WWETAC Projects

Project Title: Estimating insect distributions in Alaskan landscapes not covered in aerial surveys

STDP ID: R10-2008-01

Status: Ongoing

Principal Investigator and Affiliation: John E. Lundquist, Forest Health Protection (FHP) and Pacific Northwest Research Station, Anchorage, AK

Collaborators and Affiliations: Mark Schultz, FHP, Juneau, AK; James Kruse, FHP, Fairbanks, AK; Dustin Wittwer, FHP, Juneau, AK; Eric Johnson, National Forest System, Juneau, AK; Robin Reich, Colorado State University, Fort Collins, CO

E-mail Contact: John E. Lundquist, jlundquist[at]fs.fed.us

Key Issues/Problems Addressed: Because of its immense size, only a small percentage of the forested area in Alaska is accessible by roads.  As a consequence, estimates of pest extent, distribution, and impacts are mostly made from data collected in surveys from airplanes.  Total acreage covered during current surveys can be only a small fraction of the actual forested land area.  An obvious source of error arises from predicting pest conditions where the plane does not fly.  The size of this error can be significant, and pest incidence, severity and distributions are almost certainly underestimated.  These statistics are nonetheless used in regional and national summaries for pest condition reporting.  Statewide and national estimates of pest impacts would be much more accurate if non-flown survey areas could be more accurately assessed. A way of predicting pest conditions between flight lines and the errors associated with these predictions would certainly be useful.

Study Objectives and Goals: The study we are proposing here focuses on the use of spatial statistics to generate predictive spatial models. As spatial statistical techniques become more acceptable and as new spatial analysis techniques are developed, it is certain that these methods will provide valuable insights in understanding the roles of insects and other disturbance agents in forest ecosystems.  One immediate practical application to Alaska is how will forest insect pests be influenced by a changing climate.

Because of its geographic location at the northern edge of various forest types where ecosystems are notably sensitive to changing environment, Alaska has been referred to as the “poster state for global warming”.  Based on the 2007 survey, the most widespread forest pest condition in Alaska is aspen defoliation, which occurs over 750,000 acres, and is associated primarily with aspen leaf miner (Phyllocnistis populiella).  This is a dramatic increase in infested acreage occuring over a period of only 5 or 10 years.  Some believe that it was caused by climate change.

Our long range goal is to work with all of the major insects in Alaska, but initially, we will focus on the two major aspen defoliating insects which have dramatically contrasting population dynamics: one (P. populiella) apparently sensitive to environment factors commonly used to characterize climate (temperature and precipitation), and another (Choristoneura conflictana) that is much less sensitive. 

The specific objectives of this study are:

1)  To determine how existing aerial survey methods and results can be integrated with recently developed spatial modeling techniques to predict insect pest distributions in remote areas of Alaska where aerial surveys are logistically difficult, expensive, and currently impractical.

2)   To adapt a spatial modeling/aerial survey approach aimed at measuring climate change impact on forest insects statewide and establish baseline conditions for future assessments of insect pest migrations and intensification.

General Description: Reich, Lundquist, and their collaborators developed methods of predicting the spatial distribution and severity of forest insect pests and diseases using Landsat satellite imagery linked to field plots and a selection of auxiliary GIS data layers.   The spatial models that were generated represented the probability of pest occurrence at multiple scales from canopy gaps to landscapes to entire forests, and were able to maintain a consistency among spatial and temporal scales (Figure 1).  The methods were tested in ponderosa pine stands in South Dakota (Reich, Lundquist, and Bravo 2004; Lundquist and Reich in review), mixed conifer stands in New Mexico (Lundquist and Reich 2006), and various forest types in Colorado.

Mountain pineArmillaria root diseaseApsen Steam Canker

Figure 1.  Probability distribution models of mountain pine beetle, armillaria root disease, and aspen stem canker in the Black Hills.  Results of study by Lundquist and Reich, in press.

