Introduction:
I have found the research to be utterly fascinating. I sincerely wish I had more time to work on it in order to test more variables. To begin, the USDA is interested in planting the Loblolly Pine tree in the Southeastern United States. Considering the climate has been changing and will continue to change, the USDA would like to understand how the forest may be affected. The Loblolly's natural range goes as far west as eastern Texas and as far north as southern New Jersey. Extreme winter temperatures and ice storms are a factor in lack of northward extension. However, with the projected increase in temperature, the natural Loblolly range may extend northward. It grows relatively quickly, compared to other trees, and does not have any long-term health problems. However, like most living things, it is susceptible to diseases, pests, and damage. For my study, I chose to focus on one pest, the Southern Pine Beetle (SPB).
It is well known that climate has an influence on the life cycle of the Loblolly. However, does climate influence the SPB? Well, numerous studies have found climate (i.e. temperature and precipitation) impact SPB outbreaks (Beal 1929, Duehl et al. 2011, Gan 2004, Mcclelland and Hain 1979,
Michaels 1984). Over the years, studies found that severely cold winters resulted in broad morality rates of the SPB (Beal 1929, Mcclelland and Hain 1979, and Michaels 1984). Those same studies found that mild winters kill less SPBs and lead to outbreaks the following spring and summer months. It is important to note that all these studies found extremely cold temperatures (i.e. -5F) lead to these mortality rates. As a consequence, I thought I would determine if there was a relationship between the number of freezing days and the number of outbreaks. I will also study how the annual number of freezing days may change with the future. I will determine the likely hood of temperatures occurring by calculating probability density functions (PDF). Lastly, to determine a level of confidence in the models, I will conduct a statistical bias test.
Data and Methods:
For this study, I used outbreak data from the USDA and temperature data calculated from one downscaled method, the Multivariate Adaptive Constructed Analogs (MACA). The outbreak data lists the number of outbreaks per county from 1960-2004. For this study, I chose to use two variables that determined outbreaks. The first is the total number of spots (tspot) in a county and the second is a measure of outbreak level determined by the size of the county (disc). "Tspot" ranges from 0 to about 1000. However, the tspot data is unreliable with missing data. Tspot data is reliable from 1992-2004. "Disc" is 5 different values. 0 equals less than 0.1 spots per thousand acres, 1 equals 0.1 to less than 1 spot per thousand acres, 2 equals 1 to less than 3 spots per thousand acres, 3 equals 3 or more spots per thousand acres, and 9 equals more than 1 spot per thousand acres. The disc data was consistent with no missing data, therefore disc data was used from 1986 to 2004. The MACA data is 6 by 6 km resolution gridded data over the entire continental United States. I used the historical observations of the annual number of freezing days from 1986-2005. So, the historical observations are an averaged value from 1986-2005. I used two models for future projections (from 2020-2099). The RCP 4.5 and RCP 8.5, which is run under different two scenarios. The RCP 4.5 is the stabilization scenario in which the total radiative forcing stabilized shortly after 2100. The RCP 8.5 increases the greenhouse has emissions over time which leads to high amounts of CO2 levels. I calculated PDFs for the historical observations (1986-2005) and 20 climate models (1950-2005) over 15 climate regions. Lastly, to determine a bias, I compared the 20 climate model baseline periods (1950-2005) to that of the historical observation period (1986-2005) over the entire U.S.
Considering the outbreak data is per county, I needed to get the averaged number of freezing data per county, too. I imported the gridded data as a raster file and applied zonal statistics in ArcMap to receive the average per county (Fig. 1). For example, the average number of days below freezing in wake county is 70 days per year or about 20 per cent of the year.
 |
| Figure 1. The historical averaged number of freezing days per county. |
I used the same method to conduct the future projections for each county over the U.S. The projections are separated into 4 ranges: 2020-2039, 2040-2059, 2060-2079, 2080-2099. So, in total there are 4 projections for each RCP scenario.
I use the programming language R to compute the average number of tspots per county from 1992-2004 (because data is missing) and the average disc per county from 1986-2004 to correlate with the historical average (1986-2005). From here, I used R to conduct a correlation test between the mean tspot and mean number of freezing days and mean disc and mean number of freezing days.
I also used R to produce PDFs for the historical observations and 20 climate models. I made PDFs for 15 climate regions over the Southeastern U.S. Overall, each climate zone has a historical PDF from 1986-2005 and 20 historical model runs from 1950-2005.
R was also used to compute the statistical bias of the climate models. The mean of all 20 models was subtracted from the historical observation. This will inform me of how the climate models over and under predicted the annual number of freezing days.
