18. For ERALClim and WorldClim inputs, species distributions were modelled using maximum entropy habitat suitability model implemented in MaxEnt, version 3.4.3 (Phillips et al., 2006) using the Dismo package for R (Hijmans et al., 2019). Species’ presence points were downloaded from the Global Biodiversity Information Facility (http://www.gbif.org; https://doi.org/10.15468/dl.hdeere), with duplicate records and records which did not fall within both a ERALClim and WorldClim raster cell being removed, leaving 9,485 presence points (Figure 4A).

The MaxEnt models were run five times using default parameters, withholding a separate 20% of presence points for model testing on each model run. The final niche model for each species represented the average habitat suitability calculated across the five model runs. Model fit was also assessed by estimating the area under the Receiver Operating Characteristic (ROC) curve (AUC), with both models having an AUC greater than 0.8.

Results highlight limited agreement in model results between the WorldClim and ERALClim datasets, where Bio10 has the both the highest contribution and permutation importance when the data is modelled with ERALClim data, while Bio5 is the most important and has highest contribution when modelled with WorldClim (Table 4). Likewise, when comparing suitability maps, while there is general agreement in the projected suitability (Figures 4B, C), there are distinct differences (Figure 4D). These differences within this case study, can have profound impact of the interpretation of results, given that SDM’s are often used to inform conservation decision making, such as species reintroduction (Barlow et al., 2021), or making predictions of range shifts under climate change (Jarvie and Svenning, 2018).

While this case study does not have the power to suggest that one dataset performs better than another, results here do highlight the importance of climate dataset choice when conducting biogeographical studies. As demonstrated here, results can vary significantly, and can have meaningful impact on the interpretation of results. It is therefore important to have access to a wide range of environmental data to enable accurate studies of the relationship between the environment and biological systems.

Within the disciplines of Biology, Ecology and Evolution, it is commonplace to correlate biological data with abiotic environmental data so that mechanistic links and genotype by environment associations can be derived. Here, we demonstrate with a simple example the utility of this dataset for applications in ecological and biological studies. Miscanthus is a perennial C4 grass that is used as a biofuel primarily for combustion and occasionally for anaerobic digestion, with approx. 20,000 ha commercially grown in the EU and some 8,286 ha grown in the United Kingdom (EU CORDIS, 2016; DEFRA, 2021).

Originating from Asia, Miscanthus has now been bred to produce a range of varieties that are more suitable for European and United Kingdom climates (Clifton-Brown et al., 2004; Heaton et al., 2010; Brown et al., 2013). Within the United Kingdom, yields of Miscanthus can vary markedly as a result of several environmental factors, of which, temperature is a key abiotic driver of yield (Purdy et al., 2013; Awty-Carroll et al., 2023). Using ERA5-Land data obtained from GEEClimT we have correlated Miscanthus yields across 8 sites within the United Kingdom. All available yield data that was geolocated were included, providing 44 data points from 8 sites. Using the same interpolation approach employed in case study 1 the latitude and longitude were extracted from these points and placed into GEEClimT, extracting all data available from the ERA5-Land data product for the period 1990 and 2022. These data were downloaded as a. csv file, and read into R v 4.2.2 for further analysis (R Core Team, 2021).

The data for mean annual soil temperature and for mean volumetric soil water (MVSW), was selected to investigate these key abiotic drivers in relation to Miscanthus harvestable yield. For each site, the average was calculated for the period spanning when the Miscanthus rhizomes were planted to when they were harvested, and for the 12 months preceding harvest. Correlations were then obtained between yield and the abiotic environmental data sets calculated using linear models. Results demonstrated that yield decreased with increasing mean soil temperature, in the 12 months preceding harvest (Effect = 66.88 ± 16.48, df = 39.52, t = 4.06, p < 0.001, Figure 5B). Conversely, yield increased with increasing MVSW (Effect = −2.14 ± 0.87, df = 38.47, t = −2.45, p = 0.019, Figure 5A). This aligns with our understanding that Miscanthus is adapted to a cool and moist climate (Beale et al., 1996; Purdy et al., 2013).