To evaluate the feasibility of this approach in Alaska, a preliminary study was carried out in 2007 in aspen stands on a 19,981 ha tract near Fairbanks that was heavily infested with aspen leaf miner, currently the most widespread insect pest in Alaska.  Receiver Operating Curves were used to assess predictive accuracy.  Figure 3 (see below) depicts the spatial distribution of various vegetation types and infested areas from the same study. The final model accounted for 45% of the variability in canopy closure, provided unbiased variance estimates (SMSE ≈1), and had prediction and confidence coverage rates close to the nominal 0.95 rate. We felt that these methods hold promise for estimating pest conditions in more remote locations like those in Alaska.  Our preliminary studies focused on a limit area sample space.  Our aim is to map most or all of the forested area in Alaska, which presents unique multi-spatial scale challenges. 

An important part of the this study aims at evaluating a multi-stage nested stratified sampling design that can be used to predict the presence and severity of forest pests throughout the state of Alaska.  The proposed sampling design takes into consideration the ability of both Landsat-5 TM and MODIS imagery to accurately describe the spatial distribution of both vegetation types and disease infestations over large geographical regions.

Status:

Status of Product Development 2009 -- 1) The severity of defoliation of willow and aspen was modeled based upon roadside surveys. There was a strong correlation between willow and aspen defoliation. 2) Vegetation modeling was completed for the Kenai Peninsula using Landsat-5 TM images. The 1986 land-cover classes were used to predict the land cover classes in 2008. 3) Mapping was completed for 2008 spruce mortality using 1986 data, Landsat-5 TM, spatial modeling techniques, and ground points.

See 2009 Progress Report and Results.

Status of Product Development 2008 -- Field data collection was completed August, 2008.  Although we have had little time to work on the data, we have produced some “preliminary” spatial models.  In this regard, binary classification trees were used to describe the presence/absence of each given vegetation type infested and non-infested as a function of topographic data and the satellite imagery using methods described by Joy et al. (2003).  A preliminary model of one of the new study sites is shown below, which displays a preliminary land cover map of an area near Fairbanks, Alaska measuring 565,651 ha in size.

This model is much larger spatial extent than those developed using 2007 data (565,651 ha vs 19,981 ha).  Accuracies ranged from a high of 0.956 (recently burned area) to a low of 0.214 (herbaceous cover). Aspen leaf minor had an overall accuracy of 0.74, while spruce had an accuracy of 0.68. The models will require some adjustments as well as an evaluation of the satellite imagery used in developing the models.  Much more work will be needed through the winter.

Spatial distribution model of vegetation types and infested and noninfested species using data collected in 2008.

Figure 2.  Spatial distribution model of vegetation types and infested and noninfested species using data collected in 2008.

Status of Product Development 2007.  During the months previous to this 2008 summer’s efforts, we continued to work with the data collected during summer 2007 with the aim of refining procedures based on Landsat imagery, to adapt these methods to MODIS, and to measure associated error.  We compared and contrasted spatial models based on MODIS and Landsat.  Statistical error and model accuracy are not subjects exciting to everyone.  An abbreviated version of our findings is shown/discussed below.  A more complete review of the procedures and methods used is attached as Appendix A.

            Model Development (Procedures explained in Appendix A).

Figure 3

Figure 3. Spatial distribution of various vegetation types and infested and non-infested aspen on study site number 1 near Fairbanks and generated using two different satellite sources.  Landsat (left) vs MODIS (right)

Table 1. Estimated areas associated with the Landsat-5 TM model for the various vegetation types on the study site near Fairbanks, Alaska.

Vegetation Type

Area (ha)

Percent

Infected Aspen

4,433

22.2

Birch/Non-infected Aspen

6,216

31.1

Open Areas

363

1.8

Spruce

8,979

44.9

Total

19,981

 

Table 2. Estimated areas associated with the MODIS model for the various vegetation types on the study site near Fairbanks, Alaska.

Vegetation Type

Area (ha)

Percent

Infected Aspen

6,306

30.6

Birch/Non-infected Aspen

3025

14.7

Open Areas

219

1.1

Spruce

11,037

53.6

Total

20,587

 

Error/Accuracy Estimates (procedures explained in Appendix A).

Table 3.  Accuracy estimate for distribution models based on Landsat imagery.

Vegetation Type

Accuracy

Area Under the Curve1

ROC
Ranking1

Infected Aspen

0.80

0.85

Good

Birch/Non-infected Aspen

0.89

0.89

Good

Spruce

0.82

0.89

Good

Open Areas

0.59

0.78

Fair

Overall

0.81

 

 

                       
Table 4.  Accuracy estimate for distribution models based on MODIS.

