^{1}

^{*}

^{2}

^{1}

^{1}

^{1}

^{2}

This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution license (http://creativecommons.org/licenses/by/3.0/).

Relationships between discrete-return light detection and ranging (LiDAR) data and radiata pine leaf area index (LAI), stem volume, above ground carbon, and carbon sequestration were developed using 10 plots with directly measured biomass and leaf area data, and 36 plots with modelled carbon data. The plots included a range of genetic types established on north- and south-facing aspects. Modelled carbon was highly correlated with directly measured crown, stem, and above ground biomass data, with r = 0.92, 0.97 and 0.98, respectively. LiDAR canopy percentile height (P30) and cover, based on all returns above 0.5 m, explained 81, 88, and 93% of the variation in directly measured crown, stem, and above ground live carbon and 75, 89 and 88% of the modelled carbon, respectively. LAI (all surfaces) ranged between 8.8–19.1 in the 10 plots measured at age 9 years. The difference in canopy percentile heights (P95–P30) and cover based on first returns explained 80% of the variation in total LAI. Periodic mean annual increments in stem volume, above ground live carbon, and total carbon between ages 9 and 13 years were significantly related to (P95–P30), with regression models explaining 56, 58, and 55%, respectively, of the variation in growth rate per plot. When plot aspect and genetic type were included with (P95–P30), the R^{2} of the regression models for stem volume, above ground live carbon, and total carbon increment increased to 90, 88, and 88%, respectively, which indicates that LiDAR regression equations for estimating stock changes can be substantially improved by incorporating supplementary site and crop data.

Radiata pine comprised approximately 90% of the 1.75 million hectares of planted forest estimated to occur in New Zealand in April 2009, a large proportion of which was established on grassland sites of moderate to high fertility predominantly between 1992 and 1998 [

A national inventory of carbon stocks in forest that existed prior to 1 January 1990 (Pre-1990 forest) and in new forest that was established on grassland after 31 December 1989 (Post-1989 forest) is providing data for New Zealand to meet its obligations under the Kyoto Protocol and the United Nations Framework Convention on Climate Change [

LiDAR has proven to be useful for assessing tree height, stand volume, and carbon stocks [

LiDAR has also been used to estimate tree height growth, stem volume increment, and net primary production [

In young stands, where live organic matter pools comprise a relatively large proportion of the total carbon stock, changes over time are due mostly to net dry matter production. Production, P can be predicted from leaf area index, solar radiation and radiation-use efficiency using the following relationship [_{0} is annual incident solar radiation,

Assuming that LiDAR point cloud metrics can provide robust estimates of the total LAI, we hypothesize, based on Equation 1, that information related to incident solar radiation level and radiation-use efficiency will improve the precision of stock change estimates. This hypothesis was tested in a pilot study at Puruki Forest by assessing the effects of site aspect and crop tree genetics on stem volume and carbon stock changes per plot in conjunction with using the best estimates of LAI we could derive using LiDAR.

This study was undertaken at Puruki Forest [

In 2006 the understorey vegetation at Puruki comprised grasses and bracken fern approximately 0.5–1 m in height and woody vegetation, including the invasive

Three trials were established using seedlings and cuttings with different levels of genetic improvement, GF# [

GF7 cuttings (clonally propagated seedlings from GF7 parents);

GF2 seedlings (seed from an unimproved land-race [

Puruki Control (PC) seedlings (from a GF7 stand that was heavily thinned to improve crown health, so GF rating is not known);

GF30 seedlings (seed from a defined set of control pollinated parents with improved growth and form);

High Density (HD) seedlings (seed from a defined set of control pollinated parents with improved growth, form and wood density).

Trial 1 is comprised of six 50 × 50 m plots installed in juxtaposition, with each plot covering a range of slopes and aspects. Each plot was block planted with an identical set of 400 genetically different cuttings (genetic type 1) at a nominal stocking of 1,600 stems/ha. All trees (excluding severely malformed stems) were pruned between April and June 2003 to a nominal prune height of 3.5 m, and thinned in May–June 2003 to approximately 850 trees/ha. Trees were pruned again in November 2004 to a nominal prune height of 6.5 m.

