MS-HyTimeXer：一种融合多尺度时序特征的树干液流预测模型

MS-HyTimeXer: a stem sap flow prediction model integrating multi-scale temporal features

摘要

【目的】树干液流是量化植物蒸腾过程和评估生态系统水分利用效率的重要指标，其变化受到环境因子、生理调控以及时间滞后效应的共同影响，传统预测方法难以有效刻画复杂非线性关系和中长期动态变化。针对现有时间序列预测模型在多尺度特征提取和植物水分传输滞后响应建模方面的不足，本文提出一种基于多尺度混合Transformer的树干液流中长期预测模型MS-HyTimeXer。【方法】基于SAPFLUXNET数据库中加拿大东部白松树干液流观测数据，选取2010年1月1日至2011年12月31日期间30 min尺度连续观测数据，共35 040组样本开展实验。以TimeXer模型为基础，引入多尺度分块嵌入、LSTM时序校准和多尺度卷积3个关键模块。其中，多尺度分块嵌入用于增强不同时间尺度特征表达能力，LSTM时序校准用于捕获环境变化与树干液流响应之间的时间滞后关系，多尺度卷积用于提取局部突变和短时动态特征。选取DLinear、Informer、PatchTST、iTransformer和TimeXer 5种模型作为对比方法，对不同模型的预测性能进行评价。【结果】实验结果表明，MS-HyTimeXer模型能够有效融合多尺度时间特征和环境因子信息，在树干液流中长期预测任务中取得最优性能。与基准TimeXer模型相比，MS-HyTimeXer的均方误差（MSE）由0.765 cm³/(cm²·h)降低至0.312 cm³/(cm²·h)，降低59.74%；决定系数（ R²）由0.869提高至0.918；平均绝对百分比误差（MAPE）降低至7.652%。在6种深度学习模型中，MS-HyTimeXer获得最低的MSE、MAE和MAPE，表现出更高的预测精度和稳定性。进一步分析表明，预测窗口长度为512时，该模型在中长期预测任务中仍保持较优性能。【结论】MS-HyTimeXer通过融合多尺度时序表示、长期依赖建模和局部动态特征提取机制，有效提升了复杂生态环境条件下树干液流预测能力，为植被蒸腾估算、森林水资源管理以及生态系统智能监测提供了一种新的技术方法。

Abstract

[Objective] Stem sap flow is a key indicator for quantifying plant transpiration and evaluating ecosystem water use efficiency. Its dynamics are jointly influenced by environmental factors, physiological regulation, and time-lag effects, making it difficult for conventional prediction methods to effectively characterize complex nonlinear relationships and medium- to long-term variations. To address the limitations of existing time series forecasting models in multi-scale feature extraction and modeling the lagged response of plant water transport, this paper proposes MS-HyTimeXer, a medium- to long-term stem sap flow prediction model based on a multi-scale hybrid Transformer. [Method] Based on stem sap flow observations of Pinus strobus in eastern Canada from the SAPFLUXNET database, continuous 30 min interval data from January 1, 2010 to December 31, 2011 were selected, yielding a total of 35 040 samples for the experiments. Built upon the TimeXer model, three key modules were introduced: multi-scale patch embedding, LSTM temporal calibration, and multi-scale convolution. Multi-scale patch embedding enhances the representation of features across different temporal scales; LSTM temporal calibration captures the time-lag relationship between environmental changes and sap flow response; and multi-scale convolution extracts local abrupt changes and short-term dynamic features. Five models, including DLinear, Informer, PatchTST, iTransformer and TimeXer, were chosen as baseline methods to evaluate the prediction performance of different models. [Result] Experimental results demonstrated that MS-HyTimeXer effectively integrated multi-scale temporal features and environmental factor information, achieving optimal performance in medium- to long-term stem sap flow prediction. Compared with the baseline TimeXer model, the mean squared error (MSE) of MS-HyTimeXer decreased from 0.765 cm³/(cm²·h) to 0.312 cm³/(cm²·h), a reduction of 59.74%; the coefficient of determination (R ²) increased from 0.869 to 0.918; and the mean absolute percentage error (MAPE) dropped to 7.652%. Among the six deep learning models, MS-HyTimeXer attained the lowest MSE, MAE, and MAPE, exhibiting superior prediction accuracy and stability. Further analysis revealed that the model still maintains competitive performance in medium- to long-term prediction tasks when the prediction window length reached 512. [Conclusion] By integrating multi-scale temporal representation, long-term dependency modeling, and local dynamic feature extraction mechanisms, MS-HyTimeXer effectively improves stem sap flow prediction capability under complex eco-environmental conditions, offering a new technical approach for plant transpiration estimation, forest water resource management, and intelligent ecosystem monitoring.