Vegetation Type

Area (ha)

Percent

Infected Aspen

6,306

30.6

Birch/Non-infected Aspen

3025

14.7

Open Areas

219

1.1

Spruce

11,037

53.6

Total

20,587

 

Table 5. Accuracy assessment of MODIS imagery in identifying vegetation types on the study site near Fairbanks, Alaska.


Tree Cover

Accuracy

Area Under the Curve1

ROC
Ranking1

Infected Aspen

0.60

0.76

Fair

Birch/Non-infected Aspen

0.73

0.77

Fair

Spruce

0.78

0.82

Good

Open Areas

0.59

0.78

Fair

Overall

0.69

 

 

1AUC = 1-0.90 – excellent; 0.80 – 0.89 – good; 0.70-0.79 – fair; 0.60-0.69 – poor; 0.50 – 0.59 – failure.


Background Citations:

Agterberg, F.P. 1984. Trend surface analysis. In:  Gaile,G.L., C.J. Willmott,C.J. (Eds.) Spatial statistics and models, Reidel: Dordrecht, The Netherlands. pp. 147-171

Benedetti, R., G. Espa and G. Lafratta. 2005. A tree-based approach to forming strata in multipurpose business surveys. Discussion Paper No. 5 2005, Dipartimento di Economia, Universita Degli Studi di Trento, Trento, Italy. 17 pp.

Binford, M.W., T.E. Lee, and R.M. Townsend. 2004. Sampling design for an integrated socioeconomic and ecological survey by using satellite remote sensing and ordination. Proceeding of the National Academy of Science 101:11517-11522.

Boyce, M.S., Vernier, P.R., Nielsen, S.E.,  Schmiegelow, F.K.A., 2002. Evaluation resource selection functions . Ecologiacl Modeling. 157:281-300

Brieman, L. J. Freidman, R. Olshen and C. Stone. 1984. Classification and Regression trees. Wadsworth and Brooks, Pacific Grove, CA 358 p.

Brockhaus, J.A. and S. Khorram. 1992. A comparison of spot and Landsat-TM data for use in conducting inventories of forest resources. International Journal of Remote Sensing 13:3035-3043

Efron, B. and R. J. Tibshrani. 1993. An introduction to the bootstrap. Chapman and Hall. New York, 456 pp.

Erdle, T.A. and D.A. MacLean.  1999. Stand growth model calibration for use in forest pest impact assessment.  The Forest Chronicle 75 (1): 141-152

ESRI, 1995. ARC/INFOâ Software and on-line help manual.  Environmental Research Institute, Inc., Redlands, CA.

ESRI. 1998.  ArcViewâ 3.1.  Environmental Research Institute, Inc., Redlands, California.

Gesch, D., Oimoen, M., Greenlee, S., Nelson, C., Steuck, M. Tyler.D., 2002. The national elevation dataset. Photogrammetric Engineering & Remote Sensing 68,5-32.

Guisan, A., and N.E. Zimmermann. 2000. Predictive habitat distribution models in ecology.  Ecological Modelling 135: 47-186

Hevesi, J.A., Istok, J.D., Flint,A.L., 1992.  Precipitation estimation in mountainous terrain using multivariate geostatistics. Part I: structural analysis. Journal of Applied Meteorology 31, 661-676.

Husser, S.M., F.J. Breidt, and W.A. Fuller. 1997. Design and estimation for investigating the dynamics of natural resources. Ecological Application 8:234-245

Hyyppa, J., M. Hyyppa, M. Inkinen, M. Engdahl, S. Linko and Y.H. Zhu. 2000. Accuracy comparison of various remote sensing data sources in the retrieval of forest stand attributes. Forest Ecology and Management 128:109-120

Joy, S. M., R. M. Reich, R. T. Reynolds. 2003. A non-parametric supervised classification of vegetation types on the Kaibab National Forest using decision trees. International Journal of Remote Sensing. 24:1835-1852

Kravchenko, A. and D.G. Bullock. 1999. A comparative study of interpolation methods for mapping soil properties. Agronomy Journal 91: 393-400

Laurent, E.J., H. Shi, D. Gatziolis, J.P Lebouton, M.B. Walters and J. Liu. 2005. Using the spatial and spectral precision of satellite imagery to predict wildlife occurrence patterns. Remote Sensing of Environment 97:249-262

Lee, N.J. and K. Nakane. 1996. Forest vegetation classification and biomass estimation based on Landsat TM data in a mountain region of west Japan. In: H.L. Gholz, K.

