Site Index Adjustment for the Vanderhoof IFPA

Transcription

1 Site Index Adjustment for the Vanderhoof IPA Prepared for: Slocan orest Products Plateau Division Vanderhoof March 31, 04 Craig arnden R.P.. Consulting orester Prince George, B.C.

2 Executive Summary The purpose of this project is to develop unbiased estimates of site index for managed stands in the Vanderhoof orest District. It is intended that these improved estimates of SI be used in future timber supply analyses for the area. The approach taken was to compare measured site index rates for randomly selected points to predictions of site index for the same points using a combination of Predictive Ecosystem Mapping (PEM) and the Moʼs correlation of site index to site series (SIBEC). Two hundred and twenty stands were sampled, split into 6 strata based on species (interior spruce and lodgepole pine) and biogeoclimatic subzones (dry SBS, moist SBS, and ESS). Observed differences between PEM/SIBEC estimates of site index and those measured through ground sampling were large. Mean differences by stratum ranged from 2.6 m for pine in the moist SBS subzones to 6.1 m for spruce in the dry SBS subzones, with an area weighted mean difference of 2.7 m for pine and 4.8 m for spruce. Overall it was observed that PEM/SIBEC estimates of site index had very low precision. Differences between the predicted and observed values were used to build a set of linear models that provide corrections to predicted values for any polygon. Separate linear models were developed for each sampling stratum. Parameters used in the models include PEM/SIBEC SI and elevation. Applying these models in a timber supply context is expected to result in approximately a 44% increase in the Long Term Harvest Level over that predicted in the 01 TSR analysis, and a 34% increase over that from the Vanderhoof IPA timber supply analysis. Due to an expected high rate of unsuitable stands for measuring site index in the general population, sampling for this project was restricted to young managed stands where stand history is known. Use of this sample frame introduces the risk that sampled stands are not truly representative of the larger target population. While the relative distribution of sampled and target stands was similar by elevation and site series, there is some uncertainty that growth rates for the last few decades as reflected by stands in the sample frame may not represent the longer term trend. Any projects using this prediction methodology should therefore employ sensitivity testing to explore variations in the corrected site indices of up to +/- 2 m.

3 Acknowledgements Valuable advice on sample design was provided by Gordon Nigh, BC Ministry of orests, Research Branch. ield data collection for this project was undertaken by Kim orest Management Ltd. and Erafor Consultants Ltd. Review comments were provided by Wendy Bergerud, Gordon Nigh, Shirley Mah, Craig DeLong and Doug Beckett of the BC Ministry of orests. Rob Vassov of RJ Vassov & Associates provided invaluable contributions to project design, data analysis and interpretation, and document reviews. My thanks to all.

4 Table of Contents 1.0 Introduction The Vanderhoof orest District land base Methods Basic Approach Sample rame Stratification Sample Size Sample methodology Data Analysis Results Comparison of PEM/SIBEC and Measured Site Indices Linear Models Discussion Differences Between PEM/SIBEC and GI Estimates of SI Precision of PEM/SIBEC Estimates Recommendations for SI Adjustment Implications of Adjusted Site Indices for Projecting Timber Supply Assessment of Risk and Uncertainty Modifications to SI Adjustments with Improved SIBEC Data Summary and Recommendations References Appendix 1. Data Summaries...19

5 Site Index Adjustment for the Vanderhoof IPA 1.0 Introduction In recent years, forest managers have become increasingly aware of biased estimates of site index that are common in the forest inventory. The biases arise primarily from estimating site index in old stands, which are prevalent in the inventory in many areas of the province. Specific causes include uncertainty in the height/age curves for old stands, height/age pairs from trees/ stands that underwent significant periods of suppression (i.e. in multi-aged stands), inappropriate selection of site trees and lodgepole pine repression, all of which lead to underestimates of site index. These biases have important implications for timber supply analysis and strategic silviculture planning, as relatively small increments of site index have significant effects on predicted forest yields. Acquiring accurate estimates of site index have thus become an important focus in many timber supply areas. Several methods of correcting site index estimates have been employed. Earlier examples included paired plot and veteran studies, where estimates of SI that were known to be biased were compared to immediately adjacent sources of estimates that are believed to be more reliable. Results from these studies have been used in some analyses to make coarse, broad scale adjustments to SI estimates, although often simply to test sensitivity to possible bias. More recently, methods have evolved to extensively sample certain management units, with the intent of developing statistically defensible, locally applicable adjustments. This document describes the results of such a project in the Vanderhoof orest District. 1.1 The Vanderhoof orest District land base The Vanderhoof orest District (igure 1) covers 1.4 million hectares of primarily low and rolling plateau land in central British Columbia. Just over 1 million ha is considered productive, with a timber harvesting land base of 785,000 ha. This forested area currently supports an AAC of 1.7 million m 3 /yr, supplying 5 major processing facilities within the District as well as several value added processors. The Nechako River basin is the major topographic feature in the District, including within it raser Lake and the east ends of rancois Lake and the Nechako Reservoir. A few low mountain ranges occur, with the awnie and Nechako Ranges in the southwest corner being the highest at just under 1900 m. Most areas of the District lie well below 10 m elevation. The Vanderhoof District is dominated by the Sub-Boreal Spruce (SBS) biogeoclimatic zone. There are 4 SBS subzones occurring, 2 of which are considered to have dry climates (SBSdk and SBSdw) and one a moist climate (SBSmc). There are also significant areas of the Engelmann Spruce Subalpine ir (ESS) zone on the tops of low mountains in the area, and small areas of the Sub-Boreal Pine Spruce (SBPS) and Montane Spruce (MS) zones in the southwest corner of the District. Page 1