Trial 2 is comprised of ten 50 × 50 m plots installed either singly or in juxtaposition, with each plot covering a range of slopes and mixed aspects. Plots were planted with alternating rows of cuttings (9 rows of genetic type 1) and seedlings (3 rows each of genetic types 3 and 4), with 10 trees per row, giving an initial stocking of 600 stems/ha. Approximately 400 trees/ha were pruned between April and June 2003 to a nominal prune height of 3.5 m. The following year approximately 350 trees/ha were pruned in November 2004 to a nominal prune height of 6.5 m. No thinning occurred in these plots. The GF30 and Puruki Control trees were beginning to overtop and hence tending to suppress some of the GF7 trees by 2010.

Trial 5 is comprised of twenty plots, each entirely planted with a single genetic type (two plots of type 2, and six plots each of types 3, 4 and 5). Eighteen plots were installed on either high (North-facing) or low (South-facing) solar radiation aspects, at an initial stocking of 1000 stems/ha, using a randomised block design. Two plots of genetic type 2 were installed on low radiation (south) aspects, owing to a limited supply of GF2 seed. Plots were mostly 40 × 40 m square plots (a 30 × 30 m measurement area with two buffer rows), although some were quadrilateral in shape (same plot area) where the terrain dictated this. Approximately 400 trees/ha were pruned between April and June 2003 to a nominal prune height of 3.5 m, and 350 trees/ha pruned again in November 2004 to a nominal prune height of 6.5 m. No thinning occurred in these plots.

Plot corners were marked with pulp boards visible in stereo-photographs acquired in February 2001 by New Zealand Aerial Mapping Ltd using standard survey methods, and plot locations accurately digitised using an analytical plotter.

Tree measurements included stem diameter at breast height (DBH) of all crop trees and total height, green crown height, prune height, and needle retention (count of needle age classes present in the lower half of the nominally live crown) of approximately every third tree per plot in Trials 2 and 5, and one tree per row per plot in Trial 1. Understorey mean height and cover of vegetation growing below the crown of height trees were assessed. In Trial 5, the number of branch clusters (or whorls) within the live crown of height trees was counted in 2006 to aid with sample tree selection for biomass and leaf area measurement, as described in the following section. The 2006 and 2010 field data were entered into the New Zealand Forest Research Institute Ltd Permanent Sample Plot system (PSP), to ensure appropriate data checking and quality control.

Leaf area was directly measured using biomass procedures, to avoid potential issues that have been identified when estimating LAI indirectly [

Carbon stocks per plot were also estimated using a system referred to as the Forest Carbon Predictor (FCP) version 3. This carbon model is conditioned by plot summary attributes (BA, MTH, stocking) calculated from field measurements undertaken in 2006 and in 2010, which ensures that accurate estimates of stem volume and carbon are obtained per plot. Stem volume was estimated using a nationally applicable volume function incorporated within the 300 Index growth model [

Significant breast height density differences have been reported previously for seedlots in Trial 5, based on the mean density of the 6 outer growth rings of core samples from 20 trees per plot sampled at age 9 years [

Natural and management related transfers of carbon from above- and below ground biomass pools to litter and dead wood pools were modelled for each plot, based on needle retention assessments, prune height data, and stocking changes following thinning operations [

Airborne LiDAR data were acquired 1200 m over Puruki forest on 16 and 17 August 2006 and 13 March 2010 using an Optech ALTM 3100EA system by New Zealand Aerial Mapping (NZAM). The aircraft operated no more than 30 km from a GPS base-station forming part of the NZGD2000 geodetic network. The LiDAR point cloud data, which were supplied in LAS version 1.0 file format, met the specified vertical accuracy of 0.15 m and horizontal accuracy of 0.30 m. (Project 06.476 Data Supply Metadata Record). The data acquisition parameters are given in

_{ib}/ha), trees per ha (trees/ha), understorey height (Under Ht), and understorey cover (Under Cover). Carbon in above ground live biomass (AG Live C), below ground live biomass (BG Live C), coarse woody debris (Dead Wood C), branch and needle litter (Litter C), live plus dead biomass (Total C) in 2006 and 2010, and the periodic mean annual increment (PMAI) in stem volume and carbon stocks between ages 9 to 13 years were estimated using the carbon model. Crown carbon, stem carbon, and above ground live carbon and LAI by 2 m vertical layer were directly measured in 10 plots.