Nakane and H. Shimoda, Editors, The use of remote sensing in the modeling of forest productivity, Kluwer Academic Publishers, Dordrecht, The Netherlands (1996), pp. 159-171

Lundquist, J.E.  2005. Landscape Pathology – Forest Pathology in the Era of Landscape Ecology. pp. 155-165.  In: J.E. Lundquist and R.C. Hamelin, eds. Forest Pathology -- From Genes to Landscapes.  APS Press, St. Paul.  175 p.

Lundquist, J.E. 2007. The relative influence of diseases and other small-scale disturbances on fuel loading in the Black Hills. Plant Disease 91:147-152.

Lundquist, J.E. and R.M. Reich. 2006. Tree diseases, canopy structure and bird distributions in ponderosa pine forests.  Journal of Sustainable Forestry 23 (2): 17- 45.

Lundquist, J.E. and R.M. Reich. Predicting the landscape spatial distribution of fuel-generating-diseases, insects, and other types of disturbances. Jnl Sustainable Forestry (in review).

Muukkonena, P. and J. Heiskanenb. 2005. Estimating biomass for boreal forests using ASTER satellite data combined with standwise inventory data. Remote sensing of Environment 99:434-447

Rabus, B., M. Eineder, A. Roth, and R. Bamler. 2003. The shuttle radar topography mission—a new class of digital elevation models acquired by spaceborne radar. ISPRS Journal of Photogrammetry & Remote Sensing 57:241-262

Reich, R. M., J. E. Lundquist, V. A. Bravo. 2004a. Spatial models for estimating fuel loads in the Black Hills, South Dakota, USA. Int. J. Wildland Fire 13:119-129
Reich, R. M., S. M. Joy, and R.T. Reynolds. 2004b. Predicting the location of northern goshawk nests: modeling the spatial dependency between nest locations and forest structure. Ecological Modeling.176:109-113

Reich, R.M. and J.E. Lundquist. 2005. Use of spatial statistics in assessing forest diseases.  pp. 127-143.  In: J.E. Lundquist and R.C. Hamelin, eds. Forest Pathology -- From Genes to Landscapes.  APS Press, St. Paul.  175 p.

Reich, R.M., C. Aguirre-Bravo and M.A. Mendoza-Briseño. 2007. An Innovative approach to inventory and monitoring of natural resources in the Mexican State of Jalisco. Journal of Environmental Monitoring and Assessment. doi: 10.1007/s10661-007-0086-4 (Online)

Reich, R.M., Davis, R.A., 1998. On-line spatial library for the S-PLUSã statistical software package.  Colorado State University, Fort Collins, Colorado.

Reich, R.M., J.E. Lundquist, and V.A. Bravo. 2004. Spatial models for estimating fuel loads in the Black Hills, South Dakota. International Journal of Wildland Fire 13:119-129.

Schloeder, C. A., N. E. Zimmermann and M. J. Jacobs. 2001. Comparison of methods for interpolating soil properties using limited data. American Society of Soil Science Journal 65:470-479

Smith, A.H., E.A. Pinkard, C. Stone, M. Battaglia, and C.L. Mohammed. 2005. Precision and accuracy of pest and pathogen damage assessment in young eucalypt plantations. Environmental Monitoring and Assessment 111: 243-256

Steininger, M.K. 2000. Satellite estimation of tropical secondary forest above-ground biomass: Data from Brazil and Bolivia. International Journal of Remote Sensing 21, 1139-1157

Stone, M. 1974. Cross-validatory choice and assessment of statistical predictions. J. Royal Statistical Society B 36:111-147

Williams, M.S. 1997. A regression technique accounting for heteroscedastic and asymmetric error. Journal of Agriculture, Biology and Environmental Statistics 2:108-129

Project ID: FY08TS53