基于遥感技术和时间序列分析法预测皖北疟疾的发病率

目的 探讨皖北疟疾的发病率与地表温度（land surface temperature，LST）、归一化植被指数（normalized difference vegetation index，NDVI）的关联性，评价用LST、NDVI对疟疾发病率自回归移动平均模型（autoregressive integrated moving average model，ARIMA）预测结果进行校正的效果。方法 以皖北五县为研究现场，收集各县2004-2011年的疟疾疫情数据及LST、NDVI等遥感图像资料，提取、合成遥感相关指标；运用SPSS 17.0软件进行统计学处理。结果 ARIMA模型对2010年各月份的预测结果较报告发病率高（平均误差=0.721/10万）。多因素分析结果显示，当地的疟疾发病率与近三个月的平均LST（lst_{_012}，β=0.295）及之前两个月的平均NDVI（ndvi_{_12}，β=0.280）有关联（P<0.001）；将二者作为校正因子（相对贡献为2:1时）对2010年的预测结果进行校正，平均误差缩小为0.018/10万。以2004-2010年的发病率数据再次拟合并筛选ARIMA模型，并以2011年的疟疾报告发病数据为参照，再次评价lst_{_012} 与 ndvi_{_12}对模型预测结果的校正效果；发现校正后的预测误差（＜0.001/10万）低于校正前的误差（0.293/10万）。结论 ARIMA模型能较好地用于该地疟疾发病率的拟合与预测，环境遥感替代指标LST、NDVI可在一定程度上改善ARIMA模型的预测效果。

Study on the prediction of malaria incidence in the northern Anhui Province based on remote sensing techniques and time series analysis

Objective To explore the relationship between malaria incidence and land surface temperature (LST) and normalized difference vegetation index (NDVI), assess the adjusted effect on autoregressive integrated moving average model (ARIMA) prediction using LST and NDVI. Methods Five counties in northern Anhui province were selected in this study. We collected the reported malaria epidemic data, LST and NDVI remote sensing images from 2004 to 2011. Then data extraction and synthesis from MODIS images were performed. SPSS 17.0 software was used for statistical analysis. Results The incidence of malaria in 2010 predicted by ARIMA models based on malaria data from 2004 to 2009 was higher than the reported incidence with an average error of 0.721/100 000. The results of multiple regression analysis showed a significant association (P<0.001) between malaria incidence and the nearly three-month average LST (lst_{_012}, β=0.295) and the average NDVI of last month and before the last month (ndvi_{_12}, β=0.280). After adjusting predictive results of ARIMA by Lst_{_012} and ndvi_{_12} (relative ratio was 2:1), the average error decreased to 0.018/100 000. The correction effect of lst_{_012} and ndvi_{_12} on the predicted malaria incidence by ARIMA model based on malaria data from 2004 to 2010 was evaluated again based on reported malaria incidence in 2011. The results indicated that the prediction error (＜0.001/100 000) after adjustment was significantly lower than that before correction (0.293/100 000). Conclusions ARIMA model could be applied to the incidence of malaria fitting and prediction. The predicted results would be better when the predicted results were adjusted by environmental remote sensing alternate index.