In agriculture, finding correlation between field variables such as yield, NDVI, etc. is a classic problem. The most popular solution for finding correlation is to use regression analysis over complete field. In a field, there are numerous soil variables such as soil composition, water content, etc. that affect the correlation and it is not always possible to consider such factors in regression model. These factors adversely affect the accuracy of the correlation model. We demonstrate that it is incomplete and inaccurate to represent such correlation using single regression model for the complete field. We propose a novel technique- Sliding Window Based Technique, which finds correlations over small areas, windows, of the field instead of modelling one correlation for entire field. We prove our claims with the help of experiments done on the field data. We evaluate the relationship over multiple window sizes and select appropriate window sizes for further analysis.