| Maize is one of the major grain crops in China.Obtaining maize yield accurately and efficiently is of great significance for the country in making agricultural decisions and optimizing the development of the food industry.Traditional method of yield estimation requires a large amount of manpower and material resources.Remote sensing method for yield estimation not only obtains data in a timely and efficient manner,but also has a low cost.Due to the characteristics of its wide monitoring range,it makes possible to estimate crop production in large areas.At present,using remote sensing technology to estimate crop yield has become the mainstream method.With the development of modern science and technology,there are more and more types of satellites,and remote sensing images collected by satellites with different spatial and temporal resolutions,different qualities,and different functions are meeting more and more demands of the society.This paper mainly combines the two satellite image data of HJ-1A/B and Landsat8 to study the remote sensing estimation method of maize yield in the study area.This paper takes the Jiutai District,Dehui City and Nong’an County of Changchun region of Jilin Province as the study area.Using the vegetation index data obtained from remote sensing images to estimate the yield through the method of statistical yield estimation and the radiation use efficiency model,and evaluating the accuracy of the model.The principle of statistical yield estimation method is clear,the method is simple and the required data is easy to obtain.The radiation use efficiency model has the scientific principle as the theoretical basis,the process is simple,the parameters are few and easy to obtain.It can meet the large-scale yield estimation.In the study of statistical methods for estimating yield,the normalized vegetation index NDVI,enhanced vegetation index EVI,ratio vegetation index RVI,and greenness G were selected as analysis parameters to establish curve statistics,multiple stepwise regression,partial least squares regression,and neural Network maize yield estimation model.The results show that the yield estimation model established by neural network is stable and has the highest precision.R~2 of the model is 0.997.The mean absolute percentage error MAPE measured at the verification point is 6.94%,and the relative root mean square error RRMSE is 7.36%.It can meet the demand for maize yield estimation in the study area.In order to study the applicability of the statistical yield estimation model on the time scale,the 2016 remote sensing parameters are used as input data,and the 2017neural network yield estimation model is used to estimate 2016 maize yield.The results show that the MAPE of the actual yield and the estimated yield of the verification point is 11.71%,and the RRMSE is 14.05%,which indicates that the yield estimation model has certain applicability on the time scale.When using the radiation use efficiency model to estimate the maize yield in the study area,an estimated model is used to predict the SOL value and using the IDW spatial interpolation method to obtain SOL spatial distribution data.Field experiments are conducted to analyze the daily variation trend of the fraction of absorbed photosynthetically active radiation FPAR,and the accuracy of the FPAR estimation results in the model was evaluated to verify its reliability.The maize yield in the study area is obtained through model estimation,and the estimation accuracy was evaluated using verification point data.The results show that the relative errors between the measured and estimated yields are mainly in the range of±20%,the maximum error is28.75%,the minimum error is-0.73%,the MAPE is 11.93%,and the RRMSE is15.12%,indicating that the model has certain reliability for estimating maize yield in the study area. |
