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 American Society of Agricultural and Biological Engineers (ASABE), 53(4): 1321-1329. 2010
Y. Huang, Y. Lan, Y. Ge, W. C. Hoffmann, S. J. Thomson
“Spatial modeling and variability analysis of soil and crop canopy coverage has been accomplished using aerial multispectral images. Multispectral imagery was acquired using an MS-4100 multispectral camera at different flight altitudes over a 40 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 geostatistics) and one non-spatial approach (multiple linear regression). Comparison of the three approaches indicated that OLS (ordinary least squares) solutions from multiple linear regression models performed worst in modeling ground-based soil conductivity and NDVI with high AIC (Akaike information criterion) (-668.3 to 2980) and BIC (Bayesian information criterion) (-642.4 to 3006) values. Spatial regression and geostatistics performed much better in modeling soil conductivity, with low AIC (2698 to 2820) and BIC (2732 to 2850) values. For modeling ground-based NDVI, the AIC and BIC values were -681.7 and -652.1, respectively, for spatial error regression and -679.8 and -646.5, respectively, for geostatistics, which were only moderate improvements over OLS (-668.3 and -642.4). Validation of the geostatistical models indicated that they could predict soil conductivity much better than the corresponding multiple linear regression models, with lower RMSE (root mean squared error) values (0.096 to 0.186, compared to 0.146 to 0.306). 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 used for this study could help deliver baseline data for crop monitoring with remote sensing and establish a procedure for general crop management.”