This analysis acts as a working example of how climate derived data from GEEClimT can be used to forecast yields and could be utilised as a tool for identifying suitable locations for Miscanthus crop production, or optimal locations for other crop species. At a local level, farmers and growers can increase their localised species-specific productivity by identifying species that are best adapted within their local environmental mosaic. Not only does this highlight the suitability for reanalysis climate data to be used to biological studies but illustrates the potential accessibility that GEEClimT allows for ease of access, opening up a wider range of data to biologists without the need for technical knowledge or specialist software skills.

Multiple reanalysis products provide data output that represent similar variables. However, for each data product, results will likely differ depending on (amongst other factors): the underlying physics of the model used to generate the output; the observations used to drive the model; the data assimilation schemes used to incorporate data into the model; and the spatial resolution that the simulations are conducted at. In this example, we compare results of 2 m air temperature output from the 2.5⁰ resolution NCEP/NCAR Reanalysis dataset (Kalnay, 1996), and the 0.1⁰ resolution ERA5-Land Monthly Aggregated dataset (Muñoz-Sabater et al., 2021) for the period 1951-2022 at annual temporal resolution. To illustrate how these results can vary along a latitudinal gradient, data were extracted using GEEClimT for every 10th degree of latitude from 80⁰S to 80⁰N along the 20⁰W line of longitude by directly defining latitude and longitude within the tool. This line of longitude was chosen to maximise data availability for the ERA5-Land dataset, given that it does not provide output over oceans. NCEP/NCAR Reanalysis Data are provided at 6 hourly intervals, while we take advantage of the ERA5-Land Monthly Aggregated product to minimise GEEClimT processing and post-processing times. For this scenario, GEEClimT took approximately 3 h to extract the NCEP/NCAR data (providing data for 17 locations, and a total of 1,874,760 data values), and less than 1 minute to extract the ERA5-Land data (providing data for 10 locations, and a total of 8,800 data values). Output data were subsequently imported into MatlabⓇ and to allow calculation of annual means for each latitude for each reanalysis product.

Results from both reanalysis products capture expected temperature variability with latitude (Figures 6A, C), and display late 20th century/early 21st century warming trends (Figures 6B, D). However, differences between the data products are observed in both the absolute temperatures provided for each location (Figure 6C) and in the magnitude of the temperature anomalies compared to the 1961–1990 mean baseline temperatures (Figures 6D–F).

In addition to previously mentioned influences on the values provided by reanalysis data, at least part of the differences observed between these two datasets can be accounted for due to the differing spatial resolutions of the reanalysis output, and the manner in which GEEClimT extracts data from them (i.e., using bilinear interpolation of the nearest grid cells to obtain values for a given location). GEEClimT output from coarser spatial resolution data (e.g., NCEP/NCAR) will therefore be influenced by grid values that cover larger areas than finer resolution datasets (e.g., ERA5-Land). It is also worth reiterating at this point that ERA5-Land output at 0.1⁰ resolution is equivalent to lapse-rate corrected data from ERA5 simulations performed at 0.25⁰ resolution.

Hydrological extremes such as floods and droughts are major climate related hazards with methods that allow estimations of their probability of occurrence and magnitude being of much use. One such measure is the Standardised Precipitation index (SPI) (Mckee et al., 1993), one of the World Meteorological Organisation’s key drought indexes used in drought prediction and assessment.

