Frequency Domain Analysis of Remotely Sensed Vegetation Indices and Crop Yield after Sinusoidal Nitrogen Application.

To determine optimum nitrogen fertilizer application rates, scientists usually perform a randomized plot study in an experimental field. Inherent soil variability can, however, cause misleading conclusions. In the worst case, no significant treatment effect can be found. In this study, varying nitrogen rates were not randomly applied but in a sinusoidal pattern across a farmers wheat field. Six Nitrogen (N) levels between 0 and 180 kg N/ha were applied in 42 plots (27 by 15 m, n = 4 - 9) along a 650 m long transect. During the growing season, leaf area index (LAI), normalized difference vegetation index (NDVI), and plant height were measured at different times. Wheat yield was measured with an automatic yield monitoring system. The following results were obtained. Wheat yield varied considerably along the treatment transect. No significant effect of nitrogen fertilizer was detected if plots were statistically treated as if they were a randomized experiment analyzed using Fisher’s ANOVA. However, when plots and all measured observations were treated as regionalized variables, fertilizer application rate was identified as an obvious source of variation in the field. Wheat yield clearly exhibited a relationship with vegetative measurements taken during the growing season. Moreover, a systematic change in soil quality across the field affected the results. Sources of yield variation were identified using spectral analysis, based on Fourier transformation. The power spectra of nitrogen application, wheat yield, and vegetative indices exhibited peaks at common wave lengths, indicating their spatial variation was associated with each other. Computer software was developed to study yield and crop sensor measurements at different spatial resolution and different data aggregation levels. First results of sample support-size reveal a narrow resolution of the signal of NDVI obtained from an active sensor.