Submitted to: BARC Poster Day
Publication Type: Abstract Only
Publication Acceptance Date: March 12, 2009
Publication Date: N/A
The USDA-NRCS is partnering with other agencies to conduct a national scale assessment of environmental benefits and effects of 2002 Farm Bill programs. One of the components of the resulting Conservation Effects Assessment Project (CEAP) is a national-scale effort to estimate the on-site benefits of conservation practices in cultivated croplands. Our group provides an independent evaluation of the cropland component of the national-scale CEAP project. As part of the assignment, we evaluate how field soil conditions are represented in the simulations. We identified that soil bulk density (BD) – a property that e.g. partially determines conditions for soil aeration, solute transport and storage as well as the outcome of soil carbon stock calculations – has been represented by values that: (1) reflect map based, aggregated (up-scaled) values; (2) that have been derived following partially unknown/unpublished methodology, and (3) that appear to show improbable or unrealistic values in for many soils. Such values need to be revised in order to obtain a realistic representation of field soil conditions. We offer a solution to this problem that uses a US-wide soil database of point measurements and a published ‘k-nearest neighbor’ lazy learning algorithm combined with a data re-sampling technique. We provide a ‘second opinion’ to existing BD values and use the statistical properties of the estimations to give grounds for the rejection of improbable values in the CEAP database. The RMSE and ME of estimations for independent measured point BD values, using the chosen algorithm options is 0.175 and -0.01 (g/cm3) respectively, that are both comparable with such measures reported in literature. The developed technique provides up to a 98.2% acceptance rate of true BD values which depends on a user defined - standard deviation based - criterion of strictness.