设为首页
加入收藏
Application of Deep Learning Method to Drought Prediction
-
摘要: 使用标准化降水蒸散指数作为农业干旱的监测指标, 以站点气象要素和大尺度环流要素为驱动变量, 建立干旱预测模型, 分析评价传统的整合移动平均自回归(ARIMA)时间序列模型以及不同深度神经网络模型(DNN)的预测效果。结果表明: DNN模型的总体预测能力优于ARIMA模型; 同基于长短期记忆网络(LSTM)提出的传统LSTM预测模型(TLSTM)相比, 改进的LSTM模型(ILSTM)通过预处理全连接层对预测因子进行非线性映射, 能够自动剔除无效信息, 提取高层次综合特征, 可使预测序列和观测序列的相关系数提升0.04~0.25, 均方根误差降低0.07~0.32, 误差绝对值的平均降低0.06~0.27;卷积神经网络(CNN)可提取影响干旱变化的大尺度环流信息, 其与ILSTM的组合深度网络模型(CLSTM)可进一步使相关系数提升0.03~0.44, 均方根误差降低0.09~0.33, 误差绝对值的平均降低0.05~0.26。CLSTM模型可应用于短期区域性干旱变化气候预测。Abstract: Drought has brought great potential threat to agriculture, ecology, economy, society and available water resources of China, while accurate drought prediction can help risk management and development of early warning system, and it can reduce the destructive impact of drought. Among many prediction methods, data-driven model is a suitable tool with small data demand and fast development speed. With the development of machine learning, especially deep neural network (DNN), deep learning method shows great ability in drought prediction, and is reported to outperform traditional time series model (e.g. integrated moving average autoregressive model, ARIMA). However, its use needs to be widely estimated, further developed and adjusted for geoscience analysis. The standardized precipitation evapotranspiration index (SPEI) is reported to meet the needs of agricultural drought monitoring and early warning under the background of climate warming. SPEI at 1-, 3- and 6-month time scale are selected as the quantitative description of agricultural drought, and DNN model driven by meteorological and circulation variable is presented to explore the ability of SPEI prediction at the lead time of 1-3 months. The traditional long short-term memory neural network (TLSTM) has been used in drought prediction, which is limited by the quality of prediction factors and noise. Therefore, an improved TLSTM model (ILSTM) is proposed. With highlight of large-scale climate characteristics, a convolution neural network (CNN) module is combined with the ILSMT model. This newly-developed model (CLSTM) can extract circulation information that contributes to the regional drought change, as well as other outputs of prediction model. Evaluation of the drought prediction capabilities in different models is based on the Pearson correlation coefficient, the root mean square error, and the mean absolute error. Results indicate that overall prediction ability of DNN models outperforms the ARIMA model. And the comparative evaluation results among DNN models show that the architecture of the model has an important impact on the prediction performance. The ILSTM model can extract comprehensive information that contributes to future drought change by nonlinear coding of input variables through the full connected layer. When the correlation coefficient can be raised by 0.04-0.25, the root mean square error can be reduced by 0.07-0.32 and the mean absolute error can be reduced by 0.06-0.27 at the validation stage with different lead time comparing with the TLSTM model. Taking advantage of the circulation information as extra inputs to the ILSTM model, the CLSTM model outperform the ILSTM model, when the correlation coefficient can be raised by 0.03-0.44, the root mean square error can be reduced by 0.09-0.33 and the mean absolute error can be reduced by 0.05-0.26. Both results show that deep learning method has great ability in short-term regional climatic drought prediction.
