Open Access

Development and evaluation of an individual tree growth and yield model for the mixed species forest of the Adirondacks Region of New York, USA

  • Aaron Weiskittel1,
  • Christian Kuehne1, 4Email author,
  • John Paul McTague2 and
  • Mike Oppenheimer3
Forest Ecosystems20163:26

https://doi.org/10.1186/s40663-016-0086-3

Received: 28 September 2016

Accepted: 14 November 2016

Published: 26 November 2016

Abstract

Background

Growth and yield models are important tools for forest planning. Due to its geographic location, topology, and history of management, the forests of the Adirondacks Region of New York are unique and complex. However, only a relatively limited number of growth and yield models have been developed and/or can be reasonably extended to this region currently.

Methods

In this analysis, 571 long–term continuous forest inventory plots with a total of 10 – 52 years of measurement data from four experimental forests maintained by the State University of New York College of Environmental Science and Forestry and one nonindustrial private forest were used to develop an individual tree growth model for the primary hardwood and softwood species in the region. Species–specific annualized static and dynamic equations were developed using the available data and the system was evaluated for long–term behavior.

Results

Equivalence tests indicated that the Northeast Variant of the Forest Vegetation Simulator (FVS–NE) was biased in its estimation of tree total and bole height, diameter and height increment, and mortality for most species examined. In contrast, the developed static and annualized dynamic, species–specific equations performed quite well given the underlying variability in the data. Long–term model projections were consistent with the data and suggest a relatively robust system for prediction.

Conclusions

Overall, the developed growth model showed reasonable behavior and is a significant improvement over existing models for the region. The model also highlighted the complexities of forest dynamics in the region and should help improve forest planning efforts there.

Keywords

Individual tree growth model Mixed species Forest vegetation simulator

Background

The forests in the Adirondacks Region of New York are a complex mixture of hardwood and softwood species that have a long and varied history of natural disturbance occurrences (Lorimer and White 2003) and human management (McMartin 1994). The region is considered an ecotone at the southernmost end of the eastern forest–boreal ecoregion with over 25 different tree species present and is particularly sensitive to variation in local climate (Beier et al. 2012). Due to variation in elevation and geology, a variety of forest communities occur in this region ranging from high alpine to northern hardwood (Leopold et al. 1988). Historically, research in the region was concentrated on spruce–fir (Picea – Abies) with increasing attention on northern hardwoods (Berven et al. 2013). The forest currently faces a number of issues including climate change (Beier et al. 2012), beech bark disease (McGee 2000), high fern cover (Engelman and Nyland 2006), and atmospheric deposition (Chen et al. 2004), which are all likely to affect future stand dynamics and management options. Consequently, there is a strong need to understand and project the potential influence of alternative forest management strategies.

Growth and yield models are important tools for effective and sustainable forest management and have a long history of development (Weiskittel et al. 2011a). However, a relatively limited number of growth and yield simulators exist for the Adirondacks Region. Aldridge (1982) developed stand–level basal area and volume projections equations, but they were only applicable to northern hardwood stands in the region. Likewise, Pan and Raynal (1995) presented individual tree volume growth equations for three plantation conifers in the region. Today, the most commonly used models in the region currently are Northeast–TWIGS (NE–TWIGS; Hilt and Teck 1989) and the Northeast variant of the Forest Vegetation Simulator (FVS–NE; Dixon et al. 2007). Recent work has suggested NE–TWIGS and FVS–NE to have some important limitations in the region (Ray et al. 2009). For example, Russell et al. (2013) indicated FVS–NE predictions of total height as well as diameter and height increment were statistically not equivalent for many of the common species in the Adirondacks. Based also on recent US Forest Service Forest Inventory and Analysis (FIA) data, Pandit et al. (2012) also indicated that FVS–NE had biased short–term predictions of stand–level basal area growth in New York, especially in the elm–ash (Ulmus–Fraxinus) and oak–hickory (Quercus–Carya) forest types.

The goal of this project was to develop an individual–tree growth and yield simulator that is specific to the Adirondacks Region of New York. Specific objectives were to: (1) test the component equations of Northeast (FVS–NE) for bias in the Adirondacks Region; (2) refit component equations (total height, bole height, diameter increment, height increment, and mortality) as necessary; and (3) evaluate and present long–term prediction behavior for forest types common to the region.

Methods

Study area

The Adirondacks Region of New York is located in Upstate New York and is characterized by a unique mountain range that varies in elevation from 350 to over 1500 m. Climate and soil conditions vary dramatically with elevation, but the average monthly summer high and winter low temperatures are 26 and –15 °C, respectively. Total annual precipitation is around 1044 mm with a nearly equal distribution of precipitation during the year (monthly mean ± SD; 87.1 ± 19.3 mm). Soils in the region are generally young, thin, sandy, acidic, and infertile.

Common forest types in the region include black spruce (Picea mariana (Miller) B.S.P) – northern white cedar swamps (Thuja occidentalis L.), red spruce (Picea rubens Sarg.) – balsam fir (Abies balsamea L.) flats, red spruce – balsam fir – red maple (Acer rubrum L.) – yellow birch (Betula alleghaniensis Britton), and sugar maple (Acer saccharum Marsh.) – American beech (Fagus grandifolia Ehrh.) – yellow birch. Other species present include white spruce (Picea glauca (Moench) Voss.), eastern white pine (Pinus strobus L.), Norway spruce (Picea abies (L.) Karst.), red pine (Pinus resinosa Ait.), eastern hemlock (Tsuga canadensis (L.) Carr.), paper birch (Betula papyrifera Marsh.), gray birch (Betula populifolia Marsh.), quaking aspen (Populus tremuloides Michx.), bigtooth aspen (Populus grandidentata Michx.), black cherry (Prunus serotina Ehrh.), northern red oak (Quercus rubra L.), and white ash (Fraxinus americana L.).

Based on their frequency and occurrence, this analysis focused on the top eight hardwoods (American beech, ash, black cherry, quaking aspen, red maple, red oak, sugar maple, yellow birch) and top six softwoods (balsam fir, eastern hemlock, red pine, spruce, white cedar, white pine). The other species present were grouped into either “other hardwoods” (OH) or “other softwoods” (OS).

Data

The data used in this analysis was obtained from long–term continuous forest inventory (CFI) plots. In addition to the privately owned Shirley Forest (SF), this data included four experimental forests maintained and managed by the State University of New York College of Environmental Science and Forestry (SUNY–ESF), namely Dubuar Memorial Forest (DMF), Huntington Wildlife Forest (HWF), Pack Demonstration Forest (PDF), and Pack Experimental Forest (PEF). Large differences in site quality, based on observed dominant height, are apparent among the experimental forests. For example, Germain et al. (2016) reported white pine site index (dominant/codominant height at base age 50 in m) values of 25 and 16 for HWF and PDF respectively. Given differences in location and sampling, each of these forests are described separately below.

Dubuar memorial forest

Dubuar Memorial Forest is a 1068 ha forest in St. Lawrence County and is near Wanakena, NY. In 1982, approximately 72 potential plot locations were selected to be sampled across the entire property with an even distribution between hardwood and softwood forest types. Exact plot locations were referenced to the north–south/east–west grid system already surveyed on the property. The plots have been measured in 1989, 1996, and 2006 with a total of 5765 observations.

Each CFI plot is a fixed–area series of concentric circles including a 0.08 ha (1/5 acre) for sawtimber [>27.9 cm (>11 in.) DBH for hardwoods and >22.8 cm (>9 in.) DBH for softwoods], a 0.04 ha (1/10 acre) for poletimber [>12.7 cm (>5 in.) DBH], and a 0.001 ha (1/300 acre) for saplings [>2.5 cm (>1 in.) DBH and <12.7 cm (<5 in.) DBH]. Measurement of saplings was only completed in 1982. A subsample of trees were selected for total and bole height measurements. Bole height (BHT) was defined as the distance between a 30 cm stump and either a 10 cm upper–stem diameter outside bark or where the central stem terminates due to forking.

Huntington wildlife forest

Huntington Wildlife Forest is an approximately 6000 ha forest in Essex and Hamilton Counties and near Newcomb, NY. In 1970, 290 plots were established in a systematic grid. The plots have been measured in 1970, 1981, 1991, 2001, and 2011 for a total of 21,170 observations. Each CFI plot is a fixed–area series of concentric circles including a 0.08 ha (1/5 acre) for sawtimber [26.9 cm (>10.6 in.) DBH] and a 0.2 ha (1/20 acre) for poletimber [>11.7 cm (>4.6 in.) DBH and <26.9 cm (<10.6 in.) DBH]. A subsample of trees were selected for total and bole height measurements.

Pack demonstration forest

Pack Demonstration Forest is a 977 ha forest in Warren County near Warrensburg, NY. In 1983, a total of 95 plots were established in a systematic grid. The plots have been measured in 1993 and 2003 for a total of 5795 observations. Each CFI plot is a fixed–area, circular 0.04 ha (1/10 acre) plot with trees >9.1 cm (>3.6 in.) DBH being measured. A subsample of trees were selected for total and bole height measurements.