关键词

蒸腾 / 树干液流 / Transformer / 深度学习 / 多尺度特征融合 / 时间序列预测

Key words

transpiration / stem sap flow / Transformer / deep learning / multi-scale feature fusion / time series forecasting

参考文献

[1] Oki T, Kanae S. Global hydrological cycles and world water resources[J].Science, 2006, 313: 1068-1072.

[2] Rodell M, Famiglietti J S, Wiese D N, et al. Emerging trends in global freshwater availability[J].Nature, 2018, 557: 651-659.

[3] 吴慧杰, 汪海娇, 许行, 等. 不同降水梯度下黄土高原油松人工林蒸腾特征及其生物物理响应[J].北京林业大学学报, 2025, 47(10): 42-53.

[4] Wu H J, Wang H J, Xu H, et al. Transpiration characteristics and biophysical response of Pinus tabuliformis plantations on the Loess Plateau of northern China under different precipitation gradients[J].Journal of Beijing Forestry University, 2025, 47(10): 42-53.

[5] Oogathoo S, Houle D, Duchesne L, et al. Vapour pressure deficit and solar radiation are the major drivers of transpiration of balsam fir and black spruce tree species in humid boreal regions, even during a short-term drought[J].Agricultural and Forest Meteorology, 2020, 291: 108063.

[6] 王彪, 陈立欣, 吴应明, 等. 干旱胁迫下环境与生理因子对晋西黄土区刺槐人工林树干液流的影响[J].北京林业大学学报, 2024, 46(4): 127-140.

[7] Wang B, Chen L X, Wu Y M, et al. Impact of environmental and physiological factors during drought stress on sap flow in Robinia pseudoacacia plantations in the loess region of western Shanxi Province of northern China[J].Journal of Beijing Forestry University, 2024, 46(4): 127-140.

[8] 刘彩霞, 陈俊辉, 秦华, 等. 有机无机肥长期配施对毛竹林土壤固碳和固氮微生物的影响[J].林业科学, 2022, 58(7): 82-92.

[9] Liu C X, Chen J H, Qin H, et al. Effects of long-term combined application of organic and inorganic fertilizers on soil CO₂- and N₂-fixing microorganisms in a subtropical bamboo forest[J].Scientia Silvae Sinicae, 2022, 58(7): 82-92.

[10] Marshall D C. Measurement of sap flow in conifers by heat transport[J].Plant Physiology, 1958, 33(6): 385-396.

[11] Feng D U, Suo L Z, Lun S, et al. Evapotranspiration measurements of community using weighting method[J].Acta Botanica Boreali-Occidentalia Sinica, 2003, 23(8): 1411-1415.

[12] Granier A. A new method of sap flow measurement in tree stems[J].Annals of Forest Science, 1985, 42(2): 193-200.

[13] Burgess S S O, Adams M A, Turner N C, et al. An improved heat pulse method to measure low and reverse rates of sap flow in woody plants[J].Tree Physiology, 2001, 21(9): 589-598.

[14] 任启文, 王玉忠, 徐振华, 等. 基于水量平衡的华北落叶松人工林承载密度研究[J].北京林业大学学报, 2024, 46(7): 1-8.

[15] Ren Q W, Wang Y Z, Xu Z H, et al. Carrying density of Larix principis-rupprechtii plantations based on water balance[J].Journal of Beijing Forestry University, 2024, 46(7): 1-8.

[16] 彭思乐, 贾文丽, 邓鹏飞, 等. 基于环境参数和深度学习模型的毛竹液流密度预测[J].农业工程学报, 2025, 41(5): 201-208.

[17] Peng S L, Jia W L, Deng P F, et al. Prediction of moso bamboo sap flow based on environmental parameters and deep learning models[J].Transactions of the Chinese Society of Agricultural Engineering (Transactions of the CSAE), 2025, 41(5): 201-208.

[18] 张延文, 王子瑄, 孙志虎, 等. 基于随机森林的长白落叶松树干液流影响因素研究[J].温带林业研究, 2022, 5(3): 21-28.

[19] Zhang Y W, Wang Z X, Sun Z H, et al. Recognition of dominant factors behind sap flow of Larix olgensis based on random forest method[J].Journal of Temperate Forestry Research, 2022, 5(3): 21-28.

[20] Tu J, Wei X, Huang B, et al. Improvement of sap flow estimation by including phenological index and time-lag effect in back-propagation neural network models[J].Agricultural and Forest Meteorology, 2019, 276−277(9): 107608.

[21] Junior L S F, Vellame L M, Oliveira A S D, et al. Transpiration of young cocoa trees under soil water restriction[J].Scientia Agricola, 2021, 78(2): e20190093.

