Fig 1.
The undulating terrain of the field near Lietzen is typical for the landscape formed by the last Ice-age formed landscape in Northeastern Germany. The 42 measurement plots were distributed along four transects in the 63 ha field with 1–21 being under conventional and 22–42 under reduced tillage. The average distance between plots within the transects was 66 m.
Fig 2.
Earthworm abundances at the measurement plots.
Average abundance of earthworms and of individuals for each species as observed between 1997 and 2007 at the measurement plots under conventional and reduced tillage. For configuration of plot sequence see Fig 1.
Table 1.
Univariate statistics of average individual counts of earthworm species observed at all measurement plots during the years 1997 till 2007.
Fig 3.
Spatial locations of the sensor readings by the three-sensor platform. Apparent electrical conductivity (ECash), pH, and light absorbance (NIRavg) in the topsoil are depicted from left to right. Quantile classification scheme with 10 levels used.
Table 2.
Univariate statistics derived from the soil sensor data collected at the Lietzen site in summer 2011.
Statistics for the variables marked with asterisk (*) were computed for the sensor values at the measurement plots interpolated with ordinary Kriging. Spectroscopic data were summarized by the average of the raw spectra (NIRavg) and two independent components extracted by ICA from the second derivatives (NIRICA1-2). Autocorrelation range (Range) and nugget-to-sill ratio (N/S) were computed by semivariogram modelling.
Fig 4.
Correlation matrix (Spearman rank correlation coefficients) between sensor variables and earthworm species.
(a) conventional tillage and (b) reduced tillage.
Table 3.
Generalized additive modelling (GAM) results for total earthworm abundances and the observed earthworm species with PSS as predictors.
GAMs were computed with log-link function and quasi-Poisson error distribution with the observed sums at the measurement plots and averaged afterwards. The spatial coordinates of the measurement plots f(x,y) were included as a bivariate function smoothed with thin plate regression splines. Model validation was performed using leave-one-out cross validation.
Fig 5.
Average earthworm abundances predicted by the generalized additive models as given in Table 3. Predictions were made on a 5x5 m² grid spanning the entire field. Quantile classification (10 levels).
Fig 6.
(a) Estimation of average total earthworm abundances based on the GAM as given in Table 3, (b) state-space modelling (SSM) with the PSS variables as covariates as given in Table 3, (c) SSM with ECash and ECadp and (d) SSM without co-variates. Confidence intervals were computed from the standard errors of prediction on a 95% significance level.