Attachment8_VolEstDescription.pdf

Chemult Ranger District | March 2025

Tiny G-Z IRSC Volume Estimation Method

The volume estimates used for this contract were produced from regression modeling of cruise per acre volumes and remotely sensed data, including both LiDAR and multispectral imagery. Cruise volumes were used from 226 cutting units across 18 designation by prescription stewardship contract areas within the Chiloquin and Chemult Ranger Districts. This method uses a multiple linear regression to calculate the relationship between the remotely sensed data and the per acre cruise volume within a given cutting unit, which outputs an equation that is then used in GIS to generate a 30-meter resolution raster that covers the same footprint as the available data, which is nearly the entire Fremont-Winema National Forest, aside from some areas where data is not available.

Statistical measures for the linear regression are available in the table below.

Table 1. Summary of cruise volume multiple linear regression model output for the three separate regressions of tons per acre, gross board feet per acre, and gross cubic feet per acre.

Dependent variable:

Tons Gross

BdFt

Gross

CuFt

(1) (2) (3)

Observations 226 226 226

R2 0.744 0.712 0.714

Adjusted R2 0.695 0.657 0.660

Residual Std. Error (df = 189) 6.520 1,115.886 215.535

F Statistic (df = 36; 189) 15.239* 12.990* 13.127*

Note: *p < 0.01

Using this raster of predicted per acre volume, estimates can be made for individual cutting units by taking the median value of the raster pixels in GIS within that cutting unit. Each pixel represents a 30-by-30-meter area on the ground. Unlike area-based cruising, which uses a grid of plots and samples the trees only on those plots, this method calculates a unit-specific volume based on the characteristics of every individual quarter acre within that unit (each 30-by-30-meter raster pixel represents approximately 0.22 acres). The remotely sensed characteristics themselves were drawn from modeled GIS products created by remote sensing specialists in the Forest Service and at Oregon State University.

In total, 36 remotely sensed variables were used in this regression model. The list of variables is described in the table below.

Table 2. List of the 36 variables used in the volume model. All are 30-meter rasters aside from the plant association group variables, which are 90-meter rasters.

Variable Title Description Source

Asp_Mn Average aspect of the ground surface

Modeled by USFS remote sensing specialists using LiDAR data from summer 2021.

Slp_Mn Slope in degrees

Elev_Mn Elevation in meters

BA_Mn Square feet of basal area per acre

Biom_Mn Live biomass in megagrams per acre

CuFt_Mn Volume of cubic feet of trees >=5 inches

DBH

BdFt_Mn Volume of board feet of trees >=9 inches

DBH

TPA_Mn Trees per acre >=5 inches DBH

SDI_Mn Stand density index

CRDI_Mn Curtis relative density index

QMD_Mn Quadratic mean diameter in inches of trees

>=7 inches DBH

CnpyBH_Mn Canopy base height in meters

CnpyH_Mn Canopy height in meters

CnpyC_Mn Canopy cover in percent

UndC_Mn Understory cover in percent bac_ge3 Square meters per hectare of conifers >=2.5 cm DBH Modeled by Landscape, Ecology, Modeling, Mapping & Analysis (LEMMA), a collaborative research group of the USFS Pacific Northwest Research Station and Oregon State University.

(https://lemma.forestry.oregonstate.edu/data).

pipo_ba Square meters per hectare of ponderosa pine pico_ba Square meters per hectare of lodgepole pine psme_ba Square meters per hectare of Douglas fir abgrc_ba Square meters per hectare of white fir / grand fir svph_ge25 Cubic meters per hectare of snags >=25 cm

DBH and >=2 m tall stph_ge25 Snags per hectare >=25 cm DBH and >=2 m tall ba_ge3 Square meters per hectare of trees >=2.5 cm DBH qmd_dom Quadratic mean diameter in centimeters of all dominant and codominant trees stndhgt

Stand height in meters, computed as average of heights of all dominant and codominant trees mndbhba Basal-area weighted mean diameter in centimeters of all live trees tph_ge3 Trees per hectare >=2.5 cm DBH tph_ge3_25 Trees per hectare >=2.5 cm DBH and

