气候变化对我国粮食生产的影响及产量灾损风险研究

发布时间：2018-03-18 11:07

本文选题：BP滤波　切入点：Fourier模型　出处：《南京信息工程大学》2016年硕士论文　论文类型：学位论文

【摘要】：中国耕地面积占全球耕地面积总数的8%,人口比例却达到20%,因此粮食生产是头等大事,而其中产量影响因素研究一直是学术界关注的重点。目前已有较多学者研究了气候变化对粮食产量的影响,但总体上从气候因子变化及气候异常变化对粮食产量的影响进行分析的文章较少,且分析的过程中未考虑到气候因子与产量问并非普通的线性关系。本文同时采用1961-2014年的粮食单产、气候因子以及1982-2014年的气象灾害数据,从统计学建模的角度提出分析气候变化对粮食生产影响的新方法。本文着重研究三个方面：一是先用HP滤波分离出趋势产量和气候产量,并将经济学中的BP滤波与Fourier模型相结合预测气候产量,用多项式滞后模型预测趋势产量,最终与传统的BP神经网络法和灰色模型法进行比较；二是采用偏最小二乘估计法(PLSR)解决气候因子自变量之间的共线性问题,又针对气候因子与产量间复杂的非线性关系,进一步建立两种非线性PLSR模型来拟合气候因子与粮食产量间的关系,一种是对气候因子利用三次B样条函数进行外部转换,使得变量之间的非线性变为拟线性进行建模；另一种是对普通的PLSR模型内部嵌入GRNN模型从而达到非线性效果；三是由于气候异常变化会对产量造成严重的损害,本文利用R/S分析法探讨了农业气象灾害未来发展趋势,并采用基于信息扩散的模糊数学方法对我国粮食生产进行产量灾损风险评估。本文根据三个方面的重点分析主要得出以下结论：(1)多项式滞后模型以及BP滤波的Fourier模型分别对趋势产量和气候产量进行预测,最终极差值为3.11%,方差为1.02%,误差较小,表明模型不仅能够较佳地应用于粮食单产预测,又能够直观地表明未来气候变化情景下,粮食产量的波动状况,结果精确科学。(2)基于三次B样条变换的非线性偏最小二乘法(Spline-PLSR)在拟合气候因子与气候产量关系时预测精度较高,与常用的C-D生产函数法相比,Spline-PLSR所需要素较少,操作简单,精度高；与GRNN内部嵌入的非线性PLSR拟合结果相比,该模型较稳定。(3)R/S法分析出我国未来低温冻害受灾比变化趋势将持续上升,干旱和风雹受灾比将持续下降,而洪涝仍将呈现出波动幅度较大的趋势；并在计算出相对灾损量序列的基础上,采用基于信息扩散的模糊数学方法分析出我国农业气象灾害的产量灾损风险随着概率的增大而减少,洪涝、干旱、风雹、低温冻害风险高值区分别出现在相对灾损率5%-40%、5%-60%、5%-15%和5%-10%。
[Abstract]:The area of cultivated land in China accounts for 8% of the total cultivated land area in the world, but the proportion of the population has reached 20%. Therefore, food production is a top priority. Among them, the research on the factors affecting the yield has always been the focus of academic attention. At present, many scholars have studied the impact of climate change on grain production. However, in general, there are fewer articles analyzing the effects of climate factors and abnormal climate changes on grain production. Moreover, the linear relationship between climatic factors and yield is not taken into account in the analysis. In this paper, the grain yield, climatic factors and meteorological disaster data from 1961-2014 and 1982-2014 are also used. A new method for analyzing the impact of climate change on grain production is proposed from the point of view of statistical modeling. This paper focuses on three aspects: first, the trend yield and climate yield are separated by HP filter. BP filter and Fourier model in economics are combined to predict climate output, and polynomial lag model is used to predict trend output. Finally, it is compared with the traditional BP neural network method and grey model method. The second is to solve the problem of collinearity between the independent variables of climate factors by using partial least square estimation method (PLSRs), and to solve the complex nonlinear relationship between climate factors and yields. Two kinds of nonlinear PLSR models are further established to fit the relationship between climate factors and grain yield. One is to use cubic B-spline function to transform climate factors into quasilinear models. The other is to embed the GRNN model inside the ordinary PLSR model to achieve nonlinear effect. Third, because the abnormal climate changes will cause serious damage to the yield, this paper discusses the future development trend of agrometeorological disasters by using the R / S analysis method. The fuzzy mathematics method based on information diffusion is used to evaluate the risk of grain production in China. According to the analysis of three aspects, the following conclusions: 1) polynomial lag model and BP filter Fourier are obtained in this paper. The model predicts the trend yield and climate yield respectively. The ultimate difference is 3.11, the variance is 1.02, and the error is small, which indicates that the model can not only be applied to the forecast of grain yield, but also can intuitively show the fluctuation of grain yield in the future climate change scenario. Results the nonlinear partial least square method (Spline-PLSRs) based on cubic B-spline transform has higher prediction accuracy in fitting the relationship between climate factors and climate yield. Compared with C-D production function method, Spline-PLSR requires less elements and its operation is simple. Compared with the fitting results of nonlinear PLSR embedded in GRNN, the model is more stable. The result shows that the variation trend of the ratio of low temperature and cold damage to disaster will continue to rise, and the ratio of drought and hail will continue to decrease in the future in China. However, flood and waterlogging will still show a trend of large fluctuation, and on the basis of calculating the sequence of relative disaster losses, The fuzzy mathematics method based on information diffusion was used to analyze the risk of agricultural meteorological disasters in China. The risk of agricultural meteorological disasters decreased with the increase of probability. The high risk areas of flood, drought, wind and hail, low temperature and freezing damage appeared in the relative damage rate of 5-405-605-5-15% and 5-10%, respectively.
【学位授予单位】：南京信息工程大学
【学位级别】：硕士
【学位授予年份】：2016
【分类号】：S42;F326.11

【相似文献】