-
图 2 ILSTM模型提前1个月预测的SPEI1，SPEI3和SPEI6与观测的相关系数
Fig. 2 Pearson correlation coefficient between prediction and observation for SPEI1, SPEI3 and SPEI6 with 1-month lead time based on ILSTM model
图 3 CLSTM模型提前1~3个月预测的SPEI1与观测的相关系数
Fig. 3 Pearson correlation coefficient between prediction and observation for SPEI1 with 1-month to 3-month lead time based on CLSTM model
表 1 3组REOF分析前8个模态的方差贡献率(单位：%)
Table 1 Variance contribution rate of the first 8 modes of REOF analysis among 3 groups (unit: %)
空间型 方差贡献率 SPEI1 SPEI3 SPEI6 A1 9.5 10.2 10.2 A2 9.1 10.4 11.5 A3 8.0 7.0 6.6 A4 6.3 6.2 6.4 A5 5.7 5.9 5.6 A6 6.2 5.7 5.0 A7 9.8 7.7 7.4 A8 8.9 10.9 11.1 
下载: 导出CSV
表 2 测试阶段不同模型在1个月提前期预测的性能评估
Table 2 Evaluation of 1-month lead time prediction performance of different models in test period
时间尺度 评价指标 ILSTM TLSTM ARIMA SPEI1 相关系数 0.59 ± 0.03 0.36 ± 0.06 0.22 ± 0.07 均方根误差 0.89 ± 0.04 1.05 ± 0.07 1.18 ± 0.09 误差绝对值的平均 0.66 ±0.03 0.86 ± 0.05 1.03 ± 0.08 SPEI3 相关系数 0.88 ± 0.01 0.83 ± 0.02 0.72 ± 0.01 均方根误差 0.50 ± 0.02 0.60 ± 0.05 0.75 ± 0.06 误差绝对值的平均 0.36 ± 0.02 0.46 ± 0.03 0.59 ± 0.05 SPEI6 相关系数 0.93 ± 0.01 0.89 ± 0.02 0.82 ± 0.03 均方根误差 0.39 ± 0.03 0.46 ± 0.05 0.59 ± 0.07 误差绝对值的平均 0.27 ± 0.01 0.33 ± 0.03 0.44 ± 0.05 注：表中数字为平均值±1倍标准差，数值均达到0.05显著性水平，下同。
下载: 导出CSV
表 3 测试阶段不同模型在2个月提前期预测的性能评估
Table 3 Evaluation of 2-month lead time prediction performance of different models in test period
时间尺度 评价指标 ILSTM TLSTM ARIMA SPEI1 相关系数 0.42 ± 0.01 0.27 ± 0.09 均方根误差 0.99 ± 0.05 1.10 ± 0.08 误差绝对值的平均 0.78 ± 0.06 0.90 ± 0.06 SPEI3 相关系数 0.66 ± 0.04 0.48 ± 0.06 0.45 ± 0.01 均方根误差 0.81 ± 0.05 1.08 ± 0.08 1.10 ± 0.08 误差绝对值的平均 0.63 ± 0.04 0.86 ± 0.07 0.89 ± 0.05 SPEI6 相关系数 0.79 ± 0.02 0.66 ± 0.04 0.66 ± 0.06 均方根误差 0.63 ± 0.05 0.81 ± 0.09 0.84 ± 0.09 误差绝对值的平均 0.47 ± 0.04 0.63 ± 0.07 0.62 ± 0.06
下载: 导出CSV
表 4 测试阶段不同模型在3个月提前期预测的性能评估
Table 4 Evaluation of 3-month lead time prediction performance of different models in test period
时间尺度 评价指标 ILSTM TLSTM ARIMA SPEI1 相关系数 0.26 ± 0.06 0.17 ± 0.05 均方根误差 1.09 ± 0.07 1.16 ± 0.09 误差绝对值的平均 0.87 ± 0.05 0.95 ± 0.07 SPEI3 相关系数 0.40 ± 0.06 0.27 ± 0.05 均方根误差 1.02 ± 0.06 1.26 ± 0.11 误差绝对值的平均 0.80 ± 0.04 1.00 ± 0.91 SPEI6 相关系数 0.69 ± 0.03 0.44 ± 0.06 0.52 ± 0.07 均方根误差 0.75 ± 0.07 1.07 ± 0.11 0.90 ± 0.09 误差绝对值的平均 0.57 ± 0.04 0.84 ± 0.08 0.71 ± 0.07
下载: 导出CSV
表 5 测试阶段CLSTM模型的预测性能评估
Table 5 Evaluation of prediction performance of the CLSTM model in test period
时间尺度 评价指标 1个月提前期 2个月提前期 3个月提前期 SPEI1 相关系数 0.82 ± 0.02 0.78 ± 0.03 0.70 ± 0.04 均方根误差 0.60 ± 0.04 0.67 ± 0.05 0.76 ± 0.06 误差绝对值的平均 0.47 ± 0.03 0.54 ± 0.04 0.61 ± 0.04 SPEI3 相关系数 0.94 ± 0.01 0.81 ± 0.04 0.73 ± 0.04 均方根误差 0.37 ± 0.03 0.61 ± 0.05 0.73 ± 0.05 误差绝对值的平均 0.28 ± 0.02 0.49 ± 0.04 0.58 ± 0.05 SPEI6 相关系数 0.96 ± 0.01 0.87 ± 0.01 0.77 ± 0.02 均方根误差 0.30 ± 0.02 0.51 ± 0.05 0.65 ± 0.04 误差绝对值的平均 0.22 ± 0.01 0.39 ± 0.03 0.51 ± 0.03
下载: 导出CSV
-
