Construction and optimization of a near-infrared spectroscopic model for the assessment of nitrogen and phosphorus in dairy farm slurry under different temperatures

Near-infrared spectroscopy (NIRS) was widely used for total nitrogen (TN) and total phosphorus (TP) detection in the slurry of animal farming. However, temperature could interfere with NIRS in practical measurements, affecting the robustness of models. Therefore, we employed three methods of temperature global model (TG), external parameter orthogonalization (EPO) and genetic algorithms (GA) to mitigate the impact of temperature on model performance. We collected 365 slurry samples from a representative dairy farm in Tianjin, China. Temperature gradient experiments were implemented in the lab, setting 5 ℃ intervals within the range of 0–40 ℃. It was found that temperature had a greater influence on the TN model compared to the TP model, with higher temperature occurring greater effects. Compared to the unoptimized TN model, the TG, EPO, and GA optimized models of TN showed reductions in RMSEP ranging from −8% to 32 %, −14 % to 35 %, and −1% to 33 %, respectively. And for the unoptimized TP model, the TG, EPO, and GA optimized models of TP exhibited a decrease in RMSEP ranging from −27 % to 16 %, −33 % to 13 %, and −24 % to 9 %, respectively. The optimal method for TN was the GA method (RMSEP = 168 mg/L), while the optimal method for TP was EPO filtering (RMSEP = 69 mg/L). The ability to rapidly and accurately predict TN and TP contents in slurry under field variable temperatures offer critical technical support for optimizing slurry management, enhancing nutrient utilization efficiency, and promoting sustainable resource recycling.