ABSTRACT

A field experiment was conducted at MARS, Dharwad during kharif 2023 on medium black soil for validation of leaf area index of maize for graded levels of fertilizers using conventional and artificial intelligence techniques. The experiment was laid out in split plot design comprising with three fertilizer levels 50, 100 (100: 50: 25 kg N: P2O5: K2O ha-1) and 150 per cent RDF as main plot and five methods of estimation of leaf area index (LAI) of maize [length × breadth method, disc method, leaf area meter, canopy analyzer and artificial intelligence (AI)] as sub plot and control (Without fertilizers).
Application of 150 per cent RDF recorded significantly higher grain (75.64 q ha-1) and stover yield (96.18 q ha-1) of maize than 50 per cent RDF (38.69 q ha-1 and 55.71 q ha-1, respectively) and it was on par with 100 per cent RDF (72.77 q ha-1 and 94.64 q ha-1, respectively).
Among the subplots there was no significant differences in grain and stover yield of maize. Among interactions, 150 per cent RDF + LAI estimation by AI method showed significantly higher grain yield (75.70 q ha-1) and stover yield (96.26 q ha-1) than control.
Among the different methods of LAI estimation, AI method showed least deviation (1.02-14.77 %) particularly at grain filling (1.02 %) followed by silking stage (2.9 %) and maximum deviation (46.1-58.0 %) was observed with disc method at all the growth stages.
Among machine learning models, random forest model outperformed other models with R² (0.67-0.94) and RMSE (0.02-0.26) at all the growth stages (Knee-high stage, tasseling stage, siliking stage and grain filling stage) compared to other models.