6 Site Index Adjustment for the Vanderhoof IPA Burns Lake Tchesinkut Lake Taltipin Lake St uart Lake raser Lake ort St James River St uar t G reat B eaver Lake River Vanderhoof rancois Lake C heslatta Lake Nechako l Tetachuck Lake K newstubb Lake Tweedsmuir Provincial Park West Road River Kluskoil Lake Prov. Park Dean Ri ver igure 1. The Vanderhoof orest Dustrict 2.0 Methods 2.1 Basic Approach This project has compared inventory estimates of site index with those actually measured in randomly selected ground samples throughout the Vanderhoof orest District. The basis for inventory estimates of site index was a combination of predictive ecosystem mapping (PEM) estimates of site series coupled with the Ministry of orestsʼ ecosystem based estimates of site index (SIBEC). The difference between the predicted and field estimates of site index at each sample point was compared across the land base and the mean difference, if significantly different from zero, indicated the need for a site index adjustment. Separate adjustments have been developed for each of lodgepole pine and white spruce, and for different strata in the population based on biogeoclimatic subzones. The project has also tested the need for separate adjustments based on elevation. The target population for this project was all spruce and pine leading stands on productive forest land in the Vanderhoof orest district. or the purposes of this project, productive forest land has been deemed as that having a site index of 7.0 or greater recorded in the inventory. The project could not, however, use this entire land base for sampling as many stands are unsuitable for making reliable site index estimates. Most existing, unmanaged lodgepole pine stands are of fire origin and are at risk of being affected by density-related repression. A higher percentage of existing, unmanaged spruce leading stands would likely have been usable, but early periods of suppression may also have lead to problematic SI estimates. or these reasons, sampling was restricted to a sampling frame of managed stands where stand history is known. Sampling only managed stands (or any other subset of the total population) leads automatically to a risk of bias. The sample can only truly be said to represent the sample frame, and no Page 2

7 Site Index Adjustment for the Vanderhoof IPA quantitative comparison can be made to the larger land base. In taking this approach, we must assume that the harvested areas are reasonably typical of the landscape as a whole. Comparison of proportional representation of the sample and entire populations by biogeoclimatic subzone variant, site series and elevational band can help provide evidence of whether or not this is really the case. Sampling only young stands with a relatively limited range of ages leads to other problems as well. In applying this methodology we are assuming that observed growth over a limited time period reflects the long term growth conditions that will be experienced in the future. While this factor is relatively intractable, it must be recognized as a source of uncertainty. 2.2 Sample rame Sampling was restricted to those stands that satisfied the following criteria: 1. Managed stands 2. Leading species is interior spruce or lodgepole pine 3. Trees on most sample plots ( 60%) will be at least 5 years breast height age 4. Trees are free of overtopping vegetation, and have been free of such competing vegetation at least since reaching breast height. 5. In the case of lodgepole pine, trees are free of repression. This means restricting sampling to stands with densities of less than approximately,000 trees/ha (thresholds are likely higher but a current density of,000 will provide a small safety margin) and stands that have not been spaced. 6. Stands have not been fertilized The first three criteria were satisfied by selecting stands within an Arc/Info GIS system that had a harvest date of 1993 or earlier for pine leading stands and 1991 or earlier for spruce leading stands. A preliminary list of 2.8 million sample points was then generated based on the intersection points of a 25 m grid. Each grid point was assigned a random number and sorted to produce a ranked list. The top candidates were further screened using ISIS data to help ensure stands were sufficiently old/tall enough for growth intercept estimates of SI, were free of overtopping vegetation and repression, and had not been spaced or fertilized. 2.3 Stratification Variations in SI adjustments on different portions of the land base may be required based on differences in the relative occurrence of factors affecting errors in inventory-based SI estimates. or the purposes of this project, three sampling strata based on biogeoclimatic subzones have been defined: 1. Dry climate portions of the SBS biogeoclimatic zone (SBSdk, SBS dw2 and SBSdw3) 2. Moist climate portions of the SBS zone (SBSmc2 and SBSmc3) 3. ESSmv1 There are small areas in the southwest corner of the District that fall within the SBPSmc and MSxv subzones, as well as small areas of the ESSxv subzone. These areas were not sampled, and no conclusions from the project can be applied to these areas. Page 3

