Agronomic research is vital to determining optimum inputs for crops to perform profitably at a local scale. However, the small-plot experiment validity is often uncertain due to on-farm variations. Furthermore, the likelihood of conducting a fully randomized trial at a local farm is low given various practical and technical challenges. We propose a new methodology with many inputs to allow for a response surface that fits the yield response to the input levels with higher accuracy to make on-farm decisions. The method makes deployment of treatments on-farm easy and allows means to quantify the spatial influence on the response variables for higher precision. Preliminary results suggest that the yield response to nitrogen is linear (R²=0.86) and methodological advancement to integrate image-based high throughput techniques to attribute yield and ground cover accumulations is promising. Strong correlations and linear dependencies between N treatment levels, yield, and the ground cover based on spectral indices indicate many possibilities to manage the yield potential agronomically. However, the departure from fully randomized trials likely intensifies the spatial heterogeneity influence on treatments up to a certain degree. Such patterns significantly contribute to residual heteroscedasticity and are likely to produce incorrect standard errors as it tends to violate the assumption var(y_i)=var(e_i)=σ². However, significant ANOVA results suggest that sufficiently large enough samples can lower standard errors and minimize the influence on estimates precision. Our primary target is to develop a statistical analysis approach that accounts for spatial-covariance to improve the accuracy and precision of the Modulated On-farm Response Surface Experiments (MORSE) for on-farm precision agronomy.