4 Forest Survey Sampling Designs: A 27 Joint Annual Forest Inventory and History Monitoring System: The North Central Perspective

W.E. Frayer and George M. Furnival Ronald E. McRoberts

Sampling designs used in national Developed jointly by two of the Forest forest survey programs have been Service’s regional research stations, changing with those programs since the system’s North Central the 1930s. That history gives context implementation is discussed. to changes now under way.

11 Adopting an Annual Inventory 33 Analytical Alternatives for an System: User Perspectives Annual Inventory System

Paul C. Van Deusen, Stephen P. Francis A. Roesch and Prisley, and Alan A. Lucier Gregory A. Reams

For the new annual system to Differences between annual and succeed, it will need increased periodic inventory approaches require participation from the states, internal revisiting previous assumptions about changes in emphasis, and greater use the spatial and temporal trends in the of technology. data.

16 Rationale for a National Annual 38 Improving Forest Inventories: Forest Inventory Program Three Ways to Incorporate Auxiliary Information Andrew J.R. Gillespie Andrew P. Robinson, David C. Hamlin, and Stephen E. Fairweather Changing from a periodic to an annual inventory system for the The potential benefits of integrating Forest Inventory and Analysis (FIA) auxiliary information into estimates of program has significant implications forest stand inventory are attractive; for traditional and new users. three popular techniques are presented.

21 Annual Forest Inventory: 44 Multistage Remote Sensing: Cornerstone of Sustainability in the Toward an Annual National South Inventory

Gregory A. Reams, Francis A. Raymond L. Czaplewski Roesch, and Noel D. Cost With much of FIA data out of date, a Inventory data from the new Southern national remote sensing program is Annual Forest Inventory System will needed. Multistage sampling designs form the basis of state, regional, and could provide cost savings and new national sustainability assessments. products.

Biometrics and Inventory History

Forest Survey Sampling Designs A History

ne of the first signals of a need vey of the forest resources of the coun- Extensive inventories of forested lands in the United for a national forest inventory try…No such survey has ever been States were begun in the early part of the 20th appears in Greeley’s (1920) made.” It was pointed out that “data… century, but widespread, frequent use was not O Timber Depletion and the Answer: have been compiled from a great vari- common until after World War II. Various sampling The original forests of the United ety of sources secured for different pur- designs have been tried; some have proved States are estimated to have covered poses by different organizations with efficient for estimating certain parameters but not 822 million acres and to have con- varying degrees of accuracy.” others. Some designs, though efficient in many tained 5,200 billion board feet of tim- The second milestone report, the respects, have been abandoned because of their ber.…There are left today about 137 Copeland Report, was prepared by the million acres of virgin timber, 112 mil- complexity.Others, while possibly not demonstrat- Forest Service in 1930 and presented to lion acres of culled and second-growth ing high efficiency, have been adopted because of Congress in 1933. It included new data timber large enough for sawing, 133 that had become available to supple- their simplicity. This is a history of these million acres partially stocked with ment the data of the Capper Report, but applications. smaller growth, and 81 million acres of devastated and practically waste there was still no “grand sampling de- land.…Three-fifths of the timber orig- sign” to acquire data. inally in the United States is gone. Early Legislation and Applications By W.E.Frayer and The bulk of the report was a call for The McSweeney-McNary Forest George M. Furnival legislation to protect forestland from Research Act of 1928 had authorized fire and to increase the area of public the Forest Service to conduct a na- land. But one small section of the re- tional forest survey, calling for “a deter- port read “legislation is needed… mination of the present and potential which will permit the Secretary of productivity of forest land.” Because Agriculture to survey the forest re- the main concern in those days was the sources of the United States, determine timber situation, the survey was pri- the present volume together with the marily a timber inventory. The survey present and possible production of began in 1930 (Andrews 1932; Wilcox each class of timber in every important 1938) in Oregon, McNary’s home forest region.…” state, as described by Doig (1976). The remarks in Greeley’s 1920 re- Planning had begun in 1929 when port were expanded in the Capper Re- Thornton T. Munger, the first director port published the same year. The of the Pacific Northwest Forest Experi- USDA Forest Service views this report ment Station, received $30,000 in as the first of a series that incorporated funding (Van Hooser et al. 1992). The new data, and it was considered a mile- first approach in the Pacific Northwest stone in appraising our timber supply. was to use available private data with However, the report admits that “a some fieldwork for verification and comprehensive and fully adequate re- supplementation. Doig described an port…would require an exhaustive sur- experiment carried out in 1930–31 in

4 December 1999 Courtesy of Weyerhaeuser Company Archives, Tacoma, Washington Tacoma, Archives, Company Courtesy of Weyerhaeuser Initial forest surveys were primarily timber inventories. Although extensive sampling began around 1930,not until the late 1940s were sampling designs developed for efficiency and suitability to local conditions.

Lewis County, Washington, in which a Two 1931 reports (Anonymous; Lentz) Forest Service Chief F.A. Wilcox line-plot survey was run to compare it described the line-plot-sample method stated in his 1938 annual report that with compilations made for the area (based on the Lewis County experi- We need in forests all the 630 mil- and to assess its potential for use in the ment) that was proposed for 25 million lion acres we now have and that are South. Plots were one-quarter acre in acres of bottomland in the Mississippi most valuable for forest growth. Yet size and spaced at 10-chain intervals on Delta states. Two interim reports were unless it is abused and neglected, we east-west strips run through forested made in 1932 (Anonymous; Lentz). probably do not and will not need areas. About 486 miles of survey line Lentz reported that 5,815 plots had more forest land. For 300 years our had been run by the end of June 1931. been taken on 4.4 million bottomland forests have been chopped, burned, The fieldwork included 3,888 sample acres, and Eldredge (1935) noted that and depleted. Yet with care and fore- plots at a cost of $10,448. It was de- field inventory of 75 million acres in thought there seems no excuse for a timber famine of national proportions. cided not to use the method in the the South would be completed by Douglas-fir region because of the April 1, 1935. A second report by El- He also stated that “the Forest Sur- rugged terrain, but the method was dredge (1937) described methodology vey indicates that we have more forests adopted in the East. using three-man crews on parallel com- than we thought…and more forest Some results for Oregon and Wash- pass lines 10 miles apart taking quar- growth.” It was also pointed out that ington were published in 1932 ter-acre plots at 10-chain intervals. He Forest Survey had begun in 1930, and (Cowlin). A later report was published stated that analysis had begun and that about one-half of the forests of the for the Douglas-fir region (Anony- some results had been published. United States had been surveyed, with mous 1934). Wieslander (1935), head In 1938, Garver reported that the about 60 percent of the resulting data of Forest Survey at the California For- forest inventory phase was well along compiled. est and Range Experiment Station in the South, Pacific Northwest, and With the outbreak of World War II, from 1935 to 1950, described the first the Lake states. He said the job was this status remained for several years. steps of Forest Survey in California. half done; 289 million acres of forest- The 1940 Yearbook of Agriculture pub- He said that mapping was in progress land had been examined. There was no lished several tables of information on and that there was a plan to take information on sampling design or in- the status of the nation’s forests by R.E. 35,000 field plots, but this was not tensity. Cunningham (1939a,b) re- Marsh, acting chief, and William H. done until after 1950. ported that the inventory phase of Gibbons, senior forester, Division of In 1930 “Cap” Eldredge was placed Forest Survey in the Lake states had Forest Economics, Forest Service. They in charge of Forest Survey in the South, been completed in 1937, and he re- had drawn on reports and unpublished headquartered at the Southern Forest ported preliminary statistics on areas manuscripts by many members of the Experiment Station in New Orleans. and volumes. Forest Service: “Where authoritative

Journal of Forestry 5 tensity of field plots was used in saw- timber stands than in poletimber stands, and a higher intensity in pole- timber than in seedling–sapling stands. In the Pacific Northwest, estimating volumes in old-growth stands pre- sented a different type of challenge than that faced in the East. Floyd John- son, a statistician at the Pacific North- west Forest and Range Experiment Sta- tion, began working on this challenge (Johnson 1950; Johnson and Hixon 1952). The approaches taken in Forest Survey were then, and are now, some- what different at the different experi- ment stations. The large ownerships and high volumes in parts of the West often required a cooperative approach between agencies and private owners, with most wanting the survey results for management planning. The smaller ownerships and generally lower vol- umes by individual owners east of the

Courtesy of USDA Forest Service Rockies enabled the stations there to Remotely sensed data has long been used to enhance statistical and field design broad inventories, produce state information. Aerial photography was a common source by the 1950s, and photo reports, and for the most part leave plots and ground plots were combined to develop double sampling designs. management planning to the individ- ual owners. Although different ap- data on forest conditions such as those the late 1940s, Forest Survey began de- proaches were needed in different geo- so far furnished by the Forest Survey veloping and applying sampling designs graphic regions, the autonomy of the have been available, they have been that were more efficient and better individual stations in the East meant used. Where such data were not avail- suited to local conditions. Cruise lines that differences in techniques appeared able, the best approximations possible, were largely abandoned in favor of per- even when there would have been ad- which are believed to be substantially manent plots located randomly or on a vantages gained by standardization. near the truth, have been made.” The grid. The actual sampling designs used Forest Service considers this data the began to appear more commonly in the Statisticians Join the Action third milestone report. literature in the 1950s. As pointed out The approach of using aerial photo in an excellent portrayal of the roots of plots and ground plots in combination A New Era forest inventory in America by Gregoire to estimate areas and volumes had been Shortly after World War II ended, (1992), one of the first (if not the first) practiced for some time and was prob- Forest Survey began in earnest. In two sampling texts of any kind was prepared ably best described by Bickford (1952). 1946 reports Chief Lyle F. Watts said specifically for forestry by Schumacher He noted that the use of photo plots to that “during 1945 and 1946 the Forest and Chapman (1942). form strata and to estimate their sizes, Service has been making a reappraisal of Becker (1950) described the Forest along with a subsample of those plots the Nation’s forest situation.” Several ta- Survey procedures used in the central being measured as field plots, is a dou- bles were presented, and the general con- states. This effort was a function of the ble sampling design. Specifically, it clusion was that there was enough forest- Central States Forest Experiment Sta- could be called stratified sampling with land but not enough timber. The survey tion in Columbus, Ohio, where Becker estimated stratum weights. He relied was credited to R.E. Marsh and the data was field supervisor for Forest Survey. heavily on Neyman (1938), and were compiled and presented by C. Ed- He noted that a line-plot system was shortly after on Cochran (1953), when ward Behre and S. Blair Hutchison. used in the central states where the ef- the first edition of his popular textbook These reports are the fourth milestone. fort began in 1946. The area estimates was published. By the 1970s, it ap- The four milestone reports pre- were based on aerial photo plots and a peared that some combination of satel- sented national results, but there is lit- subsample of these plots were field- lite imagery, aerial photos, and ground tle information on how the results were sampled to obtain volume estimates. plots would be useful in multiphase or obtained. Most survey efforts based on He also indicated that a type of opti- multistage estimators. Frayer et al. extensive sampling of the nation’s for- mum allocation was used (at least a dis- (1979) and Jeyaratnam et al. (1984) ests began in 1930 or later. Starting in proportional allocation). A higher in- were among many works that appeared

6 December 1999 in the literature at that time. same time, components of change were Simple as it seemed, the use of aerial receiving prime attention (Hall 1959; photos to set up strata sometimes was Beers 1962). A sweeping change to confusing and was not always used to point sampling was accompanied by best advantage. The usefulness of pho- the development of estimation proce- tographs that were a few years old was dures for growth components on re- often questioned, inefficiencies oc- measured points. Most stations were curred when many strata were used, now using a cluster of 10 points and photo interpreters spent a long roughly covering an acre. Although it time classifying the photo plots. Early was not discussed in detail other than work by Bickford showed the advan- in internal Forest Service documents, tages of optimum allocation. However, this cluster was based generally on tests over time, strata change, objectives conducted in the southeastern states. may change, personnel most certainly In the East, it was decided that it change, and all of these factors argued would be reasonable to sample approx- for a simple, easily understood ap- imately 20 trees on each cluster. Be- proach. Proportional allocation be- cause it was generally accepted that 75 came more common over time. square feet of basal area was the mini- mum for a fully stocked stand, some

Sampling Designs Get Attention stations started using 37.5-factor America Association of Consulting Foresters Courtesy of The early 1960s brought changes in prisms (75 square feet divided by 20 Forest surveys have recently broadened design and measurement techniques at trees times 10 points equals 37.5). to include the biological and several experiment stations. Point sam- All of these activities in the early ecological status of forest resources. pling had become very popular (Bitter- 1960s may have been the result of the PC-based resources (such as GPS and lich 1948; Grosenbaugh 1952, 1958; fifth milestone report prepared by the GIS technologies) help to provide and Beers and Miller 1964). Remeasured USDA Forest Service (USDA-FS combine archival and current data. plots were acknowledged as the most 1958). In preparing information for this precise way to estimate growth and 713-page report, data were compiled in New Legislation, New Goals change (Hall 1959). Shiue (1960) and a number of ways: for states that had The 1970s saw continued emphasis Shiue and John (1962) proposed sys- been surveyed since January 1, 1947, placed on Forest Survey. The original tematic sampling with multiple ran- information was based on the survey enabling legislation, the McSweeney- dom starts. To the purist, this approach data; for 10 states in which surveys were McNary Forest Research Act of 1928, had some appeal because it satisfied sta- in progress, the data collected were sup- was first amended by the Forest and tistical theory and, at the same time, plemented with some additional data; Rangeland Renewable Resources Plan- provided a consistent way to locate and for other states, special surveys were ning Act of 1974 and later by the Na- plots on maps, topo sheets, and photos. conducted to gather some information. tional Forest Management Planning The approach was never adopted be- This large effort, coupled with the fact Act of 1976. Finally, the Forest and yond the North Central Forest Experi- that there were now competent statisti- Rangeland Renewable Resources Re- ment Station, however, and the station cians at the experiment stations and for- search Act of 1978 replaced earlier leg- abandoned it to be more consistent estry schools, probably helped provide islation and authorized a continuing, with other stations. Sampling with par- the impetus for the myriad studies and comprehensive, nationwide survey and tial replacement, the most complicated publications of the early 1960s. analysis of all renewable natural re- design ever used by Forest Survey, was Another USDA Forest Service mile- sources. The overall result was more re- implemented at the Northeastern For- stone report was released in 1965. Tim- sponsibility and more funding for For- est Experiment Station (Ware 1960; ber Trends in the United States was est Survey. Forest Survey by this time Ware and Cunia 1962; Bickford et al. based on more complete data than any was known as Forest Resources Evalua- 1963). In time this method was also previous report. It included a descrip- tion Research and later as Forest Inven- dropped, primarily because of its com- tion of a stand-projection procedure tory and Analysis. (For the purposes of plexity (Scott and Kohl 1992). Forest- called the Timber Resource Analysis this article, we use the term Forest Sur- ers in the West were adopting some of System (TRAS) used to standardize vey throughout.) Satellite data were the procedures used in the East, such as data to a common year for publication now readily available (Langley 1971). combining aerial photo information and to provide projections for the fu- Combinations of high-altitude photog- and ground plots (MacLean 1963). ture. Some simulation studies using raphy, low-altitude photography, and Fixed-radius plots were mostly this procedure later showed that rates field samples were used for effective abandoned in favor of point samples of change (harvest and growth compo- three-phase sampling in Forest Survey that provided for precise estimates of nents) are especially critical for such a (Kent et al. 1979; Johnston 1982). It volume by sampling trees with proba- projection procedure to have any preci- was assumed that sampling of many re- bility proportional to basal area. At the sion (Frayer and Jones 1970). sources—not just timber—was needed

Journal of Forestry 7 (Frayer 1974, 1978a; McClure et al. samples in favor of a cluster of four BEERS, T.W. 1962. Components of forest growth. Jour- nal of Forestry 60:245–48. 1979; Furnival 1979). Some (for ex- 1⁄ 24-acre plots spread out over 1.5 BEERS, T.W., and C.I. MILLER. 1964. Point sampling: Re- ample, Scott 1979) thought that in- acres. Field crews map any changes in search results, theory and applications. Research Bul- terim data were needed, as survey cy- land use or distinct changes within for- letin No. 786. West Lafayette, IN: Purdue University. cles varied from less than 10 years in est conditions on the cluster. This ends BICKFORD, C.A. 1952. The sampling design used in the some states to almost 20 years in oth- an era of nearly 40 years in which the forest survey of the northeast. Journal of Forestry 50:290–93. ers. There was talk about producing most popular plot was a cluster of 10- BICKFORD, C.A., C.E. MAYER, and K.D. WARE. 1963. annual estimates (Frayer 1978b). An- point samples. An efficient sampling design for forest inventory: The other milestone report was published, northeastern forest resurvey. Journal of Forestry Outlook for Timber in the United States AFIS and SAFIS Studies 61:826–33. BITTERLICH, W. 1948. Die winkelzahlprobe. Allgemeine (USDA-FS 1973). A stand projection Two important studies are now Forst-und Holzwirtschaftliche Zeitang 59:4–5. system was again used to bring data to under way. In northern Minnesota, the COCHRAN, W.G. 1953. Sampling techniques. New York: a common year (Larson and Goforth Annual Forest Inventory System John Wiley & Sons. 1974). Peden et al. (1973) described (AFIS) is an attempt to use remote COWLIN, R.W. 1932. Areas of types in Oregon and Washington counties. Journal of Forestry 30:504–5. how variance estimators could be used sensing (satellite data in this case) to CUNNINGHAM, R.N. 1939a. Preliminary forest survey with these projections. stratify plots into classes with different figures from the lake states. Journal of Forestry Forest Survey in many ways had probabilities of disturbance. Those 37:66–69. matured. Numerous studies were con- with higher probabilities of distur- ———. 1939b. Lake states forest survey. Journal of For- estry 37:698–700. ducted over the next two decades. bance would have a higher probability DOIG, I. 1976. When the Douglas-firs were counted: Some were done within the survey of being sampled in a given year, The beginning of the forest survey. Journal of Forest units and many were done in conjunc- whereas other plots would be updated History 20:21–27. tion with the Survey Techniques Pro- using models. ELDREDGE, I.F. 1935. The south’s forest survey. Journal of Forestry 33:406–11. ject, which has been located at the The Southern Annual Forest Inven- ———. 1937. Sawtimber and cordwood volumes in Rocky Mountain Forest and Range Ex- tory System (SAFIS) is similar, but central and southwestern Mississippi. Journal of For- periment Station since the late 1970s. probabilities of field plot selection will estry 35:504–5. The maturity of Forest Survey was be equal (proportional allocation). Val- FRAYER, W.E. 1974. The increasing complexity of large forest inventories. Journal of Forestry 72:578–79. evidenced by publications for states, ues for unmeasured plots may be im- ———. 1978a. Objectives of multi-resource invento- parts of states, and periodic milestone puted or estimated from models. Thus, ries in relation to design considerations. In Integrated reports for the nation, as required by the results at first may not be as precise inventories of renewable natural resources, 267–69. enabling legislation. Sampling designs as they might be with the AFIS ap- General Technical Report RM-55. Fort Collins, were now mostly in place for the vari- proach, assuming the stratification by CO: Rocky Mountain Forest and Range Experiment Station. ous stations, although there continued satellite imagery is successful, but the ———. 1978b. Potential uses of sampling with partial to be differences between stations. SAFIS group is avoiding the long-term replacement in national inventories. In Proceedings of complexities of using unequal proba- IUFRO Conference on Forest Inventories, ed. T. Cunia, New Developments bilities with stratum boundaries chang- 16–22. Bucharest, Romania: International Union of Forest Research Organizations. If you were to characterize the de- ing over time. Rapid changes in land FRAYER, W.E., F.A. GRAYBILL, S. JEYARATNAM, D.C. scriptions in each region, you could say use as well as volumes in the South BOWDEN, B.M. KENT, and D.C. JOHNSTON. 1979. that most states have been inventoried probably reduce the gains possible Multilevel sampling designs for resource inventories. Fort with a double sampling design, using from optimum allocation. Collins: Colorado State University. FRAYER, W.E., and D.B. JONES. 1970. Effect of estimated photo plots for stratification and Developments occur rapidly in this inputs on the output of a stand-projection system—a ground plots for volume measure- field. Both of these studies are the Monte Carlo approach. Fort Collins: Colorado State ments. You can readily see many differ- forerunners of an annual inventory University, Department of Forest and Wood Sciences. ences. This is one factor that led to the system mandated by the 1998 Farm FURNIVAL, G.M. 1979. Forest sampling—past perfor- mance and future expectations. In Forest resource in- formation of a blue ribbon panel on Bill that became law while this paper ventories proceedings, 320–26. Fort Collins: Colorado forest inventory and analysis. Its report was in review. State University. (AFC 1992) offered many recommen- GARVER, R.D. 1938. The nationwide forest survey. Jour- dations, including “increase consistency nal of Forestry 36:889–92. Literature Cited GREELEY, W.B. 1920. Timber depletion and the answer. and compatibility among FIA (for- AMERICAN FOREST COUNCIL (AFC). 1992. Report of the Department Circular 112. Washington, DC: US De- merly Forest Survey) units.” A second blue ribbon panel on forest inventory and analysis. partment of Agriculture. blue ribbon panel was assembled in Washington, DC. GREGOIRE, T.G. 1992. Roots of forest inventory in 1997. Its report will recommend an an- ANDREWS, H.J. 1932. Forest survey in the Douglas fir re- North America. Paper presented at the Inventory nual inventory system and express gion. Journal of Forestry 30:264–75. Working Group session of the 1992 Society of Amer- ANONYMOUS. 1931. Second phase of forest survey gets ican Foresters National Convention. strong support for standardizing plot under way. Journal of Forestry 29:616–17. GROSENBAUGH, L.R. 1952. Plotless timber estimates. configuration. ———. 1932. Preliminary forest survey completed in New-fast-easy. Journal of Forestry 50:32–37. The Forest Service has already south. Journal of Forestry 30:624–25. ———. 1958. Point sampling and line sampling. Proba- ———. 1934. Forest survey statistics available. Journal bility theory, geometric implications, synthesis. Occa- moved to standardize plots to a cluster of Forestry 32:879. sional Paper 160. Asheville, NC: USDA Forest Ser- of four fixed-radius plots. In 1995, a BECKER, M.E. 1950. Forest survey procedures for area and vice, Southern Forest Experiment Station. decision was made to abandon point volume determination. Journal of Forestry 48:465–69. HALL, O.F. 1959. The contribution of remeasured sam-

