Evaluation of Vegetation Indices for Sugarcane Yield Modeling with Emphasis on Growth Pattern Based on Satellite Imagery: (Case Study: Khouzestan Imam Khomeini Agro Industry)

The aim of this study is to determine the growth pattern and to investigate the vegetation indices power for sugarcane yield modelling at field scale in Imam Khomeini Agro-industry. For this purpose, the vegetation indices extracted from Landsat7 satellite images were investigated using time series analysis. Overall, 306 Landsat7 satellite images from March 2004 to February 2017 were used. All of the images were converted to surface reflectance via FLAASH algorithm. The average values of 13 vegetation indices related to the study region extracted from satellite images and converted to seven days' time-series via interpolation. In order to eliminate the noise, all series were reconstructed using the Savitzky-Golay algorithm. Thus, 13 different time series of vegetation indices were made for 523 sugarcane fields. Then the growth pattern was drawn via averaging NDVI time series and it was divided into three growth periods. Then the accumulative values of vegetation indices related to the first and second periods of growth stage were extracted since 2004 to 2017. Therefore, 3286 samples were prepared overall, of which 2628 samples were used for modelling and 658 samples for evaluation. The samples extracted from time series were evaluated by simple linear regression model against the average observed yields. The result showed that the accumulative vegetation index of GNDVI for the first growth period with R2=0.47, RMSE=11.70 ton/ha and the accumulative vegetation index of NDI for the second growth period with R2=0.56, RMSE=10.62 ton/ha are a better indeces for sugarcane yield estimation as compared to the other vegetation indices. Also, the sum of GNDVI and NDI indeces for summation of first and second growth periods had a better result (R2=0.65, RMSE=9.47 ton/ha) than that's where one index at one period was used. Finally, the sugarcane yield of 658 samples was estimated for evaluation and the R2 and RMSE of the best model was obtained to be 0.58 and 10.99 ton/ha, respectively. The results of this study confirm the suitability of the GNDVI and NDI indeces for monitoring sugarcane growth during the first and second growth stages.