8 Site Index Adjustment for the Vanderhoof IPA 2.4 Sample Size or lodgepole pine and white spruce, 40 samples respectively were planned for each of the SBS strata, and in the ESS, for a total of 2 samples. Based on issues related to rejecting samples and finding suitable replacements, final sample sizes were slightly different. 2.5 Sample Methodology Each sample consisted of five sub-plots distributed along a strip line at 25 m intervals, starting at the randomly selected point and proceeding on a randomly selected bearing. Details of the field sampling procedures are available in the sample plan (arnden 03). In addition to the primary and alternate points listed in Appendix 2 of the sample plan, a further list of alternate sample points was drawn from the sorted sample frame for the spruce stratum in the drier SBS subzones. Once sampling had commenced, it was found that a high percentage of stands in this stratum were of partial cut origin (harvested prior to the mid-1960ʼs), and were unsuitable for site index measurements. The additional sample points were further screened using Mo opening records in the Vanderhoof District Office to ensure a higher rate of acceptance in the field. 2.6 Data Analysis or each sample point, a predicted site index estimate was determined using the combination of site series derived through predictive ecosystem mapping (PEM) and the Ministry of orests ecosystem/site series correlation (SIBEC). In cases where the PEM product provided more than one possible site index, the corresponding SIBEC estimates of SI were probability weighted. In a few cases, site series units identified in the PEM product did not have corresponding estimates available in the SIBEC database, and assumptions had to be made as follows: ecosystem units identified in the PEM as SBSdk B were assumed to have the same site index as the SBSdw3 05 site series (both are pine-black spruce types) ecosystem units identified in the PEM as SBSdk SI were described as combinations of the 01 and 05 site series, and were assumed to occur in a 50:50 mix ecosystem units identified in the PEM as SBSdw3 SI were described as combinations of the 01 and 04 site series, and were assumed to occur in a 50:50 mix ecosystem units identified in the PEM as SBSmc3 SI were described as combinations of the 01 and 05 site series, and were assumed to occur in a 50:50 mix ecosystem units identified in the PEM as SBSmc2 E were described as non-treed, and were assumed to have a site index of zero (this only occurred only as a small percentage of a single polygon) All SIBEC estimates were taken from the latest SIBEC release (SIBEC RDM Version: 02), which provides a combination of 1 st and 2 nd approximation estimates depending on data availability. Observed site index estimates for each sample were determined by averaging the estimates for each of the five sample tree sub-plots. or each sample tree, site index estimates were calculated using the growth intercept equations reported by Nigh (1997, 1999). Page 4

9 Site Index Adjustment for the Vanderhoof IPA Linear models were developed to compare observed site index estimates with predicted values, using elevation as an additional independent variable. 3.0 Results 3.1 Comparison of PEM/SIBEC and Measured Site Indices Site indices measured on the ground are considerably higher than those predicted using PEM/ SIBEC (igure 2 and Appendix 1). Mean differences by stratum range from 2.6 for pine in the moist SBS subzones to 6.1 for spruce in the dry SBS subzones. In all cases, there is a larger disparity between observed and predicted values for spruce than for pine. Also in all cases, the hypothesis that the mean difference is equal to zero is rejected at the 95% confidence level. Direct comparisons of predicted and observed site indices by stratum are provided in igure 3. The differences from igure 2 are repeated, whereby the combination of PEM and SIBEC consistently provides lower estimates of site index than do the sampled growth intercept estimates. Also evident is a very weak correlation between predicted and observed SI values. This suggests that the combination of PEM and SIBEC, at least in their current forms, have extremely limited value in discerning site-to-site variation in actual site indices. igure 2. Comparison of site indices derived using predictive ecosystem mapping and SIBEC values versus those observed for the same set of points using growth intercepts. In the box plots, the tops and bottoms of the filled boxes represent the 75 th and 25 th percentiles respectively, the ends of the whiskers represent the 90 th and th percentiles, and the circles and stars represent outliers. The paired mean difference between PEM/SIBEC estimates and GI estimates (along with standard errors in brackets) can be found at the bottom of each chart cell. Actual estimates for each sample can be found in Appendix 1. In all cases, a paired t-test found significant differences between predicted and observed means (p<<0.001) Predicted SI (PEM/SIBEC) Observed SI (GI) Site Index 50 yrs BHage) N = ESS Dry SBS Moist SBS ESS Dry SBS Moist SBS Pine Pine Pine Spruce Spruce Spruce Strata Mean Difference Page 5

10 Site Index Adjustment for the Vanderhoof IPA Another trend evident from igure 3 is that predicted SI values are closest to observed values on the better sites. This trend becomes more obvious if a ratio of observed SI to predicted SI is plotted over predicted SI for each stratum (igure 4). Samples with low predicted site indices consistently have higher observed/predicted ratios than those with higher predicted site indices. This partially reflects that the variation in mean SI by site series is far less in the field measurements than is predicted by the SIBEC database (igure 5). This trend is particularly evident for lodgepole pine, for which less than a 3 m range in mean site index was observed over most site series in the SBS subzones, while the range predicted using the SIBEC estimates is much greater. While the highest SIBEC estimates tend to be close to the mean site indices observed in the field, the lowest SIBEC estimates differ from observed means by as much as 8 m. The high variation in the differences between the SIBEC estimates and the observed field means by site series likely also contributes to the poor correlations between predicted and observed SI values illustrated in igure 2 (note the large preponderance of sites with a predicted spruce site index of m in the ESS). igure 3. Comparison of predicted (PEM/SIBEC) and observed (GI) site indices by sampling stratum. or illustrative purposes, a least squares regression model has been fit through each distribution (solid line through scatter). Assuming a 1:1 relationship, such models would have a slope of 1.0 and an intercept of 0 (corner-to-corner lines). The best R 2 value for any of the models was 0.071, and none of the slopes were statistically different from zero (p<0.05, one-tailed). ESS - Pine Dry SBS - Pine Moist SBS - Pine Observed SI 25 Observed SI 25 Observed SI Predicted SI 25 Predicted SI 25 Predicted SI ESS - Spruce Dry SBS - Spruce Moist SBS - Spruce Observed SI 25 Observed SI 25 Observed SI Predicted SI 25 Predicted SI 25 Predicted SI Page 6