8 December 1999 Sampling Procedures across the United States

The following procedures are in place for each USDA Forest Southern Research Station (SRS) Service research station and are based largely on an excel- Until recently, the Southern Forest Experiment Station lent in-house report (USDA-FS 1992). Although the current had responsibility for states in the mid-South.This station intense interest in monitoring resources has resulted in was recently combined with the Southeastern Station into many changes since 1992, most are still evolving and are yet the SRS, and the work handled by both will now be head- to be described in the literature. Each station’s website is quartered in Asheville, North Carolina. accessible from the Forest Service’s main site at http:// Estimates of timberland area are based on forest-to-non- www.fs.fed.us/links/research.shtml. forest interpretation of plots on aerial photographs.These plots represent approximately 230 acres.The land-use inter- Pacific Northwest Research Station (PNW) pretations are field checked at sample locations represent- The PNW Station covers the West Coast,including Alaska ing approximately 3,840 acres. After using these checks to and California. Responsibility for California was originally as- adjust the photo interpretations, an estimate of the propor- signed to the station in California (now the Pacific Southwest tion of forest-to-nonforest area is made for each county.The Research Station).Two approaches are used. In Alaska, sam- proportion of forest area is combined with US Census land ple populations are first identified by broad vegetation clas- area data to derive county-level forest area statistics. sification based on satellite digital data.Within these popula- Descriptive forest resource statistics are derived from tions, the primary sample consists of a random selection of measurements at permanent sample plots located at the in- satellite pixels transferred to aerial photographs. Items clas- tersections of a three-square-mile grid; each plot represents, sified on the primary photo samples include land class, own- on average, 5,760 acres.The sample plots are remeasured at ership, forest type, and timber volume class. Secondary sam- each survey to allow assessment of change (i.e., growth, re- ples for ground examination are selected from the primary movals, mortality estimates) and of current resource status. samples. All strata are sampled but the sampling intensity on nonforest strata is less than in forested strata. In other Pa- Southeastern Forest Experiment Station cific Coast states, the primary sample is defined by a sys- In the first phase of a two-phase design conducted by this tematic grid of permanent, mapped points. At each grid station (now part of the SRS), a large number of cluster sam- point, aerial photos are used to classify the land into strata ples are interpreted from aerial photographs for forest, non- similar to those used for Alaska. Secondary samples for forest, and non-census water land use. In phase two, a smaller ground examination are selected systematically from the pri- set of cluster samples are centered over each permanent mary sample locations. Secondary sample intensity can be ground sample and classified in the same manner as de- varied to meet special objectives. scribed above and then checked on the ground.The clusters checked on the ground are used to adjust the area estimates Rocky Mountain Research Station (RMRS) from the photo sample. A linear regression is fitted to de- The RMRS has responsibility for the Rocky Mountain velop a relationship between the photo and ground classifi- West and the Southwest.The general approach is a stratified cation of the sub-sample.The entire photo estimate in phase double-sampling design.The primary sample is defined by one is thus adjusted for change in land use since the date of points on a systematic 1,000-meter grid. Each grid point is photography and for misclassifications. located on an aerial photograph for interpretation. Items The second-phase plots are permanent sample plots. All identified for stratification include ownership, land class, and plots on timberland are used for volume-per-acre estimates, forest type group.The interpreted items are used to define number of trees, and stand attributes, as well as estimates of sampling strata.The secondary ground sample is a subset of growth, removals, and mortality. the primary sample at 5,000-meter intervals. A supplemen- tal 5,000-meter field grid is available for sampling intensifica- Northeastern Research Station (NRS) tion as required by cooperators, and additional samples can The former Northeastern Forest Experiment Station has be selected from the 1,000-meter primary grid. responsibility for the states from Ohio to the west and Maryland to the south. A primary sample is obtained from a North Central Research Station (NCRS) grid of photo points overlaid on aerial photographs of the in- The former North Central Forest Experiment Station has ventory area. Interpretation of each photo point is for land responsibility for the Midwestern and Lake states (the Cen- use and timber volume class stratification. A secondary sam- tral States Station was phased out in the 1960s). Using a sys- ple is taken for on-the-ground examination; samples include tematic grid of 121 plots per township (36 square miles or all ground plots measured at the last occasion and new 9,324 hectares) on aerial photographs, each photo plot is ground plots that are added to make the ground sample pro- classified stereoscopically based on land use, forest type, size, portional to the primary sample. Data from all plots, new and and density. Ground plots are a systematic subsample of the remeasured, are combined to calculate a single estimate of photo plots. current volume.

Journal of Forestry 9 ple plots to the precision of growth estimates. 1959. NEYMAN, J. 1938. Contribution to theory of sampling WARE, K.D. 1960. Optimum regression sampling design Journal of Forestry 57:807–10. human populations. Journal of the American Statistical for sampling of forest populations on successive occa- JEYARATNAM, S., D.C. BOWDEN, F.A. GRAYBILL, and Association 33:101–16. sions. PhD dissertation, Yale University. W.E. FRAYER. 1984. Estimation in multiphase designs PEDEN, L.M., J.S. WILLIAMS, and W.E. FRAYER.1973. A WARE, K.D., and T. CUNIA. 1962. Continuous forest in- for stratification. Forest Science 30:484–91. Markov model for stand projection. Forest Science ventory with partial replacement of samples. Forest JOHNSON, F. 1950. Estimating forest areas and volumes 19:303–14. Science Monograph No. 3. of large tracts. Journal of Forestry 48:340–42. SCHUMACHER, F.X., and R.A. CHAPMAN 1942. Sampling WATTS, L.F. 1946. Timber shortage or timber abundance? JOHNSON, F.A., and H.J. HIXON. 1952. The most effi- methods in forestry and range management. Duke Uni- Report of the chief of the Forest Service. Washington, cient size and shape of plot to use for cruising in old versity School of Forestry Bulletin 7. Durham, NC. DC: USDA Forest Service. growth Douglas-fir timber. Journal of Forestry SCOTT, C.T. 1979. Midcycle updating: Some practical ———. 1946. Gauging the timber resource of the United 50:17–20. suggestions. In Forest resource inventories proceedings, States. Washington, DC: USDA Forest Service. JOHNSTON, D.C. 1982. Theory and application of se- 362–70. Fort Collins: Colorado State University. WIESLANDER, A.G. 1935. First steps of the forest survey lected multilevel sampling designs. PhD dissertation, SCOTT, C.T., and M. KOHL 1992. Experiences with de- in California. Journal of Forestry 33:877–84. Colorado State University signing the forest survey of the northeastern United WILCOX, F.A. 1938. Report of the chief of the Forest Service. KENT, B., D. JOHNSTON, and W.E. FRAYER. 1979. Appli- States. In Proceedings of the 100th IUFRO World Con- Washington, DC: USDA Forest Service. cation of three-phase sampling for stratification to gress, Berlin. multiresource inventories. In Forest resource inventories SHIUE, C.J. 1960. Systematic sampling with multiple proceedings, 993–1,000. Fort Collins: Colorado State random starts. Forest Science 6:42–50. University. SHIUE, C-J., and H.H. JOHN 1962. A proposed sampling LANGLEY, P.G. 1971. Multistage sampling of earth re- design for extensive forest inventory: Double system- W.E. Frayer (e-mail: [email protected]) sources with aerial and space photography. In Moni- atic sampling for regression with multiple random is dean, Michigan Technological Uni- toring earth resources from aircraft and spacecraft, starts. Journal of Forestry 60:607–10. versity, School of Forestry and Wood 129–41. SP-275. Washington, DC: National Aero- USDA FOREST SERVICE (USDA-FS). 1920. Timber de- nautics and Space Administration. pletion, lumber prices, lumber exports, and concentration Products, 1400 Townsend Drive, LARSON, R.W., and M.H. GOFORTH. 1974. TRAS: A of timber ownership. Report on Senate Resolution Houghton, MI 49931; George M. Fur- timber volume projection model. Technical Bulletin 311, 66th Congress, 2nd session (The Capper Re- nival is professor emeritus, Yale School of 1508. Washington, DC: USDA Forest Service. port). Washington, DC. Forestry and Environmental Studies, LENTZ, G.H. 1931. The forest survey in the bottom-land ———. 1933. A national plan for American forestry. Sen- hardwoods of the Mississippi delta. Journal of Forestry ate Document 12, 73rd Congress, 1st session (The New Haven, Connecticut. 29:1,046–59. Copeland Report). Washington, DC. ———. 1932. Forest survey of the Mississippi delta re- ———. 1958. Timber resources for America’s future. For- gion. Journal of Forestry 30:1,015. est Resource Report No. 14. Washington, DC. MACLEAN, C.D. 1963. Improving forest inventory area ———. 1965. Timber trends in the United States. Forest statistics through supplementary photo interpreta- Resource Report No. 17. Washington, DC. tion. Journal of Forestry 61:512–16 ———. 1973. Outlook for timber in the United States. MARSH, R.E., and W.H. GIBBONS. 1940. Forest-resource Forest Resource Report No. 20. Washington, DC. conservation. In 1940 yearbook of agriculture, 458–88. ———. 1992. Forest resource inventories: An overview. Washington, DC: US Department of Agriculture. Washington, DC. MCCLURE, J.P., N.D. COST, and H.A. KNIGHT. 1979. VAN HOOSER, D.D., N.D. COST, and H.G. LUND 1992. Multiresource inventories—a new concept for forest The history of the forest survey program in the survey. Research Paper SE 191. Asheville, NC: USDA United States. In Proceedings of the 100th IUFRO Forest Service. World Congress, Berlin.

10 December 1999 Biometrics and Inventory Policy Adopting an Annual Inventory System UUserUserUserser PerspectivesPPerspectiveserspectives

he Forest Inventory and Analy- the Forest Service budget has been al- The nation’s Forest Inventory and Analysis (FIA) sis (FIA) program is one of the located to the inventory of the nation’s program is moving from a periodic to an most important services pro- forest resources, an amount smaller annual forest inventory system.The annual T vided by the USDA Forest Service. For than some regional resource assessment system will better meet the increasing demand the past 70 years, FIA has provided the and planning efforts. for current information and is backed by only comprehensive, scientifically cred- These concerns led to the conven- legislative mandate in the 1998 Farm Bill.For ible data on the extent and condition ing of a blue ribbon panel on FIA in the annual system to succeed,FIA will need of forest resources in the United States. 1991 (BRP I) and a second panel in increased state-level participation,internal These data paint a picture of the na- 1997 (BRP II). The panels were orga- changes in emphasis,and greater use of tional forest resource and serve as the nized by the American Forest & Paper foundation of large-scale policy studies Association and included user repre- technology.FIA personnel should be proud of such as those required by the Resources sentatives across the forestry profes- the public support that led to this opportunity Planning Act (RPA). FIA data are fre- sion: universities, federal and state and respond quickly and decisively to a clear quently used by government agencies, agencies, landowner associations, con- mandate for change. industry, and others in regional and servation organizations, forest products subregional analyses that influence firms, and research centers. The panels major economic and ecological man- assessed the needs and desires of a wide By Paul C. Van Deusen, agement decisions. spectrum of users of FIA data and re- Stephen P. Prisley, and Regardless of their views about for- ports. The broad basis for consensus in Alan A. Lucier est utilization or preservation, most these panel reports lends authority to FIA users agree that these data are es- their conclusions. The BRP II report sential to monitoring a healthy and identified five key findings or recom- productive forest ecosystem. A current mendations: and accurate forest ecosystem inven- 1. Elevate the priority of FIA in the tory is prerequisite to substantive dis- Forest Service program. cussion of issues like sustainability, na- 2. Move to an annual inventory and tional forest policy, carbon sequestra- analysis as quickly as possible. tion, changes in growth and productiv- 3. Fulfill the congressional mandate ity, changes in land use and demo- of reporting on all lands. graphics, ecosystem health, and eco- 4. Concentrate on the collection of nomic opportunities in the forest “core” data until the annual inventory sector. system is fully operational. The FIA user community has be- 5. Develop a strategic plan that will come increasingly concerned by the in- accomplish all of the above with a spec- ability of the USDA Forest Service to ified target date. improve the timeliness of FIA informa- The BRP reports and increasing tion. Obstacles to producing more public concern led Congress to under- timely information have included in- score the importance of this program. creasing the number of variables being Specific wording in the 1998 Farm Bill measured in the field and a flat budget. mandated the annual system and other In recent years less than 1 percent of BRP II recommendations, including

Journal of Forestry 11 the key recommendation that FIA and the development of a cohesive team of more is needed. FIA should still issue the National Forest Health Monitoring project leaders who share this vision state reports every five years or so, but Program (FHM) merge their fieldwork and cooperate to see it accomplished. regular users will be able to continu- to increase the efficiency and effective- The current success in implement- ously monitor the website and answer ness of both programs. ing an annual inventory in the South most of their questions without special and North Central regions is the result assistance from FIA. Ideally, through Annual System of strong state-level participation, as the Internet users will have access to The BRP reports, the 1998 Farm well as leadership and risk-taking in the the same software that FIA uses to do Bill, increasing user demands, and pos- FIA units and national FIA program of- internal analyses. Going online will ex- itive Forest Service response have cre- fice. State crews are collecting much of pand the user base and free FIA per- ated an opportunity for FIA to be sonnel to focus on other things. transformed into a system that can Substantial confusion exists between continue to fulfill its mandate and There can be the change to annual plot-taking and a meet users needs. But the rebirth of concurrent change in the design of the FIA is still incomplete, and compla- no complete picture plots. The Forest Service has recently cency on the part of users and the For- of the country’s changed from a variable-radius plot est Service could destroy the momen- scheme to a fixed-radius mapped plot, tum for change and leave us with a forest ecosystem but this change is independent of the crippled system. The 1998 Farm Bill annual system (Scott and Bechtold did not include specific budget recom- unless all lands 1995). Annual inventory can be con- mendations and the Forest Service and are included. ducted no matter which plot design is in USDA have not yet indicated they are place, but the simultaneous change to a prepared to increase budget share for nationally consistent plot layout has in- FIA in proportion to its importance. the data with joint federal and state creased the cost of inventory, regardless Progress in implementing the an- funding. The old model of predomi- of the timing of field sampling. For ex- nual system has been significant in nant federal control and funding for ample, to compute components of in- some regions. The North Central and FIA can be modified to include greater ventory change, the Southern FIA unit Southern regions have been conduct- state-level participation; in fact, the an- (Thompson 1998) recently has been ing annual inventory pilot studies since nual system will facilitate this modifica- measuring all plots twice: once using the 1992 and 1995, respectively, and are tion. FIA may someday evolve into an old system (for consistency in computa- therefore farther along than other re- organization whose primary missions tions) and again under the new fixed-ra- gions. The Southern region has imple- are ensuring data quality, controlling dius plot system (to establish a new basis mented the annual system in several data management, and analyzing re- for successive measurements). The draft states and has a plan for implementa- source conditions and trends. Asking strategic plan for FIA (Anonymous tion in the remaining southern states states to assume more responsibility for 1999) denotes the increasing costs for over the next few years. Likewise, the fieldwork may seem alien to a field-ori- inventory under the annual system, but North Central region has begun imple- ented program such as FIA, but rapidly does not specifically identify the already mentation in Minnesota and is formu- growing needs for more timely and increased costs of conversion to the new lating additional plans. The Northeast- complete resource information demand mapped-plot design. ern region began implementation in that new approaches be considered. Even when conversion to the Maine this year, owing largely to the ef- mapped-plot design is complete, the forts by forestry leaders in that state. Implementing the Annual System cost of fieldwork will be a significant The 1998 Farm Bill sets a goal of The challenges inherent in moving and growing concern. Therefore, FIA implementing the annual system in all to an annual system are real and signif- needs to find ways to generally hold regions on all forestlands within five icant. For example, FIA reporting will these costs down. One option is to years. It is surprising that many na- be forever changed by the annual sys- measure fewer variables on a plot. A tional forests and other federal forest- tem and the Internet. With a periodic plot currently takes about one day to lands are not currently inventoried by system, it was clear that analysis and complete, which should be considered FIA. There can be no complete picture reports should be issued immediately an upper limit that cannot be ex- of the country’s forest ecosystem unless following completion of a state survey. ceeded. Cutting out enough variables all lands are included. All regions With the annual system, every year is to get back to doing two plots per day should be using compatible field proce- the same so there is no need for annual is a good goal but may be unreachable. dures, analysis methods, and software as reports on each state. FIA will need to One alternative (Anonymous 1999) is called for by BRP II and Forest Service place more emphasis on making data more judicious selection of core vari- planning documents (Anonymous and software available on the Internet ables to be measured on every plot, and 1999). Achieving this goal will take so that users can create their own cus- allowances for measuring some vari- strong leadership at high levels in the tom reports. FIA has already made sub- ables on subsets of all plots. Careful Forest Service and USDA, along with stantial progress in this direction, but consideration of necessary precision

12 December 1999 levels might lead to different alloca- the program began. Given the rate of indicating a need to change forest policy tions of field effort. For example, there technological change, the future may in the state. An annual inventory system are fairly well understood and defensi- arrive sooner than we expect, so antici- would provide rapid and ongoing ad- ble guidelines on needed precision for pation is in order. justments to the baseline to evaluate the estimates of wood volume, and these First, we believe that the ongoing dynamics of the forest resource. guidelines typically have driven the se- change to an annual system is more For decades inventory specialists lection of sampling intensity. But once than cosmetic. A simple view is that have wanted to use remote sensing a sampling intensity is chosen to de- the annual system represents nothing technologies to gather more of their in- liver the desired precision for volume more than changing the order in ventory information, but field mea- estimates, it is not necessary that all which plots are visited. We think an surements will still be required for the other variables (soil types, textures and annual inventory is a giant step to- foreseeable future. The current plot de- erosion classes, ground cover, tree dam- ward recognizing the increasing im- sign of four circular 1⁄ 24-acre plots 120 age codes and classes, recreation use) be portance of the forest ecosystem to so- feet apart, works well for field crews in recorded at exactly the same intensity. ciety. It requires nonfederal partners some situations, but we suspect that it Another alternative to save field to play a greater role in conducting is not ideal as ground truth for re- sampling costs is to measure fewer plots the inventory and makes possible on- motely sensed data with a resolution of each year. The current plan for the an- going participation by FIA users in less than 10 meters. If remotely sensed nual system is to measure 20 percent of Forest Service assessments of the state data is to take its rightful place in the the total FIA plots in each state. This is of the forest. The periodic system FIA tool box, the field plots may have a good starting point, but after a while made it difficult for states and other to be redesigned. If ground truth be- FIA could seek ways to reduce this per- users to remain engaged during the comes a primary function of the plots, centage while maintaining acceptable 10-year hiatus between inventories. it will likely be necessary to character- precision. A 20-percent sample may be The periodic system of inventory has ize larger fixed areas (not necessarily by appropriate and necessary in the fast- been troublesome when major distur- measuring every tree) to ensure spatial changing forests of the Southeast, but bances such as hurricanes and ice storms correspondence between remote sens- far less is needed in areas where change have affected the forest resource. For ex- ing data and ground-truth data. In the is occurring at a slower pace. ample, Hurricane Hugo struck the early days of FIA, continuous strip Variability (precision) is determined South Carolina coast just one year after cruising was used. Strip cruising now by sample size and the estimation the sixth forest inventory of the state seems extravagant, and we suspect that process used. The Forest Service should was published (Tansey and Hutchins today’s system of placing one plot every develop estimators for the annual sys- 1988). Immediately the data on the cur- 5,000 to 6,000 acres and making little tem that will take advantage of the an- rent forest conditions were invalidated, use of satellite remote sensing will seem nual nature of the data. Estimating extravagant in the future. mean values of variables independently In fact, FIA is planning to make from one period to the next worked more use of satellite imagery. Initially, well under the old periodic design, but Given the rate of it may be used as a substitute for the it would be wasteful using the annual technological aerial photos that are currently used to system. Data from the previous few estimate the proportions of forest and years contain much information about change, the future nonforest acreage by county using a the current state of the forest, and may arrive sooner double-sampling procedure. More am- should be incorporated into the esti- bitious applications of remote sensing mation process. There is also an oppor- than we expect. are possible, but substantial progress tunity to incorporate models into the may be hindered by the current plot process. Plots that are not measured in design and lack of a far-reaching tech- the current year can be updated with and an interim special inventory was nology vision. Such a vision might lead models to improve current-year esti- conducted. A subsequent report to radical changes in the FIA system mates (Fairweather and Turner 1983; (Sheffield and Thompson 1992) pro- and yield significant improvements in Hansen 1990; Van Deusen 1997). In vided valuable information on the dam- information quality and cost-effective- the ideal future of FIA, analysts will age to the state’s forests. A year later, the ness. We envision a future FIA system combine field measurements, remote seventh survey of the state was com- that provides annual estimates of forest sensing, and models to provide effi- pleted and reported by Conner (1993, area by type that are highly accurate for cient and accurate annual updates of 1998). However, many of the findings survey units and perhaps individual the status of the US forest ecosystem. in these reports were dramatically af- counties. Field plots will continue to fected by the hurricane, as they were play critical roles that include ground Longer-Term Views drawn from changes observed between truth for remote sensing and precise The mandate for an annual inven- 1986 and 1992 field measurements. Yet measurements of growth components, tory is the most significant opportunity these findings have been cited (Huck- tree size, condition variables, and other for change and innovation in FIA since aby 1999), 10 years after the event, as tree and stand attributes.

