The Use Of A Ground Based Remote Sensor For Winter Wheat Grain Yield Prediction In Northern Poland
The aim of the research was to investigate if algorithms developed for winter wheat,
cv. Trend, yield predictions, based on ground measured GNDVI, differ significantly between 2 sequent years. The research was conducted in Pomerania, northern Poland (54° 31' N 17° 18' E) on sandy loam soils. The strip-trial design was used to compare the effect of 6 N treatments: 0, 50, 100, 150, 200 and 250 kg ha-1, applied as one dose at the beginning of vegetation, on grain yield variation. A light sensor Crop Circle ACS-210 (Holland Scientific, Lincoln, NE) was used to collect reflectance data at 590 and 880 nm, which were thenused to calculate the Green Normalized Difference VegetationIndex (GNDVI), at the following growth stages: GS-31, GS-39 and GS-69. At harvest time (GS 92) plants from the area of 1m2 were hand cut and used to determine grain yield at 15% of moisture. To perform a validation of the algorithm from 2008, in 2009 GNDVI was also measured on farm field (22ha) sown with the same wheat cultivar, then grain yield was determined using yield monitor. Relationships between grain yields and GNDVI were evaluated using linear regression. Statistical comparison between slopes of regression, for the years, was based on results of multiple regression analysis (year was treated as binomial independent variable).
Comparison of slopes of the relationship grain yield vs. GNDVI, for the same measurement dates, indicated significant differences between years at two earlier growth stages. This was probably due to definitely varied weather conditions at the time of intensive wheat growth. As the plants developed throughout the growing season, the differences between N treatments with respect to GNDVI increased. The later the reflectance data were collected the higher the relationship between GNDVI and grain yield was obtained, respectively in 2008 and 2009, R2 equaled 88,6 and 77,9%. However, predicted grain yields for 2009 were overestimated by 33% when the algorithm for grain yield estimation vs. GNDVI from GS-69, year 2008 was used.