11 Site Index Adjustment for the Vanderhoof IPA igure 4. Ratios of predicted (PEM/SIBEC) and observed (GI) site indices plotted over predicted site index by sampling stratum. or illustrative purposes, a least squares regression model has been fit through each distribution (solid line through scatter). In all cases, the slope of the line was significantly different from zero (p<0.05, one-tailed). Ratio of Observed SI to Predicted SI ESS - Pine Dry SBS - Pine Moist SBS - Pine Predicted SI Ratio of Observed SI to Predicted SI Predicted SI Ratio of Observed SI to Predicted SI Predicted SI Ratio of Observed SI to Predicted SI ESS - Spruce Predicted SI Ratio of Observed SI to Predicted SI Dry SBS - Spruce Predicted SI Ratio of Observed SI to Predicted SI Moist SBS - Spruce Predicted SI A large portion of the site to site variation evident in igure 3 results from variables that can t be accounted for by site series alone. Elevation for example, was expected to have a significant effect. We know that there is a negative trend in site index with increasing elevation, due to elevational influences on climate. Biogeoclimatic subzones deal with this climatic gradient on a coarse scale, but there will still be variation within a subzone. If this variation can easily be accounted for within a predictive model, it will help provide a more precise estimate of site indices for individual polygons, and help eliminate bias in the event that sampling for this project is biased toward a particular elevational range relative to the entire land base. This latter factor is of particular concern in the ESS, where it might be surmised that most of the older clearcuts being sampled are at lower elevations. Relationships between elevation and site index are plotted for each stratum in igure 6. The slopes of these relationships are significantly different from zero (p<0.05, one tailed) for all but pine in the drier SBS subzones (Appendix 1). Page 7

12 Site Index Adjustment for the Vanderhoof IPA Interior Spruce N = Site Index (m) ESS mv1 01 Site Index (m) ESS mv1 03 ESS mv1 04 ESS mv1 05 SBS dk 01 SBS dk 04 SBS dk 05 SBS dk 06 SBS dw2 07 SBS dw3 01 SBS dw3 06 SBS dw3 07 SBS mc2 01 SBS mc2 05 SBS mc2 06 SBS mc2 07 SBS mc2 08 SBS mc2 SBS mc3 01 SBS mc3 04 SBS mc3 05 SBS mc3 07 SBS mc3 09 Lodgepole Pine na N = ESS mv1 01 ESS mv1 02 ESS mv1 03 ESS mv1 04 ESS mv1 05 SBS dk 01 SBS dk 03 SBS dk 04 SBS dk 05 SBS dk 06 SBS dw2 01 SBS dw3 01 SBS dw3 04 SBS dw3 05 SBS dw3 07 SBS mc2 01 SBS mc2 02 SBS mc2 03 SBS mc2 05 SBS mc2 06 SBS mc3 01 SBS mc3 03 SBS mc3 04 SBS mc3 05 SBS mc3 06 SBS mc3 07 SBS mc3 09 Site Series igure 5. Comparison of SIBEC estimates of site index to the range of SI values observed for individual tree subplots grouped by subplot level observations of site series (trees from the same sample may fall within different site series). SIBEC estimates of site index for each site series are indicated by the black dots. Box plots for observed SI indicate the th, 25th, 75th and 90th percentiles. Page 8

13 Site Index Adjustment for the Vanderhoof IPA igure 6. Relationships between elevation and observed site index. ESS - Pine Dry SBS - Pine Moist SBS - Pine Observed SI 25 Observed SI 25 Observed SI Elevation (m) Elevation (m) Elevation (m) ESS - Spruce Dry SBS - Spruce Moist SBS - Spruce Observed SI 25 Observed SI 25 Observed SI Elevation (m) Elevation (m) Elevation (m) 3.2 Linear Models Based on the relationships explored in Section 3.1, linear regression models were fit to test the value of PEM/SIBEC SI and elevation as independent variables, both alone and in combination. Statistical details for these models can be found in Table 1. Using PEM/SIBEC SI alone as a predictive variable has little value, at least at the current time given the variability of site-to-site SIBEC estimates relative to observed values. The regression coefficients for PEM/SIBEC SI (either alone or in combination with elevation)n were never significantly different from zero (p 0.05). In all cases, elevation was a superior predictor variable than was PEM/SIBEC SI, although elevation still explained only a small percentage of the variation, and the coefficients for elevation by strata were marginally significant to non-significant (p 0.05). The use of PEM/ SIBEC SI as a second variable in addition to elevation almost always increased the r 2 values by a small amount, but either slightly increased or was neutral with regard to standard errors. The high standard errors for the regression coefficients relative to the coefficient values indicate a low degree of influence on predicted outcomes. Overall, the variation in predicted outcomes over a reasonable range of input values (elevation and PEM/SIBEC SI) is quite limited, and provides only marginal gains over using a grand mean for each stratum. Page 9

