Within-Field Yield Prediction in Cereal Crops Using LiDAR-Derived Topographic Attributes with Geographically Weighted Regression Models

Abstract

Accurate yield estimation and optimized agricultural management is a key goal in precision agriculture, while depending on many different production attributes, such as soil properties, fertilizer and irrigation management, the weather, and topography.The need for timely and accurate sensing of these inputs at the within field-scale has led to increased adoption of very high-resolution remote and proximal sensing technologies. With regard to topography attributes, greater attention is currently being devoted to LiDAR datasets (Light Detection and Ranging), mainly because numerous topographic variables can be derived at very high spatial resolution from these datasets. The current study uses LiDAR elevation data from agricultural land in southern Ontario, Canada to derive several topographic attributes such as slope, and topographic wetness index, which were then correlated to seven years of crop yield data. The effectiveness of each topographic derivative was independently tested using a moving-window correlation technique. Finally, the correlated derivatives were selected as explanatory variables for geographically weighted regression (GWR) models. The global coefficient of determination values (determined from an average of all the local relationships) were found to be R2 = 0.80 for corn, R2 = 0.73 for wheat, R2 = 0.71 for soybeans and R2 = 0.75 for the average of all crops. These results indicate that GWR models using topographic variables derived from LiDAR can effectively explain yield variation of several crop types on an entire-field scale.

Keywords

• LiDAR
• agricultural yield prediction
• geographically weighted regression
• topographic derivatives