Journal of Forestry 13 Conclusions and of the broad support they have re- straddle stand boundaries. Forest Science Monograph The nation’s forest inventory system ceived to enable their renewed man- 31:46–61. SHEFFIELD, R.M., and M.T. THOMPSON. 1992. Hurri- is in transition as a result of changing date. They should make the most of cane Hugo: Effects on South Carolina’s forest resource. information needs. The former peri- this opportunity to create a new na- Research Paper SE-284. Asheville, NC: USDA Forest odic inventory had begun to fail in part tional inventory system. Service, Southeastern Forest Experiment station. because of inadequate budgets, but also TANSEY, J.B., and C.C. HUTCHINS Jr. 1988. South Car- because it was designed to meet the olina’s forests. Research Bulletin SE-103. Asheville, Literature Cited NC: USDA Forest Service, Southeastern Forest Ex- needs of an earlier era. Timely forest in- periment Station. ANONYMOUS. 1999. Draft strategic plan for forest inven- THOMPSON, M.T. 1998. Forest statistics for Georgia, ventory information is critical to the tory and monitoring. Washington, DC: USDA Forest 1997. Resource Bulletin SRS-36. Asheville, NC: long-term health and vitality of the re- Service. USDA Forest Service, Southern Research Station. source and the forestry community. At CONNER, R.C. 1993. Forest statistics for South Carolina, VAN DEUSEN,P.C. 1997. Annual forest inventory statis- 1993. Research Bulletin SE-141. Asheville, NC: the agency level, FIA needs to be ele- tical concepts with emphasis on multiple imputation. USDA Forest Service, Southeastern Experiment Sta- vated in importance relative to other Canadian Journal of Forest Research 27:379–84. tion. programs. Internally, FIA needs to ———. 1998. South Carolina’s forests, 1993. Resource break with tradition and become more Bulletin SRS-25. Asheville, NC: USDA Forest Ser- oriented toward data analysis and man- vice, Southeastern Forest Experiment Station. agement while maintaining and im- FAIRWEATHER, S.E., and B.J. TURNER. 1983. The use of Paul C. Van Deusen (e-mail: pvandeus@ simulated remeasurements in double sampling for proving field procedures. National con- successive forest inventory. In Proceedings, Renewable tufts.edu) is principal research scientist, sistency will require better cooperation Resource Inventories for Monitoring Changes and National Council of the Paper Industry and coordination among field units Trends. August 15–19, 1983, Corvallis, Oregon, eds. for Air and Stream Improvement, Inc. and a stronger national leadership. J.F. Bell and T. Atterbury, 609–12. Corvallis: Oregon (NCASI), Tufts University, Department State University, College of Forestry. FIA must seek additional state in- HANSEN, M.H. 1990. A comprehensive sampling system of Civil Engineering, Medford, MA volvement in data collection to free its for forest inventory based on an individual tree 02155; Stephen P. Prisley is associate pro- staff for other tasks. We are confident growth model. PhD dissertation. University of Min- fessor of GIS and forest inventory, De- that FIA can successfully transition to nesota. partment of Forestry, Virginia Tech, HUCKABY, L. 1999. Environmental group targets Jasper an annual system as mandated in the timber. The Beaufort Gazette, February 23. Blacksburg, Virginia; Alan A. Lucier is 1998 Farm Bill. FIA personnel should SCOTT, C.T., and W.A. BECHTOLD. 1995. Techniques senior vice-president, NCASI, Research be proud of their accomplishments and computations for mapping plot clusters that Triangle Park, North Carolina.

So much has changed, and yet much remains the same. Take a journey back in time with

THE USE OF THE NATIONAL FORESTS by Gifford Pinchot, 1907

Reprinted in commemoration of the SAF Centennial “Many people do not know what National Forests are. Others may have heard much about them, but 42-page hardcover book have no idea of their true purpose and $5 plus $2 shipping and handling use.… It is the object of this publica- tion to explain just what they mean, To order by mail To order by credit card what they are for, and how to use send a check or money order to call (301) 897-8720, ext. 106, them.” Society of American Foresters or fax order to (301) 897-3690. Attn.: Linda O’Keefe Or visit SAF’s website at 5400 Grosvenor Lane http://www.safnet.org/market/ FROM THE USE OF THE NATIONAL FORESTS, Bethesda, MD 20814-2198 storebooks.htm WRITTEN BY GIFFORD PINCHOT IN 1907

14 December 1999 Biometrics and Inventory Policy Rationale he USDA Forest Service Forest vice budget of $2.73 billion. The USDA Forest Service Forest Inventory and Inventory and Analysis (FIA) Alternatives for reducing the cycle Analysis (FIA) program is changing to an an- program provides periodic in- with available funds, such as reducing nual inventory system that will operate at re- T formation on status and trends on a the sample intensity or scope of data duced intensity simultaneously in all states variety of parameters describing for- collected, are not acceptable to many every year.This system will provide annual in- ests and forest use: area and location program customers. Program invest- ventory updates in all parts of the country every of forests; structure and composition ments in remote sensing tools have led year,and will make it easier for partners (mainly of forests in terms of species, sizes, and to interesting new products, but cur- state forestry agencies) to collaborate in pro- volume; rates of tree growth, mortal- rent remote sensing technology is not gram planning and implementation.The ity, and removals; patterns of owner- capable of providing the required level ship of forestlands; and information of detail demanded by customers at the change has significant implications for tradi- on harvest efficiency and wood prod- state and regional scales at which we tional and new users of the national inventory uct flows throughout the United operate. For example, FIA is required system. States (USDA-FS 1992). This infor- to report area estimates by detailed mation is of vital interest to numerous classifications of forest type, stand size, customers including managers, poli- and stand volume that are not possible By Andrew J.R. Gillespie cymakers, environmental organiza- to classify with sufficient accuracy tions, business interests, consultants, without substantial levels of ground- scientists, the media, and citizens who truthing. Nevertheless, we will con- are interested in status, trends, stew- tinue to conduct research into ways to ardship, and sustainability of the na- integrate new remote sensing tools into tion’s forested ecosystems. This pro- our inventory system to increase effi- gram is referred to as a strategic inven- ciencies and to develop useful new tory to distinguish it from more local, products. project-level inventories aimed at pro- In recent years, as budgets have re- viding specific information for plan- mained relatively flat or declined (in ning management actions. real terms), FIA units have tended to FIA has historically conducted for- divert ever-increasing shares of their est inventory on a state-by-state cycle. budgets to collecting field data in an at- Under this model, we divide the coun- tempt to reduce or at least maintain the try into zones and, within each zone, current inventory cycle. These steps conduct statewide inventories one have proved insufficient to maintain state at a time for all forestland outside the inventory cycle at the desired eight of national forests. (National forest to 10 years, and have compounded the managers are responsible for forest in- problem by slowing the analysis and ventory within national forests; many publication of inventory results. The contract with FIA to provide consis- result has been increasing customer tent coverage across states.) Past inven- concern that the FIA program is not tory cycles have ranged from six to meeting present customer needs, nei- eight years in the South and 11 to 18 ther in terms of frequency of data nor years in the rest of the country. In in diversity of analytical products. 1998, the total federal appropriation These concerns have been expressed in for FIA was $19.8 million, which was two reviews of the FIA program by a approximately 10.5 percent of the cross section of program customers total research budget of $188 million, (American Forest & Paper Association or 0.7 percent of the total Forest Ser- 1992, 1998) who have encouraged us

16 December 1999 for a National Annual Forest Inventory Program

to reassess the existing program to im- Service or from partners. Several south- be provided to implement the man- prove program performance. ern states have already stepped forward dated program. We are confident, how- One way to address some of the to invest significant amounts of their ever, that we can collaborate with our problems facing the program is to own money and staff time in support partners to craft a program that ad- change to an annual inventory system. of the program (USDA-FS 1999b). dresses customers’ concerns in an effi- This approach has two major attrac- Other FIA units have until now cient fashion, to the best ability of tions. First, it guarantees that all states been interested observers. However, available resources. will have at least some new data avail- the Agricultural Research, Extension, Changing from the periodic ap- able every year, which addresses con- and Education Reform Act of 1998 proach to the annual approach has cerns about data currency. Second, it (PL 105-185) directs all FIA units to tremendous implications for many as- makes it easier for partners to help change to an annual inventory system pects of the FIA program, including lo- share the costs of the program through over the next five years. This law also gistics, cost efficiencies, partnerships, a steady annual level of participation, significantly changes and expands the and the quality and utility of the re- which removes the program from total FIA mission in other ways, requiring sulting information products. The im- fiscal dependency on federal dollars. more data collection on a wider array plications will be viewed differently de- FIA has been seriously considering of parameters, mandating consistency pending on whether one is involved in an annual approach to forest inventory in methods and data across all lands in- implementing the program or is simply since 1992, when we began develop- cluding national forests, and requiring using the results. The following discus- ment of a prototype annual inventory analysis and reporting by states at five- sion touches on some of the major im- system in Minnesota in cooperation year intervals. These requirements plications for program participants and with the Minnesota Department of imply other necessary program changes customers alike. Natural Resources. Satellite imagery that are described in the Strategic Plan analysis was used to differentiate be- for Forest Inventory and Monitoring Logistics and Cost-Efficiency tween plots that could be modeled (1999a). For purposes of this article, Changing to an annual inventory rather than visited, and plots where we assume that FIA will implement the will have different implications for lo- sufficient change had taken place to full extent of sampling required by the gistics and cost-efficiency in different warrant a field visit. By visiting a legislation: 20 percent per year per parts of the country. Under the peri- smaller set of plots annually, and lever- state. odic approach, field personnel must re- aging that information with remotely It is significant that this legislation locate every year or two as they finish sensed data and models, it was hoped was written and passed with little di- one state and begin another. This re- that a state-level report could be pro- rect input from the Forest Service, duces retention of experienced crew duced at more-frequent intervals— which is largely a reflection of how people, who soon tire of the nomadic four years or less—for the same cost. frustrated so many program clients life and seek positions with more sta- In 1996 we began testing and im- have been with the slow rate of change bility. Where we can operate year- plementing a simpler approach in co- within the FIA program. This high round (mainly in the southern half of operation with a coalition of industry level of interest is not to be taken the country and in some flat northern and state partners in the South. We di- lightly. Some aspects of the legislation, states), an annual approach allows for vided the existing set of field plots into such as that mandating visitation of 20 stability in field staff by permanently five overlapping panels, with the intent percent of all plots in all states every stationing them within a certain work- of measuring one full panel each year year (essentially a five-year inventory ing area. This will eliminate the cost of so that each plot would be measured cycle for every state), may not be the constantly relocating people, and will once every five years. Because federal most efficient way to spend public re- provide field staff with an opportunity funding levels could only support a sources. It is also important to note for more normal lives, which should in seven- or eight-year cycle, this ap- that the 1998 act is authorizing legisla- turn lead to greater retention of experi- proach would require additional tion and not appropriation legislation; enced staff and lower training costs for sources of funds, either from the Forest there is no guarantee that funding will replacements. Some of these savings

Journal of Forestry 17 will be needed to offset the increased nual system, we intend to merge the proach, we have more choices for para- travel costs, as crews will now cover sets of plots so that FHM plots will be meters of interest. One approach every part of every state every year. a subset of the annual FIA panel of would be to use some kind of moving In the rest of the country, the com- plots and will be visited by a single average, combining the most recent bination of snow and steep terrain crew during the summer window to observation for all n plots taken over make fieldwork seasonal. In these areas, collect both FIA and FHM data. This the past five years. This would yield an logistics for an annual inventory are merger would allow more-efficient use estimate of the mean value over the more complicated and expensive be- of the FHM funds (about $7 million past five years, which is not the same as cause large numbers of people must be in 1999) for focusing on extended eco- the mean value in the present year. An- moved into these areas to complete the logical data. In addition, the merger other approach currently under consid- 20-percent plot coverage across the re- will reduce the likelihood of multiple eration includes using a variety of gion within the operable window. For visits to the same plot that might modeling or imputation procedures to example, in the interior West, we esti- annoy private landowners, and will in- constantly update the past four years of mate a need for 112 person-years of crease analytical effectiveness by pro- data to the current year, to provide a field crew effort per year to cover the viding maximal linkage between the full data set of n observations for the entire region, or roughly double the two databases. Because many states are current year. Such modeling ap- staffing needed to implement an equiv- partners in both FIA and FHM, com- proaches will rely on historical data or alent five-year cycle under a periodic bining these programs will enable other auxiliary information, which will inventory system (D. VanHooser states to reduce some overhead associ- themselves add components of vari- 1998, pers. commun.). Where crews ated with participation. ance. Research will be needed to quan- once would concentrate on measuring The financial implications of the tify and incorporate the additional all plots in a smaller area, they will now annual inventory system are not lim- variance elements. need to cover a larger area, on average ited to fieldwork. We also expect to Information may be accurate and driving (or walking, in roadless areas) gain some efficiencies in data analysis. precise, yet still not be useful to users. past two plots three miles apart for Data from adjacent states will now be Under the periodic approach, highly every plot they measure. Providing the available over common time periods, accurate and precise inventory infor- coverage required will mean higher which will eliminate the need for in- mation is available only periodically— travel costs. Costs could be reduced by vesting time and energy in complicated at present, every eight to 15 years for concentrating annual fieldwork in sur- and sometimes arbitrary updating each state. For users who need current vey units; that is, dividing a state into processes as a precursor to analyses that information, this is not sufficient if too five subunits and measuring all the span multiple states. The annual ap- many years have passed since the in- plots within each subunit, one subunit proach also offers a low-cost opportu- ventory data were collected, or if a per year. But we believe this approach nity for additional reports to reflect sig- major disturbance event has occurred is not consistent with the intent of the nificant environmental events that since the last inventory. The advantage legislation, which aims to provide an- occur randomly over time, such as of the annual approach for data users is nual statewide updates of forest inven- floods, ice storms, hurricanes, or fire. that it yields some new information tory data. At present we plan to pro- The annual approach provides a con- each year, with the information having vide uniform coverage by dividing all stant platform for responding to un- an average age of 2.5 years (half the plots in a given state into five panels, predictable events without having to length of the measurement cycle) at where each panel provides full state commission or fund a special study. any given time. Users who want con- coverage at approximately even density. tinuous access to relatively recent data One immediate opportunity for in- Information Quality will prefer the annual approach, and creased efficiency is through the Both the periodic and the annual users who can wait longer for more- merger of the FIA program with the inventory approaches are designed to precise data will prefer the periodic ap- field plot portion of the Forest Health provide unbiased estimates of parame- proach. The preference for an annual Monitoring (FHM) program (USDA- ters of interest. However, the parame- approach to inventory will likely be FS, 1998). Currently FHM is a related ters estimated are not necessarily the greatest for systems where the rate of program that collects data on forest same for each approach. Under the pe- change is greatest; for relatively stable health parameters in all implemented riodic approach, the parameter esti- systems, a periodic approach is proba- states on an annual basis during a 10- mates are assumed to describe the state bly more efficient. week summer measurement window. of the forest at some specific point in There is some redundancy between time. The inventory is generally as- Customer Confidence the programs: for example, FHM col- signed to the year in which the bulk of Regardless of how accurate and pre- lects a set of mensurational data that is the data were collected, although in re- cise the data, the information will not largely duplicated on FIA plots, and it ality inventory for some larger states be useful if the ultimate consumers of involves many of the same manage- may involve data collected over two to the information do not believe it is re- ment and supervisory staff that also three years. liable. Accurate information presented manage FIA. As FIA changes to an an- Under the annual inventory ap- in a manner that undermines its own

18 December 1999 credibility is not useful. With the peri- port permanent inventory staff. Over The annual approach allows those odic approach to inventory, we have time, normal employee turnover tends relationships to mature. With opera- maximum precision at fixed intervals. to reduce the number of staff familiar tions in every state every year, FIA and We can say with confidence that the with the FIA program, methods, and partner staff will have the continuous data reflect observations of a trend at opportunities. In addition, partners contact required to build long-term fixed points in time, and that changes who want to seek resources to invest in working relationships. Partners want- that occur between the points are re- FIA are forced to make infrequent re- ing to contribute to enhancing the base flected with some accuracy in the peri- quests to state legislators for large sums, program will be able to seek permanent odic observations. rather than seeking a more modest in- budget allocations and staff to do so, Reporting updated estimates on an vestment on a continuous basis. The and states that do not choose to con- annual basis will inevitably invite com- periodic nature of past inventory pro- tribute resources will still be guaran- parison of the present estimates to the grams often leads to periodicity in rela- teed a base level of federal service. Be- previous years, and will cause conster- tionships between FIA partners, result- cause of annual fieldwork in each state, nation and distrust if there is deemed ing in a program that is forever locked partners will have the opportunity in to be a significant variation from year in a less-productive “still-getting-to- any given year to inject additional re- to year. While statistically understand- know-you” kind of relationship. sources into the program for collecting able, such behavior could nonetheless cause users unfamiliar with technical issues to mistrust and doubt the results. This is more of a risk under the annual Key Forest Inventory Websites paradigm than it is under the periodic, where the lack of mid-cycle data pre- Information about and generated by the national and regional forest inven- vents users from making the same tory efforts described in this article and throughout this issue is available comparison. Auxiliary information in online. the form of models or other assump- USDA Forest Service Forest Inventory and Analysis (FIA) tions may be incorporated to increase http://www.srsfia.usfs.msstate.edu/wo/wofia.htm the precision of the annual estimates, This site links to all other online resources sponsored by this national but such procedures will themselves program, including databases, documents for downloading, and the sites for add additional components of variance the five regional offices.These two online databases are available: that will need to be quantified and in- • National FIA Database Retrieval System, from which one can create corporated into variance estimates. standard or custom output tables by summarizing information from the Eastwide/Westwide database Partnerships • Timber Product Output (TPO) Database Retrieval System, developed in As the Forest Service increases its ef- support of the 1997 Resources Planning Act (RPA) Assessment. forts at collaborative stewardship, FIA An online library contains various documents, including the 1998 busi- is increasingly relying on partnerships ness report, the strategic plan, and the 1992 and 1998 blue ribbon panel re- to accomplish the FIA mission. For ports. Program news and updates in a series of monthly newsletters also are purposes of this article, partnership is available online. defined as a relationship where two or more parties share objectives and pool Forest Health Monitoring (FHM) resources to reach those objectives. http://willow.ncfes.umn.edu/fhm/fhm_hp.htm State forestry agencies in particular Data from 1990 to 1998 are available in various forms: the data itself have historically partnered with FIA by (through 1997), regional highlights for 1998, regional summaries, and fact contributing office space, staff time, sheets. Some of the information here is becoming out of date, having been vehicles, and other resources that allow written up before the strategic plan, and has not yet been updated. Much of FIA work to proceed at a faster pace. this content will be added to the FIA website in the near future. Many states have also contributed re- Annual Forest Inventory System (AFIS) sources to FIA for purposes of collect- http://www.ncfes.umn.edu/4801/afis.html ing additional data beyond the base AFIS is now undergoing development and pilot application in Minnesota. program, for example to intensify the AFIS combines remote sensing, actual measurements of a relatively small plot network or to collect special inter- number of tree stands, and computer models to provide an annual update of est variables on some or all plots. the state’s forest conditions. However, it is often difficult for Southern Annual Forest Inventory System (SAFIS) partners to participate more fully in the http://ncasi1.nerc.tufts.edu:443/projects/safis/ FIA program under the periodic inven- A small site for an important regional project that is the result of a part- tory approach because the long inter- nership between southern states and FIA’s Southern Research Station. vals between repeated inventory activi- ties make it difficult for states to sup-

