Developing Predictive Soil Organic C and N Models for Glaciated Soils Using Quantitative Color Sensor Measurements.

Soil organic carbon (SOC) and total nitrogen (N) are typically measured in laboratory settings and are vital to soil fertility and management. Advances in color sensor technology allow for affordable, rapid assessment of SOC both in the laboratory and in the field, but little has been done in the way of predicting total N from soil color data. The objective of this study was to assess the relationship between SOC and total N and to use an inexpensive color sensor to develop SOC and total N prediction models for multiple soil types collected at different depths from the Willsboro Farm in northeastern New York State. One hundred fifty-six dried soil samples made up of Alfisols, Entisols, and Inceptisols were analyzed for CIEL*a*b* color using a Nix Pro^TM color sensor. Soil sample horizon lower depth, L*, a*, and b* for each sample were used as predictor variables in regression analyses to develop SOC and total N prediction models for Alfisols, Entisols, and all soil samples combined. The natural log of SOC and total N were used as dependent variables. Resulting residual plots, root mean square errors (RMSE), and coefficients of determination (R²) were used to asses model fit for the SOC and total N prediction models. Cross validation was conducted for SOC and total N models for each sample set to determine how effective the prediction models were and the mean squared prediction error (MSPE) was calculated. The final lnSOC models containing sample horizon depth and soil color parameters resulted in R²=0.771 and MSPE=0.503 for all soil samples, R²=0.708 and MSPE=0.786 for Alfisols, and R²=0.739 and MSPE=0.159 for Entisols. The final lnTN models containing sample horizon depth and soil color parameters resulted in R²=0.784 and MSPE=0.008 for all soil samples, R²=0.794 and MSPE=0.013 for Alfisols, and R²=0.700 and MSPE=0.002 for Entisols. The final lnSOC models containing soil color parameters only resulted in R²=0.633 and MSPE=0.816 for all soil samples, R²=0.587 and MSPE=2.103 for Alfisols, and R²=0.623 and MSPE=0.246 for Entisols. The final lnTN models containing soil color parameters only resulted in R²=0.656 and MSPE=0.009 for all soil samples, R²=0.625 and MSPE=0.014 for Alfisols, and R²=0.647 and MSPE=0.003 for Entisols. Our results suggests that there may be a potential for rapid assessment of total N in certain soil types based on a difference in prediction model parameters for SOC and total N using soil color data.