Study of soil moisture prediction method based on soil temperature and moisture persistence

doi:10.11755/j.issn.1006-7639(2023)-05-0783

Abstract

Abstract:

Soil temperature and moisture are the important parameters in land surface process, and they are also important physical parameters in boundary conditions of atmospheric numerical model. This paper tried to obtain spatial-temporal evolution of soil moisture of the model through the machine learning method according to the memory characteristics of soil moisture. Considering the influence of soil temperature on soil moisture, the soil temperature and moisture of ERA5 reanalysis at depths of 0-7, 7-28, 28-100, 100-289 cm are used as predictors to predict changes of soil moisture on a monthly and seasonal scale based on convolutional neural networks (CNN). The results show that the method proposed in this paper is reliable and can effectively predict soil moisture 6 months in advance. The mean bias of predicted soil moisture in the shallow layer (0-28 cm) and deep layer (28-289 cm) is less than 0.05 and 0.02 m³·m^-3, respectively. In the humid area, the mean bias is basically within 0.03 m³·m^-3, showing a good effect.The prediction method and results presented in this paper can be used for both soil drought prediction and the initial and boundary conditions for numerical models.

Key words: convolutional neural network, soil temperature, soil moisture, influencing factors, prediction

摘要：

土壤温、湿度是陆面过程的重要参数，也是大气数值模式下边界条件的重要物理参量。由于土壤湿度的观测站点较少，土壤温湿度的空间资料较少，另外，土壤温湿度作为干旱预测的主要内容，需要知道未来时刻的土壤温湿度变化。因此，如何获得未来时刻土壤温湿度的时空变化具有重要意义。本文根据土壤湿度的记忆性特点，通过机器学习方法试图获得模式中土壤湿度的时空变化。采用卷积神经网络算法（Convolutional Neural Networks，CNN），考虑土壤温度对土壤湿度的影响，选取ERA5 0~7、7~28、28~100、100~289 cm深度层土壤温、湿度作为预测因子，对月、季尺度上土壤湿度变化进行预测。结果表明，本方法能提前6个月对土壤湿度进行可靠有效地预测；预测的浅层（0~28 cm）与深层（28~289 cm）土壤湿度平均偏差分别小于0.05、0.02 m³·m^-3；在湿润区，平均偏差基本在0.03 m³·m^-3以内，表现出较好的效果。本文的预测方法和结果，既可用于土壤干旱的预测，也可作为数值模式初边界场的形成。

关键词: 卷积神经网络, 土壤温度, 土壤湿度, 影响因素, 预测

CLC Number:

P426.1

WEI Sentao, WANG Chenghai, ZHANG Feimin, YANG Kai. Study of soil moisture prediction method based on soil temperature and moisture persistence[J]. Journal of Arid Meteorology, 2023, 41(5): 783-791.

魏森涛, 王澄海, 张飞民, 杨凯. 基于土壤温、湿度记忆性的土壤湿度预测方法研究[J]. 干旱气象, 2023, 41(5): 783-791.

Figures/Tables 8

Fig.1 The schematic of the convolutional neural network prediction model of soil moisture

Fig.2 The loss function of the training dataset and the test dataset

Fig.3 The distribution of difference between the predicted soil moisture of 0-7 cm depth and the real value from July to December 2021 （Unit： m3·m-3）（a） July，（b） August，（c） September，（d） October，（e） November，（f） December

Fig.4 The distribution of difference between the predicted soil moisture of 7-28 cm depth and the real value from July to December 2021 （Unit： m3·m-3）（a） July，（b） August，（c） September，（d） October，（e） November，（f） December

Fig.5 The distribution of difference between the predicted soil moisture of 28-100 cm depth and the real value from July to December 2021 （Unit： m3·m-3）（a） July，（b） August，（c） September，（d） October，（e） November，（f） December

Fig.6 The distribution of difference between the predicted soil moisture of 100-289 cm depth and the real value from July to December 2021 （Unit： m3·m-3）（a） July，（b） August，（c） September，（d） October，（e） November，（f） December

Fig.7 The fitting relation between the predicted soil moisture with different depth layers and real values from January to June 2021

Fig.8 Variation of mean values of predicted soil moisture and real values in different representative regions from July to December 2021 （a） northern China and Mongolia，（b） southern South America，（c） eastern North America，（d） highlands of South Africa

