Page Banner

United States Department of Agriculture

Agricultural Research Service

Research Project: Pesticide Application Technologies for Spray-drift Management, Maximizing In-field Deposition, and Targeted Spraying

Location: Crop Production Systems Research Unit

Title: Spatial Modeling and Variability Analysis for Modeling and Prediction of Soil and Crop Canopy Coverage Using Multispectral Imagery from an Airborne Remote Sensing System

Authors
item Huang, Yanbo
item Lan, Yubin
item Ge, Yufeng -
item Hoffmann, Wesley
item Thomson, Steven

Submitted to: Transactions of the ASABE
Publication Type: Peer Reviewed Journal
Publication Acceptance Date: August 2, 2010
Publication Date: September 8, 2010
Citation: Huang, Y., Lan, Y., Ge, Y., Hoffmann, W.C., Thomson, S.J. 2010. Spatial Modeling and Variability Analysis for Modeling and Prediction of Soil and Crop Canopy Coverage Using Multispectral Imagery from an Airborne Remote Sensing System. Transactions of the ASABE. 53(4):1321-1329.

Interpretive Summary: A multispectral (visible and near infrared) camera, which is stabilized by a control system for adjustment with aircraft’s roll, pitch and yaw rotations, has been mounted on a single-engine aircraft. The aircraft flew over a 115 ha cotton field and multispectral imagery was acquired using the camera at different altitudes. The acquired images were geo-registered and processed for analysis. The correlations between the aerial images and ground-based soil conductivity and NDVI (Normalized Difference Vegetation Index) measurements are spatially varied. Conventional statistical techniques cannot effectively model the relationships because these techniques do not consider any spatial connection of one data point with other points (typically neighborhood points). Spatial statistical techniques are needed to handle the spatial variability in the data. This study has compared two spatial analysis approaches: spatial regression and regression kriging with one conventional non-spatial approach: multiple linear regression. The results indicated that spatial regression and regression kriging performed significantly better than the multiple linear regression (0.25) for predicting soil conductivity with much higher R2 values for goodness of fit (0.70), and produced moderate improvements from the conventional multiple linear regression (0.08) for predicting ground-based NDVI with higher R2 value of 0.26 (spatial error regression) and 0.44 (regression kriging). The results also illustrated that the aerial images could be a useful, informative data source for spatial modeling and prediction of ground soil and canopy coverage variability.

Technical Abstract: Based on a previous study on an airborne remote sensing system with automatic camera stabilization for crop management, multispectral imagery was acquired using the MS-4100 multispectral camera at different flight altitudes over a 115 ha cotton field. After the acquired images were geo-registered and processed, spatial relationships between the aerial images and ground-based soil conductivity and NDVI (Normalized Difference Vegetation Index) measurements were estimated and compared using two spatial analysis approaches: model-driven spatial regression and data-driven regression kriging, and one non-spatial approach: multiple linear regression. In comparison of the three approaches, the solutions of OLS (Ordinary Least Squares) from multiple linear regression performed poor with low R2 values (0.08-0.25). Spatial regression and regression kriging performed much better for predicting soil conductivity with R2 values near 0.70 although the spatial regression methods resulted in higher RMSE (Root Mean Squared Error) values for soil conductivity (0.30), which was caused by systematic bias between measured and predicted values. For predicting ground-based NDVI, R2 value of 0.26 and 0.44 were resulted for spatial error regression and regression kriging, respectively, which were only moderate improvements from OLS (0.08). The results indicated that the aerial images could be used for spatial modeling and prediction, and they were informative for spatial prediction of ground soil and canopy coverage variability. The methods of the study could help deliver baseline data for cropping in the experimental field and establish a procedure for general crop management.

Last Modified: 10/1/2014
Footer Content Back to Top of Page