Sugarcane Growth Monitoring And Yield Forecasting Based On HJ Satellite Data

Guangxi Xinbin district is the largest production base of sucrose in China, where sugarcane production has become the local economic leading and pillar industry. So, it is of great significance to obtain sugarcane’s information timely and accurately including planting area, growth status and yield. These intelligence play an important role in sugarcane production management, sucrose market stability, sucrose safety and economic development. Currently, there has been very little study on dynamic monitoring and yield prediction of sugarcane based on "3S" technology in China and especially lacks of domestic satellite data application.In this paper, the sugarcane planting area is extracted by using remote sensing methods combining supervised classification and stepwise iteration method based on multi-temporal HJ-1A/B CCD images. Then, Combined with the agronomic parameters observed in field, sugarcane growth monitoring model is established according to standard deviation and deviation of the normally distributed random variable theory. Lastly, the relationship between the vegetation index and yield’s data is explored and sugarcane yield estimation model is build. The main results show as follows:(1)Sugarcane cultivate area is the foundation of sugarcane growth monitoring and yield estimation, which is extracted utilizes remote sensing methods and multi-source data. Firstly, interpretation keys of typical land cover types are found in terms of spectral characteristic analysis. Secondly, the possible planting area of sugarcane is identified by the use of supervised classification, land layers and slope data. Finally, Sugarcane growing region is drew accurately by the combination of NDVI thresholdvalues of multidate remote sensing images based on the interpretation key sand stepwise iteration method. The field survey data and statistical data of town’s agricultural department are used to precision validation with the overall accuracy 91.18%, Kappa coefficient 0.8101, and the average relative error is 10.17% compared with statistical data.(2)Sugarcane growth monitoring model is established according to the random variable theory. Concerning the current situation of lacking unified grading standard on sugarcane growth evaluation in China, the relationships and differences between remote sensing indicators and the agronomic parameters observed in field are analyzed in this paper. Then, enhanced vegetation index (EVI) and normalized difference vegetation index (NDVI) are selected as sugarcane status’s indicators for mature period and stem elongation period respectively. Deviation from mean is added in the growth model as sugarcane’s growth obeys the normal distribution, followed by differences between standard deviation and deviation from mean are used to evaluate sugarcane growth condition quantitatively in Xingbin district from 2009 to 2013.Sugarcane overall yieldfluctuation, sugarcane growth status inferred from field survey point data measured on Nov.2013and application results in part counties are utilized to verify the model efficiency. The results show that this model could meet the requirement of sugarcane growth monitoring at a county level and may be applied to different sugarcane planting region and remote sensing data.(3)The sugarcane yield estimation model is made by the use of remote sensing vegetation indicators. The regression relationship between sugarcane growth vegetation index and yield is explored and sugarcane yield estimation model is established during critical growth period and the whole growth period. Verified results based on the model itself and statistical yield of town’s agricultural department show that the yield estimation model of the whole growth period is the best one with optimal fitting coefficient and lowest relative error. Yield estimation model of Key growth period in mid stem elongation exhibits the highest precision which linear equations model performs better than others. Yield estimation model of mature period performs is the worst with a largest relative error and the lowest fitting coefficient.