Journal of Forestry 19 additional data about some issue rele- FIA will move to an annual approach and analysis, then we will be able to vant to their needs, without having to for the next generation of fieldwork, in create a collaborative FIA program wait years until the next inventory for advance of development and testing of that will deliver better information for their state. For organizations looking to the necessary technical program com- many years to come. increase their involvement in FIA, the ponents such as compilation and annual approach to inventory is clearly analysis approaches. We have some Literature Cited preferable to the periodic approach. In work to do to catch up. AMERICAN FOREST &PAPER ASSOCIATION. 1992. The re- fact, full implementation of the 20- Simply changing the order in which port of the first blue ribbon panel on forest inventory and analysis. Washington, DC. percent program envisioned by Con- we visit field plots is unlikely, by itself, ———. 1998. The report of the second blue ribbon panel gress likely will depend on significant to address customer dissatisfaction on forest inventory and analysis. Washington, DC. partner contributions to augment a with the timing and scope of FIA pro- USDA FOREST SERVICE (USDA-FS). 1992. Forest Ser- base federal program. Significant part- gram outputs. Nevertheless, the transi- vice resource inventories: An overview. Washington, DC. ner contributions are already forth- tion to an annual inventory system, if ———. 1998. Forest health monitoring 1998 field meth- coming from many eastern states. made simultaneously with other criti- ods guide. Research Triangle Park, NC. cal changes in the FIA program, in the ———. 1999a. A strategic plan for forest inventory and Conclusion long run will be in the best interest of monitoring. Washington, DC. ———. 1999b. 1998 annual business report for the The change to an annual inventory the largest number of FIA customers USDA Forest Inventory and Analysis Program. Wash- program offers us the opportunity to whose greatest needs are for current ington, DC. simultaneously make other key information and a flexible program changes that are needed to improve the framework. If we can simultaneously FIA program, such as better integra- address the existing problems of in- tion with FHM, increased collabora- consistency in methods and incom- Andrew J.R. Gillespie (e-mail: agillesp/ tion with partners, and increased con- pleteness in coverage, and if we can [email protected]) is leader, Forest Inventory sistency in approaches across all owner- form partnerships to make available National Program, USDA Forest Service, ships. It appears that the political mo- the resources needed to increase the 201 14th Street SW, PO Box 96090, mentum has already determined that timing and scope of data collection Washington, DC 20090-6090.

20 December 1999 Biometrics and Inventory The South Annual Forest Inventory Cornerstone of Sustainability in the South

With many competing uses and large regional he ecological and economic ested in “doing the right thing,” what- sustainability of southern for- ever that may be. shifts in forestland use,the sustainability of ests is in question. Legitimate Changes in the management of southern forests is being questioned.The new T concern spans many public groups, public lands have significantly reduced Southern Annual Forest Inventory System from those concerned about maintain- the level of timber harvest on national (SAFIS) is being implemented to address re- ing biological diversity and the region’s forest lands (USDA-FS 1998). Timber gional,state,and national questions regarding reservoirs of plant and animal genetic removals in the South are projected to past,current,and projected changes in the material, to forest landowners who increase sharply over the next several southern forest.The annual inventory manage forests to meet economic and decades in response to harvest reduc- system will provide the information needed to societal needs, to average citizens inter- tions on western public lands (USDA- closely monitor and quantify the landscape dy- namics of southern forests.These annual in- ventory data will form the basis of state,re- gional,and national forest sustainability assessments.

By Gregory A. Reams, Francis A. Roesch, and Noel D. Cost

Hurricanes can instantaneously affect millions of acres of forestland.In 1989, Hurricane Hugo reduced the inventory of softwood growth stock by 21 per- cent in South Carolina (Sheffield and Thompson 1992). Annual forest inven- tories will provide nearly real-time estimates of change following cata-

strophic events. Courtesy of USDA Forest Service

Journal of Forestry 21 Virginia

Kentucky

North Carolina Tennessee Oklahoma Arkansas South Carolina

Alabama Georgia Louisiana Texas Mississippi

Florida

Figure 1. By the end of 1999, eight southern states will be conducting joint state and federally sponsored annual forest inven- tories. All 13 southern states will likely implement the annual survey design by the year 2000.

FS 1995). Current analyses of southern ease, air pollution, and forest fragmen- (RPA) projections of future demands timber projections indicate that for tation. Humankind is suspected of con- for timber products, the consumption some regions in the South, timber re- tributing to climate change, which re- of pulp, paper, and paperboard will movals exceed growth (Cubbage et al. sults in ecosystem stress. continue to rise and may increase by as 1995). The question of whether the Forest sustainability issues are not much as 50 percent by the year 2040 South can maintain or increase current restricted to federal lands. Because (Haynes et al. 1995). levels of production cannot be an- nonfederal forests account for two- Estimating and maintaining cur- swered at this time (Nilsson et al. thirds of the nation’s forested area, they rent forest resources information is 1999). After many decades of sustained will play the predominate role in deter- fundamental to providing real-time inventory growth, southern inventories mining the sustainability of not only monitoring of forest ecosystems. The have leveled off. Increases in invento- the nation’s forests but of our planet as national Forest Inventory and Analysis ries are doubtful given the changing well. The role of nonfederal forested (FIA) program of the USDA Forest demographics and rapid urbanization lands is especially critical in the South, Service is the primary source of infor- of several historically important tim- site of 41 percent of the nation’s tim- mation on the status, trends, and use ber-producing regions. berland; 92 percent of these lands are of the nation’s forests on public and The sustainability of timber and nonfederal. The South provides 67 privately owned lands. The FIA pro- wood fiber supply is only one of many percent of the nation’s pulpwood, 50 gram is administered regionally by five components of the forest sustainability percent of its plywood, 40 percent of research stations, with the Southern issue. Fundamental sustainability de- its hardwood lumber, and 33 percent Research Station responsible for main- pends on ecosystem processes that work of its softwood lumber. Few of these taining current inventories in 13 at varying spatial and temporal scales production statistics are projected to states, Puerto Rico, and the Virgin Is- and humankind’s use and alteration of decrease. The entire country is relying lands (fig. 1). the dynamics and functioning of forest less on the timber resources on public To address the uncertainty of forest systems (Smith 1970). An incomplete lands, resulting in an increased depen- sustainability in the South, the Ameri- yet significant list of ecosystem stresses dence on privately held timber, partic- can Forest & Paper Association exerted by humans include land con- ularly in the South. According to the (AF&PA), Southern State Foresters, version, introduction of insects and dis- most recent Resources Planning Act and the Southern Governors’ Associa-

22 December 1999 tion have recognized the need for a then the program has continued to because the survey is implemented on continuous forest inventory system. provide an unbiased public database a state-by-state basis. This approach The AF&PA was instrumental in con- that be can used by all citizens to esti- can lead to the undesirable circum- vening the second blue ribbon panel mate trends in forest area, distribu- stance of bordering states having data (BRP II) on FIA in October 1997. Key tion, species composition, and other that varies in age by a decade or more. recommendations of BRP II include vital forest statistics. Public users of southern FIA data have elevating the priority of the FIA pro- In the 1970s three pieces of legisla- noted that these data are accurate for gram within the Forest Service, initiat- tion—the Forest and Rangeland Re- two or three years but become increas- ing annual inventories, reporting on all newable Resources Planning Act of ingly unreliable over time. With the forestlands, and exploring partnerships 1974, the National Forest Manage- rapid changes in the status and condi- for delivery of the program (AF&PA ment Act of 1976, and the Forest and tions of forestlands in the South, this 1998). Since BRP II, the Southern Rangeland Renewable Resources Re- system is inadequate. In addition, this State Foresters and the USDA Forest search Act of 1978—expanded the ob- system makes it difficult to observe Service have collaboratively phased im- jectives of the forest survey to include trends, because plots can be remea- plementation of an annual forest in- measurements related to wildlife, ecol- sured at intervals of up to 15 years. In ventory throughout the South. The ogy, aesthetics, and recreation and those cases it is entirely possible that Southern Annual Forest Inventory Sys- other types of human impacts. The important changes and trends are ei- tem (SAFIS) is the result of this part- Farm Bill of 1998 mandated the Forest ther missed or documented many years nership between southern states and Service to implement annual surveys after they occur, which deprives deci- the Southern Research Station’s FIA whereby 20 percent of FIA’s one-sixth- sionmakers of the opportunity to im- program. acre ground plots are measured each plement changes in management or The initiation of SAFIS is an ac- year. These laws are in place to ensure policies. knowledgement that the need for cur- the availability of accurate data and in- There are benefits and costs associ- rent information on changes in south- formation for determining the sustain- ated with an annualized inventory sys- ern forests has never been greater. The ability of forest resources. tem. Some of the benefits include cur- need for maintaining current inventory Since FIA’s inception, the most rent and uniform information across information is evidenced by the Agri- basic goal of the program has been to all states owing to a continuous and cultural Research, Extension, and Edu- provide a strategic survey that estimates seamless sampling program that pro- cation Reform Act (PL 105-185) (the total forest area and gives inventory es- vides annualized monitoring of impor- Farm Bill) of 1998, which congression- timates of broadly defined forest types. tant resource trends across the entire ally mandates FIA to implement an an- Examples include strata means and to- South. Because identical sampling and nual inventory system nationwide. tals of forest types based on tree species modeling efforts are performed each composition, average stand age, tree year, catastrophic events such as hurri- Past and Future Directions size, and ownership. Changes in soci- canes and insect and disease epidemics A chronology of congressional man- ety’s information needs have greatly in- can be observed and accounted for on dates for forest resource assessments is fluenced FIA and the use of its data an annual basis. The greatest benefit is useful for understanding the nearly 70- over the years and will continue to do that SAFIS provides data and methods year metamorphosis of the USDA For- so. For example, key global climate for producing annual estimates, which est Service’s FIA program. The Mc- change issues over sources and sinks of provide critical information for effec- Sweeney-McNary Forest Research Act greenhouse gases in forests now are tive policy and forest management de- of 1928 directed the Forest Service to being investigated through the use of cisions in the South. conduct periodic assessments of the na- FIA data (Heath and Birdsey 1997). On the cost side, compared to the tion’s forest resources. The mission of The move toward national imple- periodic system SAFIS requires addi- this act was to estimate forest area, tim- mentation of an annual forest survey tional resources. In the periodic system, ber volume, growth, and cut. The for- has gained significant justification over it takes an average of two years to col- est inventories were charged with pro- the last several years. Annualized forest lect the survey information for a state. viding the information needed to for- inventory systems fill the need for inte- Before SAFIS, the southern FIA pro- mulate policies and principles for sus- grated assessments that rely on the best gram worked in two out of 13 states at tained forest use. and most current data for identifying any point in time. Under the periodic The McSweeney-McNary Forest trends, relating trends to likely or sus- survey paradigm, about 8.3 percent of Research Act led to the creation of the pected causes and consequences, and the plots in the South were measured in USDA Forest Service’s Southern For- providing possible outcomes of alterna- comparison to 20 percent as mandated est Survey program in the 1930s. (For- tive actions. by the Farm Bill. Additional resources est Survey later became the FIA pro- Regional assessments of resource are required to ensure the quality of the gram.) These initial forest surveys were trends have proved difficult under the survey process, manage data, perform key sources of information for the de- traditional periodic survey system. statistical analyses, and publish reports. velopment of a fledgling pulp and Using the periodic system it takes 10 to SAFIS would not exist without the paper industry in the South. Since 12 years to inventory the entire South, partnership between the Southern State

Journal of Forestry 23 an equal probability design permits greater adaptability and flexibility. To minimize sample design obsolescence, structure should be employed sparingly and with awareness of its undesirable effects; variable probability sampling designs and other complex sampling schemes are less amenable to the multi- ple and changing objectives that long- term monitoring programs must ad- dress, and therefore should be avoided (Overton and Stehman 1996). Simplicity is desirable for many rea- sons. Not only will sample elements change over time (for example, a pine plot becomes a parking lot) but so will the overall objectives. Adding to the call for simplicity is the growing recog- nition that data collected from feder-

Light damage ally funded monitoring programs should be accessible to the public at Moderate and heavy damage large (Cowling 1992). With a relatively simple sample design, it is more likely that valid results and conclusions can be reached by various public users of the FIA database. Figure 2. The distribution of forestland in South Carolina damaged by Hurricane The simplicity and resiliency needs Hugo,by degree of damage. of the southern FIA program have re- sulted in the use of an equal probability Foresters and the Southern Research the efficient estimation of one or two systematic sample design (Roesch and Station’s FIA program, which combines closely related parameters of interest. Reams 1999). This interpenetrating funds with intellectual and physical Fortunately for the continuity as- design uses five annual panels, whereby human resources. The annual survey is pect of FIA, the types of measurements plots measured in year 1 will be remea- currently under way in eight states (fig. that are necessary to estimate forest re- sured in year 6. In this mode of opera- 1). Fiscal year federal dollars in 1999 sources are as valid today as they were tion the survey cycle will always be one for the operation of southern FIA in- 30 years ago, and it is unlikely they will year, and the plot cycle will be five clude $8.2 million to the Southern Re- change tomorrow. Facilitating recog- years. If funding difficulties occur it is search Station, $1.2 million in federal nized objectives and, as new resource likely that a smaller proportion of the cost-share to participating states, and questions emerge, introducing new plots will be measured each year. combined state contributions (for Al- ones into a long-term forest resource Under the annual survey system, abama, Arkansas, Georgia, Kentucky, sampling design is of great importance, core data will be compiled each year Louisiana, South Carolina, Tennessee, both technically and politically. Both into a standard set of tables for each and Virginia) of $2.5 million. aspects must be addressed because soci- state and released in hard copy and ety’s information needs are essential to electronic formats. Data will be re- Design of SAFIS defining the objectives of a federal in- leased within six months of the end of In designing inventory systems it is ventory and monitoring program. an annual measurement period. important to recognize that definitions Another dominant consideration in Every five years a complete analyti- of sustainability change over time and planning a long-term monitoring pro- cal report will be produced for each vary according to location and inter- gram is the inevitability that a highly state. For the 13 states served by the ests. Changes in forest type and condi- efficient sample design—one that opti- Southern Research Station, two or tion have accelerated, and the rapid mizes on one or very few resources of three state reports will be prepared per pace of change likely will continue. interest—will go out of date. Examples year by the FIA program in collabora- The combined effect of real change in forest inventory include the use of tion with state, federal, academic, and and definitional changes calls for a re- stratification and variable probability other knowledgeable individuals. silient and simple sampling frame. This of selection based on volume or value Each state report will document goal is in direct contrast to many oper- per unit area. Design features that in- the following: (1) the current status of ational timber inventories in which the volve complex sample structure create the forest based on the last five years sampling strategy is specifically tied to potentially serious difficulties, whereas of data; (2) trends in forest status and

24 December 1999 condition over the preceding 20 years, New design features give rise to dif- tives are becoming increasingly neces- with emphasis on comparing the most ferent and improved methods of analy- sary. Well planned and executed annual recent data with data from the previ- ses. However, new estimation methods survey systems can provide the basic ous period; (3) timber product output usually undergo both a research and a initial baseline and monitoring infor- information for the state; (4) analysis user group development and phasing mation to address the many scientific and discussion of the probable forces period. Annualized estimates that are and societal issue–driven assessments causing the observed conditions; and most similar to the periodic system es- of sustainability. (5) projection of the likely trends in timates will provide the foundation of Currently, the importance of forests key resource attributes over the next first-generation annual inventory esti- and forest management to the global 20 years, under a range of plausible mates (Roesch and Reams 1999). First- carbon cycle is a controversial subject scenarios. generation estimates will use rolling or being negotiated for the Kyoto Proto- In the transition period from peri- moving average techniques based on col to the United Nations Framework odic design to full implementation of averaging the last five years (panels) of Convention on Climate Change. FIA the annual five-panel design, the fol- data (Reams and Van Deusen 1999; survey data are used to estimate US lowing options for analysis and report- Roesch and Reams 1999). Because in- forest carbon stocks so that sources ing are being considered: (1) produce ventory estimates are based on the five- and sinks of carbon can be identified. estimates based only on those plots year moving average, the perceived This represents a relatively new and measured each year; (2) average the danger of mistaking fluctuations in es- unique use of FIA data and is certainly new panel information with the previ- timates of inventory because of ran- not a traditional one. Such use of these ous periodic information using moving dom sampling with real change is min- data would not have been predicted by average models; (3) complete the first imized. Based on a minimum number inventory specialists even a decade two or more annual panels (at least 40 of assumptions, moving average meth- ago. FIA data are the very foundation percent of all FIA plots) before report- ods have been shown to be identical to of US carbon stock estimates, and in ing current inventory information. estimation techniques used by FIA all likelihood they will continue to These options are being discussed be- under the periodic system (Reams and provide the basic monitoring data for cause circumstances may dictate the re- Van Deusen 1999). Second-generation carbon stock changes both regionally porting of information before the five- methods that involve tree- and plot- and nationwide. FIA is the only na- year analytical report is prepared. level modeling and other modeling up- tional public database that can esti- Assuming a state does not receive dating techniques will be incorporated mate and provide continuous moni- new inventory information until all based on performance and utility to toring of forest carbon stocks in the five panels are measured, some south- the program (McRoberts 1999; Reams United States (Heath and Birdsey ern states will be relying on informa- and Van Deusen 1999; Roesch and 1997; Joyce et al. 1997). tion that is 15 years old. At the very Reams 1999). FIA data will continue to provide least the southern FIA program ex- There are circumstances in which the basic information at the forest pects that a number of FIA customers the five-year moving average will overes- area, plot, and tree level for all types of will either use estimators developed by timate or underestimate current inven- regional, national, and international FIA for items 1 through 3 above or de- tory. These situations are most obvious forest assessments. National resource velop their own. Southern FIA believes when there is either an abrupt shift in assessments such as RPA (Powell et al. it should provide statistical methods inventory or a strong trend in the vari- 1993) and the recently completed for developing interim estimates for able of interest. For example, if a hurri- Southern Appalachian Assessment public users. cane similar to Hugo hit South Carolina (SAMAB 1996) rely heavily, in many Once the new rotating panel design during the measurement of panel 3, in- cases exclusively, on FIA data to de- has been fully implemented, the in- ventory estimates based on a five-year scribe and estimate current conditions creased flexibility in inventory estima- moving average would overestimate in- and trends of forests within a region. tion techniques will be realized. Sev- ventory in the eastern half of South Annual information will allow for eral approaches have been presented Carolina(fig. 2). In this case, basing esti- continued monitoring of forest re- by the scientific community and are mates on the two panels measured after source trends and suspected causes ad- under investigation by FIA scientists the hurricane (panels 4 and 5) would be dressed by the Southern Appalachian and the external users of FIA informa- a reasonable alternative. Other possible Assessment. tion (Van Deusen 1996; Reams and solutions would be to use more-sophis- In the dual realm of strategic inven- Van Deusen 1999; McRoberts et al., ticated time series models that can more tories and landscape-scale assessments, in press; Reams and McCollum, in readily account for trend or discontinu- annual survey systems provide the in- press). Some of the methods can be ities in inventory. formation essential for monitoring re- implemented immediately, and several source conditions and trends. Annual others will need further research and Improved Assessments inventory systems that are cost-effec- pilot testing before implementation is Large-scale assessments of forest sus- tive, are publicly entrusted, and pro- considered (McRoberts 1999; Roesch tainability related to one or more vide unbiased information of forest re- and Reams 1999). major public policy themes or initia- source trends, are requisite for sound