[1] Dai A. Drought under global warming: A review. WIREs Clim Change, 2011, 2(1): 45-65. doi: 10.1002/wcc.81 [2] Song Y, Tian J, Linderholm H W, et al. The contributions of climate change and production area expansion to drought risk for maize in China over the last four decades. Int J Climatol, 2021, 41(Suppl Ⅰ): 2851-2862. [3] Yao N, Li Y, Lei T, et al. Drought evolution, severity and trends in mainland China over 1961-2013. Sci Total Environ, 2018, 616/617: 73-89. doi: 10.1016/j.scitotenv.2017.10.327 [4] Zhang L, Zhou T. Drought over East Asia: A review. J Climate, 2015, 28(8): 3375-3399. doi: 10.1175/JCLI-D-14-00259.1 [5] Yao N, Li L, Feng P, et al. Projections of drought characteristics in China based on a standardized precipitation and evapotranspiration index and multiple GCMs. Sci Total Environ, 2020, 704: 135245. doi: 10.1016/j.scitotenv.2019.135245 [6] Zhao T, Dai A. Uncertainties in historical changes and future projections of drought. Part Ⅱ: Model-simulated historical and future drought changes. Climatic Change, 2017, 144(3): 535-548. doi: 10.1007/s10584-016-1742-x [7] 郭建平. 农业气象灾害监测预测技术研究进展. 应用气象学报, 2016, 27(5): 620-630. doi: 10.11898/1001-7313.20160510Guo J P. Research progress on agricultural meteorological disaster monitoring and forecasting. J Appl Meteor Sci, 2016, 27(5): 620-630. doi: 10.11898/1001-7313.20160510 [8] Ding Y B, Gong X L, Xing Z X, et al. Attribution of meteorological, hydrological and agricultural drought propagation in different climatic regions of China. Agric Water Manag, 2021, 255(1): 106996. [9] Vicente-Serrano S M, Beguería S, López-Moreno J I. A multi-scalar drought index sensitive to global warming: The standardized precipitation evapotranspiration index. J Climate, 2010, 23(7): 1696-1718. doi: 10.1175/2009JCLI2909.1 [10] Hao Z, Singh V P. Drought characterization from a multivariate perspective: A review. J Hydrol, 2015, 527: 668-678. doi: 10.1016/j.jhydrol.2015.05.031 [11] Li X, He B, Quan X, et al. Use of the standardized precipitation evapotranspiration index(SPEI) to characterize the drying trend in Southwest China from 1982-2012. Remote Sens, 2015, 7(8): 10917-10937. doi: 10.3390/rs70810917 [12] Tian Y, Xu Y P, Wang G Q. Agricultural drought prediction using climate indices based on support vector regression in Xiangjiang River Basin. Sci Total Environ, 2018, 622/623: 710-720. doi: 10.1016/j.scitotenv.2017.12.025 [13] Wang H, Rogers J C, Munroe D K. Commonly used drought indices as indicators of soil moisture in China. J Hydrol, 2015, 16(3): 1397-1408. [14] Agana N A, Homaifar A. A Deep Learning Based Approach for Long-term Drought Prediction//Proc Southeast Con 2017, 2017: 1-8. [15] 沈学顺, 苏勇, 胡江林, 等. GRAPES_GFS全球中期预报系统的研发和业务化. 应用气象学报, 2017, 28(1): 1-10. doi: 10.11898/1001-7313.20170101Shen X S, Su Y, Hu J L, et al. Development and operation transformation of GRAPES global middle-range forecast system. J Appl Meteor Sci, 2017, 28(1): 1-10. doi: 10.11898/1001-7313.20170101 [16] Hao Z, Singh V P, Xia Y. Seasonal drought prediction: Advances, challenges, and future prospects. Rev Geophys, 2018, 56(1): 108-141. doi: 10.1002/2016RG000549 [17] Wang D, Borthwick A G, He H, et al. A hybrid wavelet de-noising and rank-set pair analysis approach for forecasting hydro-meteorological time series. Environ Res, 2018, 160(1): 269-281. [18] Xu L, Chen N C, Zhang X. A comparison of large-scale climate signals and the North American Multi-Model Ensemble(NMME) for drought prediction in China. J Hydrol, 2018, 557: 378-390. doi: 10.1016/j.jhydrol.2017.12.044 [19] Deo R C, Şahin M. Application of the artificial neural network model for prediction of monthly standardized precipitation and evapotranspiration index using