Analyzing spatiotemporal variability of heterotrophic soil respiration at the field scale using orthogonal functions

Soil CO2 efflux was measured with a closed chamber system along a 180 m transect on a bare soil field characterized by a gentle slope and a gradient in soil properties at 28 days within a year. Principal component analysis (PCA) was used to extract the most important patterns (empirical orthogonal functions, EOFs) of the underlying spatiotemporal variability in CO2 efflux. These patterns were analyzed with respect to their geostatistical properties, their relation to soil parameters obtained from laboratory analysis, and the relation of their loading time series to temporal variability of soil temperature and moisture. A particular focus was set on the analysis of the overfitting behaviour of two statistical models describing the spatiotemporal efflux variability: i) a multiple regression model using the k first EOFs of soil properties to predict the n first EOFs of efflux, which were then used to obtain a prediction of efflux on all days and points; and ii) a modified multiple regression model based on re-sorting of the EOFs based on their expected predictive power. It was demonstrated that PCA helped to separate meaningful spatial correlation patterns and unexplained variability in datasets of soil CO2 efflux measurements. The two PCA analyses suggested that only about half of the total variance of efflux could be related to field-scale spatial variability of soil properties, while the other half was “noise” attributed to temporal fluctuations on the minute time scale and short-range spatial heterogeneity on the decimetre scale. The most important spatial pattern in CO2 efflux was clearly related to soil moisture and the driving soil physical properties. Temperature, on the other hand, was the most important factor controlling the temporal variability of the spatial average of soil respiration.

► CO2 efflux from a 18 point transect on a bare soil field with gentle slope. ► Repeated on 28 dates, empirical orthogonal functions (EOF) to identify patterns. ► First EOF exhibits strong spatial dependence, later EOFs contain increasingly noise. ► Spatiotemporal variability modelled best from soil properties via first EOF. ► Temperature controls spatial average, moisture spatial pattern.