<=25cm DBH tph_ge25_50 Trees per hectare >=25 cm DBH and

<=50cm DBH tph_ge50 Trees per hectare >50 cm DBH

Simpson_PPLP Ponderosa pine and lodgepole pine intermix

Plant association group (PAG) determined by the Forested Plant Associations of the Oregon

East Cascades by Mike Simpson (2007)

Simpson_MCD Dry mixed conifer

Simpson_MCM Moist mixed conifer

Simpson_XER Xeric pine

Simpson_LP Lodgepole pine

Eastside_Screens Status of whether a raster pixel is in the

Eastside Screens Northwest Forest Plan boundaries

Comparisons between estimates from the model and cruises across 19 sales found an average difference of 11.1% between model MBF and cruise MBF and 10.5% for model tons and cruise tons (Tables 3, 4). Of these sales, 6 have both final harvested volumes and cruise data that allow for comparison of their accuracy to the weight scale (this excludes the Hawks DxP SPA Stewardship, which has harvest data but no actual cruise data). The average difference between harvest MBF and the cruise was 14.7% and the model was 15.5%. This includes comparisons with GNA sales, which have some differences in cruising. Of the 6 sales with MBF volume, 3 have scaled volumes in tons, with an average difference from the cruise of 12.3% and from the model of 5.8%.

Because portions of the remotely sensed data were collected during the summer of 2021 (primarily the LiDAR) comparisons can generally only be made for sales that began cutting after that time. Available below are tables of these comparisons for MBF and tons.

Table 3. Awarded sales across the Fremont-Winema National Forest along with cruise volumes and modeled estimates in MBF.

Award Date

Sale Name Cruise

Volume

(MBF)

Model Volume

(MBF)

Difference from

Cruise (%)

Final Harvest Volume (if available) (MBF)

2024-12-18 Gate DxP Stewardship IRTC 6,406 6,310 -1.5%

2024-09-30 Cox Peak 1 DxP SPA Stewardship 11,128 10,436 -6.6%

2024-07-02 BMCF and North 46 DxP IRSC Stewardship 17,600 22,578 22.0%

2023-09-27 Dill SPA Stewardship 6,708 6,146 -9.1%

2023-09-27 CHI WUI DxP IRSC Stewardship 3,340 3,411 2.1%

2023-09-27 Kirk SPA Stewardship 10,879 12,611 13.7%

2023-09-15 Hog DxP IRSC Stewardship 16,059 13,138 -22.2%

2023-09-05 Anders DxP IRTC Stewardship 18,918 16,151 -17.1%

2023-08-23 Hawks DxP SPA Stewardship 8,000 9,320 14.2% 10,632

2023-02-02 GNA Coleman 14,746 14,813 0.5%

2023-02-01 GNA Box 3,672 2,711 -35.4% 3,226

2023-01-19 Spring Creek RD DxP IRTC 480 492 2.5% 408

2022-11-30 Solomon DXP IRTC Stewardship 10,562 8,697 -21.5% 9,198

2021-12-22 Edge DxP IRTC Stewardship 5,643 4,995 -13.0%

2021-05-13 Cave DxP IRSC Stewardship 4,281 4,426 3.3% 4,707

2020-12-17 GVR DxP IRTC Stewardship 4,348 4,126 -5.4%

2020-07-02 Salt DxP IRTC Stew Reoffer 800 888 9.9%

2020-05-08 GNA Rhymno Reoffer 3,017 3,360 10.2% 2,346

2020-04-30 GNA Sick Kid 3,989 4,024 0.9% 4,618

Table 4. Awarded sales across the Fremont-Winema National Forest along with cruise volumes and modeled estimates in tons.

Award Date

Sale Name Cruise

Volume (Tons)

Model Volume (Tons)

Difference from

Cruise (%)

Final Harvest Volume (if available) (Tons)

