Zhang Zhitao ^{a, b, c}, Lan Yubin ^{b *}, Wu Pute ^c, Han Wenting ^c

　　(a. College of Water Resources and Architectural Engineering, Northwest A&F University, Yangling, Shaanxi, 712100, China.  b. Aerial Application Technology, USDA-ARS-SPARC -AATRU, College Station, TX 77845, USA. c. Institute of Water Saving Agriculture in Arid Area of China, Northwest A&F University, Yangling, Shaanxi, 712100, China.)

　　Due to the factors of air temperature, humidity, solar radiation, soil moisture, etc., Normalized Difference Vegetation Index (NDVI) of soybean varies dynamically in a day. The soybean NDVI values are continuously monitored in hours during soybean seeding, flowering & podding and maturating phases by way of UAV. Results show that the trend of NDVI change every day in the three stages is taken on as a reserve parabola. The NDVI value reaches the maximum at 8am or 9am and then decreases gradually with its minimum at 2pm and then it increases to some extent. A model for NDVI change in a day and continued days of soybean is built. The test of the models with independent data indicates that the precision meets the demands, with the root mean square error (RMSE) of each day being 3.95, 5.45 and 2.86 for seeding stage, bean pod stage and maturation period respectively. The prediction RMSEs of the soybean NDVI models in the fifth day are 5.75, 2.65 and 5.51 respectively and the prediction RMSEs in the sixth day are 9.74, 2.82 and 14.04 respectively according to the data from the first four days. Meanwhile, the air temperature, humidity, solar radiation, wind speed are measured to set up a Regression Model to test the impact of the meteorological factors on soybean NDVI and analyze the quantitative relations among them by using Partial Least Squares (PLS), Stepwise Regression and Ridge Regression. Results show that among the main meteorological factors solar radiation and air temperature are the major ones affecting the soybean NDVI values, the wind speed is the minor one, while the influence of humidity can be neglected. In contrast, the Ridge Regression ranks the highest predictive accuracy, with the Root Mean Square Error (RMSE) of 0.034、0.018 and 0.016 and R-Square of 0.820、0.908 and 0.934 in three stages of seeding, flowering & podding and maturating, followed by a less accurate predictive level of Stepwise Regression and the least accuracy of PLS.