hydrometeorological parameters and climate indices in eastern Australia. Atmos Res, 2015, 161/162(1): 65-81. doi: 10.1016/j.atmosres.2015.03.018 [20] Abbot J, Marohasy J. Input selection and optimisation for monthly rainfall forecasting in Queensland, Australia, using artificial neural networks. Atmos Res, 2014, 138(1): 166-178. [21] Moreira E E, Coelho C A, Paulo A A, et al. SPI-based drought category prediction using loglinear models. J Hydrol, 2018, 354(1): 116-130. [22] Han P, Wang P X, Zhang S Y, et al. Drought forecasting based on the remote sensing data using ARIMA models. Math Comput Modell, 2010, 51(11): 1398-1403. [23] Xu L, Chen N, Zhang X, et al. An evaluation of statistical, NMME and hybrid models for drought prediction in China. J Hydrol, 2018, 566(1): 235-249. [24] Dikshit A, Pradhan D, Huete A. An improved SPEI drought forecasting approach using the long short-term memory neural network. J Environ Manage, 2021, 283: 111979. doi: 10.1016/j.jenvman.2021.111979 [25] Mishra A K, Desai V R. Drought forecasting using feed-forward recursive neural network. Ecol Modell, 2006, 198(1/2): 127-138. [26] Ochoa-Rivera J C. Prospecting droughts with stochastic artificial neural networks. J Hydrol, 2008, 352(1/2): 174-180. [27] 闵晶晶, 孙景荣, 刘还珠, 等. 一种改进的BP算法及在降水预报中的应用. 应用气象学报, 2010, 21(1): 55-62. doi: 10.3969/j.issn.1001-7313.2010.01.007Min J J, Sun J R, Liu H Z, et al. An improved BP algorithm and its application to precipitation forecast. J Appl Meteor Sci, 2010, 21(1): 55-62. doi: 10.3969/j.issn.1001-7313.2010.01.007 [28] Mekanik F, Imteaz M A, Gato-Trinidad S, et al. Multiple regression and Artificial Neural Network for long-term rainfall forecasting using large scale climate modes. J Hydrol, 2013, 503(30): 11-21. [29] Hossain M, Rekabdar B, Louis S J, et al. Forecasting the Weather of Nevada: A Deep Learning Approach//Proc 2015 Int Joint Conf on Neural Netw, 2015: 1-6. [30] Han X, Wei Z, Zhang B, et al. Crop evapotranspiration prediction by considering dynamic change of crop coefficient and the precipitation effect in back-propagation neural network model. J Hydrol, 2021, 596: 126104. doi: 10.1016/j.jhydrol.2021.126104 [31] de Oliveira e Lucas P, Alves M A, de Lima e Silva P C, et al. Reference evapotranspiration time series forecasting with ensemble of convolutional neural networks. Comput Electron Agric, 2020, 177: 105700. doi: 10.1016/j.compag.2020.105700 [32] Lecun Y, Bengio Y, Hinton G E. Deep learning. Nature, 2015, 521(7553): 436-444. doi: 10.1038/nature14539 [33] Reichstein M, Camps-Valls G, Stevens B, et al. Deep learning and process understanding for data-driven Earth system science. Nature, 2019, 566(7743): 195-204. doi: 10.1038/s41586-019-0912-1 [34] 李颖, 陈怀亮. 机器学习技术在现代农业气象中的应用. 应用气象学报, 2020, 31(3): 257-266. doi: 10.11898/1001-7313.20200301Li Y, Chen H L. Review of machine learning approaches for modern agrometeorology. J Appl Meteor Sci, 2021, 31(3): 257-266. doi: 10.11898/1001-7313.20200301 [35] 孙健, 曹卓, 李恒, 等. 人工智能技术在数值天气预报中的应用. 应用气象学报, 2021, 32(1): 1-11. doi: 10.11898/1001-7313.20210101Sun J, Cao Z, Li H, et al. Application of artificial intelligence technology to numerical weather prediction. J Appl Meteor Sci, 2021, 32(1): 1-11. doi: 10.11898/1001-7313.20210101 [36] 韩丰, 龙明盛, 李月安, 等. 