An analysis of growth prediction of forest attributes using

Forest Vegetation Simulator (FVS)

A final internship report submitted to Edna Bailey Sussman Foundation

-Karun Pandit

  1. Background

Forest monitoring has been recognized as one of the most important programs in forest management with an increasing concern over the issues like climate change, biodiversity conservation and protection of forest ecosystems. Large-scale field monitoring programs used to quantify carbon sequestration can be cost prohibitive, but increased efficiencies may be attained by supplementing measurements with model estimates. Researchers have been trying to explore efficient models that could predict forest growth attributes. Application of these growth models would supplement forest monitoring system and help forest managers draw inferences on various aspects of forest stand dynamics and structure more efficiently.

Forest Vegetation Simulator (FVS) was developed by US Forest Service to predict forest stand dynamics throughout USA. FVS is an individual-tree, distance-independent, growth and yield model. FVS has been evolved into a collection of different variants to represent different geographic conditions across the country. FVS can simulate a wide range of silvicultural treatments for most major forest tree species, forest types, and stand conditions.

The New York State Department of Environmental Conservation (DEC) manages over 775,000 acres of state forest land with the goal of maintaining forest health and diversity, with an added benefit of increased carbon sequestration to mitigate climate change. Information on the accuracy of FVS for the prediction of forest growth would be important for DEC. The Forest Inventory and Analysis (FIA) Program of the US Forest Service provides the information needed to assess forests throughout the country.The primary objective of this study was to investigate the reliability of FVS growth estimates in order to see the possibility of its use by authorities like New York Department of Environment Conservation (DEC) to monitor state forest lands. Additionally, FIA data were analyzed to observe the accuracy in the predicted forest growth results.

  1. Data and Method

The study was based on forest inventory data at least from two different time intervals. One set of data was collected from DEC managed forest land. Stand level data was obtained from DEC managed Cayuga county. Thirty one stands which were inventoried at different time intervals were considered for the growth study. These time intervals however were not same for all the stands. Also due to lack of plot level information FVS was run to produce stand level growth estimates but not at the plot level. Forest Inventory and Analysis (FIA) data was the second set of data studied. This data had information at plot level and forest type could also be identified. Data was downloaded from FIA website and later processed to run into FVS. Data from 55 counties covering all 8 units in New York State was downloaded and processed. A total of 366 plots were selected from the units and counties that had FIA data from 2002 and 2008.

FORMAT4FVS and SUPPOSE software were used to convert the file format and run the FVS growth model. First different Excel files were prepared for each plot and for each inventory year and saved as CSV format files. These files were then converted into FVS format using FORMAT4FVS tool. FVS files were read into SUPPOSE software and simulation was carried out. Northeast variant of the FVS was used in the analysis to make the model assumptions more realistic to the area of study. Attributes like number of tree, basal area, total cubic feet volume, merchantable cubic feet volume and board feet volume were produced as estimates. For the purpose of comparison, percent error was calculated for basal area and total cubic feet volume using predicted estimates from the model and observed data of the same year.

  1. Results

3.1DEC data

Percent error in the growth estimate of basal area and total cubic feet volume for each of the stand was calculated. Mean percent error for all the stands was 28.28 for basal area and 17.44 for total cubic volume feet. The results show that there is overestimate from the FVS growth model.Scatterplot of the percent errors of both basal area and total cubic feet volume shows almost similar pattern. Percent errors for most of the stands are within a range of about -10 to +50 except some outliers. One reason for high overestimation could be due to lack of information on the type of management carried out between the two time intervals. Model could not have fully recognized the exact type of management undertaken in these stands.

1a. Percent Error for Basal Area for each stand / 1b. Percent Error for Total Cubic Feet volume for each stand

3.2FIA data

The average percent error for basal area estimation was 11.03 and that for total cubic feet volume was 9.32. We found the percent error for FIA data quite lower than what we found for the DEC data. When we look by forest type, on average higher percent errors were observed for Elm/Ash forest and Oak/Hickory Forest. The smallest percent error was observed for Maple and Pine category. Hence, the FVS model prediction seems quite close for Maple and Pine forests compared to other forest types. We could observe higher range of percent error for Elm/Ash and Spruce/Fir forest which are evident by the whiskers in the box plot.

Forest Type / Elm/Ash / Maple / Oak/Hickory / Spruce/Fir / White/Red Pine
N / 30 / 197 / 48 / 33 / 57
Mean / 16.82 / 6.87 / 15.07 / 7.79 / 8.62
Maximum / 170.17 / 99.10 / 73.94 / 120.80 / 92.15
Q3 / 21.59 / 15.11 / 26.77 / 11.83 / 13.18
Median / 6.58 / 2.58 / 13.92 / 4.23 / 3.15
Q1 / 0.84 / -3.74 / 1.05 / -3.23 / -2.28
Minimum / -48.96 / -45.39 / -24.04 / -31.37 / -11.76
Standard Deviation / 37.62 / 16.39 / 19.46 / 25.26 / 17.82

Figure 2. Percent error in basal area estimation for different forest types

Total cubic feet volume estimate revealed almost similar results as of basal area estimate. Maple and Pine categories had the lowest of errors while Oak/Hickory and Elm/Ash category had the highest of errors. Despite the number of plots studied, maple forest had very narrow range of error. The ranges between first and third quartiles were smaller for maple, spruce/fir and pine forests.

Forest Type / Elm/Ash / Maple / Oac/Hickory / Spruce/Fir / White/Red Pine
N / 30 / 197 / 48 / 33 / 57
Mean / 11.88 / 5.57 / 13.25 / 9.86 / 6.04
Maximum / 153.03 / 87.85 / 115.32 / 145.84 / 149.56
Q3 / 26.32 / 12.96 / 24.03 / 12.42 / 11.82
Median / 3.68 / 0.76 / 5.82 / 6.12 / -1.23
Q1 / -8.34 / -6.59 / -3.94 / -2.46 / -6.46
Minimum / -73.62 / -53.80 / -26.29 / -34.32 / -22.43
Std / 41.28 / 20.52 / 27.50 / 31.07 / 26.30

Figure 3 Percent error in total cubic feet volume estimate for different forest types

  1. Discussion

Current result from the analysis shows that the prediction from FVS growth model is pretty good. We should try to accommodate the overprediction in the result. Results from DEC data were not much encouraging basically because we could not properly take into account some of the underlying facts about the types of forest management activities in between the time intervals. But in FIA data this was not a serious issue since most of the management activities were identified properly. For certain forest types like Maple and Pine, percent error was quite low than other types of forest types. Thus, we can make more reliable estimation for these types of forests from FVS. Since we used Northeast variant of FVS for the analysis, this variant might have provided best results for the species which are most prevalent in Northeast geographical region.

  1. Future Work

Further analysis on FIA data could help in drawing inferences separately for each of the eight forest management unit. This information would be useful to understand if there is better prediction for certain units compared to others. Also some analysis at individual tree level would be sought for to see the accuracy of prediction with respect to given diameter class of trees.

Acknowledgments

I would like to thank my advisor Dr. Eddie Bevilacqua for helping me in successfully carrying out this internship study. I am also thankful to Justin A Perry from DEC for his help in this study. I am grateful to Edna Bailey Sussman Foundation for providing me this wonderful opportunity to do study on the subject matter that I am much interested in. The foundation will also be acknowledged in any future presentations and paper based on this study.