References 25

[1]	陈浩, 宁忱, 南卓铜, 等, 2017. 基于机器学习模型的青藏高原日降水数据的订正研究[J]. 冰川冻土, 39(3): 583-592.
[2]	李若麟, 保鸿燕, 李课臣, 等, 2016. 全球土壤湿度的记忆性及其气候效应[J]. 冰川冻土, 38(6): 1 470-1 481.
[3]	聂红梅, 2019. 基于多源数据和机器学习算法的土壤水分预测[D]. 西安: 西北大学.
[4]	任萍, 陈明轩, 曹伟华, 等, 2020. 基于机器学习的复杂地形下短期数值天气预报误差分析与订正[J]. 气象学报, 78(6): 1 002-1 020.
[5]	宋海清, 孙小龙, 李云鹏, 2020. 欧洲中期天气预报中心第五代全球再分析土壤湿度资料在内蒙古的适用性评估[J]. 科学技术与工程, 20(6): 2 161-2 168.
[6]	孙浩, 吴梦碟, 2023. 基于 BP+ Elman 神经网络的草原土壤湿度预测研究[J]. 建模与仿真, 12(1): 573-581.
[7]	王澄海, 师锐, 左洪超, 2008. 青藏高原西部冻融期陆面过程的模拟分析[J]. 高原气象, 27(2): 239-248.
[8]	王一昕, 2019. 基于多源土壤湿度数据的中国高敏感区探究[D]. 南京: 南京大学.
[9]	杨凯, 2017. 一个基于通用陆面模式 CLM 的全水热耦合模式的建立及模拟检验[D]. 兰州: 兰州大学.
[10]	周飞燕, 金林鹏, 董军, 2017. 卷积神经网络研究综述[J]. 计算机学报, 40(6): 1 229-1 251.
[11]	CAI Y, ZHENG W, ZHANG X, et al, 2019. Research on soil moisture prediction model based on deep learning[J]. PloS one, 14(4), DOI: 10.1371/journal.pone.0214508
[12]	CIFUENTES J, MARULANDA G, BELLO A, et al, 2020. Air temperature forecasting using machine learning techniques: a review[J]. Energies, 13(16), DOI: 10.3390/en13164215
[13]	GARCIA D H, 2021. Analysis and precision of the Terrestrial Surface Temperature using Landsat 8 and Sentinel 3 images: study applied to the city of Granada (Spain)[J]. Sustainable Cities and Society, 71, DOI: 10.1016/J.SCS.2021.102980
[14]	GUO Y, CAO X, LIU B, et al, 2020. El Niño index prediction using deep learning with ensemble empirical mode decomposition[J]. Symmetry, 12(6), DOI: 10.3390/sym12060893
[15]	HAGEMANN S, STACKE T, 2015. Impact of the soil hydrology scheme on simulated soil moisture memory[J]. Climate Dynamics, 44: 1 731-1 750.
[16]	HAM Y G, KIM J H, LUO J J, 2019. Deep learning for multi-year ENSO forecasts[J]. Nature, 573(7775): 568-572. DOI
[17]	HINTON G E, OSINDERO S, TEH Y W, 2006. A fast learning algorithm for deep belief nets[J]. Neural Computation, 18(7): 1 527-1 554.
[18]	LECUN Y, BENGIO Y, 1998. Convolutional networks for images, speech, and time series[M]. Cambridge, MA: MIT Press.
[19]	PRAKASH S, SHARMA A, SAHU S S, 2018. Soil moisture prediction using machine learning[C]// 2018 Second International Conference on Inventive Communication and Computational Technologies (ICICCT), IEEE: 1-6.
[20]	SENEVIRATNE S I, KOSTER R D, 2012. A revised framework for analyzing soil moisture memory in climate data: derivation and interpretation[J]. Journal of Hydrometeorology, 13(1): 404-412. DOI URL
[21]	SHINODA M, NANDINTSETSEG B, 2011. Soil moisture and vegetation memories in a cold, arid climate[J]. Global and Planetary Change, 79(1/2): 110-117. DOI URL
[22]	TAN J, NOURELDEEN N, MAO K, et al, 2019. Deep learning convolutional neural network for the retrieval of land surface temperature from AMSR2 data in China[J]. Sensors, 19(13), DOI: 10.3390/s19132987
[23]	TANG Y, DUAN A, 2021. Using deep learning to predict the East Asian summer monsoon[J]. Environmental Research Letters, 16(12), DOI: 10.1088/1748-9326/AC34BC
[24]	WANG C, YANG K, 2018. A new scheme for considering soil water-heat transport coupling based on Community Land Model: model description and preliminary validation[J]. Journal of Advances in Modeling Earth Systems, 10(4): 927-950. DOI URL
[25]	YOSINSKI J, CLUNE J, BENGIO Y, et al, 2014. How transferable are features in deep neural networks?[C]// Proceedings of the 27th International Conference on Neural Information Processing Systems, 2: 3 320-3 328.