Quantitative determination of nutrient content in poultry manure by near infrared spectroscopy based on artificial neural networks

Abstract

Excessively applied manure contains a considerable amount of nutrient content such as nitrogen and phosphorus that could potentially pollute groundwater and soil. The present paper evaluated the use of nonlinear regression methods, such as artificial neural networks (ANN), for developing near infrared reflectance spectroscopy calibration models to predict nutrient content in poultry manure. Four representative nutrient ingredients (ammonia nitrogen, AN; total potassium, TK; total nitrogen, TN; total phosphorus, TP) in poultry manure were selected for evaluating ANN feasibility using 91 diverse samples in which three-fourths of the samples were used as a training set and one-fourth as a validation set. The performance of the ANN models was compared with the partial least squares (PLS) models. We found that the ANN models for all 4 nutrient contents consistently gave better predictions than PLS models. The ratios of prediction to deviation of 2.62 (AN), 1.51 (TK), 2.75 (TN), and 2.01 (TP) with the PLS models were improved to 3.02 (AN), 1.74 (TK), 3.41 (TN), and 2.71 (TP) with the corresponding ANN models. These findings demonstrated that the near infrared reflectance spectroscopy model based on the ANN method may be an appropriate tool to predict nutrient content in poultry manure.