Journal of Forestry 25 strategic planning, management, and MCROBERTS, R.E. 1999. Joint annual forest inventory New York: Springer-Verlag. and monitoring system: The north central perspec- SOUTHERN APPALACHIAN MAN AND THE BIOSPHERE conservation of the nation’s forests. tive. Journal of Forestry 97(12):27. (SAMAB). 1996. The Southern Appalachian assessment MCROBERTS, R.E., M.R. HOLDAWAY, and V.C. LESSARD. technical report. Report 5 of 5. Atlanta: USDA Forest Literature Cited In press. Comparing the STEMS and AFIS growth Service, Southern Region. AMERICAN FOREST & PAPER ASSOCIATION (AF&PA). models with respect to the uncertainty of predictions. USDA FOREST SERVICE (USDA-FS). 1995. The 1993 1998. Report of the second blue ribbon panel on forest In Proceedings of the IUFRO Conference: Integrated RPA timber assessment update. General Technical Re- inventory and analysis. Washington, DC. tools for natural resources inventories in the 21st century. port RM-GTR-259. Fort Collins, CO. COWLING, E.B. 1992. Challenges at the interface be- St. Paul, MN: USDA Forest Service, North Central ———. 1998. Report of the Forest Service: Fiscal year tween ecological and environmental monitoring: Im- Research Station. 1997. Washington, DC. peratives for research and public policy. In Ecological NILSSON, S., R. COLBERG, R. HAGLER, and P. WOOD- VAN DEUSEN,P.C. 1996. Annual forest inventory statis- indicators, eds. D.H. McKenzie, D.E. Hyatt, and V.J. BRIDGE. 1999. How sustainable are North American tical concepts with emphasis on multiple imputation. McDonald, 1,461–480. London: Elsevier Applied wood supplies? Interim Report IR-99-003. Laxenburg, Canadian Journal of Forest Research 27:379–84. Science. Austria: International Institute for Applied Systems CUBBAGE, F., T. HARRIS JR., R. ABT, G. PACHECO, R. Analysis, A-2361 ARMSTER, and D. ANDERSON. 1995. Southern timber OVERTON, W.S., and S.V. STEHMAN. 1996. Desirable de- supply: Surplus or scarcity? Forest Farmer sign characteristics for long-term monitoring of eco- Gregory A. Reams (e-mail: greams/srs_fia@ logical variables. Environmental and Ecological Statis- 54(3):28–39. fs.fed.us) is section head, Remote Sensing HAYNES, R.W., D.M. ADAMS, and J.R. MILLS. 1995. The tics 3:349–61. 1993 RPA timber assessment update. General Technical POWELL, D.S., J.L. FAULKNER, D.R. DARR, Z. ZHU, and and Statistical Techniques, Francis A. Report RM-259. Fort Collins, CO: USDA Forest D.W. MACCLEERY. 1993. Forest resources of the United Roesch is section head, Data Acquisition, Service, Rocky Mountain Forest and Range Experi- States, 1992. General Technical Report RM-234. Fort and Noel D. Cost is SAFIS coordinator, Collins, CO: USDA Forest Service, Rocky Mountain ment Station. USDA Forest Service, Southern Research HEATH, L.S., and R.A. BIRDSEY. 1997. A model for esti- Forest and Range Experiment Station. mating the US forest carbon budget. In USDA Forest REAMS, G.A., and J.M. MCCOLLUM.In press. The use of Station, 200 Weaver Boulevard, PO Box Service global change research program highlights: multiple imputation in the Southern Annual Forest 2680, Asheville, NC 28802. 1991–95, eds. R. Birdsey, R. Mickler, D. Sandberg, Inventory System. In Proceedings of the IUFRO Con- R. Tinus, J. Zerbe, and K. O’Brian, 107–9. General ference: Integrated tools for natural resources inventories Technical Report NE-237. Radnor, PA: USDA For- in the 21st century. St. Paul, MN: USDA Forest Ser- est Service, Northeastern Forest Experiment Station. vice, North Central Research Station. JOYCE, L.A., R.A. BIRDSEY, J. MILLS, and L. HEATH. REAMS, G.A., and P.C. VAN DEUSEN. 1999. The South- 1997. Progress toward an integrated model of the ef- ern annual forest inventory system. Journal of Agricul- fects of global change on United States forests. In tural, Biological, and Environmental Statistics 4(4): USDA Forest Service global change research program 345–59. highlights: 1991–95, eds. R. Birdsey, R. Mickler, D. ROESCH, F.A., and REAMS, G.A. 1999. Analytical alter- Sandberg, R. Tinus, J. Zerbe, and K. O’Brian, 93–96. natives for an annual inventory system. Journal of For- General Technical Report NE-237. Radnor, PA: estry 97(12):33. USDA Forest Service, Northeastern Forest Experi- SMITH, F.E. 1970. Analysis of ecosystems. In Analysis of ment Station. temperate forest ecosystems, ed. D.E. Reichle, 7–18.

26 December 1999 Biometrics and Inventory North Central

Joint Annual ForestForest InventoryInventory and Monitoring System The North Central Perspective

he Renewable Forest and Range- and implementing this system was a The USDA Forest Service is developing proce- land Resources Planning Act of joint effort of NCRS, the Rocky dures for annual forest inventories to establish T1978 requires that the USDA Mountain Research Station, and the the capability of producing annual estimates Forest Service conduct periodic inven- Minnesota Department of Natural Re- of forested area,timber volume,related tories of forestland in the United sources (MN DNR). Shortly after the variables,and changes in these variables.The States to determine its extent and con- system was implemented, the South- inventory system (JAFIMS) features an annual dition and the volume of standing ern Research Station (SRS) imple- sample of measured field plots; remote timber, timber growth, and timber de- mented an annual inventory system sensing; a database of plot and tree informa- pletions. Five separate Forest Inven- that was both similar and dissimilar in tory and Analysis (FIA) programs, lo- key aspects to the NCRS system. Al- tion; logistical procedures for supporting field cated in USDA Forest Service research though the NCRS effort was initiated crews; and an optional function,mechanisms stations, conduct these inventories and before the SRS effort, the political and for updating the status of plots measured in publish summary reports for individ- industrial support generated by SRS previous years.The discussion focuses on ual states. was primarily responsible for placing system implementation in the North Central The quality of periodic inventory annual forest inventories on the na- region. estimates decreases over time because tional FIA agenda. of factors such as changes in land use With passage of the 1998 Farm Bill, and tree growth, mortality, and re- formally known as the Agricultural Re- By Ronald E. McRoberts movals. Quality is further degraded by search, Extension, and Education Re- the effects of conducting inventories in form Act of 1998 (PL 105-185), Con- heavily forested states over multiple gress required that the Forest Service years. FIA clients recognize these defi- conduct annual forest inventories in all ciencies and have proposed solutions, states. The Farm Bill established fur- such as increasing the sampling inten- ther requirements: (1) each year, 20 sity, reducing the period between in- percent of plots are to be measured in ventories, and conducting mid-cycle each state; (2) the annual data are to be updates. Although these solutions made available each year; and (3) might resolve some of the deficiencies, statewide resource reports are to be they are expensive to implement and published every five years. In addition, are a piecemeal approach to dealing the Farm Bill required integration of with the problems inherent in periodic FIA and the Forest Health Monitoring inventories. (FHM) program (Eagar et al. 1991; In the early 1990s, scientists in the White et al. 1992) at the level of plot FIA program at the North Central Re- measurement. FHM is a national pro- search Station (NCRS) formulated gram that uses data from ground plots, concepts that led to implementation aerial surveys, and other sources to pro- of the first large-scale annual forest in- duce annual estimates of the status, ventory system under the auspices of changes, and trends in indicators of the USDA Forest Service. Planning forest health.

Journal of Forestry 27 One result of the Farm Bill has been tablishing a plot in each smaller hexa- tation of the system of the FHM hexa- the virtual merger of the NCRS and gon: (1) if an FHM plot fell within a gons and the lack of relationship be- SRS efforts. Scientists from the two sta- hexagon, it was selected as the grid tween the locations of the hexagons tions have agreed on a common state- plot; (2) if no FHM plot fell within a and the locations of permanent FIA ment of objectives, a common set of sys- hexagon, the plot from the existing field plots. tem functions, and a common name, network of permanent FIA plots the Joint Annual Forest Inventory and Intensification Sample Monitoring System (JAFIMS). The Some states contribute additional common name has been selected to dis- funding to intensify inventories as a tinguish the inventory system developed Plots are way to increase precision, address bio- by the two stations from other ap- logical issues such as growth declines, proaches to annual forest inventories. classified into or investigate the effects of weather “Joint” connotes that the system is phenomena such as droughts, blow being developed and implemented by three categories: downs, and ice storms. Several options more than one station, and “monitor- forested, are available for selecting intensifica- ing” connotes integration with FHM. tion plots. First, if a state wants simply The primary objective of JAFIMS is nonforested, and to increase the precision of the overall to maintain the capability of producing questionable. inventory, a systematic distribution of annual statewide estimates of forested supplementary plots across the entire area, timber volume, related variables, state is appropriate. In this case, sup- and changes in these variables. The sys- plementary plots are established in all tem designed to accomplish this objec- (Vasilevsky and Essex 1977; Schmidt hexagons, and the intensification sam- tive features several distinct functions: 1998) that was nearest the hexagon ple consists of the supplementary plots (1) an annual sample of measured field center was selected as the grid plot; and from a panel whose number is offset by plots; (2) remote sensing for area esti- (3) if neither FHM nor existing FIA a constant number from the panel cur- mation and stratification; (3) a user- plots fell within the hexagon, a new rently measured for the federal base friendly, publicly accessible database of permanent FIA plot was established at sample. plot and tree information; and (4) lo- the hexagon center and selected as the A second option is to select plots gistical field procedures for implement- grid plot. This grid of plots is called the that satisfy species, spatial, or other ing the inventory. In addition, JAFIMS federal base sample and is considered conditions. Minnesota has experi- features an optional function: (5) an equal probability sample; its mea- mented with this option in a unique mechanisms for updating plot and tree surement in the North Central region manner: intensification plots have been information for plots that have not is funded by the federal government. considered for selection on the basis of been measured in the current year. Al- The federal base sample was system- vegetation disturbance. The underlying though these functions generally char- atically divided into five interpenetrat- assumption is that the growth and acterize Forest Service annual forest in- ing, nonoverlapping panels. Each year mortality of trees on well-established, ventory systems, the options selected the plots in a single panel are selected undisturbed plots can be predicted ad- for implementing them may vary by for measurement with panels selected equately for intervals of up to 20 years region. on a five-year rotating basis. Before the using models such as the Stand and field measurement of plots, remotely Tree Evaluation and Modeling System Annual Sample sensed images are examined to classify (STEMS) (Belcher et al. 1982). The The characterization of USDA For- plots into three broad categories: intensification sample would consist of est Service forest inventories as “an- forested, nonforested, and question- supplementary plots selected according nual” is based on the measurement of a able. Whereas nonforested plots receive to three ordered criteria: (1) plots that proportion of plots each year and the at most a cursory check to ensure cor- experienced substantial recent vegeta- capability of producing annual FIA es- rect classification, field crews visit plots tion loss; (2) plots that have not been timates, not on a complete annual in- in the forested and questionable cate- measured in the past 20 years; and (3) ventory of all permanent plots. FIA gories. They measure individual tree at- plots randomly selected from among precision standards require a sampling tributes such as diameter, crown ratio, undisturbed plots. intensity of one plot for approximately and mortality, and record plot level at- every 6,000 acres in the North Central tributes such as land use, forest type, Remote Sensing region (USDA-FS 1970). To satisfy and ownership. Remote sensing techniques are ap- this requirement, the geographical The federal base sample is consid- plied in forest inventories for area esti- hexagons established for the FHM pro- ered an equal probability sample of the mation, for forest–nonforest stratifica- gram were divided into 27 smaller FIA total surface area of a state, with the tion, for post-measurement stratifica- hexagons, each of which contains ap- basis for inference residing in the sam- tion for variance reduction purposes, proximately 5,900 acres. A grid of field ple design. Equal probabilities for plot and optionally for disturbance detec- plots was established by selecting or es- selection result from the random orien- tion. Where available, the Gap Analy-

28 December 1999 sis Program (GAP) (NCASI 1996; plots predicted to be unchanged were sheets, entering and editing remotely Scott and Jennings 1997) is used for found to have experienced no vegeta- sensed and field data, tracking the area estimation and stratification. tion loss in 95 percent of cases. For progress of inventories, calculating es- GAP analyses are generally conducted purposes of disturbance-based sam- timates, creating files for public access, by state agencies or universities, are pling, virtually no cost is associated and storing information for future in- based on two-date, same-year satellite with erroneously selecting a plot for ventories (Hansen 1998). Although in- imagery from the Landsat Thematic measurement that was predicted to be dividual FIA programs may accom- Mapper (Bauer et al. 1994), and clas- disturbed but was found by the field plish these tasks in somewhat different sify pixels across entire states into crew to have experienced no vegetation ways, standardization of the FIA plot strata related to land use, vegetative loss; the plot is simply treated like design and field procedures, agreement cover, and tree density. other undisturbed, measured plots. on a common set of estimates and a The estimate of surface area in each Therefore, although the 67-percent common table format for reporting stratum is calculated as the product of prediction success rate is rather low, purposes, and selection of a common the total number of pixels classified there is little penalty for an incorrect database management system are lead- into the stratum and the 900-square- prediction. However, the penalty asso- ing to greater overall uniformity. meter area per pixel. The area repre- ciated with erroneously predicting a Several points of agreement regard- sented by each plot in a stratum, called plot to be undisturbed is potentially ing database operations have emerged the area expansion factor, is calculated much greater. Such plots likely will not among FIA programs in recent years. as the ratio of the pixel-based area esti- be selected for inclusion in the intensi- First, establishment of the Eastwide mate and the number of plots in the fication sample, and their predistur- (Hansen et al. 1992) and Westwide stratum. Thus, while area expansion bance plot volume will be erroneously (Woudenberg and Farrenkopf 1995) factors across an entire state will aver- carried forward. Fortunately, the pre- database file formats has provided age approximately 5,900 acres by de- diction success rate for this category of common, well-documented, easy pub- sign, there will be some variation plots is very high at 95 percent. lic access to FIA database information. among strata. Second, acceptance of a set of common Remote sensing techniques for clas- Database Operations estimates and of a common table for- sifying plots with respect to vegetation The database consists of plot and mat for reporting purposes establishes change have been developed by MN tree information for all permanent FIA uniformity among FIA programs and DNR to facilitate the optional sample plots and is crucial to inventory esti- provides linkage between resource pub- intensification scheme and to identify mation, analysis, and reporting. Be- lications and the databases. plots that have been harvested between cause many FIA users are more inter- measurement years (MN DNR 1999). ested in the database than in the pub- Estimation For each 30m × 30m Landsat Thematic The properties of the statistical esti- Mapper pixel, the difference between mators used to calculate annual FIA es- the digital values is calculated for each timates depend on the sampling de- spectral band for two sets of imagery Establishing signs used to collect the data. Regard- obtained in different years. The differ- less of the estimation technique em- ences for selected bands are combined the Eastwide and ployed, data resulting from the mea- to calculate index values for each pixel Westwide formats surement of some plots from the fed- using an algorithm that maximizes the eral base sample will be available each correlation between the index and provides easy year. Therefore, the simplest way to ground vegetation change. The mean public access calculate annual FIA estimates is to use and standard deviation of these index only the data from the panel of plots values are calculated, and a pixel-based to FIA database measured in the current year. Such es- map is constructed based on five cate- timates reflect current conditions and gories of deviations of individual pixel information. are based entirely on measured plots, values from the mean. The pixel-based but their precision will be unacceptable map is overlaid on the array of plots, for some variables because of the small and each plot receives a disturbance lished assessments and reports, extracts annual sample size. An alternative is to value. of the database are designed to be a use the data for all plots obtained from The accuracy of disturbance detec- public, accessible, and user-friendly the five most recent panels of measure- tion for plots whose disturbance value medium for transferring information. ments and employ a moving average predicts substantial vegetation loss has FIA programs use database opera- estimator. The advantage of this alter- been partially assessed using plot sheet tions to accomplish a variety of tasks native is that precision is increased be- comments recorded by field crews. such as selecting plots to be measured, cause data for all plots are used for esti- Plots predicted to be disturbed were retrieving and verifying data from pre- mation; the disadvantage is that the es- found to have experienced vegetation vious inventories, preparing field data timates do not reflect current condi- loss in 67 percent of cases, whereas recorders, producing field crew plot tions but rather an average of condi-

Journal of Forestry 29 tions over the past five years. Another These regional models were developed ditional field crew supervisors must be alternative is to update to the current from data collected primarily from hired or additional levels of supervi- year data for plots measured in previ- long-term research plots and have gen- sion must be established. Existing su- ous years and then base the estimates erally been accepted for application in pervisors must travel extensively (or on data for all plots. If the updating the North Central region. Neverthe- use technology to provide oversight re- procedures are unbiased and suffi- less, research to improve the efficacy of motely) or field crews must be granted ciently precise, this alternative increases the growth models has been under- greater independence. the precision of the estimates without taken with several objectives: (1) to cal- •Because field crews will be in mul- the adverse effects of using out-of-date ibrate the models using FIA data rather tiple states simultaneously, uniformity information. Thus, a variety of estima- than data from research plots; (2) to must be developed and maintained tion procedures are available under use current statistical techniques that with respect to field manuals, data JAFIMS, with the moving average were unavailable when the STEMS recorder programs, and editing pro- agreed on as the default. models were developed; and (3) to in- grams. Regardless of the alternative se- corporate a climatic component into •Many smaller, permanent loca- lected, FIA estimates for the federal the models. The hypotheses underlying tions for stationing field crews must be base sample for areas such as counties, the third objective are that incorporat- arranged, the preferred solution being FIA units, or states are obtained as the collocation with other Forest Service sum over strata of within-stratum esti- units or land management agencies. In mates. Within-stratum estimates are addition, each location must be pro- calculated as the sum over all plots Logistically, annual vided with computer, communication, falling within both the stratum and the and support services that otherwise selected area of the product of plot- inventories may could be transported with the field level estimates and the stratum area ex- permit long-term crews as they move. pansion factor. For intensification sam- •Some states will not require a large ples obtained using different designs, cost savings, enough annual sample to justify per- such as disturbance-based sampling, but the initial manent stationing of a full-time crew. separate estimates must be calculated But the requirement to measure plots and combined with the federal base implementation may systematically distributed throughout sample estimates. these states each year may substantially be costly. increase travel costs. Alternatives in- Updating clude contracting for part-time crews Two options for updating plot in- in those states, having crews from adja- formation are under investigation: im- ing a long-term climatic component cent states satisfy the requirement, and puting missing values from a pool of will provide greater spatial precision, maintaining a small number of crews data obtained under similar condi- and incorporating an annual climatic in permanent travel status to cover sev- tions, and predicting individual tree component will provide greater tempo- eral such states. growth using models. The first ral precision. Unpublished analyses in- •Quality assurance and quality method is referred to as imputation dicate that the bias in the models is control issues will require greater atten- (Rubin 1987) and is being investigated both negligible and less than that for tion because of less-intense supervi- by SRS. Imputation is a two-step the STEMS models and that the effects sion, variability among field proce- process: (1) plots not measured in the of imprecision in the model predic- dures, and diversity among field crews current year are matched with a pool tions are very small relative to the vari- and their funding sources. of similar plots measured in the cur- ation among measured plots within a From a logistical perspective, annual rent year; and (2) estimates of current stratum. inventories may permit long-term cost year properties for each nonmeasured savings, but the initial coordination plot are obtained by substituting the Field Logistics will be difficult and the initial imple- properties for a plot selected randomly The requirement to measure plots mentation may be costly. from the pool of similar plots. This every year in all states creates both op- method is particularly appropriate portunities for and obstacles to effi- Conclusion when a large proportion of plots are ciency. The primary opportunities re- In 1998 the initial panel of plots measured each year or when a large late to the advantage of stationing field based on the grid of hexagons was se- number of plots are measured in the crews in permanent locations. The ob- lected for Minnesota, Missouri, Indi- current year that are similar to previ- stacles, however, require complex coor- ana, and Iowa; logistical procedures ously measured plots. dination: were implemented; and field measure- The FIA program at NCRS has •Supervising field crews becomes ments were initiated. Although some used the STEMS (Belcher et al. 1982) more difficult, because they are dis- issues remain to be resolved and analy- growth models to update plots and tributed across an entire region rather ses of efficiencies and precision remain trees not measured in the current year. than concentrated in a few states. Ad- to be completed, JAFIMS appears to

30 December 1999 be a viable solution for satisfying the demands of FIA users for more precise, timely, and accessible FIA information.