[22] Nalevanková P, Fleischer P, Jr., Mukarram M, et al. Comparative assessment of sap flow modeling techniques in European beech trees can linear models compete with random forest, extreme gradient boosting, and neural networks[J].Water, 2023, 15(14): 2525.

[23] 姚琴娟，王伟博，胡云昊，等. 基于改进时间全卷积网络的中短期树干液流预测[J].中南林业科技大学学报, 2026, 46(1): 104-114.

[24] Yao Q J, Wang W B, Hu Y H, et al. Enhanced temporal fully convolutional network for short-and mediumterm sap flow prediction[J].Journal of Central South University of Forestry & Technology, 2026, 46(1): 104-114.

[25] Liu Y F, Gui D W, Chen X P, et al. Sap flow characteristics and water demand prediction of cash crop in hyper-arid areas[J].Agricultural Water Management, 2024, 295: 108767.

[26] Li Y, Ye J, Xu D, et al. Prediction of sap flow with historical environmental factors based on deep learning technology[J].Computers and Electronics in Agriculture, 2022, 202: 107400.

[27] Li Y, Guo L, Wanag J, et al. An improved sap flow prediction model based on CNN-GRU-BiLSTM and factor analysis of historical environmental variables[J].Forests, 2023, 14(7): 1310.

[28] Zhang B, Zhang D, Feng Z, et al. Assessment of the potential of indirect measurement for sap flow using environmental factors and artificial intelligence approach: a case study of Magnolia denudata in Shanghai urban green spaces[J].Forest, 2023, 14(9): 1768.

[29] Li Y, Hu Y, Wang W, et al. Patch-TS: a fast and accurate patch mixer-based model for medium- and long-term sap low prediction with environmental factors[J].Forests, 2025, 16(4): 606.

[30] Wang J, Renninger H J. SapFlower: an automated tool for sap flow data preprocessing, gap‐filling, and analysis using deep learning[J].New Phytologist, 2025, 246(5): 2324-2345.

[31] 付照琦, 胡旭, 田沁瑞, 等. 晋西黄土区2种典型森林树种夜间液流特征及对环境因子的响应[J].植物生态学报, 2024, 48(9): 1128-1142.

[32] Fu Z Q, Hu X, Tian Q R, et al. Nocturnal sapflow characteristics of two typical forest tree species and responses to environmental factors in the loess region of West Shanxi, China[J].Chinese Journal of Plant Ecology, 2024, 48(9): 1128-1142.

[33] 马世杰, 焦炳忠, 韩磊, 等. 基于BP神经网络优化模型的不同退化程度刺槐树干液流动态模拟[J].水土保持研究, 2026, 33(2): 333-344.

[34] Ma S J, Jiao B Z, Han L, et al. Simulation of stem sap flow dynamics of Robinia pseudoacacia with different degrees of degradation based on BP neural network optimized model[J].Research of Soil and Water Conservation, 2026, 33(2): 333-344.

[35] Poyatos R, Granda V, Flo V, et al. Global transpiration data from sap flow measurements: the SAPFLUXNET database[J].Earth System Science Data, 2020, 13(6): 2607-2649.

[36] Lai L O, Kaplan J O. A fast mean-preserving spline for interpolating interval data[J].Journal of Atmospheric and Oceanic Technology, 2022, 39: 503-512.

[37] Wang Y, Wu H, Dong J, et al. TimeXer: empowering transformers for time series forecasting with exogenous variables[J].Advances in Neural Information Processing Systems, 2024, 37: 469-498.

[38] Liu Y, Hu T G, Zhang H R, et al. Itransformer: inverted transformers are effective for time series forecasting[J].arXiv preprint arXiv:2310.06625, 2023.

[39] Nie Y Q, Nguyen N H, Sinthong P, et al. A time series is worth 64 words: long-term forecasting with transformers[J].arXiv preprint arXiv:2211.14730, 2022.

[40] Zhou H, Zhang S, Peng J, et al. Informer: beyond efficient transformer for long sequence time-series forecasting[J].Proceedings of the AAAI Conference On Artificial Intelligence, 2021, 35(12): 11106-11115.

[41] Zeng A, Chen M, Zhang L, et al. Are transformers effective for time series forecasting?[J].Proceedings of the AAAI Conference on Artificial Intelligence, 2023, 37(9): 11121-11128.

[42] Profillidis V A, Botzoris G A. Modeling of transport demand: analyzing, calculating, and forecasting transport demand[M]. Oxford: Elsevier,2018.

[43] Efrosinin D, Kochetkova I, Stepanova N, et al. The Fourier series model for predicting sapflow density flux based on tree talker monitoring system[J]. InInternational Conference on Next Generation Wired/Wireless Networking, 2020, 12044: 198-209.

基金资助

国家自然科学基金项目(32071842)

国家重点研发计划(2022YFD2200304)