Nonlinear Hierarchical Models for Predicting Red Clover Biomass Using NDVI and Factors Affecting Biomass Performance.

Cover crops generate important benefits to soil ecosystems, especially improving several soil properties for crop growth. Understanding of how variations in topography and soil properties affect cover crops will contribute to increase its adoption among farmers, thus allowing producers to tailor cover crop management in their fields. However, these relationships are not well established yet and quantification of biomass across topographical conditions is time consuming and expensive task. Our first objective was to examine whether normalized difference vegetative index (NDVI) can be used to predict biomass of red clover planted as a cover crop in a corn-soybean-wheat rotation system, thus providing a relatively inexpensive method of obtaining dense cover crop biomass measurements. The second objective was to evaluate effects of topography and soil properties on red clover growth. Data were collected from three field sites in southwest Michigan in August and September 2007. An optical active sensor was used to record NDVI of 0.5 x 0.5 m or 1.0 x 1.0 m quadrants at each site, after which biomass samples were collected. Sensor NDVI readings were also collected at a density of approximately 570 points ha^-1 across each of 2 of the sites. Elevation data were used to derive topographical attributes. At one of the field sites, data on total soil C and N, soil texture, and bulk density were obtained in 2004-2006.  Results indicated that the use of NDVI is a promising tool for predicting red clover biomass on a field scale. We used hierarchical non-linear models to improve predictions between and within fields, being a Richards model with random coefficients the one with best performance. NDVI explained 84% of biomass variability in August and 70% in September. The final model also includes a term for unequal variances, allowing accurate predictions at lower NDVI values. Clover biomass more closely related to soil properties than to topographical features. Biomass was positively correlated with total soil C (r=0.61), total soil N (r=0.68), and silt content (r=0.45) and was negatively correlated with sand content (r=-0.46). Clover growth better at sites with lower elevation (r=-0.34) and slope(r=-0.33). Relative elevation and slope were significant included in all models for each field and sampling date. Results from the first year indicate relative elevation and slope are the factors more related to biomass performance. Results from the second year will be reported on the conference.