International workshop on

Koli (c) Metla/Erkki Oksanen

Development of Models and Forest Soil Surveys for Monitoring of Soil Carbon

April 5-8, 2006 at Koli, Finland

Validation of modelled soil organic carbon pools by DRIFT-PLS

Jens Leifeld1, Uwe Franko2 and Elke Schulz2

1 Swiss Federal Research Station for Agroecology and Agriculture, Zürich, Switzerland
2 UFZ Centre for Environmental Research, Leipzig-Halle, Germany

Pool-based models for the turnover of soil organic carbon (SOC) are often successfully applied to describe soil organic matter dynamics under different management. Model validation is typically obtained by comparing measured with modelled bulk SOC contents. Because pools are mathematically defined and regarded as conceptual, a direct comparison of pools with measurable soil properties is challenging. A linkage between modelled pools and measurable soil attributes would be valuable in particular for the experimental validation of short to medium-term C pools that respond relatively fast to management. Moreover, assigning chemical or physical attributes to soil C-pools could contribute to an improved process understanding.

We have used the process-oriented model CANDY to simulate SOC in long-term field experiments at Bad Lauchstädt, Germany. The soil carbon module of CANDY consists of several C pools (i.e., ‘inert’, ‘stabilised’, ‘active’, ‘plant residues’) that differ in their turnover time. The selected field treatments included different types and levels of fertilisers at different days of the year and represent steady-states at high and low SOC contents and transitional states from high to low or vice versa. Measured and modelled C contents agreed reasonably well (R2 = 0.85 – 0.94). Pool size values were taken as y-variables to perform partial least squares (PLS) regression on diffuse reflectance infrared FT (DRIFT) spectra of bulk soils. Prediction for cross-validated samples were of reasonable quality (R2 = 0.82 - 0.90; relative standard error of prediction 6 - 7%) for bulk SOC and ‘inert’ C, the latter contributing approximately 75% of the total SOC. Prediction of the ‘active’ and ‘stabilised’ SOC pools required separate calibration of high and low fertiliser treatments. For the so partitioned samples, these pools could be predicted with similar accuracy as above. The latent variables of the PLS regression were markedly different, indicating that the corresponding chemical constituents of a single pool are not the same for all of the treatments. Prediction of ‘plant residues’ failed, most likely because they accounted for less than one percent of the SOC and thus contributed only little to IR reflectance.

The prediction of modelled pools by DRIFT-PLS offers a tool for model validation. The results demonstrate that the pools are not purely conceptual, but have some chemical analogy. Apart from errors in measuring and modelling, we suggest that some of the unexplained variance is attributable to SOC-stabilisation processes that are not directly related to the chemical composition of the soil and are thus not detectable by spectroscopic methods, for example.

   Päivitetty:   02.03.2006 / EKel Metla : Events   Palaute Metlan etusivulle
. .