Measuring greenhouse gas emissions in natural and controlled ecosystems can be notoriously tricky, both because of spatial heterogeneity and the highly dynamic systems which are responding to drivers (temperature, moisture, sunlight, etc.) on multiple time scales. This often makes direct comparison between measurement locations and across differing flux measurement instruments and methodologies impractical, if not impossible.
Consequently the question of what can be expected for spatial variability when measuring soil respiration at a site often comes up – and how does it or how should it impact the design of the study? Here we briefly review the various techniques that have been used to estimate the spatial heterogeneity in the field and discuss how you can include these considerations in your next field campaign.
Just how variable can sites be?
In order to understand how spatially variable soil respiration (Rs) is in practice, many researchers use chamber based soil gas flux measurements at varying spatial scales and variograms (or semivariograms) to estimate the degree of autocorrelation. For example, Das Gupta and Mackenzie (2016), studying a fire-chronosequence of boreal aspen, found that soil respiration (Rs) in the 1 year post-fire stand had a spatial autocorrelation range of 6 m in May and August but much coarser autocorrelation range (>23 m) was recorded for the duration of June and July. The spatial range of Rs in the 9 years post-fire stand (canopy closure) varied between 4.4 m (May) to 5.6 m (June) to 4.8 m (July), while no spatial autocorrelation was detected in August. In the mature stand soil respiration had a spatial autocorrelation range ≤5 m throughout the growing season except in May (8 m). The researchers also found that like many boreal ecosystems, forest floor depth was the main contributor to driving longer distance autocorrelations at these study sites.
On a much smaller spatial scale, research by Stoyan et al. using soil cores from poplar and winter wheat plots (samples taken from a winter wheat field within a 2m radius from poplar tree bole) showed that spatial autocorrelation only happens within a distance of about 21.9 cm. This contrasts starkly with the results of Das Gupta and Mackenzie, but keeping in mind that in a relatively homogenous system, like that studied by Stoyan et al., we would expect there to be no larger length scale drivers of soil respiration (like the forest floor depth in the example above).
Finally, bringing these two highly contrasting estimates of spatial autocorrelation together, research by Prolingheuer et al. (2010) showed that while spatial autocorrelation values were large for Rs at their study site (0-35.1 meters); when the authors broke the total efflux down by soil respiration type (heterotrophic versus autotrophic) they found that the majority of the long range autocorrelation was coming from the plant component of the respiration, and that heterotrophic respiration is autocorrelated on scales less than 1 meter generally.
How can you account for variability in your studies?
Obviously “accounting for variability” can mean different things, depending on what the purpose of your study is. For this short blog article, I am going to assume the purpose of the field study is to estimate the site-wide mean of soil respiration, so it can be correlated to environmental variables and upscaled to stand or ecosystem level fluxes.
Research by Davidson et al. (2000) using a well replicated automated chamber system (36 individual chambers), in both forest and pasture ecosystems, showed that selection of 8 random chambers from the population, will be placed within 50% of the population mean ~100% of the time. However, if being within 10% of the population mean is desired, the probability is reduced to between 63.3% and 71.9% of the time by randomly selecting eight sampling locations. Some recommendations about the number of chambers to use at a site with a similar soil respiration mean and standard deviation to Harvard forest are given by Davidson et al. (2000), which are summarized in the table below.
Based on this study by Davidson et al. (2000), our first recommendation is to choose an acceptable confidence threshold and interval about the population mean for your study, and then scale the necessary respiration measurement equipment from there. Obviously the most cost effective approach will be to use a single survey chamber, however, this can result in convolution of the space and time components of the measurements – possibly misrepresenting the spatial variability with temporal variability or vice versa – and also requires significant field labour to make the measurements. Alternatively it’s possible to work backward based on the available equipment (which might be a more pragmatic approach) and decide what fundamental research questions can be possibly answered, for example with 10 chambers and a 95% confidence interval.
Finally, knowing a priori if there is a highly spatially autocorrelated factor at the site before designing the sampling scheme may also help reduce the uncertainty. For example at sites with strongly varying water tables, or differences in the forest floor depth (as was discussed above), the sampling plan can be adjusted to take into account these long scale effects, while still sampling randomly perpendicular to that larger scale effect.
Hopefully the summary presented here gives a taste for the spatial variability at field sites and the expected variability between respiration measurements taken with the same device. Keep in mind that the results shown here are only treating spatial variability, don’t forget that temporal variability and methodological biases can also impart similar amounts of noise to your data, and should be considered carefully especially when studying small differences in soil respiration rates.