The SPI requires only precipitation as an input data source which can be useful in data sparse environments. Use of reanalysis data (that provide data averaged over a grid cell or region), can be a useful addition to this in providing an indication of conditions over an area, rather than just at an individual point. The SPI is based on the probability of precipitation within a chosen time interval and is outputted as a monthly value. This is achieved through first fitting precipitation input data to a gamma distribution that are then transformed to a normal distribution. SPI values are computed over running time intervals (Mckee et al., 1993), with traditional intervals of 3, 6, and 12–48-month timescales being representative of meteorological, ecological, and longer-term hydrological drought (affecting groundwater and reservoir reserves) respectively. The monthly SPI value calculated represents the deviation from the recent long-term mean (i.e., the time interval), with negative numbers representing relatively dry periods and positive relatively wet periods–the more extreme the value, the more extreme the period. In addition to drought, the ability to identify periods of extreme precipitation provides a means to assess the meteorological conditions preceding flooding.

Here, we provide examples of the use of both GEEClimT and GEECE to obtain precipitation data (from ERA5 (Muñoz Sabater et al., 2019) and a 24 model CMIP ensemble respectively) in the investigation of recent hydrometeorological extremes and investigate the potential impacts of climate change on droughts in areas previously determined to be at risk. CSV files of data for each location were obtained through copy and pasting the polygon coordinates of each area of interest directly into GEEClimT and GEECE.

First, we focussed on the California (United States) drought of 2012–2016, focussing on El Dorado Forest where drought conditions were so severe tree mortality reached almost 50% (Fettig et al., 2019). SPI was computed for a 12-month interval (SPI-12) representing hydrological drought from 1980–2021 (Figure 7A). In this example SPI clearly identifies the 2012 to 2016 drought as the most intense (most negative SPI) of the study period, second only to the drought in the early 2000s.

ERA5 Monthly Aggregated precipitation data extracted from GEEClimT was then used to investigate a wetter period for a different location, with the late December 2015–January 2016 floods in Aberdeenshire, Scotland chosen (Figure 7B). For this example, an interval of 6 months was chosen (SPI-6) as it represents soil water stores, and the flooding was linked to an ongoing wet period and lack of soil water storage capacity (Soulsby et al., 2017). Here while the late December 2015–January 2016 wet period is clearly visible; it was not the most intense event (second to March-July 2012). The SPI index does however clearly show the large magnitude of the wet period, as all bars between December 2015 to April 2016 exceed an SPI of +1.

Finally, we applied the SPI technique to CMIP6 ensemble data obtained via GEECE. ERA5-Land, CMIP6 historical scenario data, SSP2-4.5 and SSP5-8.5 data were downloaded for Dorset, United Kingdom, which has been identified as being susceptible to drought with projected changes in climate (Arnell et al., 2021). A mean of the 24 member model ensemble was used at a monthly time step to calculate SPI over a 24-month interval (SPI-24), representative of long-term hydrological drought. The historical period (defined here as 1955–2005), and future projections using SSP2-4.5 (medium emissions) and SSP5-8.5 (high emissions) scenarios for 2020-2050 were then compared to identify changes in drought duration (months of negative SPI), magnitude (sum of all SPIs within the drought), and intensity (maximum SPI during drought) (Table 5)).

While caution should be applied as this analysis only considers precipitation and no other variables associated with other elements of climate change, the analysis suggests that under medium emissions scenarios that the intensity of long-term extreme (<-1.5 intensity) hydrological drought remains similar to the historical period, though there will likely be increases in the duration, intensity and magnitude of these hydrometeorological hazards in the future. Mismatch between the ERA5 and the CMIP6 historical scenario may result from using the ensemble mean values as input for the SPI, which will dampen monthly variability that may be captured by individual simulations. Similar caveats apply to results driven by the SSP2-4.5 and SSP5-8.5 ensemble means, highlighting that users should carefully consider how or if the assumptions made in datasets extracted from GEEClimT and GEECE may impact the results of subsequent analysis.

The data used within this analysis was identified and processed for download in <15 min before being used in the relatively simple SPI analysis. This highlights the use of the tool in accessing data from anywhere on the planet - something traditionally difficult and time consuming–and applying it to real world, beneficial analyses enabling assessment of conditions before, during, and after extreme hydrometeorological events, as well as understanding potential future scenarios.