การประมาณการผลผลิตข้าวด้วยค่าดัชนีพืชพรรณโดยภาพถ่ายรายละเอียดสูงหลายช่วงคลื่นจากอากาศยานไร้คนขับ

Abstract

Multi-spectral high resolution imagery from an unmanned aerial vehicle (UAV) of rice (Chai Nat 1 varieties) in the experimental plots at Mae Hia Agricultural Research, Demonstration and Training Center, Faculty of Agriculture, Chiang Mai University was analyzed spatially to crop spectral parameters with a program of geographic information system (GIS) in terms of vegetation index (VI) in three types i.e. (1) normalized difference vegetation index (NDVI), (2) green normalized difference vegetation index (GNDVI), and (3) triangular greenness index (TGI), including two physical collaborating parameters i.e. (4) canopy cover (CC) and (5) plant height (PH). These parameters were tested collaboratively with the rice yield data from crop cutting as the model development in terms of simple and multiple linear regression equations to estimate rice yield in experimental plots. Plant spectral parameters received from UAV were analyzed spatially as vegetative index and rice physical values of NDVI, GNDVI, TGI, CC and PH at the 90 days after planting, with the mean values of 0.02254, -0.01069, 0.43816, 0.77915 and 0.73154, respectively. Those rice parameters were applied collaboratively as the predicting variables (predictors, X1 - X5) in the stage of rice yield model development. The results showed that the most reliable model was developed with crop spectral parameters from unmanned aerial vehicle (UAV) in terms of multiple linear regression equation that is Y = 1261.497 -1017.303X1 + 2353.014X2 -1306.652X3+ 231.248 X4 –113.286, with the determination coefficient of 0.812. As the stage of model development, the model could estimate an averaged rice yield of 738.18 g/m2 (1,181.08 kg/rai). Meanwhile, in the stage of model validation, the model provided an averaged rice yield of 746.40 g/m2 (1,194.24 kg/rai), less than the observed yield of 15.11 g/m2. Finally, the model then was conducted to predict rice (Chai Nat 1 varieties) yield at the 5 neighboring rice growing plots, adjusted the standard rice seed humidity at 14% getting an averaged rice yield of 1,305.37 kg/rai, less than the observed yield of 289.11 kg/rai. In conclusion, the developed model still could not estimate closely rice yield to the actual values. Therefore, for the efficient rice model in the future, it is possible that some more other parameters of rice in the experimental plots are necessary jointly in the model other than vegetation indices and physical factors from UAV. However, this study can be useful for the approach, analyzing stages, and application of using data from UAV to develop rice yield model continually