Pack experimental forest

Pack Experimental Forest is located in St. Lawrence County near Cranberry Lake. In 1989, a total of 27 plots were established in a systematic grid. The plots have been measured in 1989, 1996, and 2006 for a total of 1706 observations. Like Dubuar Memorial Forest, the Pack Experimental Forest uses CFI plots that are a fixed–area series of concentric circles including a 0.08 ha (1/5 acre) for sawtimber [>27.9 cm (>11 in.) DBH for hardwoods and >22.9 cm (>9 in.) DBH for softwoods] and a 0.04 ha (1/10 acre) for poletimber [>12.7 cm (>5 in.) DBH]. A subsample of trees were selected for total and bole height measurements. In contrast to the other experimental forest mentioned above, the Pack Experimental Forest has not received any active forest management in recent decades due to limited access.

Shirley forest

Shirley Forest is a privately owned forest with several locations in Essex County west of Lake Champlain with a total of 93 potential plots established in the 1960s and 1970s. The majority of plots have been measured in 1962/63, 1967/68, 1973/74, 1977/79, 1984/86, 1995/98, 2002/05, and 2011/2014 for a total of 11,060 observations. The circular fixed–area CFI plots were 0.08 ha (1/5 acre) in size and trees with DBH > 12.7 cm (>5 in.) were measured. No height and bole height measurements were taken. SF has received active forest management on a regular basis with harvest operation taking place approximately every 10 to 20 years.

Analysis

All available measurements from the five studied forests not taken after cleaning or harvesting operations were standardized, merged into a common format, and converted to metric units. This resulted in a dataset with 45,496 observations with 16.6% and 15.7% having total and bole height measurements, respectively (Table 1). Table 2 displays the same tree measurement statistics of Table 1 by the key tree species. Missing individual tree total and bole height values were imputed by using a plot– and species–specific mixed–model as outlined by Robinson and Wykoff (2004). Stand– and tree–level attributes were then summarized for each plot and included stem density (TPH, # ha–1), additive stand density index (SDI, # ha–1), quadratic mean diameter (QMD, cm), total basal area (BA, m2∙ha–1), basal area in larger trees (BAL, m2∙ha–1), and percent basal area in hardwood species (PBAHW, %) (Table 3).
Table 1

Individual tree attributes by forest, which are Dubuar Memorial Forest (DMF), Huntington Wildlife Forest (HWF), Pack Demonstration Forest (PDF), Pack Experimental Forest (PEF), and Shirley Forest (SF)

Attributea

N

Mean

SD

Minimum

Maximum

Dubuar Memorial Forest (DMF)

 DBH

5765

28.39

9.76

12.70

79.76

 HT

1058

21.22

4.95

3.96

36.58

 BHT

1053

15.23

5.08

1.83

30.18

 ∆DBH

2423

0.24

0.16

0

1.02

 ∆HT

435

0.19

0.18

0

0.98

 PS

2831

87.35

13.00

21.43

100.00

Huntington Wildlife Forest (HWF)

 DBH

21170

31.62

13.97

11.18

100.84

 HT

4302

21.45

5.01

1.52

38.10

 BHT

3938

15.42

4.98

1.52

29.57

 ∆DBH

11357

0.28

0.17

0

1.27

 ∆HT

1827

0.18

0.15

0

0.91

 PS

13167

88.05

11.53

0

100.00

Pack Demonstration Forest (PDF)

 DBH

5795

21.17

11.64

9.14

88.39

 HT

869

20.08

6.15

5.49

37.80

 BHT

851

13.83

6.34

2.44

30.18

 ∆DBH

2083

0.18

0.13

0

0.99

 ∆HT

327

0.17

0.14

0

0.82

 PS

2422

88.62

8.43

61.73

100.00

Pack Experimental Forest (PEF)

 DBH

1706

29.12

11.61

12.70

80.01

 HT

359

21.49

3.62

10.97

31.09

 BHT

359

14.79

3.52

3.96

25.60

 ∆DBH

803

0.26

0.15

0

1.12

 ∆HT

147

0.24

0.24

0

0.96

 PS

1037

80.12

12.08

40.00

100.00

Shirley Forest (SF)

 DBH

11060

23.94

9.47

9.40

95.50

 ∆DBH

8772

0.31

0.24

0

1.49

 PS

9334

94.72

11.80

0

100.00

Overall

 DBH

45496

27.92

12.77

9.14

100.84

 HT

7557

21.17

5.24

1.52

38.10

 BHT

7150

14.63

5.27

1.22

30.18

 ∆DBH

25438

0.28

0.19

0

1.49

 ∆HT

2736

0.18

0.16

0

0.98

 PS

28791

89.17

12.01

0

100.00

a DBH is diameter at breast height (cm), HT is total height (m), BHT is bole height (m), ∆DBH is annual diameter increment (cm∙yr–1), ∆HT is annual height increment (m∙yr–1), and PS is probability of survival (here percentage of live trees per plot by measurement period, %)

Table 2

Tree–level attributes by species examined in this analysis

Attributea

N

Mean

SD

Minimum

Maximum

American beech (AB)

 DBH

6350

25.11

10.13

9.14

69.85

 HT

928

18.48

4.44

2.44

32.00

 BHT

840

11.50

4.43

1.52

26.52

 ∆DBH

2974

0.30

0.16

0

1.04

 ∆HT

331

0.23

0.19

0

0.91

Ash (AS)

 DBH

883

27.41

12.76

9.14

81.28

 HT

263

22.67

5.17

6.40

33.53

 BHT

250

14.31

5.52

1.83

28.35

 ∆DBH

521

0.30

0.21

0

0.89

 ∆HT

60

0.22

0.17

0

0.70

Black cherry (BC)

 DBH

666

28.70

12.65

9.14

80.01

 HT

212

20.05

5.12

7.62

29.26

 BHT

206

13.22

4.95

2.74

23.47

 ∆HT

58

0.16

0.15

0

0.61

Balsam fir (BF)

 DBH

945

21.74

8.31

9.14

55.12

 HT

131

17.12

5.36

5.18

28.65

 BHT

121

11.43

5.87

1.83

21.64

 ∆DBH

365

0.31

0.20

0

1.14

 ∆HT

30

0.29

0.17

0

0.82

Eastern hemlock (EH)

 DBH

4180

28.70

13.72

9.14

95.76

 HT

635

19.59

5.06

5.49

34.14

 BHT

605

14.92

5.22

1.22

28.65

 ∆DBH

2750

0.30

0.17

0

1.02

 ∆HT

308

0.17

0.14

0

0.67

Other hardwoods (OH)

 DBH

3112

22.37

9.30

9.14

72.39

 HT

309

19.39

5.95

4.57

33.53

 BHT

297

12.41

5.68

1.22

26.21

 ∆DBH

2145

0.23

0.18

0

1.02

 ∆HT

48

0.22

0.19

0

0.74

Other softwoods (OS)

 DBH

1281

26.94

8.53

9.14

58.67

 HT

440

21.47

5.39

5.49

36.58

 BHT

431

13.90

4.73

2.44

24.99

 ∆DBH

456

0.23

0.14

0

0.83

 ∆HT

87

0.20

0.16

0

0.76

Quaking aspen (AS)

 DBH

792

24.35

9.96

9.14

65.53

 ∆DBH

455

0.36

0.23

0

1.14

 ∆HT

14

0.21

0.23

0

0.64

Red maple (RM)

 DBH

4549

28.24

10.63

9.14

72.90

 HT

827

21.39

4.14

1.52

30.18

 BHT

783

14.42

4.13

1.52

24.99

 ∆DBH

2502

0.25

0.15

0

0.62

 ∆HT

304

0.19

0.19

0

0.96

Red pine (RN)

 DBH

1693

28.75

9.71

9.14

61.47

 HT

313

23.00

4.05

9.75

35.05

 BHT

310

17.47

4.53

2.44

28.65

 ∆DBH

822

0.20

0.13

0

0.81

 ∆HT

122

0.21

0.17

0

0.98

Red oak (RO)

 DBH

907

25.14

9.30

9.14

73.41

 ∆DBH

478

0.35

0.23

0

1.30

 ∆HT

35

0.13

0.14

0

0.46

Sugar maple (SM)

 DBH

6289

33.44

16.01

9.14

94.23

 HT

1672

23.40

4.72

7.62

38.10

 BHT

1604

16.09

5.04

2.13

28.35

 ∆DBH

3799

0.26

0.17

0

1.27

 ∆HT

592

0.16

0.15

0

0.83

Spruce (SP)

 DBH

3051

25.73

10.99

9.14

64.01

 HT

498

18.96

5.69

4.27

32.61

 BHT

443

14.88

5.66

1.52

28.04

 ∆DBH

1499

0.18

0.12

0

0.53

 ∆HT

203

0.14

0.13

0

0.52

White cedar (WC)

 DBH

423

28.28

8.31

11.68

56.13

 ∆DBH

332

0.17

0.12

0

0.72

 ∆HT

12

0.19

0.13

0

0.49

White pine (WP)

 DBH

5790

27.07

13.52

9.14

95.50

 HT

340

23.25

7.18

5.79

37.80

 BHT

329

17.98

7.11

2.44

30.18

 ∆DBH

3754

0.38

0.27

0

1.49

 ∆HT

122

0.21

0.17

0

0.65

Yellow birch (YB)

