| 引用本文: | 曹秀佳,谷 健,马宁宁,等.基于粒子群优化小波神经网络模型的春玉米生育阶段干旱预测[J].灌溉排水学报,2021,(3):125-133. |
| CAO Xiujia,GU Jian,MA Ningning,et al.基于粒子群优化小波神经网络模型的春玉米生育阶段干旱预测[J].灌溉排水学报,2021,(3):125-133. |
|
| 摘要: |
| 【目的】为更好地开展区域性作物生长季气候干旱预测，指导春玉米高效节水补灌生产。【方法】采用皮尔逊相关系数方法选取了与干旱指数最相关的因子，利用阜新市阜蒙县1965—2019年逐日气象数据，探索建立了粒子群算法优化的小波神经网络模型（PSO-WNN），将春玉米不同生育阶段的水分亏缺指数结果进行对比验证模型精度，并利用模型模拟预测未来5 a干旱发生情况。【结果】通过模型验证，春玉米5个生育阶段（播种—出苗阶段、出苗—拔节阶段、拔节—抽雄阶段、抽雄—乳熟阶段、乳熟—成熟阶段）的均方根误差（RMSE）分别为0.041 9、0.017 4、0.048 1、0.029 7、0.042 1，决定系数R2分别为0.840 2、0.985 3、0.899 0、0.957 5、0.917 7，且预测结果与实际干旱等级相符。【结论】文中构建的模型适用于阜新地区春玉米干旱预测，未来5 a该地区春玉米在播种—出苗阶段可能无旱或轻旱，出苗-拔节阶段可能发生中旱甚至特旱，生育后期干旱程度逐渐减弱，拔节—抽雄和抽雄—乳熟两个阶段出现轻旱概率较高，乳熟—成熟阶段出现干旱的概率较低，程度较小，表明未来几年该地区春玉米生产应该更多关注出苗—拔节阶段的旱情。 |
| 关键词: 干旱预测；小波神经网络；春玉米；作物水分亏缺指数 |
| DOI：10.13522/j.cnki.ggps.2020531 |
| 分类号: |
| 基金项目: |
|
| Predicting Droughts in Growth Season of Spring Maize with the Wavelet Neural Networks using Particle Swarm Optimization Training |
|
CAO Xiujia, GU Jian, MA Ningning, LIU Yongqi, WANG Zihao, YIN Guanghua
|
|
1. Institute of Applied Ecology, Chinese Academy of Sciences, Shenyang 110016, China; 2. University of Chinese Academy of Sciences, Beijing 100049, China; 3. Tillage and Cultivation Research Institute, Liaoning Academy of Agricultural Science,Shenyang 110116, China; 4. Shenyang Agricultural University, Shenyang 110016, China
|
| Abstract: |
| 【Background】 Drought is the most common natural hazard occurring more frequently over the recent years, due to climate change, and could have calamitous impact on agriculture. In northeast China, damage caused by long-lasting and frequent droughts is especially severe and has been in increase over the past years. Forecasting drought occurrence in plant growth season is hence important to safeguard agricultural production.【Objective】This paper is to present and test a model for drought forecast in efforts to offer a guidance to water-saving irrigation for spring maize in the northeast of China.【Method】The Pearson correlation coefficient method was used to select the factors that impact the drought index most, based on daily meteorological data measured from 1965-2019 at Fumeng county in Fuxin City of Liaoning province. We then forecasted the crop water deficit index at different growth stages with the wavelet neural network model using the particle swarm optimization.【Result】The root mean square error (RMSE) of the drought forecasted by the model at sowing-seedling, seedling-joining, jointing-tasseling, tasseling-milking, and milking-maturity stages was 0.041 9, 0.017 4, 0.048 1, 0.029 7 and 0.042 1 respectively, and their associated determination coefficient was 0.840 2, 0.985 3, 0.899 0, 0.957 5 and 0.917 7 respectively. These were consistent with the ground-truth data, proving that the model is suitable for drought forecast in these areas. There were no or only mild droughts in the sowing-seedling stage, but moderate or even extreme drought may occur during the seedling-jointing stage. As the crop grew, the occurrence of severe drought became increasingly unlikely especially in the milking-maturity stage, although mild drought might still occur in the jointing-tasseling and tasseling-milking stages.【Conclusion】The spring maize in the studied area is most prone to drought during the seedling-jointing stage, and our results are of significance to precision irrigation, mitigating detrimental impact of droughts on agricultural production. |
| Key words: drought forecast wavelet neural network; spring maize; crop water deficit index |
