Background & Summary

The study of the connections between organisms and climate is a classic question in ecology and has a long history of study1,2. It has become one of the dominating topics in ecology in the past 20 years3–5, in part because of the acknowledgement of the issue of anthropogenic global climate change, but also because of the development of high resolution, gridded climatic datasets for terrestrial environments. The latter technical breakthrough has facilitated the rise of correlative methods for species distribution modelling3,5,6. This modelling approach began with the BIOCLIM method7 (see also Ferrier8) and an associated dataset of climatic indices9 derived from interpolations of weather station observations10, which are now available on a global scale11.

Correlative species distribution modelling methods are able to capture statistical relationships between occurrence records and climate, and are powerful tools for defining and projecting climatic envelopes. However, these methods only implicitly capture the processes connecting organisms to climate12. An explicit understanding of such processes is not only of great scientific interest, but is also important for applied issues in managing biodiversity under climate change. There is, therefore, rising interest in approaches that can mechanistically connect the functional traits of organisms (e.g., thermal tolerances, desiccation resistance, metabolic processes) to climate13–15. Such approaches, however, are constrained by the need to define the actual microclimatic environments experienced by an organism at the appropriate time-scale16. The environments experienced will often differ dramatically from those represented by gridded climatic layers, particularly in the terrestrial setting. For example, the air-temperature–based climate layers represent conditions approximately 1.5 m above the ground. In contrast, the body temperature experienced by an organism depends on its height above or below the ground, together with the biophysical interaction of its thermal properties (e.g., size, shape, solar reflectance) with the local air temperature, wind speed, radiation and humidity levels as defined by its habitat (e.g., slope, aspect, shading) (Figure 1). If those conditions are known, potential body temperatures and rates of water exchange can be computed using the principles of biophysical ecology17–20 and then used to define climatic constraints on the behaviour, distribution and abundance of organisms15,21–24.

Figure 1
figure 1

Schematic of microclimatic processes relating to heat and water transfer of an organism, after Porter et al.20

Full size image

To facilitate such mechanistic studies, here we present a gridded dataset of hourly estimates, for the middle day of each month, of all the microclimatic conditions needed to compute heat and water balances of organisms in terrestrial environments at the global scale (Data Citation 1). This microclimatic dataset, which we have called ‘microclim’, complements the ‘worldclim’11 (http://www.worldclim.org/) and ‘cliMond’25 (https://www.climond.org/) gridded datasets of macroclimate for correlative species distribution modelling.

Methods

Model description

Microclimatic calculations were made with the microclimate model of the ‘Niche Mapper’ biophysical modelling package. This model was first described in Porter et al.20 and Beckman et al.26, with further descriptions in Porter and Mitchell27, Fuentes and Porter28 and Kearney et al.29 It is a Fortran program that includes routines for hourly calculations of solar and infra-red radiation intensities, above-ground profiles of air temperature, wind velocity and relative humidity, and substrate temperature profiles at ten nodes from the surface down to a user-specified maximum depth (typically around 2 m). These calculations can be made as a function of shading by vegetation, soil properties, thermal properties and terrain (slope, aspect, hill shade). To generate the layers presented in this paper, we used a version of the Niche Mapper package (‘NicheMapR’), which has been set up to run in the R-environment30 and is being prepared for public release.

The model requires as input the maximum and minimum daily values of air temperature, wind speed, relative humidity and cloud cover, the timing of the maxima and minima relative to dawn or solar noon, soil properties (conductivity, specific heat, density, solar reflectivity, emissivity) as well as the roughness height, slope and aspect. Clear sky solar radiation is computed based on latitude and longitude using the algorithm described in McCullough and Porter31. Substrate temperatures are computed based on a one-dimensional partial differential equation that uses above- and below-ground boundary conditions20. The substrate surface temperature is computed via a heat balance equation, accounting for surface heat exchange via radiation, convection, conduction and evaporation. For these simulations we assumed flat ground.

Model parameters and input data

For substrate temperature calculations, we used nodes of 0, 2.5, 5, 10, 15, 20, 30, 50, 100 and 200 cm for three different substrate types: soil, rock and sand (see Table 1 for respective substrate parameters). For the ‘soil’ substrate type, a 5 cm cap of organic material was simulated (see Kearney et al.29). The deep substrate temperature (at 200 cm) was assumed to be the mean annual air temperature, based on the monthly daily maximum and minimum air temperatures. All calculations were made for six different levels of shading by vegetation: 0, 25, 50, 75, 90 and 100%. The model produces an hourly output of air temperature, wind speed and relative humidity at the reference height (10 m for wind, 1.2 m for air temperature and relative humidity) and at a second height specified by the user. For the latter we used 1 cm, and report how to interpolate to other heights in the Usage Notes section below.