 DBH

4585

31.88

14.15

9.14

100.84

 HT

989

21.37

4.18

7.62

34.14

 BHT

931

14.52

4.29

2.13

25.60

 ∆DBH

2586

0.25

0.17

0

1.27

 ∆HT

412

0.16

0.15

0

0.64

a DBH is diameter at breast height (cm), HT is total height (m), BHT is bole height (m), ∆DBH is annual diameter increment (cm∙yr–1), and ∆HT is annual height increment (m∙yr–1), see Table 8 for probability of survival (PS, %)

Table 3

Key plot– and tree–level attributes by forest

Attributea

N

Mean

SD

Minimum

Maximum

Dunbar Memorial Forest (DMF)

 TPH

197

496.3

253.0

37.1

1779.2

 SDI

197

540.4

243.1

26.9

1211.5

 QMD

197

29.3

5.9

13.1

46.4

 BA

197

29.4

13.8

1.1

70.6

 BAL

197

17.1

8.2

0.3

42.8

 PBAHW

197

56.4

44.6

0

100.0

 CSI

72

17.2

0.3

16.4

18.1

Huntington Wildlife Forest (HWF)

 TPH

1207

447.0

238.8

12.4

1803.9

 SDI

1207

466.9

183.0

14.3

1090.5

 QMD

1207

34.3

7.0

11.7

67.8

 BA

1207

26.2

10.6

0.5

67.4

 BAL

1207

14.9

6.6

0

41.8

 PBAHW

1207

79.5

27.0

0

100.0

 CSI

281

16.9

0.5

15.6

18.7

Pack Demonstration Forest (PDF)

 TPH

171

837.4

381.4

24.7

2223.9

 SDI

171

710.0

255.2

11.2

1346.5

 QMD

171

24.6

5.7

14.1

44.8

 BA

171

38.4

15.3

0.5

89.6

 BAL

171

27.7

11.5

0

66.0

 PBAHW

171

34.4

31.6

0

100.0

 CSI

77

17.3

0.5

16.4

18.3

Pack Experimental Forest (PEF)

 TPH

74

413.6

121.2

160.6

741.3

 SDI

74

465.5

99.2

269.1

695.8

 QMD

74

31.6

4.8

23.9

48.8

 BA

74

25.9

6.1

15.5

43.6

 BAL

74

15.5

4.2

8.6

29.8

 PBAHW

74

94.5

20.0

0

100.0

 CSI

25

17.0

0.3

16.3

17.5

Shirley Forest (SF)

 TPH

328

416.6

214.3

37.1

1630.9

 SDI

328

405.9

172.4

18.9

1074.4

 QMD

328

26.1

4.7

15.8

40.4

 BA

328

21.7

9.8

0.8

51.8

 BAL

328

14.2

6.5

0.3

36.0

 PBAHW

328

46.2

33.9

0

100.0

 CSI

93

17.6

1.4

15.7

20.3

a TPH is trees per ha (# ha–1), SDI is stand density index (# ha–1), QMD is quadratic mean diameter (cm), BA is total stand basal area (m2∙ha–1), BAL is basal area in larger trees (m2∙ha–1), PBAHW is the percent basal area in hardwood species (%), and CSI is the climate site index (m)

Since tree age measurements were not available, climate site index (CSI, m) of Weiskittel et al. (2011b) was used in place of traditional site index, which is similar to the Acadian variant of FVS (Weiskittel et al. 2012). CSI is a base–age 50 years estimate of site index based on downscaled climate variables (http://forest.moscowfsl.wsu.edu/climate/) and a random forest nonparametric regression model built on field–observed values (Table 3). The general approach to model and derive CSI has been described in detail previously (Jiang et al. 2015). Since FVS–NE requires traditional site index, it was assumed that CSI was an equivalent estimate for the fastest–growing species on the plot. FVS–NE then uses the estimate for the fastest–growing species on the plot to convert to species–specific values based on nine species groups (Dixon et al. 2007).

To evaluate the suitability of FVS–NE component equations, an equivalence test (Robinson and Froese 2004) with 15% allowable error was conducted. Equations were then fit to the Adirondacks data and included individual tree total and bole height, diameter and height increment, and mortality. Each of these equations was fit to the primary species using nonlinear mixed effects modeling (NLME) with each forest being treated as random (Pinheiro et al. 2016). Extending the random effects structure by including plots and trees did not improve prediction accuracy. The specific equations are further described separately below. All analyses were conducted using the programming software R version 3.3.1 (R Core Team 2016).

The primary species included in this analysis were American beech (AB), ashes (AS, including white and black ash (Fraxinus nigra Marshall)), black cherry (BC), balsam fir (BF), eastern hemlock (EH), quaking aspen (QA), red maple (RM), red pine (RN), red oak (RO), sugar maple (SM), spruces (SP, including black, red, and white spruce), northern white cedar (WC), white pine (WP), yellow birch (YB, also including river birch (Betula nigra L.)), other hardwood species (OH), and other softwood species (OS). However, species specific equations could not be derived for all attributes.

Total height

Based on the findings of Rijal et al. (2012b), a modified Chapman–Richards equation form was used for the prediction and imputation of total tree height (HT, m):
$$ \mathrm{H}\mathrm{T}\kern0.5em =\kern0.5em {b}_{10}\cdot {\left(1\hbox{--} \exp \left(\hbox{--} {b}_{11}\cdot \mathrm{D}\mathrm{B}\mathrm{H}\right)\right)}^{\left({b}_{12}+{b}_{13}\cdot \ln \left(\mathrm{BA}+1\right)+{b}_{14}\cdot \ln \left(\mathrm{B}\mathrm{A}\mathrm{L}+1\right)\right)} $$
(1)
where DBH is diameter at breast height (cm), BA is total basal area (m2∙ha–1), and BAL is the basal area in larger trees (m2∙ha–1). Unlike Rijal et al. (2012b), climate site index was not found to be a significant predictor in preliminary analysis. In addition, an additive term (e.g. 1.3 m) for when DBH is equal to zero was not included as prior work has suggested that this constraint may cause poorer model performance across the full range of DBH (Newton and Amponsah 2007).

Bole height

Although Rijal et al. (2012a) modeled height to crown base and our Adirondacks data contained measurements of bole height (defined as the first live branch or a 10.2 cm (4 in) top), its applicability was still tested. However, rather than using a modified Chapman–Richards equation form similar to Rijal et al. (2012a), a modified logistic similar to Hann et al. (2003) was used for the prediction and imputation of bole height (BHT, m):
$$ \mathrm{B}\mathrm{H}\mathrm{T}\kern0.75em =\kern0.5em \frac{\mathrm{HT}}{\left(1\kern0.5em +\kern0.5em \exp \left(\begin{array}{c}\hfill {b}_{20}+{b}_{21}\cdot \mathrm{H}\mathrm{T}+{b}_{22}\cdot \ln \left(\mathrm{B}\mathrm{A}\mathrm{L}+0.1\right)+{b}_{23}\cdot \ln \left(\mathrm{BA}\right)+\hfill \\ {}\hfill {b}_{24}\cdot \ln \left(\frac{\mathrm{DBH}}{\mathrm{HT}}\right)+{b}_{25}\cdot \mathrm{C}\mathrm{S}\mathrm{I}\hfill \end{array}\right)\right)} $$
(2)
where CSI is climate site index (m), and all other variables are defined above.

Diameter increment

A variety of model forms and potential covariates were tested for individual tree diameter increment predictions, but it was found that it was best modeled using an equation form similar to Hann et al. (2003):
$$ \varDelta \mathrm{D}\mathrm{B}\mathrm{H}\kern0.5em =\kern0.5em \exp \left(\begin{array}{c}\hfill {b}_{30}+{b}_{31}\cdot \ln \left(\mathrm{D}\mathrm{B}\mathrm{H}\right)+{b}_{32}\cdot \mathrm{D}\mathrm{B}\mathrm{H}+{b}_{33}\cdot \left(\mathrm{B}\mathrm{A}\mathrm{L}\right) + \hfill \\ {}\hfill {b}_{34}\cdot \sqrt{\mathrm{BA}}+{b}_{35}\cdot \mathrm{C}\mathrm{S}\mathrm{I}\hfill \end{array}\right) $$
(3)
where ∆DBH is the annual diameter increment (cm∙yr–1) and all other variables have been defined above. Preliminary analysis indicated that neither bole nor crown ratio (CR, (HT–BHT)/HT) were effective predictors. In addition, variables expressing species composition such as PBAHW and relative density (Stout and Nyland 1986; Woodall et al. 2005) were not found to be influential.

Since annual parameters were desired but the observed variables were on longer growth intervals (2–17 years) parameters were annualized using an iterative technique of Weiskittel et al. (2007). Based on Cao (2000) the right side of the equation was a function which summed the estimated annual ∆DBH estimates over the number of growing seasons during the observed growth period using the updated parameter estimates from the NLME optimization algorithms. For each growing season during the growth period, DBH was subsequently updated, while BAL and BA were linearly interpolated between their beginning values and ending values. Final equations were fitted using initial parameter estimates derived using non–linear least square modeling and assuming SI was constant over time. Although this assumption of linear growth is likely too simplified for highly irregular and longer remeasurement intervals (>10 years), the iterative approach used in this analysis does produce model behavior similar to a more sophisticated optimization approach and is more effective than using the remeasurement interval as a covariate (e.g. Juma et al. 2014).

