Severe drought and conventional farming affect detritivore feeding activity and its vertical distribution

Soil invertebrates are key to decomposition, a central ecosystem process related to soil health. In many temperate areas climate change will decrease soil water content, which strongly modulates biological activity. However, data are lacking on how shifts in rainfall patterns affect soil biota and the ecosystem processes they provide. Here, we used the bait-lamina test to experimentally assess how a severe drought event influenced detritivore feeding activity, during a wheat growing season, in soils under long-term organic or conventional farming. Additionally, biotic and abiotic soil parameters were measured. Feeding activity was reduced under extreme drought and conventional management, although no climate-management synergies were found. Vertical migrations of Collembola and Oribatida partially explained the unexpectedly higher bait consumption at shallower depths in response to drought. Exploratory mixed-effects longitudinal random forests (a novel machine learning technique) were used to explore whether the relative abundances of meso‑, microfauna and microbes of the decomposer food web, or abiotic soil parameters, affected the feeding activity of detritivores. The model including meso‑ and microfauna selected four Nematoda taxa and explained higher variance than the model with only microbiota, indicating that detritivore feeding is closely associated with nematodes but not with microbes. Additionally, the model combining fauna and microbiota explained less variance than the faunal model, suggesting that microbe-fauna synergies barely affected detritivore feeding. Moreover, soil water and mineral nitrogen contents were found to strongly determine detritivore feeding, in a positive and negative way, respectively. Hence, our results suggest that severe drought and conventional farming impair the feeding activity of soil detritivores and thus, probably, decomposition and nutrient mineralization in soils. Furthermore, machine learning algorithms arise as a powerful technique to explore the identity of potential key drivers relating biodiversity to ecosystem functioning.