Table 1 Microclimate model parameters.
Full size table

Climatic data were drawn from a global 10 arc-minute dataset of monthly mean (1961–1990) daily maximum and minimum air temperature and monthly mean daily relative humidity, wind speed and cloud cover32. For computations of solar radiation, an aerosol attenuation profile is required31. For this we used the 5-degree resolution Global Aerosol Data Set (GADS)33, modifying the original GADS Fortran program to give output for the full spectral profile for a single location, and using the average of the summer and winter values of aerosol attenuation. For estimates of elevation (from which air pressure was derived for calculating properties of air) we used the 10 arc-minute worldclim grid (http://www.worldclim.org). Monthly soil moisture estimates were obtained from the Climate Prediction Center (NOAA/OAR/ESRL PSD, Boulder, Colorado, USA, http://www.esrl.noaa.gov/psd/). The simulations were run on an IBM iDataplex x86 supercomputer (1120 Intel Sandybridge compute cores running at 2.7 GHz) administered by the Victorian Life Sciences Computation Initiative.

Data Records

The microclim dataset (Data Citation 1) is provided in netCDF format, with one file per variable per month, each file comprising 24 layers (one for each hour of the day from 0:00 h to 23:00 h, so layer 13 is 12:00 midday) for the globe (852 rows, 2159 columns, latitudinal extents −58.57 to 83.43 degrees, longitudinal extents −180.00 to 180.00 degrees). The variables are summarized in Table 2. Shade-specific values are provided for all substrate temperatures as well as for radiant sky temperatures and 1 cm estimates of air temperature and relative humidity. Substrate-specific values are provided for 1 cm air temperature and relative humidity estimates, since they depend on the substrate temperatures.

Table 2 The microclim datasets. For the naming convention, the part in italics denotes the particular scenario as follows: mon is month of the year (1 to 12, where 1=Jan, 12=Dec), shd is shade level in % (0, 25, 50, 75, 90, 100), sub is substrate (soil, rock, sand).
Full size table

Example layers for midday surface temperature in January and June are depicted in Figure 2, and example hourly profiles are provided in Figures 3 and 4, for all variables in January and June for the cool-temperate Australian town of Hobart. Solar zenith angles (Figure 3a and b) are zero when the sun is directly overhead and 90 when it is below the horizon. The solar radiation layers (Figure 3c and d) include the influence of cloud cover as well as the effects of aerosols, with the latter imposing a slightly blocky nature to the layers due to its coarser spatial resolution (5°). The wind speed 1 cm above the ground (Figure 3e and f) is computed from the 10 m reference wind speed based on a standard logarithmic vertical profile and the assumed roughness height (Table 1)20. The 1 cm air temperature (Figure 3g and h) is similarly adjusted based on both the 1.2 m reference temperature and the substrate surface temperature, and the 1 cm relative humidity (Figure 3i and j) is temperature-corrected to the 1 cm air temperature based on the relative humidity and air temperature at the 1.2 m reference height (specific humidity is assumed to be homogeneous in the vertical column). Note that our assumption of ‘neutral conditions’ in the formation of wind speed and air temperature profiles will not be valid for very windy regions of the world, and will result in steeper gradients than would be observed in nature.

Figure 2
figure 2

Substrate (soil) surface temperature for 12:00 h in January and June.

Full size image
Figure 3
figure 3

Hourly profiles for January and June for the microclim variables (excluding substrate temperature) for a cool-temperate site—Bushy Parks Estate near Hobart, Tasmania, Australia.

Full size image
Figure 4
figure 4

Hourly substrate temperature profiles with depth, for January and June for the microclim variables for a cool-temperate site—Bushy Parks Estate near Hobart, Tasmania, Australia. Examples are shown for the three generic substrate types included in the microclim dataset—soil, sand and rock.

Full size image

The ‘sky temperature’ (Figure 3g and h) is used to compute the downward longwave radiation flux. All objects emit longwave radiation at a rate proportional to the 4th power of their temperature, QIR=σε T4, where σ is the Stefan Boltzmann constant (W m−2 K−4), ε is the emissivity and T is the object’s temperature in Kelvin. Clear sky temperature is calculated based on the 1.2 m air temperature using a sky emissivity approximated as ε S K Y =1.72 ( e A / T A ) 1 7 (equation 10.10 in Campbell and Norman17) where TA is the 1.2 m air temperature in Kelvin and eA is the vapour pressure of the air in kilopascals. The shade and cloud cover components are calculated assuming an emissivity of 1 and that they are at the 1.2 m air temperature.

Finally, the substrate temperatures at different depths (Figure 4) reflect the influences of the specified soil properties (Table 1) and all the aforementioned variables (day length, radiation levels, wind speeds and air temperature) except relative humidity, as a dry surface was assumed.

