Estimating Wheat Canopy Cover by Remote Sensing.

Quantifying crop phenological development prior to attainment of full canopy using light transmission methods to estimate leaf area index (e.g., LI-COR 2000 Plant Canopy Analyzer (LI-COR Biosciences, Lincoln, NE) is challenging. Canopy cover based on a vertical overhead perspective, in which land area is classified as either crop canopy or other (bare soil, crop residue, etc.), offers an alternative approach for sparse canopies. We recorded digital images (3 bands: nir, red, green, pixel size=1.6 mm) from 2 m above the crop canopy of winter wheat (Triticum aestivum L.) grown under a range of soil moisture and tillage systems. We classified the pixels using two unsupervised multispectral techniques: (1) density slicing (DS) of NDVI (normalized difference vegetation index = [nir-red]/[nir+red]), where pixels classified as canopy had an NDVI≥0.3 and (2) spectral cluster analysis (SCA) using an iterative self-organizing data algorithm (ISODATA). Overall accuracy of the SCA method was estimated at approximately 95% across a wide range of canopy conditions from sparse to dense. The NDVI threshold (0.3) was chosen based on a digital grid overlay using 100 random points, but these reference data were inferior to those used for the SCA method because of the difficulty of manually classifying a single random reference point overlying very small pixels. Therefore, we considered SCA as the more accurate method and compared DS to SCA by correlation (r=0.96). Either method has application as a remotely sensed proxy for canopy development in wheat growth models. While SCA was considered the more accurate method for small pixels, density slicing NDVI is more efficient to implement and, with calibration against SCA, could be used with confidence.