中国农业受灾与成灾面积预测研究

STUDIES ON PREDICTING MODEL SYSTEM FOR SIMULATING AGRICULTURAL AREAS AFFECTED AND INJURED BY NATURAL DISASTERS

文中运用SAS软件建立了中国农业受灾面积非线性回归预测模型及成灾面积一元回归预测模型.前者运用SAS软件的时间序列方法、回归分析方法和时延神经元网络方法,分别建立了4个农业受灾面积预测模型.在运用各模型进行模拟预测的基础上,运用组合方法建立组合预测模型,得到农业受灾面积的集成预测结果,从而克服了单个预测模型的偶然误差,提高了预测结果的精确性和可靠性.2000年农业受灾面积的预测值为5322.37×104 hm2;后者根据中国1951～1999年农业受灾、成灾面积资料,用SAS的REG过程拟合受灾面积与成灾面积的关系,预测未来成灾面积.2000年农业成灾面积预测值为2479.47×104 hm2.经检验,模型预测效果良好,根据预测结果可以采取有针对性的减灾措施,减少灾害损失.

It is reported that economic loss caused by natural disasters has been increasing in recent years in China, so some more accurate methodologies should be developed to predict and mitigate the disaster loss. Since agriculture in China is still one of the most vulnerable departments, the agricultural areas affected and injured by natural disasters are then chosen for modeling and analyzing. The nonlinear and unitary regression models are developed by using SAS software, for predicting agricultural areas affected and injured by natural disasters. As to the former, the affected area, four predicting models are established by using STEPAR, ARIMA, Time Delay Neural Network BP1 and BP2 methods. Then, four integrated models are built up for diminishing accidental errors caused by each specific model, and in turn improving prediction precision. The affected area by natural disasters is first calculated by the four predicting models, respectively. Then the modeling results are nonlinearly averaged by the four integrated models to derive the final prediction results. The simulated results are quite consistent with recorded results from 1994 to 1999. Afterwards the model system is formally used for prediction in 2000. According to the model prediction, the affected area by natural disasters in 2000 is 5322.37 × 104 hm2. As to the injured area by natural disasters, a linear regression model with SAS REG process is developed for fitting the relationship between the affected and injured areas. The simulated results agree in general with recorded results from 1994 to 1999, but are not as well as the affected area predictions, which might indicate the linear regression model structure is not sufficient, and the data sets not long enough. The injured area in 2000, derived from the modeling results, is 2479.47 × 104 hm2. The prediction results of the model system are quite consistent with the observed ones in that year, which means that the model system, with certain improvements in future, can be used for agricultural disaster prediction and as reference to make disaster reduction decisions.