Height increment

A variety of height increment models including the one used in FVS–NE and those presented in Russell et al. (2014) were tested. The following form was shown to perform best:
$$ \varDelta \mathrm{H}\mathrm{T}\kern0.5em =\kern0.5em \exp \left(\begin{array}{c}\hfill {b}_{40}+{b}_{41}\cdot \mathrm{H}\mathrm{T}+{b}_{42}\cdot \ln \left(\mathrm{H}\mathrm{T}\right) + {b}_{43}\cdot \left(\frac{\mathrm{DBH}}{\mathrm{HT}}\right)+\hfill \\ {}\hfill {b}_{44}\cdot \left(\mathrm{B}\mathrm{A}\mathrm{L}\right)+{b}_{45}\cdot \mathrm{BA}+{b}_{46}\cdot \ln \left(\mathrm{BA}\right)\hfill \end{array}\right) $$
(4)
where ∆HT is the annual height increment (m∙yr–1) and all other variables have been defined above. Like diameter increment, the parameters were annualized using an iterative technique of Weiskittel et al. (2007). Initially, the model was fit for each species individually. However, no or poor model convergence resulted, likely because of small sample sizes for most species and high variability (Table 2). Instead, the model was fit by treating b 40 and b 41 as random parameters that varied by species similar to the approach outlined in Russell et al. (2014). CSI and variables describing species composition were not found to be significant predictors.

Mortality

A logistic function was used to model the probability of individual tree survival:
$$ \mathrm{P}S = \kern0.5em {\left(1 + \exp \left(\hbox{--} \left(\begin{array}{c}\hfill {b}_{50}+{b}_{51}\cdot \mathrm{D}\mathrm{B}\mathrm{H}+{b}_{52}\cdot \ln \left(\mathrm{D}\mathrm{B}\mathrm{H}\right)+{b}_{53}\cdot \mathrm{C}\mathrm{R}+{b}_{54}\cdot \mathrm{B}\mathrm{A}\mathrm{L}+\hfill \\ {}\hfill {b}_{55}\cdot \sqrt{\mathrm{BA}}+{b}_{56}\cdot \mathrm{C}\mathrm{S}\mathrm{I}\hfill \end{array}\right)\right)\right)}^{-1} $$
(5)
where PS is the probability of annual survival and all other variables have been defined above. The parameters were annualized using an iterative technique of Weiskittel et al. (2007).

Evaluation

For all developed equations presented above, biological consistency with expectation, parameter significance, sign, and magnitude as well as degree of multicollinearity were evaluated. Equation fit statistics assessed were mean bias (MB; observed – predicted) and root mean square error (RMSE). MB and RMSE were computed using the fixed–effects only. For the mortality equation, a receiver operator curve (ROC) was constructed and area under the curve (AUC) computed (Hein and Weiskittel 2010). To determine the optimal cutpoint for the species–specific survival equations (Rota and Antolini 2014), the function ‘coords’ of the R package ‘pROC’ was used (Robin et al. 2011).

Simulations

To evaluate the long–term behavior of the equations, a simulation model was constructed by linking all of the component equations and using them to project each plot from its initial to final measurement. Missing total tree heights at the start of each plot specific simulation run were predicted using Eq. [1], while subsequent height growth was predicted with Eq. [4]. Due to its importance on long–term simulations, the prediction of tree mortality was handled in two ways, namely an expansion factor method and a fixed cutpoint (Weiskittel et al. 2011a). In the expansion factor method, the tree’s expansion factor was annually multiplied by the probability of survival (Eq. [5]), while in the fixed cutpoint method an entire tree record was killed if the probability of survival exceeded the optimal cutpoint derived from the species specific survival equations. Plots that were harvested or showed signs of either excessive mortality (>20% of total basal area) or excessive ingrowth (≥5 trees) were included up until the occurrence of the event. MB and RMSE were computed for key stand–level attributes like stem density and basal area. Finally, some common forest types in the region were projected for 50 years assuming similar initial starting conditions.

Results

Total height

A total of 7575 observations were available for the total height equation with most of the observations being sugar maple (22.1%), yellow birch (13.1%), American beech (12.3%), and red maple (10.9%). For most species, an equivalence test suggested that the observed values and the predicted values from FVS–NE were not statistically different at the 15% levels of allowable error (Table 4). Species–specific equations were developed and resulted in RMSE between 2.5 to 4.5 m. For a given set of covariates, sugar maple had the greatest total height, while eastern hemlock had the smallest (Fig. 1).
Table 4

Results of the equivalence test of the Northeast (NE) Variant of FVS for total height (m) and the parameter estimates and fit statistics for Eq. [1]. Mean bias (MB; observed – predicted) and root mean square error (RMSE) were computed using the fixed effects only. Both mean difference and MB were computed as observed – predicted. The null hypothesis for the equivalence test is that μ 1 ≠ μ 2

SPPa

N

FVS–NE

Equation [1]

Mean ± SE difference (m)

Result of equivalence test

b 10

b 11

b 12

b 13

b 14

MB (m)

RMSE (m)

AB

928

–0.45 ± 0.11

rejected

33.6020

0.0201

1.0584

–0.0498

–0.0889

0.0009

2.9021

AS

263

3.33 ± 0.21

not rejected

42.0612

0.0093

0.7372

–0.0916

0.0011

2.8813

BC

212

2.00 ± 0.26

rejected

30.2327

0.0243

1.4215

–0.2522

0.0814

2.6475

BF

131

1.33 ± 0.27

rejected

42.0696

0.0121

0.8860

–0.0720

0.0011

2.8830

EH

635

–1.43 ± 0.12

rejected

48.0279

0.0117

1.2979

–0.0981

–0.0429

–0.4499

2.7909

OH

309

0.23 ± 0.22

rejected

35.4457

0.0242

1.0128

–0.0755

0.1742

3.7033

OS

440

2.82 ± 0.22

not rejected

26.1875

0.0819

3.5310

–0.3375

0.7962

4.5155

RM

827

2.49 ± 0.11

rejected

29.9463

0.0331

1.2740

–0.0892

–0.0878

–0.0496

2.7841

RN

313

4.19 ± 0.16

not rejected

29.9419

0.0752

3.5272

–0.5136

–0.4386

2.7722

SM

1672

2.21 ± 0.08

rejected

38.0384

0.0211

1.0325

–0.0544

–0.0639

–1.0922

2.9732

SP

498

0.36 ± 0.13

rejected

42.3324

0.0217

1.4738

–0.0453

–0.1076

–0.4566

2.5105

WP

340

1.85 ± 0.20

rejected

37.7213

0.0289

1.8887

–0.1368

–0.1392

0.3715

3.1214

YB

989

0.58 ± 0.10

rejected

29.5755

0.0286

1.0028

–0.0299

–0.0886

0.0001

2.7534

a SPP Species are AB American beech, AS ash, BF balsam fir, EH eastern hemlock, OH other hardwoods, OS other softwoods, RM red maple, RN red pine, SM sugar maple, SP spruce, WP white pine, and YB yellow birch

Fig. 1

Predictions of total height (HT, m) and bole height (BHT, m) across diameter at breast height (DBH, cm) for different species using Eqs. [1] and [2], respectively. Other equation covariates were set at mean levels, which were 10 m2∙ha–1, 25 m2∙ha–1, and 17 m for basal area in larger trees, total basal area, and climate site index, respectively. Species are American beech (AB), balsam fir (BF), eastern hemlock (EH), red maple (RM), sugar maple (SM), spruce (SP), white pine (WP), and yellow birch (YB)

Bole height

A total of 7150 observations were available for the bole height equation with most of the observations being sugar maple (22.4%), yellow birch (13.0%), American beech (11.7%), and red maple (11.0%). For all species, an equivalence test suggested that the observed values and those predicted by FVS–NE were statistically different at the 15% levels of allowable error (Table 5). Species–specific equations were developed and resulted in RMSE between 1.6 to 3.8 m. For a given set of covariates, sugar maple and white pine showed the highest bole height for the majority of DBH range examined, while eastern hemlock showed the lowest (Fig. 1)
Table 5

Results of the equivalence test Northeast (NE) Variant of FVS for bole height (m) and the parameter estimates and fit statistics for Eq. [2]. Mean bias (MB; observed – predicted) and root mean square error (RMSE) were computed using the fixed effects only. Both mean difference and MB were computed as observed – predicted. The null hypothesis for the equivalence test is that μ 1 ≠ μ 2

SPPa

N

FVS–NE

Equation [2]

Mean ± SE difference (m)

Result of equivalence test

b 20

b 21

b 22

b 23

b 24

b 25

MB (m)

RMSE (m)

