一种基于趋势单产和遥感修正模型的混合估产模型

【目的】在分析国内外农作物估产方法的相关研究进展基础上,将传统统计估产方法和遥感估产方法相结合,提出一种新的混合估产模型。【方法】该模型由趋势单产、遥感修正单产和随机误差项三部分组成,其中趋势单产利用历史长时间序列的单产统计数据,通过多项式回归的方法结合ARIMA模型修正得到,遥感修正单产利用3个作物关键生育期NDVI和实测单产多元回归得到。为验证所提出估产方法的可行性和精度,利用2015年冬小麦关键生育期的三景环境卫星遥感影像和冬小麦实测地块单产数据以及近30年(1985—2014年)北京市各区县的冬小麦单产数据,对2015年的北京市的冬小麦单产进行估算,与真实值(2015年单产统计数据)对比。【结果】混合估产模型对北京市的冬小麦单产预测精度达到98.7%,各区县估产精度均超过90%,除房山(90.3%)外,各县单产预测相对精度均超过95%;传统趋势单产模型对北京市的冬小麦单产预测精度达到94.75%,但在区县尺度上,传统估产模型预测精度较低,对房山区的估产精度不足80%;引入ARIMA模型可以提高传统趋势单产模型的精度。修正后的趋势单产模型冬小麦单产预测精度平均提高了1.59%。本文建立的遥感修正模型,利用三景遥感影像修正结果最优,此方法使冬小麦估产精度整体提升3.55%,尤其是房山、平谷等区县,精度明显提升。【结论】该模型在市级尺度和县级尺度上预测冬小麦单产均取得较高精度,充分考虑冬小麦时间尺度和空间尺度上的变化,对农作物估产有一定的指导意义。

A Hybrid Yield Estimation Model Based on the Trend Yield Model and Remote Sensing Correction Yield Model

【Objective】 This paper analyzed the advantages and disadvantages of current crop yield estimation methods and proposed a novel hybrid yield estimation model which combines statistical yield estimation and yield estimation methods. 【Method】The model consists of three parts, trend yield estimate (Yt), remote sensing correction yield (Ys) and random error. The trend yield estimation was firstly calculated by using the polynomial regression method based on a long time series data of historical yield and then corrected by ARIMA model, which was set up by using the bias between the trend yield estimates and the historical yields. After that, a multiple linear regression model was set up to further reduce the estimation errors by using the bias between the trend yield estimates (Yt) and the reference yields as dependent variable and NDVI in critical growth period of crop as independent variables. In order to verify the feasibility and accuracy of the new hybrid estimation model, this paper estimated the yield of winter wheat in Beijing in 2015 based on three HJ Imagery obtained in winter wheat growing season, winter wheat yield of 30 sampling fields in 2015, and a nearly 30 years time series data of winter wheat yield (1985-2014) of Beijing. The estimation results from the hybrid yield estimation model was then compared with the true yield (2015 statistic winter wheat yield).【Result】The accuracy of winter wheat yield by using novel hybrid yield estimation model was 98.7% at city level and above 90% at country level. Except Fangshan(90.3%), the relative accuracy of yield estimation at the other countries was above 95%. The accuracy of winter wheat yield by using traditional trend yield model in Beijing was 94.75%, but the accuracy by using traditional trend yield model at country level was low, especially was lower 80% in Fangshan. ARIMA model was used for improving the accuracy of the traditional trend yield model. The accuracy of winter wheat yield improved in average by introducing the ARIMA model. For the remote sensing correction model established in this paper, using three remote sensing images for improving the accuracy was better, and this method improved the accuracy of winter wheat yield by 3.55%, especially the accuracy had a significant ascension in Fangshan and Pinggu.【Conclusion】The accuracy of winter wheat yield by using the novel hybrid estimation model is good at city level and county level. The model considers the change of time and spatial and can be used in crop yield estimation.