November 09, 2021

Print Friendly, PDF & Email

Assessing and Predicting Cyclone Effects on Forests

Researchers evaluated what factors affect forest disturbance intensity across multiple regions and the potential to develop a cyclone impact model

Disturbance intensity maps of forest impacts from hurricanes Katrina, Rita, Yasi, and María from high-resolution airborne remote sensing images validated with field observations.

[Reprinted under a Creative Commons Attribution NonCommercial-NoDerivatives 4.0 International (CC BY-NC-ND 4.0) from Feng, Y., et al. "Multi-Cyclone Analysis and Machine Learning Model Implications of Cyclone Effects on Forests." International Journal of Applied Earth Observation and Geoinformation 103, 102528 (2021). DOI: 10.1016/j.jag.2021.102528]

The Science

Scientists used satellite images of the impacts of multiple tropical cyclones to study what factors contribute to different impacts on forests brought by hurricanes. Scientists found that a 40 m/s wind speed threshold affects a cyclone’s impact, but discovered little consistency in the influence of other variables. Each cyclone interacted with the landscape in a unique way. In addition, the researchers discussed the difficulties for building a model that can predict the location or damage of future cyclones.

The Impact

Disturbance from cyclones impacts the structure and function of forests. Therefore, it is important to understand how forests in different regions were affected by past cyclones and gain improved insights for future cyclones. This study reveals the links between remote sensing of forest disturbance intensity and the factors of wind and rainfall, forest structure, terrain features, and soil properties at the landscape scale, and discusses the possibility of using machine learning to help predict the impact of hurricanes on forests.


This study addressed the importance of climate variables, terrain features, and forest properties in predicting tree damage caused by cyclones. Wind, elevation, and pre-disturbance vegetation condition are strong predictors. Cyclones interacted with the landscape in unique ways, and there are no consistent rules can be applied to all the cyclones. Machine learning technologies were used to build cyclone impact models, and this study showed the limitations of machine learning models in cyclone effects prediction. The models worked well on hold out test data, but they had weak predictability on unseen cyclones. The authors believe that finer scale data can be helpful to build local models that work with similar ecosystems and landscapes; however, the complexities of cyclone effects coupled with landscapes, soils, states of affected systems, and climate change lead to questions regarding the existence of an omnipotent cyclone impact model that works for the globe.

Principal Investigator

Yanlei Feng
Lawrence Berkeley National Laboratory

Program Manager

Brian Benscoter
U.S. Department of Energy, Biological and Environmental Research (SC-33)
Environmental System Science


This research was supported by the Next Generation Ecosystem ExperimentTropics (NGEETropics) project (DE-AC02-05CH11231), which is funded by the Office of Biological and Environmental Research (BER) within the U.S. Department of Energy’s (DOE) Office of Science.


Feng, Y., et al. "Multi-Cyclone Analysis and Machine Learning Model Implications of Cyclone Effects on Forests." International Journal of Applied Earth Observation and Geoinformation 103 102528  (2021).