Literature Cited BAUER, M.E., T.E. BURK, A.R. EK, P.R. COPPIN, S.D. LIME, T.A. WALSH, D.K. WALTERS. W. BEFORT, and D.F. HEINZEN. 1994. Satellite inventory of Min- nesota forest resources. Photogrammetric Engineering and Remote Sensing 60:287–98. BELCHER, D.W., M.R. HOLDAWAY, and G.J. BRAND. 1982. A description of STEMS—the stand and tree evaluation and modeling system. General Technical Re- port NC-79. St. Paul: USDA Forest Service, North Central Forest Experiment Station. EAGAR, C., M. MILLER-WEEKS, A.J.R. GILLESPIE, and W. BURKMAN. 1991. Summary report: Forest health moni- toring—New England/mid-Atlantic 1991. NE-INF- 115-92. Radnor, PA: USDA Forest Service. HANSEN, M.H. 1998. Database design considerations for the Forest Inventory and Analysis program. Paper pre- sented at Conference on Environmental Monitoring Surveys over Time, April 20–22, University of Wash- ington, Seattle. HANSEN, M.H., T. FRIESWYK, J.F. GLOVER, and J.F. KELLY. 1992. The eastwide forest inventory database: User’s manual. General Technical Report NC-GTR- 151. St. Paul: USDA Forest Service, North Central Forest Experiment station. MINNESOTA DEPARTMENT OF NATURAL RESOURCES (MN DNR). 1999. Changeview. Available online at www.ra.dnr.state.mn.us/changeview/. NATIONAL COUNCIL OF THE PAPER INDUSTRY FOR AIR AND STREAM IMPROVEMENT (NCASI). 1996. The na- tional gap analysis program: Ecological assumptions and sensitivity to uncertainty. Technical Bulletin 720. Re- search Triangle Park, NC. RUBIN, D.B. 1987. Multiple imputation for nonresponse in surveys. New York: John Wiley & Sons. SCHMIDT, T.L. 1998. Wisconsin forest statistics, 1996. Re- source Bulletin NC-183. St. Paul: USDA Forest Ser- vice, North Central Research Station. SCOTT, J.M., and M.D. JENNINGS. 1997. A description of the national gap analysis program. Available online at www. gap.uidaho.edu/gap/new/Publications/GapDescription/. USDA FOREST SERVICE (USDA-FS). 1970. Operational procedures. Forest Service Handbook 4809.11, Chap- ter 10: 11.1-1–11.1-3. Washington, DC. VASILEVSKY, A., and B.L. ESSEX. 1977. An accurate way to select sample plots on aerial photos using ground control. In Proceedings, 1997 Midwest Mensurationists Meeting, 28–29. General Technical Report NC-46. St. Paul: USDA Forest Service. WHITE, D., J. KIMERLING, and S.W. OVERTON. 1992. Cartographic and geometric components of a global sampling design for environmental monitoring. Car- tography and Geographic Information Systems 19:5–22. WOUDENBERG, S.W., and T.O. FARRENKOPF. 1995. The westwide forest inventory database: User’s manual. Gen- eral Technical Report INT-GTR-317. Ogden, UT: USDA Forest Service, Intermountain Research Station.

Ronald E. McRoberts (e-mail: mcrob001@ maroon.tc.umn.edu) is mathematical statistician, USDA Forest Service, North Central Research Station, 1992 Folwell Avenue, St. Paul, MN 55108.

Journal of Forestry 31 Biometrics and Inventory Methodology Analytical Alternatives for an Annual Inventory System

he USDA Forest Service South- been conducting an annualized inven- Methods for analyzing data from the Southern ern Research Station (SRS) tory in Minnesota in cooperation with Annual Forest Inventory System (SAFIS) are Forest Inventory and Analysis the Minnesota Department of Natural discussed.Differences between the annual T Unit (FIA) has initiated an annualized Resources. More recently, the Agricul- inventory approach and the more traditional forest inventory sampling design, the tural Research, Extension, and Educa- periodic approach require that we revisit the Southern Annual Forest Inventory Sys- tion Reform Act of 1998 (PL 105-185) previous assumption that there are no impor- tem (SAFIS). SAFIS was introduced to directed the entire Forest Service to tant spatial and temporal trends in the data. improve estimation of both the current move toward an annualized inventory. Over the next few years,the USDA Forest resource inventory and changes in the Although this article addresses SAFIS resource. Under the previous periodic directly, recent developments have led Service Southern Research Station will be inventory system, individual states to SRS and NCRS scientists joining evaluating models of varying complexity to were inventoried over a two- to three- forces to investigate the challenges and determine the most efficient estimation year period, about every 10 years. opportunities arising from this transi- approach for each variable,at all Many factors, including rapid land use tion to annual inventories. spatiotemporal scales of interest. changes and the intense forest dynam- The plot arrangement for the SAFIS ics in the southern United States, con- sample design resulted from an intensi- tributed to diminished confidence in fication of the National Forest Health By Francis A. Roesch and inventory estimates that were more Monitoring (FHM) grid, which has Gregory A. Reams than a few years old. It was decided been described as a component of a that an annualized inventory system, in global environmental monitoring sam- which data is collected statewide every ple design (Overton et al. 1990; White year, would provide more timely and et al. 1992). The sample plots are lo- useful estimates. We will discuss some cated in a systematic triangular grid of the analytical proposals for data with five interpenetrating panels. One from this system. panel per year is measured for five con- Before the SAFIS effort, the North secutive years. Every five years the Central Research Station (NCRS) had panel measurement sequence reiniti- ates. If panel 1 was measured in 1998, it will also be measured in 2003, 2008, and so on. Panel 2 would then be mea- sured in 1999, 2004, 2009, and so on. The panels will be as well dispersed as possible if we apply them according to the pattern in figure 1. Note that in a triangular grid the cells are hexagonal in shape. The result of this pattern is that each element has no immediate neighbors from the same panel. Implementation of SAFIS requires a transition from one of two variations of a periodic system to the rotating panel design described above. The first of the two variants of the periodic sample de- sign, that found in the western states within the SRS area of responsibility, consists of a collection of three-square- mile grids placed randomly within Figure 1. An interpenetrating pattern for a five-panel design.No element has each survey unit. The survey units are another member from the same panel as an immediate neighbor. of such a size that there are typically

Journal of Forestry 33 The different methods involve varying degrees of compro- mise between simplicity and the desire to sacrifice as little of the historical trend information as possible. The options we consider here are to: 1. eliminate all the old plot locations and start over with a triangular grid 2. delete plot locations until a roughly regular grid results, at the same intensity as the desired grid 3. use a coarse mapping to assign existing plots to the nearest grid point (fig. 2). Subsequent to the coarse mapping, we could: a. delete any extra plots in each grid cell and establish new plots at the center of every empty cell (fig. 3), or b. assign residual plots within one grid cell of an empty Figure 2. Example of a coarse mapping of existing plot cell to the empty cell and establish new plots at the cen- locations. ter of any still-empty cell (fig. 4). Option 1 is the most expensive and option 2 is the least expensive. The variations of option 3 are considered a com- promise because they do not result in a regular grid of points at a fine scale, but they do at a coarse scale. Also, they pro- vide a formal mechanism for assigning the existing locations to a regular grid of cells, and could be analyzed, with caveats, as though the sample consisted of a regular grid. The advantage of option 1 is that we would be starting fresh with nothing messy or complex to compensate for in the future. Also, the entire SRS would operate under a single sample design with no conflicts in concept or analytical pro- cedure. The disadvantages are that it would result in a gap in observations of trend and would necessitate replication of all pre-field work with respect to point location identification and classification. Figure 3. Example of a coarse mapping of existing plot Option 2 has the advantage of being quick and easy to locations after deletions of unneeded plots.Cells containing implement. In addition, trend information will benefit from an “N”would require new plots under option (3a). the continuity of plot locations. The drawbacks include the fact that some clumping of plot locations will occur and, if several within a state. The second variant of the periodic spatial relationships are modeled, there could be potentially sample design, occurring in the eastern SRS states, under- large differences in the analytical procedures between eastern went a number of changes over past decades. Unfortunately, and western states for some variables. not all of these changes have been well documented. The re- Option 3a also has the advantages that trend information sulting pattern of plot locations on the landscape is a some- will benefit from the continuity of sample locations and what irregular grid, of a higher spatial density than desired. some of the work that goes into point location identification SAFIS has presented a few challenges with respect to in- and classification will be reusable. Less clumping of plot lo- ventory goals that can sometimes be at odds. For instance, cations will occur than with option 2 and there will be at an obvious goal would be to ensure that the transition is as most small differences in analytical procedures within the smooth as possible, and another goal would be to implement station. As with option 2, the actual location of plots would the new design as quickly as possible while minimizing cost. not be regularly spaced at the finest scales of measurement. The goal of a smooth transition can conflict with the goal of We note that option 3b retains more of the original plot lo- quickly implementing the new design. To ensure a smooth cations than option 3a. Therefore, trend information will transition, we must maintain temporal consistency and con- benefit to a greater extent under option 3b. Most of the work tinuity for trend estimation. We could argue that this would that has gone into point location identification and classifi- be easier to accomplish if we retained as many of the old plot cation will be reused. Again, as with option 3a, small differ- locations as possible. Although it is true that there is a cost ences of analytical procedures may be necessary. associated with establishing new field plot locations, the quickest and easiest implementation of the new design Analysis would occur if an entirely new grid of sample points is es- As we discuss the different analytical approaches for the tablished across the SRS area of responsibility. SAFIS design, we assume that option 3a above will be used There are numerous methods we might use to choose ex- to assign existing plots to their enclosing cells and that new isting sample point locations for retention in the new design. plots will be established at the centers of all empty cells. We

34 December 1999 The focus here is on estimation of a per acre value (V ) for condition class k under different assumptions of spatial and temporal trend. If we assume that there is no time or spatial trend at the observed scales, then our data model would have the sim- plest form possible, and the overall mean for the five-panel series would provide the best estimator of a per-acre value (V ) for condition class k:

T I J A 1 ∑∑∑ ijtk Vk = Xijtk t = 1 i = 1 j = 1 ASk AP

Otherwise, if we were willing to ignore any spatial trend, Figure 4. Option (3b) would utilize extra plots in adjacent we could calculate the mean within each panel for an esti- cells for assignment to empty cells. mate each year:

recognize that there will be some demand for analyses during H J 1 ∑∑Aijtk years 1 through 4 of the annual design in a given area. For Vtk = Xijtk i = 1 j = 1 these transition years, FIA will provide composite estimates AStk AP formed by the appropriate weighting of estimates from the complete periodic inventory combined with estimates from where AStk = sum of the plot areas sampled in condition class the incomplete annual inventory. In this article, we examine k at time t. procedures to be used once a full series of observations is available; that is, when all five panels have been measured at This approach would, however, provide an inadequate least once. sample for many variables. Rather, we should explore differ- The major difference among the analytical procedures ent models for the time trend to efficiently use the entire being proposed is in the extent to which spatial patterns and five-panel sample. The simplest model, that of no time time trends are ignored. That is, in extensive inventories, trend, would weight the panels equally: such as SAFIS, one may or may not wish to make the usual assumption that spatial location or actual year of measure- 5 ∑ ment within a single panel series is unimportant in the analy- Vk = .2Vtk (1) sis. We will briefly discuss some of the analytical procedures t = 1 being proposed while keeping this perspective at the fore- front. To do this we must provide notation that will allow This is the method used by FIA for periodic inventories the use of a full spatial-temporal model; however, we will oc- in states that required more than one year to inventory. An casionally be able to collapse the model along one or more advantage to using this approach initially is that the current dimensions. software used by FIA would be applicable. One of the measured variables for each plot will be the Because the time duration of measuring all five panels is proportion of plot area in each condition class. A condition somewhat longer than the duration of one to three years per class is defined as the combination of variables that identify state that it took for the periodic inventories, equal weight- different strata. Forest condition classes are at least an acre in ing of plots across panels may have the tendency to mask size and identified by land use, forest type, stand origin, temporal trends. One suggested solution for this problem stand size, stand density, and ownership class (Anonymous has been to form an estimator in which panels that were- 1998). Assume that we seek estimates for each condition measured more recently are weighted more heavily than class observed in a survey unit. Also assume that the hori- those measured earlier. This estimator would take the fol- zontal and vertical positions of cell centers are numbered lowing form: from west to east and south to north, respectively. Let:

hi = horizontal position i (i = 1,…,I ) 5 ∑ vj = horizontal position j (j = 1,…,J ) Vk = wtVtk (2) t = 1 tt = time t (t = 1,…,5) ck = condition class k (k = 1, …,K ) Xijtk = the per-acre value observed at hi , vj , and tt , for ck And w1 through w5 would be weighted so that the sum is Aijtk = the area in acres sampled in ck at hi , vj, and tt equal to 1, such as .1, .1, .2, .3, and .3, respectively. If one used the preceding weighting scheme, it would be analogous 1 If ck occurs at hi, vj , and tt , Cijtk = {0 Otherwise to stating that one has three times as much confidence in panels 4 and 5 being fair representations of today’s condition AP = plot area as panels 1 and 2.

Journal of Forestry 35 2,400

a Robust estimator 2γ b Classical estimator 2γˆ 1,900

1,400

900

400

Ð100 012345678910 11 12 13 14 012345678910 11 12 13 14

Directed distance (h)

Figure 5. (a) Estimated NE/SW variogram for percent basal area spruce/fir. (b) Estimated NE/SW variogram of residuals from the median polish.

A less arbitrary approach would be to attempt to model potentially important relationships in the data, and will at the time trend within a panel series. Van Deusen (in review) times provide a simple explanation for high variability in a presents a mixed estimator that can incorporate increasing measurement of interest. levels of constraints on the derivatives of the time trend, al- To perform a spatial analysis at a particular scale, the first lowing one to model various levels of complexity in the time step is usually a coarse mapping, which is accomplished by trend. The mixed estimator literally mixes two models: the segmenting an area with a specific size grid and pooling the first describes the relationship of observations within each plots within each segment (see Cressie 1991). Next, the me- panel (or time period) and the second describes the time dian polish technique is often used to decompose the value trend. The mixed estimation approach is both powerful and in each cell at each time (Xijtk ) into its assumed components practical for most variables of interest to FIA. A slightly more of an effect common to all cells, spatial effects in two direc- complex formulation than that given by Van Deusen would tions, and a residual: be appropriate to satisfy FIA’s charge to recognize changes in condition class within field plots. This might be considered Xijtk = Ctk + Hitk + Vjtk + Rijtk necessary because, although each plot samples the same amount of surface area, the area of a condition class sampled where: by a plot can vary from zero to the size of a plot. Therefore, Ctk = the “common” effect at time t (t=1,...,5) individual plot averages for a particular condition class have Hitk = the ith horizontal effect (i=1,...,I) different bases of support and should probably be weighted Vjtk = the jth vertical effect (j=1,...,K) accordingly. Rijtk = the residual in cell i,j at time t Finally, as pointed out in Roesch (1994) for the case of forest health monitoring, some variables will display spatial For T time periods, the result is a 1 ×T vector C of com- trends within condition classes in extensive inventories. Spa- mon effects, two matrices (H and V ) of directional effects, tial analyses are of interest any time the measurement of a and a matrix R of residuals. variable is likely to be different solely because of the spatial Although there are ways other than the median polish to location of the observation. In these cases, a larger class of accomplish this decomposition, we do not want the effects models, which include spatial correlation, should be used. to be overly influenced by any outliers present. See Cressie Along these lines we could fully analyze the effect of all of (1991) for a defense of the two-way median polish when the spatial dimensions or we could implicitly undermine the outliers are a potential concern. The matrix R can be evalu- importance of one (or more) of the dimensions by collaps- ated for special cases in the same manner as the more famil- ing it down into the remaining dimensions (as is usually iar residual analysis for regression. Subsequent to the median done for elevation). Having the ability to discover and re- polish we can obtain residuals that are not time-detrended move spatial correlation makes it easier to investigate other by adding the “common” effect for each time period back

36 December 1999 into the residuals: ity to determine the most efficient estimation approach for each variable, at each spatial scale of interest. Wijtk = Rijtk + Ctk Literature Cited We could then treat the matrix W as an independent set ANONYMOUS. 1998. Field instructions for southern forest inventory, version #2. Asheville, of time-series observations, and analyze the time trend. NC: USDA Forest Service, Southern Research Station. CRESSIE, N. 1991: Statistics for spatial data. New York: John Wiley & Sons. One way to ensure that suspected spatial trends have been OVERTON, W.S., D. WHITE, and D.L. STEVENS. 1990. Design report for EMAP (Envi- removed is to estimate the variogram at a series of directed ronmental Monitoring and Assessment Program). EPA/600/3-91/053. Washington, distances. The variogram is the variance of the difference in DC: US Environmental Protection Agency, Office of Research and Development. values, separated by a specific distance and direction, ob- ROESCH, F.A. 1994. Spatial analysis for monitoring forest health. In Proceedings of the section on statistics and the environment of the American Statistical Association, 104–9. served at defined points in space. If s represents an observa- 1994 Joint Statistical Meetings, August 14–18, 1994, Toronto. tion point, h represents a directed distance, and X(s) repre- VAN DEUSEN,P.C. In review. Modeling trends with annual survey data. sents the value of the variable at point s, then the variogram WHITE, D., J. KIMBERLING, and S.W. OVERTON. 1992. Cartographic and geometric is defined as 2γ(h) = var(X(s +h)–X(s)). By plotting estimates components of a global sampling design for environmental monitoring. Cartography of the variogram for different values of h, we can determine and Geographic Information Systems 19:5–22. the magnitude of spatial correlation for a variable at different scales. The classical estimator of the variogram is: Francis A. Roesch (e-mail: froesch/[email protected]) is mathe- matical statistician, and Gregory A. Reams is section head, Re-

N(h) mote Sensing and Statistical Techniques, USDA Forest Service, γ 1 ∑ 2 2ˆ(h) = (xi – x(i+h)) Southern Research Station, 160A Zillicoa Street, PO Box 2750, N(h) i = 1 Asheville, NC 28802. where: N(h) = the number of distinct pairs of points separated by directed distance h x(i+h) = the estimate (or observation) of the variable at the point separated from point i by directed distance h

Any trend in variogram estimates will show the spatial correlation in the variable. For example, figure 5, taken from Roesch (1994), shows the variogram plots before and after a median polish was used to remove spatial correlation from the data. The relatively flat variogram of figure 5(b) shows that the median polish had effectively removed the spatial correlation from the data represented by the variogram in fig- ure 5(a).