AB

840

4.93 ± 0.13

not rejected

–1.3767

–0.0613

–0.0408

–1.2310

0.9215

–0.0268

1.8829

AS

250

2.70 ± 0.27

not rejected

–4.5957

–0.0449

–0.2274

1.7595

–0.0026

2.5730

BC

206

2.93 ± 0.25

not rejected

–7.2046

–0.0490

–0.2216

2.6474

0.0077

2.2417

BF

121

6.03 ± 0.45

not rejected

–5.7095

–0.0991

–1.1598

2.5493

–0.0657

1.8356

EH

605

8.48 ± 0.16

not rejected

–12.3299

–0.0654

–0.0418

–1.1709

4.7187

–0.0341

1.6112

OH

297

3.15 ± 0.23

not rejected

–2.1695

–0.0546

–0.1961

–0.5721

1.2364

–0.1407

1.9742

OS

431

5.03 ± 0.18

not rejected

0.0110

–0.0160

–0.0184

–0.1176

–1.0851

3.8629

RM

783

5.27 ± 0.10

not rejected

0.7482

–0.0322

–0.1254

–0.4321

0.7071

2.1710

RN

310

9.04 ± 0.22

not rejected

0.0821

–0.0206

–0.1456

–0.4580

0.3897

2.1899

SM

1604

5.11 ± 0.09

not rejected

–2.1520

–0.0229

–0.0701

–0.2583

0.8273

0.4317

2.5170

SP

443

8.85 ± 0.21

not rejected

0.8253

–0.0604

–0.0350

–1.1086

0.6008

1.6807

WP

329

8.98 ± 0.29

not rejected

–4.3500

–0.0437

–0.0279

–0.7877

1.6198

–0.0769

1.6524

YB

931

4.48 ± 0.09

not rejected

–1.2995

–0.0442

–0.0316

–0.0660

–0.8534

0.8015

–0.0228

1.7532

a SPP Species are AB American beech, AS ash, BF balsam fir, EH eastern hemlock, OH other hardwoods, OS other softwoods, RM red maple, RN red pine, SM sugar maple, SP spruce (SP), WP white pine, and YB yellow birch

.

Diameter increment

A total of 25,438 observations were available for the diameter increment equation with most of the observations being sugar maple (14.9%), white pine (14.8%), American beech (11.7%), and yellow birch (10.2%). An equivalence test suggested that for all species, the observed values and the predicted values from FVS–NE were statistically different at the 15% levels of allowable error (Table 6). Species–specific equations were developed and resulted in RMSE between 0.11 and 0.22 cm∙yr–1. For a given set of covariates, red maple and American beech showed the greatest diameter increment at smaller tree sizes (DBH < 25 cm), while white pine and eastern hemlock showed the greatest diameter increment for larger tree sizes (DBH > 25 cm; Fig. 2). Spruces had the slowest diameter increment across a range of tree sizes.
Fig. 2

Predictions of annual diameter (ΔDBH, cm∙yr–1), height increment (ΔHT, m∙yr–1), and annual probability of survival across diameter at breast height (DBH, cm) for different species using Eqs. [3], [4], and [5] respectively. For all equations, basal area in larger trees, total basal area, crown ratio, and climate site index were fixed at 10 m2∙ha–1, 25 m2∙ha–1, 0.33, and 17 m, respectively. For height increment, total height (HT, m) was varied as fixed percentage of the DBH (HT = DBH × 0.6). Species are American beech (AB), balsam fir (BF), eastern hemlock (EH), red maple (RM), sugar maple (SM), spruce (SP), white pine (WP), and yellow birch (YB)

Table 6

Results of the equivalence test of Northeast Variant (NE) of FVS for diameter increment (cm yr–1) and the parameter estimates and fit statistics for Eq. [3]. Mean bias (MB; observed – predicted) and root mean square error (RMSE) were computed using the fixed effects only. Both mean difference and MB were computed as observed – predicted. The null hypothesis for the equivalence test is that μ1 ≠ μ2

SPPa

N

FVS–NE

Equation [3]

Mean ± SE difference (m)

Result of equivalence test

b 30

b 31

b 32

b 33

b 34

b 35

MB (cm)

RMSE (cm)

AB

2974

0.23 ± 0.003

not rejected

–3.4430

1.4219

–0.0727

–0.0250

–0.0054

0.1508

AS

521

0.20 ± 0.009

not rejected

–6.1840

0.0195

–0.0020

–0.0211

1.800

0.0339

0.2030

BF

365

0.23 ± 0.010

not rejected

–2.5924

1.2056

–0.0556

–0.0064

–0.2013

0.0488

0.1972

EH

2750

0.19 ± 0.003

not rejected

–1.7508

0.5775

–0.0113

–0.0057

–0.1795

0.0126

0.1437

OH

2145

0.12 ± 0.004

not rejected

–1.5197

0.3231

–0.0225

–0.0286

–0.0048

0.1685

OS

456

0.16 ± 0.007

not rejected

–3.4203

1.0599

–0.0301

–0.0183

–0.0541

–0.0060

0.1234

QA

455

0.21 ± 0.011

not rejected

–2.3591

1.0659

–0.0653

–0.0469

0.0239

0.2122

RM

2502

0.18 ± 0.003

not rejected

–4.1465

0.2889

–0.0148

–0.0156

0.8415

0.0157

0.1445

RN

822

0.09 ± 0.005

not rejected

–7.6829

2.7843

–0.1017

–0.0210

0.0359

0.1259

RO

478

0.24 ± 0.011

not rejected

–2.3680

0.8776

–0.0537

–0.0159

0.0356

0.2218

SM

3799

0.19 ± 0.003

not rejected

–2.6059

0.4614

–0.0176

–0.0156

–0.1351

0.3294

0.0411

0.1616

SP

1499

0.11 ± 0.003

not rejected

–4.5654

0.3646

–0.0098

–0.0258

0.7272

0.0482

0.1259

WC

332

0.10 ± 0.007

not rejected

–2.8205

0.4290

–0.0042

–0.0111

0.0011

0.1108

WP

3754

0.25 ± 0.004

not rejected

–4.5354

0.9649

–0.0208

–0.0230

–0.0986

0.6226

0.0237

0.2197

YB

2586

0.11 ± 0.003

not rejected

–1.0956

0.4043

–0.0186

–0.0123

–0.1808

0.0043

0.1595

a SPP Species are AB American beech, AS ash, BF balsam fir, EH eastern hemlock, OH other hardwoods, OS other softwoods, QA quaking aspen, RM red maple, RN red pine, RO red oak, SM sugar maple, SP spruce, WC white cedar, WP white pine, and YB yellow birch

Height increment

A total of 2744 observations were available for the height increment equation with most of the observation being sugar maple (21.6%), yellow birch (15.3%), American beech (12.1%), and eastern hemlock (11.2%). For all species except eastern hemlock, an equivalence test suggested that the observed values and those predicted by FVS–NE were statistically different at the 15% level of allowable error (Table 7). As mentioned in the Methods section, a preliminary species–specific equation did not converge for the majority of species so an equation was fitted by using species as random effect. This resulted in an adequate fit with RMSE ranging from 0.16 to 0.25 m∙yr–1.

For a given set of covariates, yellow birch and American beech had the greatest height increment for smaller tree sizes (DBH < 25 cm), while spruce had the smallest height increment (Fig. 2). On larger trees (DBH > 25 cm), white pine and balsam fir had the greatest height increment and American beech and yellow birch had the lowest. Since small–sized suppressed yellow birch individuals were not available for our analysis, predicted height growth thus might be too optimistic.
Table 7

Results of the equivalence test of the Northeast (NE) variants of FVS for height increment and the parameter estimates and fit statistics for Eq. [4]. Mean bias (MB; observed – predicted) and root mean square error (RMSE) by species are presented. Both mean difference and MB were computed as observed – predicted. The null hypothesis for the equivalence test is that μ1 ≠ μ2. Columns marked with an asterisk (*) were those that were treated as a random effect and presented as the fixed + species random effect

SPPa

N

FVS–NE

Equation [4]

Mean ± SE difference (m)

Result of equivalence test

b 40*

b 41*

b 42

b 43

b 44

b 45

b 46

MB (m)

RMSE (m)

