Datasets and analysis for: Microclimatic buffering in forest, agricultural and urban landscapes through the lens of a grass-feeding insect
Data files
May 12, 2023 version files 3.60 MB
-
aggregated_data2.csv
-
README.md
-
stats_on_temprh2018edit.R
-
temprh2018.R
-
temprh2018edit.csv
Abstract
Using this dataset, we aimed to identify the microclimatic offsets that accurately represent the environment in which a small arthropod spends most of its life. As a case study, we selected grassy sites that corresponded to the microhabitat of grass-feeding insects in general, and larvae of the butterfly Pararge aegeria in particular, as this insect recently expanded its habitat use from forest (edges) to agricultural and urban environments. We tested to what extent local microclimates and microclimatic buffering capacity differed between tufts of grass in forest and in two anthropogenic (i.e. agricultural and urban) landscape settings by measuring microclimatic variables with sensors at the level of the grasses. We compared temperature, relative humidity and vapour pressure deficit (VPD) during an exceptionally warm and dry summer period with parallel data from weather stations and tested for differences between microclimatic profiles among the three landscape settings. Using this approach, our findings stress the functional implications of landscape-specific microclimatic profiles at the appropriate organism-centred scale.
Methods
Sample sites
We selected four study areas consisting each of an urban, an agricultural and a forest location (latitudinal range: 50° 51' 0" N ± 6' 0"; study areas: Aalst, Brussels, Leuven and Tienen). Urban built area of the cities ranges from 13.6 km2 (Tienen) to 128.86 km2 (Brussels). Each of the 12 sample locations was chosen at a 61.8 acres (25 hectares) grid area and were characterized by their dominant land cover type (i.e. i) buildings and other impervious surfaces, ii) meadows and pastures or iii) tree cover; land cover of interest > 70% at a grain size of 500 m2). Tree types within the sample locations all consisted of deciduous trees. In each of these locations, P. aegeria females have been repeatedly observed since 2016 (S. Braem, A. Kaiser, C. Turlure and T. Merckx, personal observations). Urban sample locations were at two city parks (Aalst and Leuven), a cemetery (Brussels) and a small woodland patch near the city centre (Tienen). Agricultural locations consisted of small woodlots, hedgerows and sunken lanes along meadows and cultivated fields. Locations differed with an altitudinal range of 80 m at most and showed usually limited topographic variation within location (maximum altitude difference between two sample sites never exceeded 20m, except in the agricultural site in Leuven where the maximum difference was 42m).
Climatic measures
Microclimatic measures were taken during a period of six weeks from the 13th of June until the 25th of July 2018. This period covers the start of summer conditions during which the species spends its life in large parts as egg or larva (i.e., in between two peaks of adult stages, which represent the first two annual generations) and may be particularly vulnerable to microclimatic conditions (Schweiger et al., 2006; Oliver et al., 2015; Pateman et al. 2016). Our Belgian study system experienced exceptionally warm and dry weather at that time (average temperature: 20.5°C during June & July 2018 vs. 17.3°C June & July 1830–2010; precipitation: 19 mm/m2 vs. 100 mm/m2 for the same 42-day period averaged for 1830–2010; Belgian Meteorological Institute KMI-RMI Brussels-Ukkel).
Within each location we measured temperature and relative humidity (RH) in two sample sites, using climatic sensors connected to a data logger (HOBO U23 v2; temperature accuracy of ± 0.2°C from 0° to 50°C and a RH accuracy of ± 2.5% from 10% to 90% and ±5% below 10% and above 90%; Onset Computer Corporation, 2010). The sensors measured relative humidity and temperature every 30 minutes. Every week for six weeks, we randomized sample sites by moving the two HOBO sensors to a different sample site, positioning them within a 100m perimeter of the previous sampling site.
Although several herbivorous insects can be found on grasses, we selected a number of grassy microhabitats that corresponded to the oviposition site search profile of Pararge aegeria as an instructive case. During summer, females prefer tufts of grass in a humid and thermally buffered, semi-canopy-shaded microhabitat (Shreeve, 1986b, Schweiger et al., 2006; Oliver et al., 2015; Pateman et al. 2016). We positioned the microclimatic sensors at tufts of host grass at 12 cm above ground and in partial or nearly full shade of the surrounding canopy layer, which were always deciduous. We did not use any casing to shield off the HOBO climatic sensors to better simulate biologically realistic conditions and thus increase sensitivity to variation in wind and direct exposure to solar radiation (Terando et al., 2017). However, to avoid direct heating through the semi-transparent lid of the HOBO sensor, sensors were provided with a green plastic roof (⌀ 90 mm), positioned at 15 mm above the sensor. Nine of the 144 data logger series were lost due to either issues with the HOBO climatic sensors or failure of data transfer from the logger.
