Publication:
Bone, Christopher, and Michael Nelson. “Improving Mountain Pine Beetle Survival Predictions Using Multi-Year Temperatures Across the Western USA.” Forests 10 (October 3, 2019): 866. https://doi.org/10.3390/f10100866.
Background
Mountain pine beetle (MPB) overwinter survival is often limited by low temperatures. Overwintering larvae are killed at -40°C, but they can also be killed in warmer conditions if there is a sudden drop in temperature. Winter low temperatures are often used to predict overwinter survival of MPB, but given the complex relationship between insect physiology and temperature, a single temperature measurement may not provide sufficient information to make good predictions.
For this study, we reimplemented the overwintering physiological model of Jacques Régnière and Brbara Bentz in Java for use with the DAYMET large-extent daily temperature data sets to predice MPB overwintering survival. We had two main goals in this study:
- 1: to examine spatial patterns of overwintering survival.
- 2: to improve the accuracy of MPB survival predictions by using overwintering survival from multiple years (instead of just the preceding year).
This project was an opportunity to build my Java skills and to work with large-scale temperature datasets.
The Data and Model
DAYMET provides daily minimum and maximum temperature over a 1km * 1km grid over continental North America. As such, it is a very large data product, and we used custom scripts to download the applicable datasets.
To characterize the spatial distribution of winter temperatures, we created a map of the mean winter minimum temperature from 1980 to 2016. To capture the amount of interannual variation in minimum winter temperatures, we also create a map showing the standard deviation of winter minimums.
Figure 1: This map shows the mean minimum annual temperature (first panel) as well as the variability in minimum temperatures (second panel) over western North America from 1980 - 2016). The right panel shows the cumulative pine trees killed by MPB over the same period.
We implemented the Régnière and Bentz model in Java for use with DAYMET data. To validate the model, we compared our implementation’s output to the survival estimates in the original paper at the set of weather stations that they reported in their paper.
In order to test how well the model survival predictions approximated reality, we used a spatial dataset of tree mortality from Meddens and Hicke (2014). We could then use beetle overwinter survival percentage as a predictor for pine tree mortality.
Main Findings
We found that single-year model output did a poor job of predicting pine mortality in the following growing season. As we used mean modeled survival rates over longer numbers of previous years, the predictions improved:
Figure 4: You can see that the relationship between modeled MPB overwinter survival becomes much clearer when more years are considered. The points follow a very defined curve when at least 8 years of previous survival were used.
Conclusions
Multi-year modeled survival rates were much better predictors of subsequent pine mortality than simple winter minimum temperatures in the previous year.