Application of support vector machine technology for the estimation of crop biophysical parameters using aerial hyperspectral observations

Karimi, Y., Prasher, S.O., Madani, A. and Kim, S. 2008. Application of support vector machine technology for the estimation of crop biophysical parameters using aerial hyperspectral observations. Canadian Biosystems Engineering/Le génie des biosystèmes au Canada 50: 7.13 7.20. This study investigated the potential of support vector machine (SVM) methodology to extract information on crop growth and various biophysical parameters from airborne hyperspectral remote sensing data. The study was carried out in a corn field, consisting of a main plot with four weed control strategies (no weed control, broadleaf control, grass control, and full weed control) and a sub-plot with three nitrogen (N) fertilization rates (60, 120, 250 N kg/ha), all replicated four times. Hyperspectral data were taken in 72 narrow wavebands (409 to 947 nm) using a compact airborne spectrographic imager (CASI) sensor. Various crop physiological parameters were measured concurrently including: leaf greenness (SPAD readings), plant height, leaf nitrogen content, and leaf chlorophyll content. The objective of the study was to evaluate the ability of SVM regression models to extract continuous vegetation variables using aerial hyperspectral observations. Several SVM models were developed to estimate the crop biophysical parameters and the grain yield. The study showed that by using reflectance data collected at the tasseling stage, crop parameters can be estimated with reasonable accuracy. Generally speaking, the coefficients of determination (r) were greater than 0.9 for biomass, yield, plant height, and SPAD with the training data set. The r was slightly lower for the test data set (0.51, 0.82, 0.91, and 0.86, respectively), which is acceptable given the small size of the data set used in the study. The results of the five fold cross validation procedure indicated that the SVM results were consistent. The results were also compared with those obtained with a stepwise approach, and the SVM results were found to be superior.