AB

331

0.059 ± 0.010

not rejected

–2.0016

–0.1519

0.8261

–0.0583

–0.0121

–0.0049

0.3741

0.0009

0.1640

AS

60

0.168 ± 0.020

not rejected

–2.4663

–0.1134

0.8261

–0.0583

–0.0121

–0.0049

0.3741

0.0171

0.1555

BC

58

–0.011 ± 0.020

not rejected

–2.6260

–0.1227

0.8261

–0.0583

–0.0121

–0.0049

0.3741

–0.0081

0.1399

BF

30

0.102 ± 0.042

not rejected

–2.7233

–0.1037

0.8261

–0.0583

–0.0121

–0.0049

0.3741

0.0253

0.1537

EH

308

0.001 ± 0.008

rejected

–2.8409

–0.1169

0.8261

–0.0583

–0.0121

–0.0049

0.3741

–0.0042

0.1337

OH

48

0.133 ± 0.025

not rejected

–2.7560

–0.1111

0.8261

–0.0583

–0.0121

–0.0049

0.3741

0.0231

0.1820

OS

85

0.029 ± 0.021

not rejected

–2.7532

–0.1081

0.8261

–0.0583

–0.0121

–0.0049

0.3741

0.0004

0.1471

QA

14

0.134 ± 0.061

not rejected

–2.6734

–0.1114

0.8261

–0.0583

–0.0121

–0.0049

0.3741

0.0189

0.2232

RM

304

0.058 ± 0.011

not rejected

–2.4157

–0.1284

0.8261

–0.0583

–0.0121

–0.0049

0.3741

0.0142

0.1801

RN

122

0.094 ± 0.015

not rejected

–2.8316

–0.0979

0.8261

–0.0583

–0.0121

–0.0049

0.3741

0.0110

0.1636

RO

35

0.033 ± 0.025

not rejected

–3.0972

–0.1040

0.8261

–0.0583

–0.0121

–0.0049

0.3741

–0.0314

0.1420

SM

592

0.093 ± 0.006

not rejected

–2.5667

–0.1180

0.8261

–0.0583

–0.0121

–0.0049

0.3741

0.0024

0.1436

SP

203

–0.023 ± 0.010

not rejected

–3.1431

–0.1069

0.8261

–0.0583

–0.0121

–0.0049

0.3741

–0.0094

0.1235

WC

12

0.043 ± 0.035

not rejected

–2.8164

–0.1139

0.8261

–0.0583

–0.0121

–0.0049

0.3741

–0.0481

0.1123

WP

122

0.141 ± 0.015

not rejected

–3.1466

–0.0836

0.8261

–0.0583

–0.0121

–0.0049

0.3741

0.0087

0.1596

YB

412

0.090 ± 0.007

not rejected

–2.0130

–0.1526

0.8261

–0.0583

–0.0121

–0.0049

0.3741

0.0010

0.1277

a SPP Species are AB American beech, AS ash, BC black cherry, BF balsam fir, EH eastern hemlock, OH other hardwoods, OS other softwoods, QA quaking aspen, RM red maple, RN red pine, RO red oak, SM sugar maple, SP spruce, WC white cedar, WP white pine, and YB yellow birch

Survival

A total of 28,791 observations were available for the survival analysis with most of the observation being sugar maple (14.6%), white pine (13.9%), American beech (12.5%), and yellow birch (10.2%). Balsam fir (28.5%), quaking aspen (19.0%), American beech (16.6%), and other softwoods (14.4%) showed the highest percentage of dead trees. A species–specific survival equation fit to the data showed a much higher AUC (0.74 ± 0.08; 0.63 – 0.95) compared to FVS–NE predictions (0.57 ± 0.09; 0.46 – 0.69, Table 8). The determined optimal cutpoint varied by species and ranged from 0.75 to 0.98. Across the various tree sizes examined, red maple showed the highest level of survival, while American beech and balsam fir displayed some of the lowest predicted survival rates (Fig. 2).
Table 8

Parameter estimates for individual tree survival (Eq. [5]), the area under the curve (AUC) and optimal cutpoint (OCP) for the developed equation and the ones used in the Northeast Variant of FVS (FVS–NE) by species

SPPa

N

FVS–NE

Equation [5]

Alive

Dead

AUC

AUC

OCP

b 50

b 51

b 52

b 53

b 54

b 55

b 56

AB

2991

596

0.6885

0.7740

0.8028

7.4069

–0.0855

2.9649

–0.1041

AS

541

59

0.5439

0.6962

0.8900

–2.6972

–0.0511

2.7220

3.1868

–0.2677

BF

364

145

0.6190

0.6936

0.7480

5.4321

–0.0618

–0.0383

EH

2752

88

0.5098

0.6301

0.9584

–1.9897

–0.0809

3.1020

OH

2159

315

0.6872

0.7313

0.8760

–0.7529

–0.0709

2.5867

-

–0.1076

OS

470

79

0.5585

0.6767

0.8718

15.1544

–0.2712

2.4283

–0.2753

QA

465

109

0.6645

0.7304

0.8296

0.2811

–0.0517

1.7626

–0.2104

RM

2587

284

0.4793

0.6926

0.8906

11.2896

–0.1113

–0.1480

RN

845

58

0.4866

0.9483

0.9365

8.9149

0.2446

–0.5346

RO

530

12

0.5741

0.7962

0.9757

–6.2021

–0.0889

4.5545

SM

3842

359

0.4556

0.7453

0.9314

3.7231

–0.1037

1.9431

–0.1096

SP

1613

240

0.5690

0.7034

0.8911

7.1204

–0.0591

–0.0794

WC

324

11

0.4753

0.8157

0.9541

10.5219

0.1672

–1.6906

WP

3782

222

0.7106

0.7980

0.9212

4.3116

–0.0242

2.0274

–0.2974

YB

2601

348

0.5150

0.7298

0.9081

8.9994

–0.0607

–0.1359

a SPP Species are AB American beech, AS ash, BF balsam fir, EH eastern hemlock, OH other hardwoods, OS other softwoods, QA quaking aspen, RM red maple, RN red pine, RO red oak, SM sugar maple, SP spruce, WC white cedar, WP white pine, and YB yellow birch

Long–term simulations

Based on the initial plot tree list, the component equations described above were used to project the plot from its first to last measurement. The simulations ranged from 7 to 41 years in length with an average of 22.1 ± 11.3 years. Tree–level mortality was estimated using either the expansion factor reduction or species–specific optimal cutpoints derived from the survival equations (Eq. [5]). At the tree–level, the equations did not show a high degree of bias with prediction of crown recession having the highest error. A minor decreasing trend in error across total tree height was apparent. Across species, the best performance was achieved for eastern hemlock and red maple, while the poorest performance was for spruce. At the stand–level, using the optimal species cutpoint for mortality was shown to be superior as the mean percent bias ranged from –2.4% to 4.6% with the error being the highest for total stem density (Table 9). Only minor trends were evident across the range of observed values or projection lengths (Fig. 3).
Table 9

Stand–level projection mean bias (MB; observed – predicted), root mean square error (RMSE), and percent bias (%Bias) for total stem density (# ha–1), quadratic mean diameter (QMD; cm), and basal area (m2∙ha–1) by tree mortality method and forest, which are Dubuar Memorial Forest (DMF), Huntington Wildlife Forest (HWF), Pack Demonstration Forest (PDF), Pack Experimental Forest (PEF), and Shirley Forest (SF)

Forest

N

Years simulated (min – max)

Stem density

QMD

Basal area

MB

RMSE

%Bias

MB

RMSE

%Bias

MB

RMSE

%Bias

Expansion Factor method

 DMF

10

8.9 (7 – 17)

6.69

64.18

5.00

–0.29

1.47

–0.90

–0.28

3.10

1.22

 HWF

59

20.1 (10 – 31)

174.81

214.62

30.08

–3.21

5.83

–10.64

5.31

7.04

18.19

 PDF

71

10.0 (10 – 10)

130.67

188.45

13.95

–0.98

1.85

–4.41

0.80

3.79

0.70

 PEF

2

8.5 (7 – 10)

87.72

89.38

17.85

–1.34

1.49

–4.98

2.59

3.21

9.04

 SF

103

8.7 (3 – 24)

22.06

48.47

6.39

–0.11

1.54

–0.59

0.81

2.38

4.39

 Overall

245

11.8 (3 – 31)

90.23

150.35

14.32

–1.13

3.20

–4.17

1.86

4.35

6.55

Optimal Cutpoint method

 DMF

10

8.9 (7 – 17)

–18.53

77.85

–1.01

–0.28

1.46

–0.86

–1.54

3.93

–4.03

 HWF

59

20.1 (10 – 31)

74.97

136.73

12.48

–2.97

5.67

–9.96

–1.09

5.28

–1.81

 PDF

71

10.0 (10 – 10)

58.82

143.28

5.84

–0.96

1.83

–4.29

–1.39

4.09

–4.99

 PEF

2

8.5 (7 – 10)

49.42

55.25

10.13

–1.30

1.47

–4.86

–0.19

0.35

–1.16

 SF

103

8.7 (3 – 24)

–5.64

53.52

–0.32

–0.08

1.55

–0.48

–0.51

2.83

–0.85

 Overall

245

11.8 (3 – 31)

32.38

109.22

4.61

–1.05

3.14

–3.92

–0.95

3.94

–2.42

Fig. 3

Prediction bias (observed – predicted) for stem density (# ha–1), quadratic mean diameter (cm), and total basal area (m2∙ha–1) over observed values (left) and years in projection (right) using the different methods for simulating individual tree mortality. The expansion factor method is where the predicted probability of survival is multiplied by the tree’s current expansion factor, while the optimal cutpoint method is where an entire tree record is killed when the predicted probability of survival falls below the species optimal cutpoint

For the different forest types tested, growth was fairly linear for the 50 year projections when using either the expansion factor method or the optimal cutpoint method (Fig. 4). Overall, the hardwood dominated stand types showed greater growth when compared to the softwood dominated ones. The total basal area values (10.5 – 17.1 and 15.7 – 17.7 m2∙ha –1) and basal area growth rates (0.19 – 0.32 and 0.27 – 0.33 m2∙ha–1∙yr–1 for the expansion factor and the optimal cutpoint method, respectively) were all within reason of the general expectations for the region.
Fig. 4