The 10 plots in Trial 5 with direct LAI measurements had a similar MTH but a higher basal area (24.1%) and stocking (17.5%) than plots in all three trials (

The measured total LAI ranged between 8.8 and 19.1, with cumulative LAI profiles shown for each plot in

LAI vertical profiles were obtained for each plot by summing measurements by 2 m depth zone, starting at the top of the canopy down to specified heights above ground. For example, LAI_6 is the cumulative LAI from top of canopy to 6m above ground, and LAI_8 is the cumulative LAI from top of canopy to 8 m above ground. LAI data below 6 m above ground were within the partially unpruned part of the pine canopy where understorey shrubs often occurred.

LiDAR metrics were calculated after normalizing the data by subtracting the bare ground elevation generated from the LiDAR bare-earth digital elevation model (DEM) at each XY coordinate of each return, from each LIDAR return elevation. Metrics were computed using the “Cloudmetrics” utility in FUSION [

Metrics calculated included canopy height percentiles of the point cloud above the specified ground height thresholds (5th, 10th, …, 95th) defined as P5, P10, …, P95 along with the mean height, several statistics describing the height distribution (standard deviation (Sd), coefficient of variation (Cv)), and canopy cover. LiDAR height percentiles were computed as follows: (1) using only the first returns within each pulse; or, (2) using all returns. Both first returns and all returns percentile metrics were computed with a 0.5 m ground height cut-off and 0.5 m cover height threshold. This cover height threshold includes returns from unpruned and partially pruned pines and from understorey vegetation greater than 0.5 m in height. Metrics were also calculated using 1 m/1 m, 1 m/3 m and 5 m/5 m as ground height/cover height cut-off thresholds. These thresholds can potentially affect the percentile heights, distribution statistics, and cover metrics, and were specified to determine the influence of returns from various understorey vegetation tiers and mean prune height on LiDAR relationships with the directly measured LAI.

Cover was computed by dividing returns above the cover height threshold by total number of returns (including those below the ground height cut-off) that were in the plot. Alternative estimates of canopy cover were computed as follows: (1) first returns from the canopy divided by first returns in the plot; and (2) all returns from the canopy divided by all returns in the plot. Use of all returns from canopy elements above the canopy height cut-off was expected to provide an improved characterisation of canopy cover [

The GLM procedure of SAS (Windows Version 9.1, SAS Institute Inc., Cary, NC, USA) was used to develop regression equations between stand attributes of interest (Leaf area index, pine biomass, stem volume increment under bark, carbon sequestration) and radiata pine genetic type, aspect, and LiDAR metrics. Least squares means for genetic type and aspect were estimated across all 36 plots, after allowing for stocking differences between plots, to quantify genetic and micro-site related gains in volume increment and carbon sequestration rates at Puruki Forest.

LiDAR metrics were selected for inclusion in regression equations if they were biophysically meaningful, provided they were not strongly correlated with each other (r ≤ 0.5). LiDAR metrics selected to explain variation in LAI included cover because cover was expected to be low when there were gaps in the canopy and would be expected to increase with increasing canopy closure. However, by age 9 years, variation in LAI at Puruki was largely due to differences in crown transparency between plots, because the canopy had more-or-less closed. Percentile heights can be expected to be higher when crowns have dense compared with sparse foliage. Percentile heights would also be expected to increase with stand height. It was therefore considered necessary to include P95, a LiDAR metric highly correlated with canopy height, in regression equations with P30 and cover, to allow LAI to stabilise following canopy closure. Because P95 and P30 were moderately strongly correlated, the independent variable used in the regression model was (P95–P30). The influence of understorey vegetation on LiDAR metrics was tested empirically by examining the influence of variously calculated LiDAR height cut-off thresholds on LAI regression equations, and by including understorey height in place of cover in regression equations.

Measured and modelled biomass in crown, stem and stem plus crown components were separately regressed and compared using identical LiDAR metrics (percentile height and cover) as independent variables. Correlations between the measured and modelled biomass estimates were examined to provide confidence in the sequestration estimates obtained using the carbon model.

Finally, multiple regression analysis was used to test whether periodic mean annual increments in stem volume, above ground live carbon, and total carbon were related to LiDAR metrics identified as important for estimating LAI, and whether aspect and genetic type explained additional variation in growth, when fitted jointly with LAI-related LiDAR metrics.