Conclusion This discussion of the proposed methods for analyzing data from the SAFIS design has shown how these proposals differ mostly in the level of simplification accepted. Traditionally, FIA has given estimates for survey units, which are fairly ex- tensive areas of land within a state. Plots in a survey unit were measured in one or at most two years. Therefore, it was not only reasonable but necessary to ignore time trend in variables during the execution of the survey. In addition, the survey units were thought to be small enough that spatial trend within the unit was not important for the variables of interest. The new sample design has two profound effects: the impor- tance of the survey unit as a logistical tool is eliminated, and the measurement of plots is spread out over five years. These effects require that we revisit the previous assumptions of there not being important spatial or temporal trends within an area for each variable of interest. Probably we will find that these assumptions are often appropriate and, in these cases, the use of the estimator in equation (1) will be valid. Over the next few years, FIA will be evaluating models of varying complex-

Journal of Forestry 37 Biometrics and Inventory Techniques

Improving FOREST INVENTORIES

hy are we doing all this the population. We will focus on esti- Precise estimates of standing inventory often cruising?” “Do we have mating the total standing volume for are needed in a very short time; for example,in to measure everything?” a forest. appraisal situations for acquisitions and divesti- W “With all the data we’ve been collecting The population is divided into tures.Opportunities to integrate auxiliary infor- for the past 20 years, isn’t there some physical or logical units, called sam- mation into forest stand inventory are consid- way to cut back on the amount of work pling units. Here, the sampling units erable,and the potential benefits are very at- we have to do?” are either plots or trees. The number tractive.This article presents three popular These management questions are of sampling units in the population is techniques for incorporating auxiliary informa- quite familiar, and warranted, because the population size. A subset of the tion into forest inventory:list sampling,stratifi- in forestry we have a wealth of data, units in the population is chosen in models, and rules-of-thumb that can some way for measurement: the num- cation,and Poisson sampling. be used to make our current inventory ber of units selected in the sample is efforts more efficient than ever before. the sample size. Each unit selected in Statistically speaking, inventory forest- the sample is measured to obtain the By Andrew P.Robinson, ers are often in a position to use auxil- variable of interest. David C. Hamlin, and iary information to design and imple- We use the sample to represent the Stephen E. Fairweather ment forest inventories. population, and we scale the sample Auxiliary information is informa- estimate to the population total using tion that is used to develop, support, an expansion factor. This expansion is or execute an inventory design. It is called estimation, and its result is the typically related to, but distinct from, estimate. The correspondence between the “item of interest,” which is the the estimate and the population para- subject of the inventory. The quality meter is uncertain because the esti- of an estimate of the item of interest mate comes from a sample, not from can be vastly improved for the same the whole population. This uncer- overall cost by using auxiliary infor- tainty is called sampling error, and is mation. Despite the widespread ac- quantified by the standard error of the ceptance of auxiliary information in estimate. sampling theory (see Cochran 1977; The standard error is used to calcu- Särndal et al. 1992), techniques that late an α% confidence interval for the feature the use of auxiliary informa- estimate. Imagine that the entire pro- tion often meet resistance in some cir- cedure (including the sample selec- cles of forest inventory. tion) was repeated on the same popu- This article presents three tech- lation a large number of times. Then niques for incorporating auxiliary in- on average α% of the confidence in- formation into forest inventory. Our tervals calculated would contain the emphasis is on the practical application true underlying total, if certain as- of estimating timber volume at the sumptions were true. Formulae for cal- stand and forest levels. As much as pos- culating all these quantities can be sible, calculations have been excluded, found in any basic statistics or forest but are available from the authors on mensuration reference. A fundamental request. criterion for success in sampling is as small a confidence interval as possible Background for a given price, or alternatively, as in- The objective of sampling is to ob- expensive as possible an interval of tain estimates of population parame- fixed length. ters, such as the total, for one or more We are concerned with a particular variables of interest, such as volume, type of auxiliary information: one or without measuring every member of more variables that are (1) closely re-

38 December 1999 Three Ways to Incorporate Auxiliary Information

lated to the value of interest, (2) avail- able for every unit in the population, Table 1. Unit-level data for list sampling example: probability propor- and (3) quick or inexpensive to mea- tional to size with replacement, sample size 8. sure. For example, if the variable of in- Cruise terest is tree volume, then potential Stand Cumulative Random estimate auxiliary variables might include diam- number Acres acres Probability number (tons) eter, height, and species, whereas for 1 45.1 45.1 0.038 stand volume candidates would be as- 2 13.8 58.9 0.011 pect, cover type, and elevation. The ju- 323 81.9 0.019 66.0 2,256 dicious and profitable use of auxiliary 415 96.9 0.012 information in sampling is as much an 5 9.9 106.8 0.008 6 35.6 142.4 0.030 124.0 2,873 art as it is science. Experience and ex- 7 64.9 207.3 0.054 perimentation are both required to dis- 8 24.6 231.9 0.020 cover what is likely to work in any 9 76.5 308.4 0.064 given situation. 10 33.7 342.1 0.028 11 1.2 343.3 0.001 List Sampling 12 18.4 361.7 0.015 13 27.8 389.5 0.023 In industrial forest management we 14 153.8 543.3 0.128 405.0, 425.7 16,478 must frequently provide accurate and 1 15 63.3 606.6 0.053 precise estimates of standing inventory 16 9.6 616.2 0.008 in a very short time. These conditions 17 74 690.2 0.062 are typical in appraisal situations for 18 8.5 698.7 0.007 acquisitions and divestitures, for which 19 1.5 700.2 0.001 time frames of two weeks for fieldwork 20 61.3 761.5 0.051 739.1 5,838 are not uncommon. Fortunately, we 21 11.6 773.1 0.010 often have access to an existing but 22 79.4 852.5 0.066 840.4 8,880 outdated stand-based inventory, so we 23 21.5 874.0 0.018 24 4 878.0 0.003 can use auxiliary information from the 25 0.9 878.9 0.001 existing inventory to acquire a quick 26 14.6 893.5 0.012 and accurate estimate of the current in- 27 41.6 935.1 0.035 ventory. 28 9.2 944.3 0.008 In list sampling the sampling unit 29 27.8 972.1 0.023 949.1 1,691 becomes the stand itself. That is, the 30 9.5 981.6 0.008 31 9.8 991.4 0.008 population consists of a list of stands, 32 68.8 1,060.2 0.057 and the sample size refers to how 33 24.8 1,085.0 0.021 many stands will be visited to measure 34 116.7 1,201.7 0.097 1,159.4 8,155 the variable of interest. In table 1 we Total 1,201.7 1.000 have a population of 34 pine stands from the southern United States, all 20 to 25 years old, comprising a little more than 1,200 acres. The total should certainly be related to the stand tree’s basal area, so work is concen- weight of pulpwood is our variable of area. Second, because we would like to trated on the larger, more-valuable interest. How could we estimate it ef- concentrate our fieldwork in the most trees. Third, we will use list sampling ficiently? important stands—the stands with the to get the PPS sample of stands. Fi- We are going to use several sam- most volume—we will select the stands nally, we will use a “mean of ratios” es- pling techniques in this case. First, we to visit with probability proportional timator: we determine the ratio of tons will consider the area of each stand to to size (PPS). This technique is analo- to acres, and use that value in concert be the auxiliary variable, because for gous to selecting trees on a prism plot with the total acreage to estimate the this age class the volume of timber with probability proportional to each total tonnage. Biometricians will rec-

Journal of Forestry 39 ognize this as the Horvitz-Thompson compute r we simply find the eight ra- (Technically this would be known as estimator. tios of y to x, and take their average. In sampling with probability propor- The first step in selecting our PPS our example r turns out to be 91.36, tional to estimated size, or PPES.) sample of stands to visit is to accumu- which can be interpreted to be 91.36 This tended to concentrate the field- late the stand areas (table 1). We do this pulpwood tons per acre. Multiplying work in the stands of highest value, ei- because the probability of selecting a by X, or 1,201.7 acres, we arrive at an ther because they were large or because stand to visit will be proportional to estimate of 109,783 tons. Later work, they carried a lot of volume per the stand’s area; that is, larger stands which included complete enumera- hectare. The end result was an estimate will have a higher chance of being vis- tion, led to a total of 108,210 tons, of total volume with a 95 percent con- ited than smaller stands. which is very close to the sample-based fidence interval of ±5 percent on a Suppose budget and time con- estimate. property of 16,880 acres. This was ac- straints dictate that only eight stands Calculation of the 95 percent confi- complished by sampling 172 stands can be visited. We pick eight random dence interval for Yest results in an in- out of a total of 1,723 stands divided numbers in the range 0 to 1201.7 and terval of ±18,931 tons, or ±17.2 per- into six carefully defined strata. for each random number we find the cent. How can this be improved? As stand associated with a cumulative area with any sampling method, we can Stratification that is larger than or equal to the ran- probably reduce the standard error by Stratified sampling is perhaps the dom number (table 1). For example, increasing the sample size, so it would2best-known application of auxiliary in- one of the random numbers was 739.1. be helpful to visit more stands on the formation in forest inventory, and ap- The first stand in the list with a cumu- ground. Also, the size of the variance of pears in mensuration texts at least as lative acreage greater than or equal to Yest depends on the variance of the ra- early as 1949 (Chapman and Meyer 739.1 was Stand 20, so it was selected tios; the more constant the ratios, the 1949). Foresters have used stratified for a ground visit. The probability of tighter the confidence interval. This sampling successfully since at least the selecting any stand each time is shown suggests that careful stratification of early 1900s (Schreuder et al. 1993), in table 1. This probability of selection our stands into homogeneous groups and have developed good institutional on any one draw is equal to the area of based not only on age but also on site knowledge of what kind of auxiliary the stand divided by the total area in index, species, stocking, etc., might be information works in stratification. the population. beneficial. Stratified sampling uses auxiliary in- Notice that Stand 14 was selected As an example, this technique was formation to divide the population of twice: because we are sampling with re- used successfully for the appraisal of a interest into two or more subpopula- placement, any of the stands have a Southern Hemisphere plantation for- tions, called strata, where each stratum chance of appearing more than once in est. One notable difference was that should be more homogeneous than the the sample. Stand 14 also happens to instead of using stand area as the aux- population as a whole. By dividing the be the largest stand in the population, iliary variable, an estimate of stand population into several more homoge- making up about 12.8 percent of the volume was used. The estimate was neous strata, we expect that the sam- total area, so it’s not surprising that it made using a yield model that pre- pling within each stratum will be more was randomly picked twice. dicted volume based on age, site index, efficient, and therefore that the esti- The next step in the process is to and original stems per hectare. These mate of the population total will be visit the selected stands on the ground, estimated stand volumes were then improved. Making the stratification and determine for each one an estimate arranged in a list, just as in table 1, and rules is something of an art, and is of pulpwood tons. These estimates are the probability of selecting any partic- where the auxiliary information is ap- shown in the column labeled “Cruise ular stand to visit on the ground was plied in stratified sampling. estimate” in table 1. proportional to its estimated volume. The most common source of auxil- We now have an auxiliary variable, x, for every stand in the population, and an observed value of interest, y, for Table 2. Stratum-level information for the stratified sampling example. a subsample of stands. When the Stratum 1 2 3 stands to visit have been selected with Acres 16 18 14 PPS, the appropriate unbiased estima- Number of plots 8 9 7 tor for the population total Y is Board feet per acre 1,938.5 30,658.2 62,812.3 Standard deviation 1,514.2 7,914.0 18,840.0 S 535.3 2,638.0 7,120.9 Yest = X r (1) y Standard error 27.6% 8.6% 11.3% Proportion 0.3333 0.3750 0.2917 Where Yest is the estimate of the Total volume 31,016 551,848 879,372 total of y (total tons of pulpwood), X is Sy 8,565 47,484 99,692 the observed total of x for the popula- 95 percent confidence tion (total acres), and r is the average interval (total) 19,752 107,416 235,734 ratio of y to x from the sample. To

40 December 1999 iary information for stratified sampling confine ourselves to stratified sampling itself, so it does not need to be known in timber inventory is aerial pho- with simple random sampling within beforehand. In forest inventory, Pois- tographs. Each stand is identified on strata, that is, stratified random sam- son sampling is most often applied to the photograph and assigned to a stra- pling. An estimate is calculated for estimate the total volume of a stand, tum based on density, species composi- each stratum and these estimates are and the sampling units are the trees. tion, average tree size, or age class from combined based on the proportion of Poisson sampling is a good way to get the photograph. This is the familiar the population in each stratum. The an estimate for a highly valued, rela- process of phototyping. All stands computational details can be found in tively infrequent component of a stand within each phototype are treated as a most forest mensuration texts. that might otherwise have to be 100 single stratum. Because density, species Now, suppose we have 48 acres to percent cruised (for example, redwoods composition, and age are all correlated cruise in a small woodlot. An hour with or black cherry). with volume, the resulting strata are an aerial photo lets us break the area The essence of Poisson sampling is usually more homogeneous with re- into three strata of 16,18, and 14 acres calibration: spect to volume than the whole popu- respectively, and a day in the woods 1. A prediction is made of the vari- lation. Digital analysis of remotely gets us the cruise data. Table 2 summa- able of interest for every unit in the sensed data can also be applied to strat- rizes the cruise for each stratum. population. ification. This method has the advan- Combining the stratum estimates 2. A sample is taken and the vari- tage of consistent application of classi- gives us a total volume estimate of able of interest is measured for each fication rules, and has worked well 1,460 mbf for the woodlot, and a 95 unit in the sample. where detailed stand descriptions are percent confidence interval of ±230 3. The estimate of the total is the not required before sampling. mbf. By way of comparison, calculat- total of all the predictions calibrated by Defining strata from photos or dig- ing a total volume from all of the the sample data. ital imagery is not always easy, espe- plots together gives us the same total If the predictions match the mea- cially in areas where many species grow (because the plots per acre are the sured values well, a variable population together in mixed-age stands. Keeping same in each stratum) but results in a can be sampled very efficiently. The a few principles in mind will help: 95 percent confidence interval of prediction is typically a visual guess, •Keep it simple. Define strata that ±541 mbf. To match the stratified which is quick and cheap, and can be can be identified consistently on pho- sample’s confidence interval without surprisingly precise (Avery and tos. The objective is to reduce variabil- stratifying would require about 70 Burkhart 1994). The sample is selected ity, not to account for all possible com- plots rather than 24. The time spent with probability proportional to the binations of species, size, age, and stratifying from the photo was clearly prediction; that is, trees with larger pre- stocking. well spent. dictions are assigned a higher probabil- •Make strata mutually exclusive ity of being included in the sample. and comprehensive. Each stand should Poisson Sampling Several different approaches can be fall in one and only one stratum, and Poisson sampling is also known as taken to estimate the total; the simplest there should be a stratum for each3probability proportional to prediction is the mean of ratios. stand. (3P) sampling. Poisson sampling is an Each unit in the population is vis- •Create a stratum for the stands efficient, variable-probability sampling ited, and sample selection proceeds by that do not easily fit your stratification strategy for focusing resources on one comparing the prediction for the unit rules. Frayer (1978) showed that this variable of interest. Poisson sampling is in question to a random number: the technique can reduce variability in the unique in that it uses auxiliary infor- unit is selected if the prediction is the remaining strata, and save time in typ- mation collected during the inventory higher of the two. This means that the ing. For example, very simple rules might specify strata for pure hard- woods, pure conifer, and mixed stands. Table 3. Unit-level information for the Poisson sampling example. In this case, the mixed stands are the Volumes are in board feet. Desired sample size is 3. stands that do not fit easily into the Guessed Random True definite strata. Tree volume number volume Ratio •Keep the photo classification and 1 200 310 210 ground classification independent. 2 360 701 350 Photo interpreters should not know 3 250 752 270 4 180 208 190 which stands will or will not contain 5 250 631 250 ground plots. Field crews should not 6 300 791 280 know the phototype classification of 7 320 111 300 0.9375 the stands in which ground plots are 8 300 24 310 1.0333 located. 9 250 478 270 Although any sample scheme may Total 2,410 2,430 Mean ratio 0.98542 be applied within each stratum, we will

Journal of Forestry 41 sample size is no longer fixed and difficult to calculate, and the expected known during the design; it is now a width of the confidence interval of the random variable. variable of interest is difficult to To implement the variable-proba- achieve. bility sampling scheme, a collection of A commonly applied extension of at least one random number per sam- Poisson sampling is the identification pling unit in the population is needed. of sure-to-be-measured trees. Before To obtain the random numbers, a prior the random numbers are generated, a guess of the population total is re- maximum volume prediction is quired. The quality of this guess is not guessed. Any number larger than that of great importance; however, if it is volume is replaced with a set of aster- underestimated, then a larger sample isks to simplify the fieldwork. Then in size will be selected than is desired, and the field, any tree with a predicted vice versa. This guess is divided by the volume higher than this limit is called desired sample size to obtain the upper a sure-to-be-measured tree, and is limit of the random numbers. The considered separately. This adaptation lower limit is 0, and the random num- is identical to stratification of the pop- bers must be integers uniformly dis- ulation into two strata according to tributed between these limits. the prediction, and measuring every The advantage of Poisson sampling tree in the sure-to-be-measured stra- is that, although it requires two levels tum. Although it has the advantage of of auxiliary information—the prior ensuring that the largest trees will all guess of the total and the prediction for be measured, this technique rather each sampling unit—it generates the flies in the face of sampling theory latter on the fly. Having the set of ran- and complicates the implementation dom numbers before going out into somewhat. the field is likely to be cheaper than measuring the auxiliary variable for Literature Cited each unit and then, in a separate sam- AVERY, T.E., and H.E. BURKHART. 1994. Forest measure- pling visit, measuring the variable of ments. New York: McGraw-Hill. BIGGS, P.H., G.B. WOOD, H.T. SCHREUDER, and G.E. interest. BRINK. 1985. Comparison of point-model based and Consider the sample data in table 3. point-Poisson sampling for timber inventory in Jarrah We have a population of nine trees, Forest. Australian Forestry Research 15:481–93. and we would like a sample size of CHAPMAN, H.H., and W.H. MEYER. 1949. Forest mensu- ration. New York: McGraw-Hill. three. We guess that the total volume COCHRAN, W.G. 1977. Sampling techniques. 3rd ed. New for the population will be 2500 bf. York: John Wiley & Sons. Therefore, to obtain a sample size of FRAYER, W.E. 1978. Stratification in double sampling: The three we generate random numbers easy way out may sometimes be the best way. Inventory Notes BLM-10. Washington, DC: US Department of from 0 to 833.3. These numbers are the Interior, Bureau of Land Management. compared with the tree volume esti- SÄRNDAL, C.E., B. SWENSSON, and J. WRETMAN. 1992. mates as they are made. In table 3, only Model assisted survey sampling. New York: Springer two trees are included in the sample: Verlag. SCHREUDER, H.T., T.G. GREGOIRE, and G.B. WOOD. trees 7 and 8. The ratio of true-to- 1993. Sampling methods for multiresource forest inven- guessed volume is calculated for each tory. New York: John Wiley & Sons. one, and the average of these two ratios is multiplied by the total of the guessed volumes: 2,410 × 0.98542 = 2,374.85 = 2,400 bf to two significant figures. Calculation of the standard error (not Andrew P. Robinson (e-mail: andrewr@ shown here) leads to 115.5 bf. (The uidaho.edu) is assistant professor, Depart- confidence interval is omitted because ment of Forest Resources, College of For- with a sample size of two it is quite estry, Wildlife and Range Sciences, Uni- large.) versity of Idaho, Moscow, ID 83843; One reason given for reluctance to David C. Hamlin is biometrician, embrace Poisson sampling is that the Mason, Bruce & Girard, Inc., Portland, sample size is variable (see Biggs et al. Oregon; Stephen E. Fairweather is forest 1985). This factor has two ramifica- biometrician, Olympic Resource Man- tions: the expected costs of a survey are agement, Poulsbo, Washington.

