L2: Accurate Forestry Time-Series Completion and Growth Factor Inference

Linlu Jiang, Meng Yang, Benye Xi, Weiliang Meng, Jie DuanForests 2025 · DOI

TL;DR

This paper introduces "L2," a novel framework for accurately completing missing forestry time-series data and inferring key growth factors. By combining advanced tensor completion with a dual-attention LSTM model, L2 significantly improves data integrity and provides deeper insights into factors influencing tree growth, crucial for sustainable forest management.

What Problem Does This Paper Address?

Forestry relies heavily on accurate data for effective resource management and monitoring. However, real-world forestry datasets are plagued by missing values due to sensor failures, transmission issues, and human errors. These data gaps, especially when extensive or continuous, severely hinder analytical accuracy and the reliability of predictive models. Furthermore, the inherent complexity and multidimensional nature of forest ecosystems make imputing these missing values and understanding the correlations between various factors a significant challenge. Existing methods often struggle with the scale and complexity of these missing data scenarios in forestry.

Methods

The L2 model is a two-stage approach:

  1. Data Completion (L2-Complementary Model):
    • Improved LRTC-TNN: This component addresses missing data by leveraging Low-Rank Tensor Completion (LRTC) with a Truncated Nuclear Norm (TNN) to capture both global consistency and local trends in the data. The TNN is an optimization over traditional nuclear norms, better handling information loss. The model is adapted for multi-class data and uses an alternating direction method of multipliers (ADMM) for optimization.
  2. LSTM-CNN with Self-Attention: For more complex and larger missing data segments, a Long Short-Term Memory (LSTM) and Convolutional Neural Network (CNN) fusion model is employed. This architecture extracts temporal and spatial features. A self-attention mechanism is integrated to dynamically weigh the importance of different time steps, enhancing the model's ability to reconstruct intricate missing patterns.
  3. Growth Factor Inference (L2 Prediction Model):
    • Dual-Attention LSTM: To predict stem diameter (a key growth indicator), an LSTM model is augmented with a dual-attention mechanism. This includes a feature attention layer that dynamically weights different input features (like soil moisture, temperature) and a temporal attention layer that weighs historical time steps. This allows the model to focus on the most relevant information for accurate prediction.
  4. SHAP Analysis: To understand which factors most influence stem diameter, the paper utilizes Shapley Additive Explanations (SHAP). This game-theory-based approach quantifies the contribution of each input feature to the model's predictions, providing interpretable insights into the drivers of tree growth.

Datasets

The study uses data from a Populus tomentosa (woolly poplar) forest. The dataset includes:

  • Growth-related data: Trunk Diameter (DIA), Soil Moisture (SM), and Liquid Flow (LF).
  • Meteorological data: Temperature (TEMP), Radiation (RAD), and Relative Humidity (RH).

The paper notes that this dataset has a significant missing rate (over 30%), with some continuous missing values exceeding 50%, making it a challenging benchmark for data imputation and analysis.

Key Insights and Ideas

  • Hybrid Approach is Key: The success of L2 lies in its hybrid nature, combining tensor completion for structured data recovery with deep learning for complex temporal and spatial patterns. This addresses different types of missing data effectively.
  • Attention is Crucial for Forestry: Both self-attention in the completion model and dual-attention in the prediction model significantly boost performance. This highlights the importance of dynamically focusing on relevant information in complex, time-series forestry data.
  • SHAP for Actionable Insights: Moving beyond just prediction, the use of SHAP provides interpretable results, identifying soil moisture as the dominant factor in stem diameter growth. This translates model performance into actionable insights for forest managers.
  • TNN for Robust Tensor Completion: The adoption of Truncated Nuclear Norm (TNN) over standard nuclear norms in LRTC offers a more robust way to handle tensor completion, especially in scenarios with partial observations.

Key Results

  • Data Completion Accuracy: The L2 model significantly improves data completion accuracy, preserving the original data structure and characteristics.
  • Prediction Performance: The L2 stem diameter prediction model achieves high-precision predictions. Ablation studies show the L2 model consistently outperforms its variants and other baseline methods across various metrics (RMSE, MASE). For instance, in Table 1, the L2 model achieves the lowest RMSE for DIA (283.448347) and SM (0.365627) compared to other configurations. Table 4 shows L2 achieving the lowest RMSE for DIA (283.448347) and SM (0.365627) among various imputation methods.
  • Feature Importance: SHAP analysis reveals that soil moisture is the most critical factor influencing stem diameter growth, contributing over 16% to the prediction, followed by relative humidity and precipitation. Wind speed has a minimal impact.

Limitations and Future Work

  • Generalizability: While tested on Populus tomentosa, the paper doesn't explicitly discuss how well the model generalizes to other tree species or forest types.
  • Computational Cost: The complex architecture, especially with multiple attention mechanisms and tensor operations, might incur significant computational costs for training and inference, though the paper mentions implementation on a high-performance server.
  • Future Work: The authors suggest that their findings provide a solid foundation for improving forestry data analysis and offer theoretical support for precision forest management strategies. Further research could explore extending this framework to other forestry applications or different types of environmental data.

Summary updated

Abstract

In forestry data management and analysis, data integrity and analytical accuracy are of critical importance. However, existing techniques face a dual challenge: first, sensor failures, data transmission interruptions, and human errors lead to the prevalence of missing data in forestry datasets; second, the multidimensional heterogeneity and environmental complexity of forestry systems not only increase the difficulty of missing value estimation, but also significantly affect the accuracy of resolving the potential correlations among data. In order to solve the above problems, we proposed the L2 model using the aspen woodland as the experimental object. The L2 model consists of a complementary model and a predictive model. The L2 complementary model integrates low tensor tensor kernel norm minimisation (LRTC-TNN) to capture global consistency and local trends, and combines long and short-term memory and convolutional neural network (LSTM-CNN) to extract temporal and spatial features, which is effective in accurately reconstructing the missing values in forestry time-series data. We also optimised the LRTC-TNN model to handle multi-class data and incorporated a self-attention mechanism into the LSTM-CNN framework to improve performance in the case of complex missing data. The L2 prediction model adopts a dual attention mechanism (temporal attention mechanism and feature attention mechanism) based on LSTM to construct a stem diameter prediction model, which achieves high-precision prediction of stem diameter variation. Then we further analyzed the effects of various factors on stem diameter using SHAP (Shapley Additive Explanations). Experimental results demonstrate that our L2 significantly improves data completion accuracy while preserving the original structure and key characteristics of the data. Moreover, it enables a more precise analysis of the factors affecting stem diameter, providing a robust foundation for advanced forestry data analysis and informed decision making.
BibTeX
@inproceedings{jiang2025l,
  author = {Linlu Jiang and Meng Yang and Benye Xi and Weiliang Meng and Jie Duan},
  title = {{L2: Accurate Forestry Time-Series Completion and Growth Factor Inference}},
  booktitle = {Forests},
  year = {2025},
  doi = {10.3390/f16060895},
}
Figure