In this paper we will show how plant communities can be utilized to improve the quality of digital soil maps, effectively reducing the amount of field work needed by soil surveys, through a combination of relatively swifter and cheaper vegetation surveys and remote sensing data.
The approach we propose is based on the spectral and textural properties of plant communities which can be summarized from high resolution remotely sensed images and LIDAR data through the use of geostatistical, spectral and geomorphometric descriptors. These descriptors are then associated with the scores obtained from the ordination of the plant communities' relative coverage. Ordination projects the high dimensional plant cover data into a lesser dimensional space, thus making easier to establish a relation between ecological space and geostatistical descriptors. Once established this relation can be exploited through the use of regression techniques in a regression kriging framework.
In this case study, we applied the proposed model to the prediction of soil organic carbon content in an alpine grassland. The use of plant communities cover almost doubled the predictive power of the model from an R2 of 0.32 to an R2 of 0.66 in cross-validation, a result which strongly advocates for the efficiency of the proposed approach.