42 December 1999 Biometrics and Inventory Remote Sensing Multistage Remote Sensing Toward an Annual National Inventory

he Forest Inventory and Analysis Forest Council 1992). The Agricultural Remote sensing can improve efficiency of statistical (FIA) program of the USDA Research, Extension, and Education information. Landsat data can identify and map a Forest Service produces a base- Reform Act of 1998 (PL. 105-185, few broad categories of forest cover and land use. T line and long-term set of scientifically Section 253) directs the Forest Service However, more-detailed information requires a sound resource statistics for the 748 mil- to produce more-timely FIA data and sample of higher-resolution imagery, which costs lion acres of forest and woodland to better utilize remotely sensed data. less than field data but considerably more than ecosystems in the United States. These Landsat data. A national remote sensing program data are used to assess the extent, health, Rates of Change would be a major undertaking, requiring unprece- productivity, and sustainability of pub- Rapid changes in forest conditions, dented partnerships between federal programs and lic and private forestlands. FIA informa- real or perceived, fuel the demand for tion is critically important at many annual FIA data. Rapid changes are dri- stakeholders. scales to effectively deal with conserva- ven by urbanization, implementation of tion challenges; influence patterns of public policies, and fluctuating eco- capital investment; and meet the needs nomics in the forest products and agri- of the forestry profession, resource man- cultural sectors over large geographic re- By Raymond L. Czaplewski agers, forest landowners, and the public. gions (from 10 million to 50 million FIA methods vary somewhat by re- acres). Examples include clearing of gion, but the following description is a forestland for agricultural or urban uses, valid generalization. The first phase uses conversion of agricultural lands into a sample of 9.4 million plots, with one forestland, harvesting of wood, and re- plot per 240 acres. Each plot is inexpen- generation of harvested forests. Other sively classified into a few categories of rapid changes are episodic, caused by land cover using high-altitude aerial hurricanes, wind, ice storms, floods, photography. The second phase uses a droughts, and insect epidemics. These subsample of 364,000 one-acre field processes cause changes in forest cover plots, 120,500 of which are forested, that can be detected with a variety of re- with one plot per 6,200 acres. A two- mote sensing technologies, the success person field crew can measure one of which depend on sensor resolution. forested field plot in one day. The For- I assume that other indicators of est Health Monitoring (FHM) program forest conditions change more slowly measures more-expensive indicators on among detailed categories of forest a subsample of 13,500 plots, 4,500 of (table 1). An example is the average which are forested, with one plot per volume and number of trees per acre 167,000 acres. Remote sensing at the by tree species and two-inch diameter first phase improves FIA estimates of class. I make the same assumption for forest area and population totals, but trends in down woody debris, nontree detailed information on forest composi- vegetation, and similar characteristics. tion and condition (table 1) primarily Many other aspects of forest health are relies on expensive field data. Forest Ser- affected by gradual changes in forest vice funding in 1999 was $37.2 million. demographics and anthropogenic stres- Although FIA is among the best sors, such as air pollution, climate programs of its kind in the world, more change, exotic species, and diseases. than half of all FIA information is out- These slow and ubiquitous processes of-date. Current FIA procedures and are measured with field plots in the funding allow a 10- to 15-year remea- FIA and FHM programs. surement cycle, but data more than five If these assumptions are approxi- years old are not reliable (American mately true, then remote sensing could

44 December 1999 be more efficient than field plots for frequent monitoring of rapid changes. Table 1. Forest Inventory and Analysis field data used in primary However, less-frequent remeasurement statistical tables. of field plots remains essential to mon- Forest conditions Number of classes itor gradual changes in forest composi- Plot-level conditions1 tion and calibrate for errors in remotely Land use2,3 5 sensed measurements. Forest type broad2,4 29 Remote Sensing Technologies detailed4 136 Remote sensing can improve effi- Stage of stand development2 4 ciency if remotely sensed data are avail- Stand density2,5 5 able when needed and if they are well Stand origin2 (natural, artificial) 2 correlated with important field mea- Land ownership 10 surements (table 1). For example, aug- Stand age 9 mentation of field data with aerial pho- Stand productivity 7 tography can be six to 15 times more ef- Number of trees2,6, wood volume2,6 continuous ficient in estimating total area of forest, Growth6, mortality2,6, removals2,6 continuous and twice as efficient in estimating total 1 wood volume (Aldrich 1979). A wide Tree-level conditions Tree species4 331 range of remote sensing technologies are Tree size (diameter at breast height) 2-inch classes used in forestry. Satellite data are corre- Tree damage 10 lated with some attributes in table 1, but Tree quality, value 5 information content increases with sen- Wood volume continuous sor resolution. Regardless, remotely sensed data contain various degrees of Growth in wood volume continuous measurement errors that require statisti- 1FIA measures many other indicators that describe landscapes, habitats, non-tree vegetation, etc. 2Photo interpretations and photogrammetric measurements with high-resolution imagery are well cal calibration with current FIA field correlated with these field measurements (Aldrich 1979). The correlation is much lower with Land- data. My discussion of satellite data is sat data and NAPP aerial photography. 3Includes timberland, other forestland, protected forest, nonforest land, and water. based on reviews by Wynne and Carter 4Any single geographic region has only 20 to 40 percent of these national categories. (1997) and Holmgren and Thuresson 5Includes overstocked, fully stocked, understocked, and nonstocked. 6Totals are produced for thousands of permutations of different tree and forest categories. (1998); my assessment of aerial photog- raphy is based on Aldrich (1979). Low-resolution satellite data include However, there is a limit to what can be six-mile swath width, and small pixel AVHRR, MODIS, OrbView-2, ERS- measured by a satellite orbiting 500 size of three to 10 feet wide, are best 2, and SPOT 4. These data are inex- miles from Earth. These data can sepa- suited for imaging small sites. These pensive and cover vast areas, having a rate forest from nonforest, and reason- satellite data have capabilities, limita- 600- to 1,800-mile swath width. Spa- ably identify a few broad types of forest tions, and costs similar to high-altitude, tial resolution is poor, with each pixel and several levels of forest density. nine-inch-square, 1:40,000 small-scale ranging between 160 and 320 acres in Landsat data can distinguish more-de- aerial photographs from the US Geo- size. These data have proved successful tailed categories of forest cover with logical Survey (USGS) National Aerial for continental scale maps of forested customized approaches (Wynne and Photography Program (NAPP). Each landscapes, global change models, and Carter 1997). These data can identify NAPP photograph covers an area five detecting hot spots of severe deforesta- recent clearcuts, but they are less suc- miles wide. These satellite and photo- tion within densely forested land- cessful with partial cuts. Landsat data graphic data can reliably distinguish a scapes. These remotely sensed data do can identify advanced regeneration of few broad types of forest in each region, not have sufficient resolution to reli- forests after land clearing. These data several stages of stand development, ably measure and monitor most indica- can identify urban centers, but they are clearcuts and many partially cut areas, tors of forest conditions in table 1. less successful with sparse urbanization. regeneration after land clearing, and Medium-resolution satellite data in- They can measure size, shape, and con- concentrations of tree mortality. Photo clude Landsat-5&7, Radarsat, SPOT- nectivity of forest patches. High-qual- interpreters can identify forest stands, 2&4, IRS-C&D and P2&5, Spin-2, ity, cloud-free data are available for land use, distance to adjacent roads and EOS AM-1, and CBERS-1&2. These most temperate regions each year or water, forest fragmentation, and many sensors have a reasonably small pixel two, which is sufficient for annual in- types of urbanization. Depending on size of 30 to 100 feet wide, and they are ventory and monitoring. scale, it would take 200,000 to 1 million relatively inexpensive for large areas, High-resolution satellite data include images to cover the United States. The having a 30- to 100-mile swath width. Ikonos-2, OrbView-3&4, EROS- USDA National Agricultural Statistics For example, the conterminous United B1&2, and Quickbird-1&2, although Service and USGS National Wetlands States is covered by 540 Landsat scenes. none are operational yet. The two- to Inventory use NAPP photographs for

Journal of Forestry 45 national mapping on a 20-year time the country, but only a few programs scale NAPP photography for 9.4 mil- frame, but this is not practical for an- consistently cover the whole country. lion photo-interpreted plots. The nual monitoring. The NAPP schedule Several of these programs use Landsat USDA Natural Resources Conserva- for image acquisition is poorly suited to data to map the conterminous United tion Service’s National Resources In- annual monitoring, but satellite data are States. Other programs use a sample of ventory (NRI) uses NAPP and small- expected to be available when needed. higher-resolution aerial photography format aerial photography for 300,000 Large-scale aerial photography ranges to produce statistical estimates. primary sampling units. Most sam- in scale from 1:2,500 to 1:12,000. The USGS Multi-Resolution Land pling units are 160 acres, with a sam- Commercial aerial survey companies Characteristics (MRLC) program uses pling intensity of 1 to 4 percent of the routinely acquire this type of custom Landsat data to map three forest cate- total land area. Accuracy of NRI data is imagery for small sites. Each photo- gories, three urban categories, three limited by quality and scheduling of graph covers an area one-tenth to two woodland categories, three agricultural aerial photography. NRI has been con- miles wide depending on scale and for- categories, and 21 other categories of ducted once every five years, but is mat. Photo interpreters could reliably land use and cover (Volgelmann et al. changing to an annual system, much identify many of the forest cover condi- 1998). The USGS Gap Analysis Pro- like FIA. The annual budget for NRI is tions in table 1. Measurements might gram (GAP) maps critical habitats to $8.5 million. Finally, USGS National include five to 10 broad types of forest; help conserve biological diversity. Wetlands Inventory uses a sparse sam- five stages of stand development; three GAP uses 18 categories of forest, al- ple of small-scale NAPP photography stand-density classes; clearcut and par- though not all occur in every region. for its estimates of status and trends, tial cut areas; regeneration success; stand Both programs use sophisticated re- but this is a minor part of its overall origin (natural, artificial); three to five mote sensing techniques that require mapping program. severity levels for tree mortality; most considerable analytical input. MRLC indicators of urbanization and fine scale began in 1995 with an annual budget The Minnesota Experience forest fragmentation; and stand size, of $10 million, and GAP began in The Annual Forest Inventory Sys- shape, and edge metrics. This type of 1994 with an annual budget of $3.6 tem (AFIS) began in 1991 as a joint ef- photography would require many mil- million. Neither program has yet cov- fort between the Minnesota Depart- lions of images to completely cover the ered the entire country. These pro- ment of Natural Resources and the nation, but sampling makes this im- grams plan to update their maps to USDA Forest Service. Lessons learned agery feasible on the national scale. compensate for changes in land cover, in AFIS are relevant to the mandate in perhaps on a 10-year cycle. Public Law 105-185. AFIS successfully Using Remotely Sensed Data Three programs use a sample of aer- used numerous Landsat scenes to clas- Numerous land management agen- ial photography to cover the United sify land cover into a few broad cate- cies use remote sensing for portions of States. The FIA program uses small- gories and detect abrupt changes over

46 December 1999 time. AFIS processed Landsat data that Much of the expensive field data Multistage Sampling was re-imaged over four-year intervals, merely verified whether or not these AFIS demonstrated that Landsat but vigorous regeneration of clearcuts plots were cleared of trees. In the be- data can improve FIA products. How- reduced the accuracy of change detec- ginning, AFIS did not use aerial pho- ever, Landsat data alone do not greatly tion. Had Landsat data been purchased tography because Landsat is less expen- reduce the required amount of field along orbital paths rather than physio- sive for large regions. During later data. Landsat provides only broad in- graphic regions, change detection stages of AFIS, a sample of aerial pho- formation about forest conditions, and could have been conducted every two tography was reconsidered because the detailed information in table 1 re- years at little extra cost. A single tech- high-resolution imagery could reduce quires field measurement. However, nician could process a Landsat scene in the cost of field data to verify change high-resolution imagery provides five to 10 days because changes in land detection from Landsat data. much more detailed data that are bet- cover were detected with simple digital methods. AFIS classifications of land cover with Landsat replaced NAPP photography for the first phase in the FIA statistical design, and image acqui- sition dates for Landsat were more compatible than NAPP for an annual system. In addition, Landsat provided maps of land cover and change that are not feasible for large regions with aerial photography or field sampling. If Landsat data suggested that an FIA plot might have been affected by timber harvest or change in land use, then the plot was remeasured by a field crew. Remeasuring consumed about half the budget for field data. Misregis- tration and other errors with Landsat data caused incorrect classification of some FIA plots as having changed.

Real or perceived rapid changes in forest cover and conditions fuel the demand for annual FIA data. The causes of such changes can be detected with a variety of remote sensing technologies and data types.Levels of sensor resolu- tion are key to successful detection; some of those levels are illustrated in this sequence of photographs of the Beaverhead-Deerlodge National Forest in Dillon,Montana. Left to right: A Landsat Thematic Mapper (TM) satellite image; a higher- resolution digital ortho quad image; a high-altitude aerial photograph; a digital infrared camera image (above). The latter was taken five years after the TM image,closer to the time when the change in forest cover was investigated. To assess conditions at the forest,plot, and tree levels costs less than field data but more than Landsat data.

All images courtesy of USDA Forest Service Remote Sensing Applications Center, Salt Lake City, Utah

Journal of Forestry 47 ter correlated with attributes in table 1. changes over the five-year interval be- resolution images each year. Expecta- A multistage statistical design can com- tween acquisition of new imagery for tions must be kept realistic, numerous bine wall-to-wall Landsat data at the each permanent sampling unit. High- details await analysis, and formidable first stage, a sample of high-resolution resolution imagery could improve sta- problems remain unsolved. Multistage imagery at the second stage, and tradi- tistical efficiency, allowing a reduction sampling with remote sensing was en- tional FIA and FHM field plots at the in the required number of FIA sam- visioned by the National Academy of third stage. The National Academy of pling units. Calibrated measurements Sciences in 1974, but the vision has Sciences recommended a similar ap- from the high-resolution images might never been implemented. However, if proach 25 years ago (Aldrich, 1979). I even replace field data for inaccessible these challenges can be overcome, a describe two enterprises that would areas. The large sampling units better partnership among existing federal implement a multistage design. match the scale of Landsat images programs could produce the world’s The first enterprise would acquire than one-acre FIA plots, thus improv- premier system to estimate national all Landsat scenes that cover the con- ing the linkage between Landsat data trends in land cover and land use, de- terminous United States. Multi-date and more accurate measurements of tect changes in health of wildlands and Landsat data would rapidly identify sampling units. This enterprise could agricultural landscapes, evaluate effec- abrupt changes in spectral reflectance cost $15 million to $25 million each tiveness of public policies, and guide that are often associated with clear- year. sustainable use of the nation’s natural cuts, landclearing, and reforestation. The latter enterprise is similar to resources. Change detection allows relatively in- the National Resources Inventory expensive updates to existing MRLC (NRI). The cost of new imagery and Literature Cited and GAP maps. FIA would use up- interpretation might be shared be- ALDRICH, R.C. 1979. Remote sensing of wildland resources: dated maps to replace its photo inter- tween FIA and NRI, which would A state-of-the-art review. General Technical Report RM-71. Fort Collins, CO: USDA Forest Service, pretation of 9.4 million first-phase make the enterprise more feasible and Rocky Mountain Forest and Range Experiment Sta- plots. Direct annual cost is estimated efficient. This partnership poses con- tion. at $1.5 million to $2 million. siderable technical challenges, such as: AMERICAN FOREST COUNCIL. 1992. Report of the blue The second enterprise would ac- incremental alignment of separate FIA ribbon panel on forest inventory and analysis. Washing- ton, DC. quire a national sample of large-scale and NRI sampling frames; complex CZAPLEWSKI, R.L. 1999. Integration of strategic inven- aerial photography or high-resolution statistical techniques for calibration tory and monitoring programs for the forest lands, satellite imagery. The resulting data and composites of multiple time-series wood lands, range lands and agricultural lands of the would detect changes in land use, par- of multivariate sample data; a sophisti- United States. In Proceedings of the North American Science Symposium, Toward a Unified Framework for tial cuts, forest management, and se- cated information management sys- Inventorying and Monitoring Forest Ecosystem Resources, vere episodic events. Sample imagery tem; capacity-building in the aerial Guadalajara, Jalisco, Mexico, November 1–6, 1998. would include 364,000 primary sam- survey industry to deliver large quanti- Fort Collins, CO: USDA Forest Service, Rocky pling units, each covering an existing ties of photography; and adjustments Mountain Research Station. HOLMGREN, P., and T. THURESSON. 1998. Satellite re- FIA field plot. Each sampling unit for cloud cover and missing data mote sensing for forestry planning—a review. Scandi- could range from 40 to 640 acres in (Czaplewski 1999). Bureaucratic chal- navian Journal of Forest Research 13:90–110. size, and the collection of sampling lenges would be equally formidable USDA FOREST SERVICE INVENTORY AND MONITORING units would encompass 1 to 10 per- (USDA Forest Service Inventory and INSTITUTE. 1998. Integrating surveys of terrestrial nat- cent of the total land area. The large Monitoring Institute 1998). Robust ural resources: The Oregon Demonstration Project. In- ventory and Monitoring Institute Report No. 2. Fort sampling units would better capture solutions to these challenges are Collins, CO: USDA Forest Service, Rocky Mountain rare features than one-acre FIA field untested, cost-effectiveness must be Research Station. plots, which encompass only 0.016 evaluated, and risks must be reduced VOLGELMANN, J.E., T. SOHL, P.V. CAMPBELL, and D.M. percent of the landscape. Each year, 20 through simulations and realistic pilot SHAW. 1998. Regional land cover characterization using Landsat Thematic Mapper data and ancillary percent of the large sampling units tests. sources. Environmental Monitoring and Assessment would be remeasured with new high- 51:415–28. resolution imagery. Photo interpreters Conclusion WYNNE, R.H., and D.B. CARTER. 1997. Will remote would delineate and classify land uses, Congress has emphasized the need sensing live up to its promise for forest management? Journal of Forestry 95(10):23–26. land cover, and forest stands within for more-current statistical informa- each sampling unit. Photogrammetry tion about the nation’s forests. Tradi- would produce more-detailed mea- tional FIA field procedures would sat- surements of forest characteristics at isfy this need at an estimated annual secondary sampling points within each cost of $82 million. Multistage remote Raymond L. Czaplewski (e-mail: czap@ 40- to 640-acre sampling unit, and sensing might save $20 million each lamar.colostate.edu) is project leader and one of these points would be a one- year and produce valuable new prod- mathematical statistician, Forest Inven- acre FIA field plot. These measure- ucts. Implementation requires an un- tory and Monitoring Environmetrics, ments would be well correlated with precedented infrastructure that can ac- USDA Forest Service, Rocky Mountain many field observations in table 1. quire and process hundreds of Landsat Research Station, 240 West Prospect Photo interpreters would measure scenes and tens of thousands of high- Road, Fort Collins, CO 80526.

48 December 1999 Perspective

The New and Improved FIA Program

Bob Goodlatte and James Garner

Determining the ecological and bio- Second, the Forest Service must The use of such technology presents logical significance of our forest re- communicate effectively with the another opportunity to involve the sources in an accurate and timely man- states. National working groups are universities. The forestry colleges at ner is one of the most important pur- important for consistency, quality con- Virginia Tech and Mississippi State suits in modern forestry. That is the trol, and other issues of general con- University already are working closely mission of the USDA Forest Service’s cern. But more important is a state- with NASA to study forest resources Forest Inventory and Analysis (FIA) specific liaison system that ensures the using remote sensing and other lead- Program. FIA is the most complete Forest Service will adapt the new con- ing-edge technologies and applica- forest census in America, providing the gressional mandate to the unique tions. The Forest Service’s work with only continuous national inventory structure, policy, budget processes, op- NASA, USGS, NOAA, and other fed- that quantifies the status of forest erating procedures, and information eral partners should focus on harness- ecosystems across all private and most and reporting needs of each state. ing such ongoing efforts. public forestland. Third, the Forest Service must pro- Finally, the Forest Service must in- Because of its fundamental impor- vide an analysis and reporting system sist that the National Forest System co- tance in measuring sustainability, FIA that matches the data collection services operate fully in the effort to coordinate is universally popular among profes- provided by the states. Most state for- the FIA program nationwide. Some sional foresters, environmentalists, in- esters, especially in the Northeast and western regions have refused to imple- dustry, private landowners, and virtu- South, have invested considerably to ment the new FIA program in favor of ally any other group that has an inter- improve data collection as part of their parochial data collection systems devel- est in forest management. Yet, despite commitment to make FIA work. Many oped with little regard for the need for such broad-based support, FIA is only have committed to deadlines for deliv- national consistency. One way to rec- now coming of age as a 21st-century ering accurate data, relying on promises tify this problem is for Congress to di- management tool. from the Forest Service. Failure by the vert funding within the Forest Service In 1998 Congress enacted legisla- agency to analyze and report data in a from the National Forest System to tion to greatly improve the FIA pro- timely manner would embarrass not Forest Service Research with instruc- gram. Congress has been clear in its only the Forest Service, but also those tions that the diverted funds be used to mandate, requiring the agency to com- state foresters who have raised the ex- implement the law. This somewhat plete annual inventory updates, reduce pectations of their customers. draconian approach can be avoided, inventory cycles to five years, use the One way for the Forest Service to however, if regional foresters demon- latest technologies, and make current keep analysis at pace with data collec- strate greater willingness to cooperate. data available to the public in a user- tion is to draw on the resources of the The improved FIA program is the friendly format. Congress has also sub- university sector. Using university ex- cornerstone of ecologically and biolog- stantially increased funding for FIA. pertise, technology, and personnel to ically sustainable forest practices in the It is now up to the Forest Service to handle data entry, editing, compila- 21st century. Congress has provided make FIA succeed. Success will only tion, and other essential tasks could the framework. Willing partners are in occur, however, if the Forest Service significantly reduce the backlog of plot place to help with the transition. The makes a number of key changes to its data editing, which has reached 1,000 future is waiting. Now is the time for current policies and practices. We rec- plots per month in the South. the Forest Service, beginning with ommend the following. Fourth, the Forest Service must Chief Dombeck, to make it happen. First, Forest Service Chief Michael begin today to invest in the technolo- Dombeck must make FIA a clear re- gies of tomorrow. The move to further Bob Goodlatte (R-VA) (e-mail: talk2bob@ search priority. So far neither the declassify high-resolution satellite im- mail.house.gov) is chair of the Subcom- agency’s budget requests nor the chief’s ages could signal an opportunity for mittee on Department Operations, Over- statements to Congress have estab- FIA by greatly improving information sight, Nutrition, and Forestry, Committee lished this priority. The chief’s com- on real-time land use change, forest on Agriculture, US House of Representa- mitment thus far has been, at best, to disturbance, and other ground change. tives, 2240 Rayburn House Office Build- maintain the status quo. With ecologi- Such technology could be applied over ing, Washington, DC 20515-4606; cal sustainability in the balance, this a wide landscape with few personnel James Garner is state forester, Common- position is unacceptable. and relatively modest investments. wealth of Virginia, Charlottesville.

64 December 1999