Degree of canopy openness was recorded for each sample site by taking a picture while poisoning the camera 12 cm above ground and facing a fisheye lens camera skywards with the horizontal part of the frame in an east-west orientation (full-frame Fisheye Converter FCON-T01 with a diagonal 130° angle of view; Olympus Tough TG-4 camera). Afterwards, we corrected for the southward position of the sun (at the highest point of the sun path, solar zenith angle ranged between 27° and 32°) by using only the 75% southside-segment of the picture. This procedure resulted in canopy segments with a zenithal angle of 41.8° and an azimuthal angle of 74.3° with the centre point 7.0° tilted to the south. We used ImageJ freeware (https://imagej.nih.gov/ij/) to transform pictures into black-and-white format and lowered brightness thresholds manually until mostly sky-obstructing vegetation was turned black and sky was turned white. Canopy openness was expressed as the percentage of white on the cropped pictures. Canopy openness was lacking for the three sample locations around Tienen.
Hourly macroclimatic measures were recorded by The Royal Meteorological Institute of Belgium (KMI-RMI). Unfortunately, we only obtained measures over the course of a four-week period and therefore had to deal with a lower sample size when testing for microclimatic differences (see below). We selected three of their nearby weather stations to retrieve these data: Bauvechevain (for all sites near Tienen and the forest site in Leuven), Melle (for all sites near Aalst), and Zaventem (for all sites near Brussels and the agricultural and urban site from Leuven). All three weather stations were located in flat and open fields. At these stations, ambient temperature and relative humidity were measured 2 m above ground under standardized measuring conditions (i.e. casings with ‘Stevenson screens’ preventing the sensors from being directly exposed to sunlight or wind).
Data processing:
1. Offset variables and microclimatic effects
We obtained temperature and relative humidity (RH) data directly from HOBO climatic sensors and we derived Vapour Pressure Deficit (VPD) from the measurements of temperature and RH (Tetens 1930; Monteith & Unsworth 2013; for details on the calculation, see von Arx et al., 2013). To quantify the microclimatic conditions at the level of the host plant in different landscapes relative to the data from the weather stations (i.e. surrounding macroclimate), we used so-called offset microclimatic variables (De Frenne et al., 2019; Haesen et al., 2021). Microclimatic offset variables were calculated for each of the 135 weeklong sample series by subtracting average and standard deviation of local measures of temperature, RH and VPD of a single weeklong sample series with the corresponding averages and standard deviations from the nearby weather stations' data. The selected nine offset variables included offset day average (7:00 – 21:00 GMT+1), offset night average (21:00 – 7:00 GMT+1) and offset standard deviation of T, RH and VPD. In all statistical models, each of these calculated variables per sample series is treated as single replicates. Correlation analysis showed a strong correlation between standard deviation and range (i.e. maximum – minimum). However, standard deviation is less sensitive for short-lasting but extreme measures.
2. Other variables to tests for landscape type differences
We used the microclimatic data to test for differences in microclimate relative to landscape type. For each of the 135 weeklong time series, we used day mean, night mean, their standard deviation, range, minimum and maximum value of temperature, RH and VPD. These 18 intercorrelated variables were standardized and integrated by a principal component analysis (PCA) using R function ‘prcomp’ (R Core Team 2018).
Next, we introduced two variables reflecting the variability of VPD that aims to exclude variation caused by repeated diurnal oscillations or by long-term trends over multiple days. These variables were interpreted as negative correlates of predictability. For each of the weeklong time series, short-term variability of VPD was calculated by taking the mean of the residual variation (see ‘random’ in Appendix S1: Figure S1), after decomposition of time-series-wide trends (see ‘trend’ in Appendix S1: Figure S1) and day-night variation (see ‘cyclical’ in Appendix S1: Figure S1), using a decomposition algorithm that relies on moving averages (i.e. ‘decompose’ function in R; Hyndman & Athanasopoulos, 2021). ‘Day-to-day variability of VPD’ represents how the standard deviation of VPD differs, on average, from day to day and was calculated as follows: for one 7-day sequence of data, the difference in standard deviation was calculated between measures of day 2 and day 3, day 3 and day 4, day 4 and day 5, and day 5 and day 6. Then, the average of those four values represented the ‘day-to-day variability’. Day 1 and day 7 were omitted from the calculation.
Usage notes
Microsoft Excel, R (version 4.2.2) and RStudio RStudio (2021.09.2 Build 382) were used to open or edit the data files.