How many samples? Soil variability affects confidence in the use of common agroecosystem soil indicators

Abstract

There is a need for accurate and easily-measured indicators suitable for characterising and monitoring agroecosystem multi-functionality. This is particularly true in intensively-farmed landscapes where it is of interest to quantify the role of small, woody vegetation features in providing ecosystem services such as carbon sequestration. However, soil variability introduced by natural and management processes can interact with sampling designs to result in inappropriate sampling intensities and high levels of uncertainty in measured indicators. This can have consequences for upscaling of ecosystem quantities and decision making. Here, we present results from a pilot study aimed at quantifying and understanding variation in ten common indicators of soil condition and function, within shelterbelt and adjacent field soils, at four dryland sheep farms in Canterbury, New Zealand. Our results demonstrate a high level of spatially-structured soil variability, driven by (1) the effects of woody vegetation on shelterbelt soils relative to field soils, (2) differences in underlying soil types among sites, and (3) possible effects of grazing animals within fields. This soil variability had clear knock-on impacts for appropriate sampling effort, depending on the soil indicator in question, the original soil sampling density, and whether the aim was to estimate population mean values or to detect differences among sites with confidence. On the whole, confidence in soil indicator estimates was highest for soil condition indicators (pH, soil moisture, bulk density), variable for carbon quantities, depending on the measure used, and lowest for soil biological process indicators (tea bag index decomposition rate, bait lamina probe micro-invertebrate activity, and dehydrogenase enzyme activity); estimation confidence was also mostly lower for shelterbelt soils due to the effect of woody roots and inputs on soil variability. Based on our results, we present indicative sample size requirements to estimate population means for these different soil indicators. Ultimately, we advocate for the use of pilot studies, such as the one presented here, to facilitate understanding of variability in soil function indicators within different agroecosystems, and how this variability is partitioned spatially within and among vegetated features.