Results:
The correlation test between mean tspot and mean annual number of freezing days resulted with weak negative correlation. Additionally, the correlation test between mean disc and mean number of freezing days resulted with weak positive correlation. This was a bit surprising. It was assumed that if a year was above average in days below freezing, it may be colder than normal as well. However, the extreme temperature data does not give the temperature value. Therefore, there's no way to know how far below freezing it was. Previous studies found a correlation between severely cold temperatures (i.e. -5F) and less outbreaks.
As expected, both RCP scenarios projected a decrease in annual freezing days across the study area. The most dramatic change occurring by the final years 2080-2099. RCP 8.5. projected more of a decrease than the RCP 4.5 This was expected because the RCP 8.5 scenario incorporated more CO2 emissions than the RCP 4.5. More specifically, both mean RCP scenario values of Eastern and Western NC decreased significantly. To compare, the mean historical amount of freezing days across Eastern and Western NC is 65 and 102 days respectively. The RCP 4.5 scenario projected the mean amount of freezing days to decrease by 25 days across Eastern NC and 31 across Western NC by 2080-2099. The RCP 8.5 scenario projected the mean amount of freezing days to decrease by 40 days across Eastern NC and 52 days across Western NC by 2080-2099. The decrease in the annual number of freezing days could allow for the Loblolly to extend northwards.
In general, the PDF plots showed model consensus among the 20 climate models. One model consistently deviated from the norm across all evaluated domains. There was less model consensus over areas of varied elevation (Fig. 2).
All 20 models differed from the observed historical data and did possess a bias. Most of the bias was over prediction across the mountainous region (Fig. 3). Overall, the models prediction of the past resembled that of the observations. Due to this, we can have more confidence in the future projections. Also, the areas the models struggled the most with does not include the natural range of the Loblolly Pine.
Conclusions:
Overall, this was an interesting study. I was a bit surprised to see such a weak correlation between outbreaks and freezing days. For future studies, I would like to incorporate other climate variables into the study. As well as seeing if there is any inter-seasonal variability. In conclusion, this study further solidified my confidence in the climate future projections. As a consequence, I deem it necessary that the forestry sector utilizes this information to help with future impacts.
R was also used to compute the statistical bias of the climate models. The mean of all 20 models was subtracted from the historical observation. This will inform me of how the climate models over and under predicted the annual number of freezing days.
Results:
The correlation test between mean tspot and mean annual number of freezing days resulted with weak negative correlation. Additionally, the correlation test between mean disc and mean number of freezing days resulted with weak positive correlation. This was a bit surprising. It was assumed that if a year was above average in days below freezing, it may be colder than normal as well. However, the extreme temperature data does not give the temperature value. Therefore, there's no way to know how far below freezing it was. Previous studies found a correlation between severely cold temperatures (i.e. -5F) and less outbreaks.
As expected, both RCP scenarios projected a decrease in annual freezing days across the study area. The most dramatic change occurring by the final years 2080-2099. RCP 8.5. projected more of a decrease than the RCP 4.5 This was expected because the RCP 8.5 scenario incorporated more CO2 emissions than the RCP 4.5. More specifically, both mean RCP scenario values of Eastern and Western NC decreased significantly. To compare, the mean historical amount of freezing days across Eastern and Western NC is 65 and 102 days respectively. The RCP 4.5 scenario projected the mean amount of freezing days to decrease by 25 days across Eastern NC and 31 across Western NC by 2080-2099. The RCP 8.5 scenario projected the mean amount of freezing days to decrease by 40 days across Eastern NC and 52 days across Western NC by 2080-2099. The decrease in the annual number of freezing days could allow for the Loblolly to extend northwards.
In general, the PDF plots showed model consensus among the 20 climate models. One model consistently deviated from the norm across all evaluated domains. There was less model consensus over areas of varied elevation (Fig. 2).
 |
| Figure 2. The probability distribution function for the annual number of freezing days over the Southern Appalachians. |
All 20 models differed from the observed historical data and did possess a bias. Most of the bias was over prediction across the mountainous region (Fig. 3). Overall, the models prediction of the past resembled that of the observations. Due to this, we can have more confidence in the future projections. Also, the areas the models struggled the most with does not include the natural range of the Loblolly Pine.
 |
| Figure 3. The mean annual number of freezing days bias map for all 20 climate models. |
Conclusions:
Overall, this was an interesting study. I was a bit surprised to see such a weak correlation between outbreaks and freezing days. For future studies, I would like to incorporate other climate variables into the study. As well as seeing if there is any inter-seasonal variability. In conclusion, this study further solidified my confidence in the climate future projections. As a consequence, I deem it necessary that the forestry sector utilizes this information to help with future impacts.
No comments:
Post a Comment