Qualitatively, the soil substrate (Figure 4a and b) shows higher amplitude fluctuations near the surface, especially because of the ‘organic cap’ simulated on top, while the rock substrate (Figure 4e and f) shows greatly dampened fluctuations due to the higher assumed conductivity and heat capacity. The sand shows fluctuations in between those of soil and rock (Figure 4c and d). Note that the 1 cm air temperature, 1 cm relative humidity and substrate temperatures for soil and sand have slightly reduced extent relative to the other layers because of the slightly reduced extent of the soil moisture dataset used in their computation.

In addition, the data record (Data Citation 1) includes model predictions (365-day interpolations) of hourly soil temperature at 177 sites across the USA plotted against 3-hourly observations of soil temperature for all available times between 2011 and 2013 from Soil Climate Analysis Network (SCAN), described in more detail below.

Technical Validation

The microclimate model was recently tested against Australian 3-hourly observations (years 2000–2009) of soil temperature (5, 10, 20, 50 and 100 cm) as well as surface (0.5 cm above the ground) minimum temperature, measured by the Bureau of Meteorology at 43 sites across the continent29. The model predicted 85% of the variation in the soil temperature observations with an average accuracy of 2–3.3 °C (~10% of the temperature range at a given depth). For that validation study, historical daily interpolated climatic grids34 as well as historical monthly soil moisture estimates35,36 were used as environmental inputs. To compare the performance of the model when driven by the environmental inputs of the present study (i.e., long-term monthly climatic averages), we ran a 3-year simulation based on a daily spline of the monthly inputs (‘spline’ function of the statistical package ‘R’ v. 2.15.2 (ref 30)) and compared it to the observations and predictions reported in Kearney et al.29 for four climatically distinct sites (Supplementary Figures S1–S4). The resulting outputs were extremely similar except that, as expected, more variation was predicted and observed with the daily resolution data at shallow depths. Also, in the original validation study we assumed evaporative cooling on rainy days (see especially the Darwin plots, Supplementary Figure S4), whereas dry ground was always assumed in the present calculations.

In addition to the Australian tests, we have compared the outputs of the model (again, splined across 365 days) for 177 sites across the USA (including Alaska and Puerto Rico) (Supplementary Table S1, see also Supporting Images S1-S177 in the data repository (Data Citation 1)). 3-hourly soil temperature estimates from these sites were extracted from the Soil Climate Analysis Network37 (SCAN, http://www.wcc.nrcs.usda.gov/scan/), using all available data from 2011–2013. We followed the same procedure as in the Australian tests29, i.e., we calculated r2 values for linear regressions of observed vs. predicted temperature as well as the root mean square deviation (RMSD) and proportional root mean square deviation (RMSDp). For these analyses the model predicted 88% of the variation in the soil temperature observations with an accuracy of 3.2–4.5 °C (~13% of the temperature range at a given depth) under the assumption of zero shade. Visual inspection of Supporting Images S1–S177 (which include photos of the stations) indicates clearly where inaccuracies are due to the presence of snow and forest cover.

Usage Notes

The microclim datasets (Data Citation 1) can be used to extract data for specific points or for selected grids, and the user can employ interpolation functions to obtain estimates at finer resolutions of time (i.e., sub-hourly, or daily rather than monthly), shade level and heights above the ground between 1 cm and 120 cm if necessary. Interpolation of wind speed and air temperature between these two heights to any other height z, assuming neutral conditions (i.e., no strong free convection), can be achieved with the following equations38

V z =( V * /k)ln( z z 0 +1)
V z V r = T z − T s T r − T s = ln ( z z 0 + 1 ) ln ( z r z 0 + 1 )

where V* is the shear velocity (m s−1), Vz is the wind speed at the new height, Vr is the reference wind speed, zr is the reference height (m), z0 is the assumed surface roughness (0.004 m in the present case), Ts is the surface temperature (°C), Tr is the air temperature at the reference height, Tz is the air temperature at the new height, and k is the Karman constant (0.4). In the Supplementary Material we include an R-script (Supplementary Script S1) with examples of how to read the files to memory, plot global grids per hour, and extract and compile a 24 h profile of microclimatic estimates for a particular location.

The monthly-mean character of the macroclimatic data on which these datasets are based must be borne in mind when applying them to ecological problems. Many ecological questions can be answered with microclimatic simulations derived from monthly temporal data resolution (e.g., ref. 39). However, the natural daily variation in surface conditions, especially air temperature, wind gusts and cloud, will impose substantially more variability than implied by these layers. For such applications, the user may impose such variability on the base conditions provided by the microclim layers, e.g., by using environmental bootstraps40,41 on local air temperature observations. The user should also be aware that we have not simulated snow cover, which will lead to inaccuracies in the soil temperature estimates for snow-covered areas unless otherwise accounted for. If periods of snow cover are known, the microclim layers could be amended for those periods (e.g., by forcing near-surface substrate temperatures to zero).

Additional information

How to cite this article: Kearney, M. R. et al. microclim: Global estimates of hourly microclimate based on long-term monthly climate averages. Sci. Data 1:140006 doi: 10.1038/sdata.2014.6 (2014).