Fifty year projections of total basal area (m2∙ha–1) of common forest types using the developed equations and the expansion factor method (left) or the optimal cutpoint method (right), respectively. Each forest type was initialized with 500 stems per ha of 5 cm diameter at breast height. Northern hardwoods was an equal mixture of American beech, black cherry, red maple, sugar maple, and yellow birch. Mixedwood was red spruce, balsam fir, eastern hemlock, and red maple with some American beech, white pine, and yellow birch. Hardwood dominated mixedwood was American beech, eastern hemlock, red spruce, sugar maple, and yellow birch. Spruce–fir was an equal mixture of red spruce and balsam fir

Discussion

To our knowledge, this effort represents the first attempt to develop an individual tree growth model specific to the range of tree species present in the Adirondacks Region of New York. Other efforts to model growth and yield in the region have only focused on plantation conifer species (Pan and Raynal 1995), hardwood species under a specific set of stand conditions like sugar maple in uneven–aged selection stands (Kiernan et al. 2009a, b), or pure hardwood stands (Aldridge 1982). Based on the assessment of the existing FVS–NE component equations, the Adirondacks Region appear to be a distinct ecological area that is deserving of a growth model specific to the present conditions. This is likely because the Adirondacks Region generally shows a higher potential productivity, a greater hardwood composition, and more intensive history of high–grading when compared to certain other parts of the Northeast. In particular, the Adirondacks Region has a disturbance regime predominated more by wind and ice (Lorimer and White 2003) and a history of highly selective logging in previously cutover old–growth hardwood stands (Berven et al. 2013), which both create relatively complex and mixed species stands.

The Adirondacks forest data used in this analysis represents the tremendous value of an intensive and well–maintained CFI system as it provided between 2 to 52 years of regular tree–level remeasurement data. The plots were well maintained, relatively large (0.04 – 0.08 ha), and representative of the conditions of the studied region. Initially, US Forest Service Forest Inventory and Analysis (FIA) data in New York was also considered in this analysis, but it was deemed of limited value given the relatively small plot size used (0.016 ha), the low number of remeasurements, and the general lack of total tree height measurements. Overall, the data used was determined to be of a sufficient size and scope with a full range of stand structure and compositions present. As common, plots in late–successional stands were not frequent and this may limit the growth model’s ability to extrapolate to those conditions. However, the lack of forest management on the Pack Experimental Forest has created some mature stands with dynamics controlled primarily by natural factors. However, these CFI were relatively smaller than plots used in research, which may influence the estimate of key stand– and tree–level variables used during modeling.

In general, the component equations fit well and showed adequate performance when conducting long–term simulations. Consistent with other tree–level growth models, total height equations fit the best, while height increment and mortality equations were the most problematic. For total and bole height, FVS–NE tended to underpredict these values, which has important consequences for future projections. In contrast, Rijal et al. (2012a, b) found that FVS–NE generally overestimated total height in the Acadian Region, while Russell et al. (2013) indicated both under– and over–estimation across the primary species in the Northeast US. The underprediction here was likely reflective of the better growing conditions in the lower elevations of the Adirondacks Region. Given the use of bole height (defined as the first live branch or a 10.2 cm (4 in) top), prediction using height to crown base or crown ratio equations of FVS–NE were expected to be biased, which was confirmed by the data. However, the fitted bole height equation fit the data well, but utilizing it to estimate a modified form of crown ratio proved not very useful for the other equations developed in this analysis.

In this analysis, the diameter and height increment equations proved particularly challenging due to remeasurement data only being available for trees > 10 cm in DBH and a general lack of height remeasurement data, respectively. However, relatively well–behaved and logical increment equations were constructed. In fact, to ensure proper fit and behavior of the height increment equation, it was estimated by treating species as a random effect rather than determining parameters by species like the diameter increment equation. Developing height increment equations by treating species as a random effect has been shown to be an effective technique when compared to fitting the equations by species (Russell et al. 2014). This is because all of the existing data can be used and species–specific parameters still extracted, which is particularly important for minor species that may not have observations across their full range. Consistent with Russell et al. (2014), this was found to be the case in this analysis. Using binary indicators for species to fit the equation was also an alternative option, but was determined to be inefficient given the number of species being examined and not further explored.

For mortality, the AUC for most species was approximately 0.75, which represents an acceptable to excellent discrimination of alive and dead trees (Hosmer and Lemeshow 2000). Thus, the equation was a significant improvement over the FVS–NE individual tree mortality equations. A variety of other model forms and implementation techniques were evaluated, but significant improvements were not achieved. Based on the observed stand–level bias, the equation appears to perform reasonably well, particularly if the whole tree approach with optimal species cutpoints was used. Like this study, a previous study has also found that using a fixed cutpoint rather than the tree expansion factor reduction method was superior (Crecente–Campo et al. 2009). In contrast, Rathbun et al. (2010) found just the opposite and suggested the tree expansion factor method to be superior, which indicates that additional research on the influence of how tree mortality is implemented in growth models, especially for long–term projections, is needed. However, both Crecente–Campo et al. (2009) and Rathbun et al. (2010) examined relatively short–term trends in mortality (1 – 17 years), while this analysis included plots simulated for over 40 years. Regardless, the findings highlights the sensitivity of models projections to mortality and likely a more sophisticated approach like a linked stand– and tree–level approach (e.g. Zhang et al. 2011) is warranted. However, a linked approach would require the development of a stand–level mortality equation, which can be problematic for multi–cohort, mixed–species stands (Weiskittel et al. 2011a).

Despite the difficulty of fitting individual component equations, the developed growth model shows good performance across a range of stand conditions. Using some of the longest validation data available in the Northeast (>50 years), the model showed only minor signs of increasing bias with greater projection lengths. The simulations for the different common forest types in the region were logical and met expectations as the hardwood dominated stands showed the greatest growth and spruce–fir the least. In this region, northern hardwood and mixedwood stands tend to occur on the best sites with deep and well–drained soils, while spruce–fir tend to occur in more poorly–drained flats. The observed annual total basal area growth rates were also consistent with typical regional values (e.g. Solomon 1977), while differences found between expansion factor and optimal cutpoint method in the 50 year projections again reflect the significance of the mortality approach applied.

Conclusions

Given the importance and uniqueness of the Adirondacks Region, an individual tree growth and yield model was developed based on existing, long–term CFI data (>50 years) for the region. The developed equations performed well despite high variability in the data and incomplete histories of past stand disturbances and harvesting practices. When the equations were combined into a growth and yield system, long–term behavior was consistent with observed trends and general expectations. Interestingly, long–term model performance was improved when using a whole–tree rather than expansion factor approach to individual tree survival. In addition, projections for the common forest types for the region indicated that mixed hardwood stands tended to outperform mixed and/or pure softwoods stands. Important next steps include evaluating the model’s behavior in stands with varying management regimes, developing equations to predict ingrowth (e.g. Li et al. 2011), and assessing alternative approaches to forecasting mortality (e.g. Zhang et al. 2011).

Declarations

Acknowledgments

Special thanks to State University of New York College of Environmental Forestry, Adrironack Ecological Center, and Frank Shirley for providing access to the data. Bruce Breitmeyer provided extensive guidance on the SUNY data and the methods used to collect it. FVS–NE code was provided by Dr. Matthew Russell. Feedback provided by Drs. Michael Saunders, Phillip Radtke, Robert Seymour and two anonymous reviewers helped to improve a previous version of this manuscript.

Authors’ contributions

AW wrote the manuscript and assisted with the analysis. CK completed the analysis and prepared all manuscript tables and figures. JPM and MO provided input on the manuscript text and modeling approaches used. All authors read and approved the final manuscript.

Authors’ information

AW is Associate Professor of Forest Modeling and Biometrics at the School of Forest Resources at the University of Maine. CK is a postdoctoral research associate in forest growth and yield modelling at the School of Forest Resources at the University of Maine. JPM is research scientist at Rayonier Forest Research Center and MO is an inventory analyst for Rayonier Forest Resources.

Competing interests

The authors declare that they have no competing interests.

Open AccessThis article is distributed under the terms of the Creative Commons Attribution 4.0 International License (http://creativecommons.org/licenses/by/4.0/), which permits unrestricted use, distribution, and reproduction in any medium, provided you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license, and indicate if changes were made.