14 Site Index Adjustment for the Vanderhoof IPA Table 1. Regression model summaries for each stratum Stratum ESS - Pine SBS Dry - Pine SBS Moist - Pine ESS - Spruce SBS Dry - Spruce SBS Moist - Spruce Parameter Parameter Coefficients Std. Error t Sig. 95% Confidence Interval Lower Bound Upper Bound Intercept PEM/SIBEC SI Intercept Elevation Intercept PEM/SIBEC SI Elevation Intercept E PEM/SIBEC SI Intercept E Elevation Intercept E PEM/SIBEC SI Elevation Intercept E PEM/SIBEC SI Intercept E Elevation Intercept E PEM/SIBEC SI Elevation Intercept PEM/SIBEC SI Intercept E Elevation Intercept PEM/SIBEC SI Elevation Intercept E PEM/SIBEC SI Intercept E Elevation Intercept E PEM/SIBEC SI Elevation Intercept E PEM/SIBEC SI Intercept E Elevation Intercept E PEM/SIBEC SI Elevation r Discussion 4.1 Differences Between PEM/SIBEC and GI Estimates of SI Differences between the stratum level mean SI values predicted using PEM/SIBEC and growth intercepts varied from 2.6 m for pine in the moist SBS subzones to 6.1 m for spruce in the ESS zone (igure 2). While neither source of estimates can be claimed as correct, there is reasonable evidence to suggest that the growth intercept estimates are more correct. A majority of the SIBEC values used in this project are first approximation estimates. This means that there were insufficient reliable samples for these site series (a minimum of 7) to make acceptable second approximation estimates. A rough comparison of the differences between first Page

15 Site Index Adjustment for the Vanderhoof IPA and second approximation estimates can be derived by looking at zonal sites 1 for subzones in which there are two or more variants, and for which first approximation estimates are reported for at least one variant, and second approximation estimates for the other(s). In 45 out of 57 such cases, the variant with a second approximation estimate has a higher reported site index than the variant with a first approximation estimate, with a mean difference of 1.9 m (SE = 0.4). Such a trend indicates a strong likelihood of overall bias in the first approximation estimates. urther evidence of such bias is provided in the data from this study. or those individual tree subplots in which second approximation SIBEC estimates were available (based on field estimates of site series), the SIBEC estimates averaged 2.1 m (SE = 0.) below the growth intercept estimates (all species and all sites). or the subplots falling in site series for which only first approximation SIBEC estimates were available, the difference was considerably greater at 4.3 m (SE = 0.). While second approximation SIBEC estimates appear considerably better than first approximation estimates, they are still statistically different (p 0.05) than estimates provided using growth intercepts. Some possible reasons include: A real difference in growth patterns may exist between the predominantly older trees sampled in various sources used in the SIBEC program as compared to the younger trees sampled in this project. Such a risk is discussed in Section Growth intercept models may be overestimating the height that will be reached at breast height age 50. Such biases have been tested for in interior spruce (i.e. J.S. Thrower and Assoc. 00, arnden 01) and largely rejected. Biased measurement errors in this project may have resulted in over-estimates of site index. The most likely source of such an error would have been the under-counting of tree rings. In order to account for the 2.1 m difference between GI estimates and second approximation SIBEC estimates, however, an average error per tree of approximately -2 years would be required. Given the emphasis placed on careful measurement, the fact that most tree ages were re-counted indoors from prepared cookies and that third party field audits were employed, a level of measurement bias even remotely approaching this amount is extremely unlikely. Any one of several potential causes of SI underestimations may have been present in any number of the projects that contributed data to the SIBEC program. Potential sources of error might include measurement bias, sampling of old trees (trees older than 1 to 1 years are generally considered poor site tree candidates), sampling of repressed trees or trees that had undergone a period of suppression, trees that were simply recognized as dominant or co-dominant rather than one of the largest 0 trees/ha, or trees that met any of several damage criteria that would lead to sample rejection. To a large extent, data sources with a high risk of such errors have been weeded out of the database supporting second approximation estimates, but some problems may remain. 1 Zonal sites for two variants in the same subzone are considered to support the same climax plant community, and are thus likely to have similar productivity. Page 11