Substantial differences in stem volume growth rates were evident among genetic types (^{3} (GF30), 331 kg/m^{3}(HD), 332 kg/m^{3} (GF2), and 332 kg/m^{3} (PC) in breast height outerwood density cores acquired at age 9 [

Growth rate was substantially influenced by aspect, with volume increment and carbon sequestration rates approximately 30% higher at plots on sunny

LiDAR metrics were in many cases highly correlated with each other. Correlations between canopy percentiles heights decreased the further these were apart, for example, for P95 the correlation was 0.9 with P70 and 0.5 with P30 in 2006. The canopy height standard deviation metric was strongly correlated with upper canopy height percentiles, but not significantly correlated with lower canopy height percentiles. The opposite was found for the coefficient of variation, which was strongly correlated with lower canopy height percentiles. Cover metrics were not significantly correlated with any of the other LiDAR metrics, apart from with each other and the lowest height percentile.

LiDAR metrics were also highly correlated with a subset of the stand attributes of interest. Mean top height was strongly correlated with upper height percentiles but not with height percentiles from P30 and below. LiDAR height metrics were also highly correlated with basal area, but more-so at lower height percentiles, peaking at around P30. Field measurements of understorey height were negatively correlated with mid-canopy height percentiles (indicating that plots with short pines had tall understorey vegetation), while the field estimates of understorey cover had moderately strong negative correlations with lower canopy height percentiles up to P60.

Correlations between LiDAR percentile height metrics and stem volume, above ground carbon, total carbon, litter carbon, carbon sequestration, and modelled LAI were all strong, with maximums at around P30. Dead wood, while only a small pool (

Carbon sequestration and measured LAI were related to mid- and upper percentile heights, the coefficient of variation of heights (Cv), mean height, and to a lesser extent cover based on first returns. Intensity data were not used in the regression models given in this paper, but may be useful in stands with large accumulations of standing dead wood [

LiDAR metrics per plot based on first returns and all returns and using low (0.5 m/0.5 m) and high (5 m/5 m) ground height/cover height cut-off thresholds are shown for 2006 data in

Results of comparisons involving LiDAR metrics calculated using various height cut-off thresholds as predictors of cumulative LAI to 6 m above ground are shown in ^{2}, due largely to an improvement in the P30 metrics relationship with leaf area. In the 5 m/5 m example, in which the cut-off height was approximately equal to the height down to which the bulk of the canopy LAI was measured, the cover metric and P30 metric were highly correlated with each other, with either metric significant when fitted on its own, although cover was superior to P30. While the LAI_6 regression model based on 5 m/5 m metrics explained 79% of the variation in LAI the model based on 0.5 m/0.5 m explained 88% of the variation. Metrics based on returns above 0.5 m height were therefore considered superior, and were used in the subsequent analyses.

The R^{2} values were higher when both percentile height and cover were based on all returns, for example, the model R^{2} increased from 88% to 91% using 0.5 m/0.5 m height cut-offs. Furthermore, field measured understorey height was highly significant in regressions of LAI down to 8 m height above ground, and partially substituted for the effect of cover, which was not considered desirable, given that understorey vegetation can vary markedly between stands and within stands over time.

The difference between upper and lower canopy percentile heights (P95–P30) calculated using all returns and cover based on first returns were significantly related to total LAI, with a model R^{2} of 80%. When substituted for LiDAR cover, understorey height was not statistically significant in these regressions.

Significant relationships were found between the measured and modelled crown, stem, and above ground (crown plus stem) carbon and P30 and LiDAR cover metrics based on all returns (

LiDAR regressions for estimating above ground live carbon stocks using P30 and cover, based on all 36 plots at Puruki, were more precise in 2006 than in 2010 (

Additional factors tested included genetic type, aspect and stocking-genetic type and stocking were highly significant, both in 2006 and 2010. The R^{2} of the base model in 2006 increased from 81%, using P30/cover, to 88% when genetic type was added and 91% when field measured stocking was added. Likewise, in 2010 the model R^{2} increased from 53%, using P30, to 91% when genetic type was added and 93% when field measured stocking was added. Similar results were obtained for stem total volume inside bark—the R^{2} of the model in 2006 increased from 81%, with P30/cover, to 87% with P30/cover/genetic type, and 90% with P30/cover/genetic type/ stocking. The corresponding model R^{2} values in 2010 were 49, 92 and 94%.