Authors’ Affiliations

(1)
University of Maine, School of Forest Resources
(2)
Rayonier, Forest Research Center
(3)
Rayonier Forest Resources
(4)
University of Maine

References

  1. Aldridge SM (1982) An analysis of northern hardwoods permanent sample plots on State Forest Lands in New York. M.S. Thesis. State University of New York, College of Environmental Science and Forestry, Syracuse, p 471Google Scholar
  2. Beier CM, Stella JC, Dovčiak M, McNulty SA (2012) Local climatic drivers of changes in phenology at a boreal–temperate ecotone in eastern North America. Clim Change 115:399–417View ArticleGoogle Scholar
  3. Berven K, Kenefic L, Weiskittel AR, Twery M, Wilson J (2013) The lost research of early northeastern spruce–fir experimental forests: A tale of lost opportunities. In: Camp AE, Irland LC, Carroll CJW (eds) Long–term silvicultural and ecological studies, Results for science and management, vol 2, Global Institute of Sustainable Forestry Research Paper 013. Yale University, New Haven, pp 103–115Google Scholar
  4. Cao QV (2000) Prediction of annual diameter growth and survival for individual trees from periodic measurements. Forest Sci 46:127–131Google Scholar
  5. Chen L, Driscoll CT, Gbondo–Tugbawa S, Mitchell MJ, Murdoch PS (2004) The application of an integrated biogeochemical model (PnET–BGC) to five forested watersheds in the Adirondack and Catskill regions of New York. Hydrol Proc 18:2631–2650View ArticleGoogle Scholar
  6. Crecente–Campo F, Marshall P, Rodríguez–Soalleiro R (2009) Modeling non–catastrophic individual–tree mortality for Pinus radiata plantations in northwestern Spain. For Ecol Manage 257:1542–1550View ArticleGoogle Scholar
  7. Dixon GE, White R, Frank J (2007) Northeast (NE) variant overview. USDA Forest Service, Forest Management Service Center, Fort CollinsGoogle Scholar
  8. Engelman HM, Nyland RD (2006) Interference to hardwood regeneration in Northeastern North America: Assessing and countering ferns in northern hardwood forests. North J Appl For 23:166–175Google Scholar
  9. Germain RH, Nowak CA, Wagner JE (2016) The tree that built America – not making the grade. J For 114:552–561Google Scholar
  10. Hann DW, Marshall DD, Hanus ML (2003) Equations for predicting height–to–crown base, 5–year diameter growth rate, 5–year height growth rate, 5–year mortality rate, and maximum size–density trajectory for Douglas–fir and western hemlock in the coastal region of the Pacific Northwest. Oregon State University, College of Forestry Research Laboratory, CorvallisGoogle Scholar
  11. Hein S, Weiskittel AR (2010) Cutpoint analysis for models with binary outcomes: a case study on branch mortality. Eur J For Res 129:585–590View ArticleGoogle Scholar
  12. Hilt DE, Teck R (1989) NE–TWIGS: An individual tree growth and yield projection system for Northeastern United States. Compiler 7:10–16Google Scholar
  13. Hosmer DW, Lemeshow S (2000) Applied logistic regression, 2nd edn. Wiley, New YorkView ArticleGoogle Scholar
  14. Jiang H, Radtke PJ, Weiskittel AR, Coulston JW, Guertin PJ (2015) Climate– and soil–based models of site productivity in eastern US tree species. Can J For Res 45:325–342.View ArticleGoogle Scholar
  15. Juma R, Pukkala T, de–Miguel S, Muchiri M (2014) Evaluation of different approaches to individual tree growth and survival modelling using data collected at irregular intervals – a case study for Pinus patula in Kenya. Forest Ecosyst 1:14Google Scholar
  16. Kiernan DH, Bevilacqua E, Nyland RD (2009a) Individual–tree diameter growth model for sugar maple trees in uneven–aged northern hardwood stands under selection system. For Ecol Manage 256:1579–1586Google Scholar
  17. Kiernan DH, Bevilacqua E, Nyland RD, Zhang L (2009b) Modeling tree mortality in low– to medium–density uneven–aged hardwood stands under a selection system using generalized estimating equations. Forest Sci 55:343–351Google Scholar
  18. Li R, Weiskittel AR, Kerhsaw JA Jr (2011) Modeling annualized occurrence, frequency, and composition of ingrowth using mixed–effects zero–inflated models and permanent plots in the Acadian Forest Region of North America. Can J For Res 41:2077–2089View ArticleGoogle Scholar
  19. Leopold DJ, Reschke C, Smith DS (1988) Old–growth forests of Adirondack Park, New York. Nat Area J 8:166–189Google Scholar
  20. Lorimer CG, White AS (2003) Scale and frequency of natural disturbances in the northeastern US: implications for early successional forest habitats and regional age distributions. For Ecol Manage 185:41–64View ArticleGoogle Scholar
  21. McGee GG (2000) The contribution of beech bark disease–induced mortality to coarse woody debris loads in northern hardwood stands of Adirondack Park, New York, USA. Can J For Res 30:1453–1462View ArticleGoogle Scholar
  22. McMartin B (1994) The Great Forest of the Adirondacks. North Country Books, Utica, New YorkGoogle Scholar
  23. Newton PF, Amponsah IG (2007) Comparative evaluation of five height–diameter models developed for black spruce and jack pine stand–types in terms of goodness–of–fit, lack–of–fit and predictive ability. For Eco Manage 247:149–166View ArticleGoogle Scholar
  24. Pan Y, Raynal DJ (1995) Predicting growth of plantation conifers in the Adirondack Mountains in response to climate change. Can J For Res 25:48–56View ArticleGoogle Scholar
  25. Pandit K, Bevilacqua E, Perry JA (2012) Evaluating bias and accuracy of Forest vegetation simulator in predicting basal area and diameter growth of major forest types within New York State. 16th Annual Northeastern Mensurationists Organization Meeting, State CollegeGoogle Scholar
  26. Pinheiro J, Bates D, DebRoy S, Sarkar D (2016) nlme: linear and nonlinear mixed effects models. R package version 3.1–128. Available at: https://cran.r-project.org/web/packages/nlme/index.html. Accessed 20 Nov 2016.
  27. Rathbun LC, LeMay V, Smith N (2010) Modeling mortality in mixed–species stands of coastal British Columbia. Can J For Res 40:1517–1528View ArticleGoogle Scholar
  28. Ray DG, Saunders MR, Seymour RS (2009) Recent changes to the Northeast Variant of the Forest Vegetation Simulator and some basic strategies for improving model outputs. North J Appl For 26:31–34Google Scholar
  29. R Core Team (2016) R: a language and environment for statistical computing. R Foundation for Statistical Computing, Vienna. https://www.R-project.org. Accessed 20 Nov 2016.
  30. Rijal B, Weiskittel AR, Kershaw Jr JA (2012a) Development of height to crown base models for thirteen tree species of the North American Acadian Region. Forest Chron 88:60–73Google Scholar
  31. Rijal B, Weiskittel AR, Kershaw Jr JA (2012b) Development of regional height to diameter equations for fifteen tree species in the North American Acadian Region. Forestry 85:379–390Google Scholar
  32. Robin X, Turck N, Hainard A, Tiberti N, Lisacek F, Sanchez JC, Müller M (2011) pROC: an open–source package for R and S+ to analyze and compare ROC curves. BMC Bioinformatics 12:1View ArticleGoogle Scholar
  33. Robinson AP, Froese RE (2004) Model validation using equivalence tests. Ecol Model 176:349–358View ArticleGoogle Scholar
  34. Robinson AP, Wykoff WR (2004) Inputting missing height measurements using a mixed–effects modeling strategy. Can J For Res 34:2492–2500View ArticleGoogle Scholar
  35. Rota M, Antolini L (2014) Finding the optimal cut–point for Gaussian and Gamma distributed biomarkers. Comput Stat Data An 69:1–14View ArticleGoogle Scholar
  36. Russell MB, Weiskittel AR, Kershaw JA Jr (2013) Benchmarking and calibration of Forest Vegetation Simulator individual tree attribute predictions across the Northeastern United States. North J Appl For 30:75–84View ArticleGoogle Scholar
  37. Russell MB, Weiskittel AR, Kershaw JA Jr (2014) Comparing strategies for modeling individual–tree height and height–to–crown base increment in mixed–species Acadian forests of northeastern North America. Eur J For Res 133:1121–1135View ArticleGoogle Scholar
  38. Solomon DS (1977) The Influence of stand density and structure on growth of northern hardwoods in New England. USDA For. Serv. Res. Pap. NE–362. p 13Google Scholar
  39. Stout SL, Nyland RD (1986) Role of species composition in relative density measurement in Allegheny hardwoods. Can J Forest Res 16:574–579View ArticleGoogle Scholar
  40. Weiskittel AR, Hann DW, Kershaw Jr JA, Vanclay JK (2011a). Forest growth and yield modeling. Wiley & Sons, ChichesterGoogle Scholar
  41. Weiskittel AR, Russell MB, Wagner RG, Seymour RS (2012) Refinement of the Forest Vegetation Simulator northeastern variant growth and yield model: Phase III. Cooperative Forestry Research Unit Annual Report. University of Maine, School of Forest Resources, Orono, pp 96–104Google Scholar
  42. Weiskittel AR, Garber SM, Johnson GP, Maguire DA, Monserud RA (2007) Annualized diameter and height growth equations for Pacific Northwest plantation–grown Douglas–fir, western hemlock, and red alder. For Ecol Manage 250:266–278View ArticleGoogle Scholar
  43. Weiskittel AR, Wagner RG, Seymour RS (2011b) Refinement of the Forest Vegetation Simulator, Northeastern Variant growth and yield model: Phase 2. Cooperative Forestry Research Unit Annual Report. University of Maine, School of Forest Resources, Orono, pp 44–48Google Scholar
  44. Woodall CW, Miles PD, Vissage JS (2005) Determining maximum stand density index in mixed species stands for strategic–scale stocking assessments. For Ecol Manage 216:367–377View ArticleGoogle Scholar
  45. Zhang X, Lei Y, Cao QV, Chen X, Liu X (2011) Improving tree survival prediction with forecast combination and disaggregation. Can J For Res 41:1928–1935View ArticleGoogle Scholar

Copyright

© The Author(s). 2016