Estimating Rice Grain Yield Potential Using Normalized Difference Vegetation Index

Normalized difference vegetation index (NDVI) measurements have the potential to improve mid-season N crop management decisions in rice (Oryza sativa L.). The objectives of this study were to determine the optimum sensing timing and establish a yield prediction model using NDVI measurements acquired with the GreenSeeker sensor. Weekly sensor readings were collected over a 5-wk period from multi-rate N fertilization trials established at six different locations from 2008 to 2010. Categorizing sensing timing by growth stage demonstrated that late sensing timings beyond panicle differentiation (PD) were impractical and reduced yield potential estimation as opposed to panicle initiation (PI) and PD timings. Regression analysis produced two viable yield potential prediction equations at PI (r(2) = 0.36) and PD (r(2) = 0.42). When sensor timings were categorized by cumulative growing degree days (GDD), 1301 to 1500 and > 2100 GDD groupings (r(2) = 0.28 and 0.37, respectively) were found to be inferior yield predictors as compared with 1501 to 1700 and 1701 to 1900 GDD groupings (r(2) = 0.41 for both). In almost all instances, normalization of NDVI data using days from seeding (DFS; NDVI/DFS) or GDD (NDVI/GDD) did not improve yield potential prediction as compared with NDVI alone. Yield potential, response index, and N response to fertilization are the three major components needed to produce a working algorithm capable of predicting mid-season N fertilization needs in rice. The four yield prediction models gleaned from this study provide the yield potential component for this algorithm. Multiple yield prediction models give crop managers freedom to select a model based on either physical growth stage or by accumulated GDD units.