16 Site Index Adjustment for the Vanderhoof IPA Overall, direct measures of site index (i.e. height/age and growth intercepts) are considered more reliable than indirect estimates such as SIBEC. This is particularly true for the current project, where sampling occurred in a population that is as close as possible to the managed stands of interest for use in timber supply analysis. 4.2 Precision of PEM/SIBEC Estimates There are several sources of variation in the determination of PEM/SIBEC site indices that lead to imprecision in the estimates. These include: There is a general lack of precision in the PEM data. ield audits of PEM projects generally find successful prediction of site series in 60 to 70 % of cases. In the remainder of cases, a prediction error occurs, with a corresponding inappropriate assignment of site index. Assuming that most of these errors result in classifying a site to an adjacent class on the edatopic grid, however, the majority of these introduced errors should be relatively minor (1 to 2 m). SIBEC estimates of site index are a mixture of first and second approximation estimates. As observed in Section 4.1 above, there appears to be significant differences in accuracy between first and second approximation SIBEC estimates, and mixing them will add noise to the system. irst approximation SIBEC estimates for the BC interior are reported in 3 m classes, Classifying data adds an artificial and variable error to all estimates. irst approximation SIBEC estimates appear to be highly imprecise to start with. In igure 5, SIBEC estimates in some cases are very close to measured site indices (within 1 m), while others vary by as much as m. A similar range of variation was found in a comparison of zonal sites in subzones for which two or more variants exist, and where first approximation estimates are reported for on or more variants, and second approximation estimates for the remainder. Mean site indices for zonal sites within a subzone should exhibit only minor variations, but differences as large as.8 m were found, with differences of 3 m or more in 23 out of 57 cases. Overall, it is suspected that much of the variability in the predictive relationships for this project results from imprecision in the first approximation SIBEC estimates. Improvement of second approximation coverage shows good promise for strengthening these relationships. 4.3 Recommendations for SI Adjustment There is little difference in the strengths of the models using elevation alone as a predictive variable versus that using elevation and PEM/SIBEC SI in combination. The use of elevation alone, however, suggests that site index is independent of site series. This result is counterintuitive and is refuted both by observations in this study and considerable other evidence (i.e. Klinka and Wang 1995, Kayahara et al. 1994, Wang et al. 1994). It is therefore recommended that an equation form be utilized that includes PEM/SIBEC SI, if for no other reason than to facilitate future adjustments to the model as SIBEC estimates of SI are improved. Page 12

17 Site Index Adjustment for the Vanderhoof IPA The combined effects of PEM/SIBEC SI and elevation can be expressed in a single equation using multiple regression to predict actual SI. The equation form for all cases is: SI a = a + b 1 x SI p + b 2 x elevation where: SI a = adjusted site index SI p = predicted (PEM/SIBEC) site index Elevation is in metres Coefficients for the resulting equations are provided in Table 2. Table 2. Model coefficient for predicting actual site indices from PEM/SIBEC site indices and elevation. Stratum a b 1 b 2 ESS Pl Dry SBS Pl Moist SBS Pl ESS Sx Dry SBS Sx Moist SBS Sx These models are suitable for determining managed stand site indices by species for individual inventory polygons. Inputs to the model will be derived using the predictive ecosystem mapping data recently acquired for the Vanderhoof orest District, the Ministry of orestsʼ SIBEC data summary, and polygon-average elevations based on the digital elevation models available with the most recent TRIM map layers. Site index adjustments resulting from this project are applicable only to stands with similar history. The adjustments should not be applied to stands of Intermediate Utilization, selection or diameter limit harvest origin. The adjustments reflect stands regenerated using wild seed, and do not reflect any genetic gains that may be realized from later use of seed from breeding programs. 4.4 Implications of Adjusted Site Indices for Projecting Timber Supply Mean site indices for each stratum comparing PEM/SIBEC and measured values are listed in Table 3. The area weighted (by BEC zone) difference for the whole TSA is 2.7 m for pine and 4.8 m for spruce. Table 3. Mean site indices by stratum (standard errors in brackets). Stratum Pl Sx Measured PEM/SIBEC Measured PEM/SIBEC ESS 17.7 (0.3) 14.4 (0.3).1 (0.3) 14.5 (0.2) Dry SBS 21.1 (0.2) 18.5 (0.2) 22.5 (0.3) 16.4 (0.3) Moist SBS 19.9 (0.2) 17.3 (0.2) 21.0 (0.4) 17.7 (0.4) Area Weighted Mean Page 13

18 Site Index Adjustment for the Vanderhoof IPA These differences assume that the relative proportions of various site series is similar on the current harvested area compared to the entire timber harvesting land base, but we know this isnʼt true (see Section 4.5.1). Harvesting over the last few decades is weighted proportionately heavier to the zonal sites, which tend to have higher than average site indices. Coupled with harvesting in the ESS that is weighted heavier to lower elevation sites also with higher than average site indices, it is likely that these land base means are higher than would be derived if adjusted site indices were applied to individual polygons and summed using a GIS system. Rough comparisons can be made to site indices used in previous timber supply analyses for the Vanderhoof portion of the Prince George TSA (Table 4). The difference in sites indices between the TSR II and Vanderhoof IPA analyses resulted from using PEM/SIBEC estimates of SI on a portion of the land base (extrapolated from mapping on two map sheets) instead of inventory height-age derivations of SI. The differences in site indices from the Vanderhoof IPA analysis to area weighted means from this project result from a combination of: a) incomplete PEM coverage and partial use of SIʼs derived from inventory height-age relationships in the IPA analysis (leading to a lower estimate of mean SI than if PEM/ SIBEC estimates had been available for the entire land base), and b) a small bias in selection of areas harvested over the past few decades towards sites with better than average site indices. c) Differences in SIBEC estimates of SI vs. sampled growth intercept estimates. If we could assume that the PEM/SIBEC estimates of SI were accurate, the true area weighted mean SI for the Vanderhoof District would likely be somewhere between the values used in the IPA analysis and those from this project. Table 4. Area weighted mean site indices used in recent timber supply analyses for the Vanderhoof orest District. Analysis Gross Area TSR II IPA Analysis Description Unit (ha) Mean SI Mean SI 5 Spruce, SI 12 79, Spruce, SI < 12 64, Pine 843, High Elevation Se/Bl 51, The PEM/SIBEC adjustments used in the Vanderhoof IPA analysis resulted in a land base level area weighted mean increase in site index of 1.2 m. In turn, the increase in site index resulted in a 9.6% increase in the projected Long Term Harvest Level (LTHL). If we assume a further 1.2 m increase in the mean SI for complete coverage by PEM, and a 3.0 m increase to correct for bias in the SIBEC estimates, it can be surmised that there is potential for a further 34% increase in the LTHL resulting from SI adjustments from this project. Actual increases in both the LTHL and current simulated harvest levels may vary depending upon results of land base level timber supply simulations. Page 14