循环神经网络在雷达临近预报中的应用. 应用气象学报, 2019, 30(1): 61-69. doi: 10.11898/1001-7313.20190106Han F, Long M S, Li Y A, et al. The application of recurrent neural network to nowcasting. J Appl Meteor Sci, 2019, 30(1): 61-69. doi: 10.11898/1001-7313.20190106 [37] Le J A, El-Askary H M, Allali M, et al. Application of recurrent neural networks for drought projections in California. Atmos Res, 2017, 188(15): 100-106. [38] Poornima S, Pushpalatha M. Drought prediction based on SPI and SPEI with varying timescales using LSTM recurrent neural network. Soft Comput, 2019, 23(6): 8399-8412. [39] Shen R, Huang A, Li B, et al. Construction of a drought monitoring model using deep learning based on multi-source remote sensing data. Int J Appl Earth Obs Geoinf, 2019, 79(1): 48-57. [40] 蔡金圻, 谭桂容, 牛若芸. 基于迁移CNN的江淮持续性强降水环流分型. 应用气象学报, 2021, 32(2): 233-244. doi: 10.11898/1001-7313.20210208Cai J Q, Tan G R, Niu R Y. Circulation pattern classification of persistent heavy rainfall in Jianghuai Region based on the transfer learning CNN model. J Appl Meteor Sci, 2021, 32(2): 233-244. doi: 10.11898/1001-7313.20210208 [41] Ham Y, Kim J, Luo J. Deep learning for multi-year ENSO forecasts. Nature, 2019, 573(7775): 568-572. doi: 10.1038/s41586-019-1559-7 [42] Wang Z, Li J, Lai C, et al. Does drought in China show a significant decreasing trend from 1961 to 2009?. Sci Total Environ, 2017, 579(1): 314-324. [43] 晏红明, 王灵. 西北太平洋副高东西变动与西南地区降水的关系. 应用气象学报, 2019, 30(3): 360-375. doi: 10.11898/1001-7313.20190309Yan H M, Wang L. The relationship between east-west movement of subtropical high over Northwestern Pacific and precipitation in Southwestern China. J Appl Meteor Sci, 2019, 30(3): 360-375. doi: 10.11898/1001-7313.20190309 [44] 晏红明, 王灵, 李蕊. 1—3月欧亚大陆热力变化及其与中国降水的关系. 应用气象学报, 2016, 27(2): 209-219. doi: 10.11898/1001-7313.20160209Yan H M, Wang L, Li R. Thermal characteristics over Eurasia in January-March and its relationship with precipitation of China. J Appl Meteor Sci, 2016, 27(2): 209-219. doi: 10.11898/1001-7313.20160209 [45] Beguería S, Vicente-Serrano S M, Reig F, et al. Standardized precipitation evapotranspiration index(SPEI) revisited: Parameter fitting, evapotranspiration models, tools, datasets and drought monitoring. Int J Climatol, 2014, 34(10): 3001-3023. doi: 10.1002/joc.3887 [46] Simonyan K, Zisserman A. Very Deep Convolutional Networks for Large-scale Image Recognition//Proc 3rd Int Conf on Learn Represent, 2015: 1-14. [47] Zeiler M D, Fergus R. Visualizing and Understanding Convolutional Neural Networks//Proc 13th Europ Conf on Comp Visi, 2014: 818-833. [48] Zhou B, Khosla A, Lapedriza A, et al. Learning Deep Features for Discriminative Localization//Proc 2016 IEEE Conf on Comp Visi and Pattern Recognit, 2016: 2921-2929. [49] He K, Zhang X, Ren S, et al. Deep Residual Learning for Image Recognition//Proc 2016 IEEE Conf on Comp Visi and Pattern Recognit, 2016: 770-778. [50] Yu Z, Chen G, Dong Y, et al. Highway Long Short-term Memory RNNS for Distant Speech Recognition//Proc 2016 IEEE Int Conf on Acoustics, Speech and Signal Process, 2016: 5755-5759. [51] Vaswani A, Shazeer N, Parmar N, et al. Attention Is All You Need//Proc 31st Int Conf on Neural Inf Process Syst, 2017: 6000-6010. -
图(3) / 表(5)
计量
- 摘要浏览量: 2624
- HTML全文浏览量: 499
- PDF下载量: 397
- 被引次数: 0
