Research Article
An Improved Model for Sap Flow Prediction Based on Linear Trend Decomposition
@INPROCEEDINGS{10.1007/978-3-031-65123-6_14, author={Bo Li and Yane Li and Hailin Feng and Bin Wu and Qiang Zhu and Xiang Weng and Yaoping Ruan}, title={An Improved Model for Sap Flow Prediction Based on Linear Trend Decomposition}, proceedings={Quality, Reliability, Security and Robustness in Heterogeneous Systems. 19th EAI International Conference, QShine 2023, Shenzhen, China, October 8 -- 9, 2023, Proceedings, Part II}, proceedings_a={QSHINE PART 2}, year={2024}, month={8}, keywords={Deep Learning Sap flow Dlinear Environment factors Short-term prediction}, doi={10.1007/978-3-031-65123-6_14} }- Bo Li
Yane Li
Hailin Feng
Bin Wu
Qiang Zhu
Xiang Weng
Yaoping Ruan
Year: 2024
An Improved Model for Sap Flow Prediction Based on Linear Trend Decomposition
QSHINE PART 2
Springer
DOI: 10.1007/978-3-031-65123-6_14
Abstract
Tree transpiration plays an important role in environmental change, which mainly assessed by sap flow. It is great significance to predict sap flow accurately for estimation of water consumption, management of water resource and making of sustainable development strategy for geohydrological balance under climate change. In this study, an improved prediction model was built with Dilnear algorithm using historical environment factors to assess short-term sap flow. Specifically, a public dataset including 17568 records was used. One record have one sap flow value and 9 environment factors which measured from January 1, 2016 to December 31, 2016 at 30 min interval in Canada for Tsuga canadensis (L.) Carriere. After data processed with missing value padding and normalization, seven factors was extracted with fusion of Pearson correlation and Grey correlation method. Then, sap flow prediction model was established with Dlinear algorithm. For comparison and analysis, five other deep learning networks of CNN, GRU, LSTM, Transformer and Informer were used to built sap flow prediction model respectively. Results shown that Dlinear based model has better performance than other models established in this paper. The coefficient of determination (R2), the Mean Absolute Error (MSE) and the Mean Square Error (MAE) achieved 0.9568, 0.0282 and 0.0017 for Dlinear based model.R2of Dlinear based model is of 20.37%, 14.9% and 21.93% higher than CNN, GRU and LSTM based model respectively. In addition, we also analyzed length of look-back window and prediction window. Results shown that when length of look-back window was set as 720, Dlinear based model has better performance withR2of 0.9568, which higher of 2.62%, 7.75%, 18.95% and 35.46% than length of 336, 192, 96, and 48. When the prediction window length was set as 96, Dlinear based model has higher ofR2and smaller ofMSEandMAEthan length of 240, 336 and 720. At the same time, we compared performance of models established with seven selected factors and all nine factors. Results shown that model with seven selected factors have slight higher ofR2than model with all nine factors. Results indicate that the model established in this paper can accurately predict sap flow, especially for short-term sap flow, which has great application value and practical significance for management of trees and forests.