19 Site Index Adjustment for the Vanderhoof IPA 4.5 Assessment of Risk and Uncertainty Two major sources of uncertainty exist for applying the results of this study. These are: uncertainty as to whether or not the sample frame adequately represents the larger land base, and uncertainty as to whether or not growth intercept estimates of site index truly represent the long term growth potential of the sampled stands Comparison of the Sample rame to the Vanderhoof District Land base A key assumption in the methodology for this project is that the sample frame is reasonably representative of the land base as a whole. or this assumption to be true, the mean difference between regenerated stand SIʼs and those predicted by PEM/SIBEC and elevation would be the same on the sampled area as on the larger land base. While this cannot be ascertained for certain, several factors can be assessed to provide an acceptable comfort level: the relative proportions of each ecosystem unit should be similar in the sampled frame and the total land base, and a similar range of elevations should exist within the sample frame as for the entire land base. The relative proportions of each ecosystem unit in the sampled and total areas of the land base are provided in Table 5. In every case, the magnitudes of the proportions for each ecosystem unit are similar between the sample frame and total land base, although in every case the area covered by the zonal ecosystem unit (01)is higher for the sample frame than for the landscape as a whole. This suggests that harvesting has preferentially targeted the mesic sites, presumably either for better timber or easier operating conditions (although an alternative hypothesis is bias in the PEM data). Given that this shift is relatively small and that PEM/SIBEC SI plays a weak predictive role, however, it is not anticipated that this will have a significant effect on reliability of predicted SIʼs. The elevational ranges of samples by subzone group are illustrated in igure 8. or the SBS subzones, there is excellent coverage of the existing elevational range. The ESS zone, however, is potentially problematic. Most of the samples are at lower elevations with only limited sampling in the 1400 to 50 m range. Given that elevation is the strongest predictor of site index in this analysis, the shortage of higher elevation samples may be quite important, leading potentially to an incorrect slope with regard to elevation in the predictive equations. Given the limited data for elevations above 1400 m in the ESS mv1 subzone, there are several options for applying SI in timber supply analysis, each with its own risks: 1) Use the regression equations developed in this project to extrapolate beyond the range of sample data. This project predicts a 0.9 m drop in spruce site index for every 0 m increase in elevation, a figure that is very similar to (and statistically indistinguishable from) another estimate (-1.1 m SI per 0 m in elevation) developed by Klinka and Wang (1995). Given that there is considerable uncertainty in the elevation coefficient, however, and no evidence to support or refute that changes with elevation continue to be linear above the elevation range sampled, this option is difficult to support. 2) Use unadjusted PEM/SIBEC estimates. This option may provide overall conservative estimates of SI given current SIBEC estimates (particularly for spruce), but given improved second approximation estimates this option would likely result in Page

20 Site Index Adjustment for the Vanderhoof IPA overestimates of SI for higher elevations (given that most of the SIBEC sampling would likely come from lower elevation stands). 3) Use unadjusted inventory height/age estimates of SI. This option will almost certainly provide consistent underestimates. Given that Mo timber supply analysts tend to be conservative in the face of weak or uncertain data, it is likely that option 3 will be their preferred approach. There is very little of the timber harvesting land base above 1400 m, however, and the overall impact will be very small, regardless of which option is utilized. Table 5. Proportion of the Vanderhoof TSA land base (gross area) covered by each ecosystem unit included in the sample design. ESSmv Total Area (192,000 ha) Sample rame (25,000 ha) SBSdk B Total Area (165,000 ha) Sample rame (24,000 ha) SBSdw Total Area (41,000 ha) Sample rame (11,000 ha) SBSdw Total Area (2,000 ha) Sample rame (46,000 ha) SBSmc Total Area (225,000 ha) Sample rame (45,000 ha) SBSmc Total Area (0,000 ha) Sample rame (,000 ha) igure 8. requency distribution of samples by elevation and subzone class Number of Samples SBS dry SBS moist ESS Elevation (m) Page 16

