How the landscape influences soil respiration : Explaining spatio-temporal patterns with interpretable machine learning

Titelangaben

Baumberger, Maiken ; Haas, Bettina ; Borken, Werner ; Nowosad, Jakub ; Giese, Laura ; Klein-Raufhake, Theresa ; Hamer, Ute ; Meyer, Nele ; Meyer, Hanna:
How the landscape influences soil respiration : Explaining spatio-temporal patterns with interpretable machine learning.
In: Geoderma. Bd. 472 (2026) . - 117904.
ISSN 0016-7061
DOI: https://doi.org/10.1016/j.geoderma.2026.117904

Volltext

Link zum Volltext (externe URL):

Angaben zu Projekten

Abstract

Soil respiration plays a crucial role in the carbon cycle by representing the greatest flux of carbon from terrestrial ecosystems to the atmosphere. The spatio-temporal variability of soil respiration within a landscape is a result of the patterns of its climatic and environmental drivers. However, despite its importance, the factors driving soil respiration variability within heterogeneous landscapes remain insufficiently understood. To investigate such relationships, we measured soil respiration and determined potential drivers at 166 sites distributed over one year across a 400 km2 study area in the Fichtelgebirge mountains, Germany. We trained random forest models and applied interpretable machine learning methods to explain and spatio-temporally predict soil respiration. Spatio-temporal patterns of soil respiration were predicted with an RMSE of 61 mg Cm−2h−1 and an R2 of 0.39. In the heterogeneous landscape that includes grasslands, arable land, and forests, spatial variability of soil respiration was large, with variations of up to 415 mg Cm−2h−1 at a single point in time. Spatial patterns of soil respiration followed the patterns of the land use types, were further differentiated by vegetation cover, and were influenced by the topographic position within the landscape. These drivers also influenced patterns of soil temperature, which was the most important driver of soil respiration. Our high-resolution predictions demonstrate pronounced spatial variability in soil respiration at the landscape scale, arising from the interaction of multiple environmental controls, and offer new insights into responses under real-world conditions. Overall, interpretable machine learning showed great potential by explaining the spatio-temporal patterns of soil respiration resulting from complex interactions of its drivers, providing insights into soil respiration on the landscape scale.