2024-12-18 Gate DxP Stewardship IRTC 39,868 40,364 1.2%

2024-09-30 Cox Peak 1 DxP SPA Stewardship 84,879 71,592 -18.6%

2024-07-02 BMCF and North 46 DxP IRSC Stewardship 109,708 137,344 20.1%

2023-09-27 Dill SPA Stewardship 40,237 38,970 -3.3%

2023-09-27 CHI WUI DxP IRSC Stewardship 19,646 21,277 7.7%

2023-09-27 Kirk SPA Stewardship 69,013 79,164 12.8%

2023-09-15 Hog DxP IRSC Stewardship 102,441 85,620 -19.6%

2023-09-05 Anders DxP IRTC Stewardship 112,162 101,870 -10.1%

2023-08-23 Hawks DxP SPA Stewardship 47,991 57,486 16.5% 63,777

2023-02-02 GNA Coleman N/A 98,035

2023-02-01 GNA Box N/A 17,838

2023-01-19 Spring Creek RD DxP IRTC 3,089 3,164 2.4% 2,719

2022-11-30 Solomon DXP IRTC Stewardship 67,288 55,828 -20.5% 57,485

2021-12-22 Edge DxP IRTC Stewardship 33,854 32,990 -2.6%

2021-05-13 Cave DxP IRSC Stewardship 26,255 28,885 9.1% 28,984

2020-12-17 GVR DxP IRTC Stewardship 27,148 25,248 -7.5%

2020-07-02 Salt DxP IRTC Stew Reoffer 5,374 5,669 5.2%

2020-05-08 GNA Rhymno Reoffer N/A 21,348

2020-04-30 GNA Sick Kid N/A 25,262

Typically, Fremont-Winema cruises of scaled sales have a sampling error between 20- 29.9% using a 95% confidence level (approximately twice the standard error), therefore the true removal volume that is being estimated by the cruise should fall within a 20-29.9% interval of the sample mean 95% of the time the sampling procedure is repeated. This is technically not a 95% chance that the true removal volume will ultimately fall within that interval, but assuming all cruised acres are cut and no other problems occur (slope is too high in a portion of the unit or a new exclusion is added, etc.), then it is highly likely. For example, the cruise of Solomon DxP IRTC Stewardship had estimated 67,288 removal tons, which ended up being an overestimate of 14.6% compared to the final removal tons (Table 4). This was within Solomon’s sampling error of 26.7%. The regression model estimated 55,828 tons, which is a 3.0% underestimate of the final harvest tons. This estimate of Solomon was performed after the actual removal, but because the remotely sensed data is from 2021, the estimate can still be made.

Because of high coefficients of variation (CV) when cruising DxP stewardship sales, cruising to a much lower error than 20-30% requires an increasingly large number of plots.

At a 110% CV, which is about the average for Fremont-Winema cruises (due to variability in plot-to-plot volume), the number of plots required for a 95% confidence interval and an error of 27.5%, 17.5%, and 7.5% would be 64, 158, and 860, respectively. If regression modeling can produce a reliable and acceptable estimate without the need for large numbers of plots, then it would greatly benefit efficiency and capacity in producing estimates for extensive DxP stewardships.

The two images on page 7 of this attachment show Solomon Butte with cutting units highlighted in light blue from the Solomon DxP IRTC contract area to the west and the Hog DxP IRSC contract area to the east. The first image shows aerial imagery from 2012 (this year was selected to show the old sale units in the vicinity more clearly) and the second shows the modeled volume raster. Blue pixels are estimated to have a volume equal to or below 3000 gross board feet per acre. Comparing the aerial imagery and the raster shows how the modeling predicts virtually no commercial volume in certain areas. The areas being shown as less than 3000 board feet per acre are primarily past FS sales from the early 2000s to 2010s, plantations, open meadows, and small diameter lodgepole pine stands.

The two images on page 8 show aerial imagery from 2022 of Hill DxP IRTC after the removal of commercial volume from units 3 and 4 and halfway through unit 2 with only the northwestern corner remaining. In the areas where cutting has already occurred, the model is estimating minimal commercial volume, therefore most of those areas show up as blue (any pixel with a value 0 to 3000). Although there is volume in the cut areas from the residual trees, this is accounted for in the regression model and theoretically there should be no further commercial volume to remove. Also visible are areas of private land east of Highway 97 that were cut in the late 1990s and currently have low commercial volume.

Figure 1. Aerial imagery of Solomon Butte with Solomon IRTC on the west side and Hog IRSC on the east.

Figure 2. Volume estimation raster for the same area as Figure 1.

Figure 3. Aerial imagery of Hill IRTC cutting units alongside Highway 97.

Figure 4. Volume estimation raster for the same area as Figure 3.