21 Site Index Adjustment for the Vanderhoof IPA Short term growth versus long term potential In using growth intercept estimates of site index in this project, we have measured the growth rates that a discrete and relatively brief period in history. Since growth is sensitive to climate, we need to know whether or not the climate during this period is reasonably typical of what will be experienced in the future. We do know that the last to years has been generally warmer than the previous century, with higher temperatures overall and possibly of greater importance, higher minima during the growing season shoulder periods (pers, Comm. Ross Benton, CS). We suspect that these increased temperatures are leading to increased growth rates, introducing the risk that site index estimates based only on recent growth patterns will over-estimate the longer term trend from the previous century. Of even less certainty are the impacts of future climates. If the recent warming trend is a peak in the normal climatic cycle, then results of this study may reflect an over-estimation of long term growth rates. If instead this temperature trend is the early stage of a prolonged period of global warming, then the study results may under-estimate future growth rates. Adding further uncertainty are the interactions between warming/cooling trends and available moisture. We can look at variations in measured site index on zonal sites between adjacent biogeoclimatic subzones to get clues as to potential impacts of climatic variation on growth rates. Data from this study indicates a 1.2 m difference in pine site index and a 1.8 m difference in spruce site index moving from zonal sites in the drier (but warmer) SBS subzones to zonal sites in the wetter (but cooler) SBS subzones. If we assume that the climatic change from one subzone to the next (which ecologists suggest is quite subtle) is similar in magnitude to the types of climatic change that can occur during normal climatic cycles, then a 1 to 2 m variance in predicted site index might be appropriate for use in sensitivity testing for using results of this study in timber supply analyses. 4.6 Modifications to SI Adjustments with Improved SIBEC Data The results of this project and the resulting SI adjustment procedure are based on the current release of SIBEC site index estimates (SIBEC RDM Version: 02). SIBEC estimates of SI, however, are constantly being updated as additional data becomes available. The current adjustment procedure is primarily one of setting the appropriate range of actual site indices to be expected on future managed stands, but it is relatively crude in that it is insensitive to site-to-site variations. As SIBEC estimates improve, however, so will the potential site-to-site sensitivity of this procedure. or these reasons, a methodology has been included with this report to easily update the coefficients in the SI adjustment equations on an as-needed basis. Associated with this report is an MS Excel spreadsheet that will automatically update the model coefficients simply by entering new SIBEC estimates. This will avoid the necessity for timber supply analysts being restricted to use of the 02 SIBEC estimates. The stipulation for using this spreadsheet is that the same version of SIBEC estimates must be used both for determining the PEM/SIBEC estimates of SI and for calculating the adjustment estimates. Page 17

22 Site Index Adjustment for the Vanderhoof IPA 5.0 Summary and Recommendations This study attempts to provide an unbiased estimate of corrections to biased inventory estimates of site index (from a combination of Predictive Ecosystem Mapping and SIBEC) for use in timber supply analyses. To a large degree, the project has been successful: random sampling of site indices in managed stands and comparisons to the biased PEM/SIBEC estimates has provided suitable data to build the required correction models. The resulting models do not, however, improve on the low precision of the PEM/SIBEC estimates. There is also uncertainty as to whether the measured site indices for young managed stands truly reflect long term site potential, or whether they more truly represent skewed growth rates resulting from atypical climatic conditions observed over the last few decades. It is recommended that: 1. The site index prediction models from this project be used in the next Timber Supply Review to determine site indices for existing and future managed stands in the Vanderhoof orest District. 2. Sensitivity analysis be employed in any timber supply exercises employing these models to test for the impacts of potential climate-induced biases of up to 2 m in site index. 6.0 References arnden, C. 01. A time trend evaluation of interior spruce growth intercept predictions: post plantations. Contract report to BC Ministry of orests, Prince George Region. arnden, C. 03. Sample plan, Vanderhoof IPA site index adjustment project. Contract report to Slocan Group, Plateau Division, Vanderhoof. 13p. J.S. Thrower and Assoc. 00. Time trends in growth intercept predictions for interior spruce. Contract report to the BC Ministry of orests, orest Practices Branch, Victoria. 23 p. Kayahara, G.J., Klinka, K. and Moss, I Influence of site quality on lodgepole pine and interior spruce site index in the sub-boreal spruce zone of British Columbia. Northwood Pulp and Timber Ltd., Prince George BC. 32p. Klinka, K. and Wang, Q Relations between site index of Engelmann spruce, lodgepole pine, and subalpine fir and the measures of site quality in the ESS zone. BC Ministry of orests and Northwood Pulp and Timber Ltd. 63p. Nigh. G.D Revised growth intercept models for lodgepole pine: comparing northern and southern models. Extension Note 11. BC Ministry of orests Research Program. 6p. Nigh. G.D Revised growth intercept models for coastal western hemlock, Sitka spruce, and interior spruce. Extension Note 37. BC Ministry of orests Research Program. 8p. Wang, Q., Wang, G.G., Coates, K.D. and Klinka, K Use of site factors to predict lodgepole pine and interior spruce site index in the sub-boreal spruce zone. Research Note No BC Ministry of orests, Research Program. 26 p. Page 18