Forest pest prediction using a hybrid Random Forest–LSTM model

Introduction Forest pest outbreaks can substantially affect forest ecosystem stability and pose challenges to forest health monitoring and risk management. However, forest pest monitoring datasets commonly contain a high proportion of zero observations and exhibit pronounced temporal variability, which can limit the predictive performance of conventional machine learning approaches. To address these challenges, this study developed a hybrid RF-LSTM framework incorporating Zero-Inflated Poisson (ZIP) features and LSTM-derived temporal features to evaluate the effects of these feature enhancements on Random Forest (RF) model performance. Methods Multi-station daily monitoring data for Coleoptera and Lepidoptera in Rui’an City, Zhejiang Province, China, were used to evaluate the proposed framework. The ZIP model was first employed to characterize the zero inflation structure of pest occurrence data, while the LSTM network was used to extract temporal representations from sequences of meteorological, seasonal, and ZIP-derived features. The ZIP-derived and LSTM-derived features were then incorporated into the RF model for evaluation. Results Compared with the conventional RF model, incorporating ZIP features increased R2 from 0.69 to 0.71 for Coleoptera and from 0.79 to 0.81 for Lepidoptera. Further integration of LSTM-derived temporal features increased R2 to 0.85 and 0.87, respectively. The proposed framework was also evaluated under different training sample ratios. R2 increased with the training sample ratio for both pest groups, although MAE and RMSE did not decrease consistently. Discussion These results indicate that ZIP based feature enhancement and LSTM-derived temporal features can improve RF model performance, although the extent of improvement varies between pest groups and training sample sizes.

Authors

Institutions

Publication Details

Journal
Frontiers in Environmental Science
Published
2026-11-17
DOI
https://doi.org/10.3389/fenvs.2026.1991679
Primary Topic
Forest Insect Ecology and Management
Type
article
Field-Weighted Citation Impact
0.00
Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
article

Forest pest prediction using a hybrid Random Forest–LSTM model

Guofei Fang, Xiumei Mo, Jiaqi Zhang, Jixia Huang et al.
1 links
Frontiers in Environmental Science
Forest Insect Ecology and Management
article

Forest pest prediction using a hybrid Random Forest–LSTM model

Guofei Fang, Xiumei Mo, Jiaqi Zhang, Jixia Huang, Chengdong Xu, Jiaqi Xu
article en

Abstract

Introduction Forest pest outbreaks can substantially affect forest ecosystem stability and pose challenges to forest health monitoring and risk management. However, forest pest monitoring datasets commonly contain a high proportion of zero observations and exhibit pronounced temporal variability, which can limit the predictive performance of conventional machine learning approaches. To address these challenges, this study developed a hybrid RF-LSTM framework incorporating Zero-Inflated Poisson (ZIP) features and LSTM-derived temporal features to evaluate the effects of these feature enhancements on Random Forest (RF) model performance. Methods Multi-station daily monitoring data for Coleoptera and Lepidoptera in Rui’an City, Zhejiang Province, China, were used to evaluate the proposed framework. The ZIP model was first employed to characterize the zero inflation structure of pest occurrence data, while the LSTM network was used to extract temporal representations from sequences of meteorological, seasonal, and ZIP-derived features. The ZIP-derived and LSTM-derived features were then incorporated into the RF model for evaluation. Results Compared with the conventional RF model, incorporating ZIP features increased R2 from 0.69 to 0.71 for Coleoptera and from 0.79 to 0.81 for Lepidoptera. Further integration of LSTM-derived temporal features increased R2 to 0.85 and 0.87, respectively. The proposed framework was also evaluated under different training sample ratios. R2 increased with the training sample ratio for both pest groups, although MAE and RMSE did not decrease consistently. Discussion These results indicate that ZIP based feature enhancement and LSTM-derived temporal features can improve RF model performance, although the extent of improvement varies between pest groups and training sample sizes.

Frontiers in Environmental ScienceVol. 14
Chinese Academy of Sciences (CN), Beijing Forestry University (CN), State Forestry and Grassland Administration (CN), Institute of Geographic Sciences and Natural Resources Research (CN)
Life on land
Forest Insect Ecology and Management
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.

Laika access coming soon