Measured LAI in the 10 plots with biomass data explained 81% of the variation in above ground carbon sequestration and 79% of the variation in stem volume increment between age 9 and 13 years. The relationships with LAI were linear, as shown previously for radiata pine [^{2} values of 52 and 56% obtained using 2006 and 2010 LiDAR data, respectively (

In addition, growth rate was significantly influenced by aspect, with R^{2} values increasing to approximately 74 and 76% using 2006 and 2010 LiDAR data (^{2} by an additional 3%, however the LiDAR metrics were not significant in 2006 (p-value averaged 0.083) but were significant in 2010 (p-value averaged 0.0058). While the effect of stocking was relatively small, the genetic types were ranked in accordance with growth expectations (

P95 could substitute for stocking effects in the above models, which suggests that more refined national models could be developed in future, given that national data would be based on data across a wide range of stand ages and sites.

Linear regression models for predicting the PMAI in stem volume and above ground live carbon between ages 9 to 13 years are shown in

Recent research has aimed to identify key LiDAR metrics that explain most of the variability in stand structural attribute across a range of forest types [

We are not aware of LiDAR studies that have tested whether genetic improvement within a species influences LiDAR regression models. Most importantly, our analysis shows that inclusion of genetic type and aspect markedly improved the precision of LiDAR regression models based on (P95–P30), which provided the best estimate of LAI from LiDAR. LAI was measured directly at Puruki to avoid potential issues involved with deriving LAI indirectly [

The results of our analysis were consistent with the productivity model (Equation 1) of Monteith [_{ib}, PMAI AG live C and LAI were linear, as commonly reported for a range of tree species. Radiation-use efficiency has proven to be a robust approach for predicting growth in a wide range of situations [

Our results are particularly relevant in the New Zealand context, because the quality of planting stock deployed in NZ's exotic forest estate has changed over time, and there has also been widespread adoption of clonally propagated cuttings (

Cumulative LAI measurements from the top of the canopy down to 6 m above ground were strongly related to P30 and cover, although these plots were of similar mean top height. Analysis of LAI profiles showed that LiDAR relationships were strong at all levels in the canopy however the model R^{2} tended to decrease with depth in the canopy. This weak tendency presumably reflects the fact that near-ground returns were absent below some tree crowns, nevertheless, an appreciable number of pulses penetrated the canopy, even when the LAI was high. For example, the plot with an LAI of 19.1 has 28% (2,886/10,450) of all returns within 3 m of the ground surface (

LiDAR metrics based on all returns generally gave stronger relationships with LAI vertical profiles than metrics based on first returns. Other researchers [

Percentile heights were lower in the canopy when the pines had thin transparent crowns with a low LAI. The influence of understorey on regressions based on P30 for estimating LAI was demonstrated by substituting the field measured understorey heights for LiDAR derived cover metrics in regressions where low canopy height thresholds were used. Returns from understorey vegetation can be expected to reduce the canopy height percentiles, especially when the overstorey pines have thin transparent crowns that allow greater penetration of LiDAR pulses to the understorey. Additional research may help resolve issues with understorey [

Crown mass of radiata pine plantations increases up to canopy closure and then decreases [

One source of uncertainty relates to temporal variation in stand health (

Tree crowns were weighed by 2 m height zone in 10 plots at Puruki forest, providing direct estimates of above ground live tree biomass and leaf area index.

Cumulative leaf area index profiles for 10 radiata pine plots at Puruki, based on measurements by 2 m height zones above ground and summed from the top of the canopy.

Quality of planting stock deployed in New Zealand's radiata pine plantations over the past 20 years, based on nursery sale statistics.

LiDAR point clouds colour coded by vegetation height for low (top left) and high (top right) leaf area index plots: (

Optech 3100EA LiDAR data acquisition parameters.

Flying height | 1,200 m above ground |

Scan angle | Clipped to 7 degrees either side of nadir |

Pulse density | 8–10/m^{2} |

Beam divergence | 0.3 mrad (1/e) |

Maximum returns/pulse | 4 |

Clipped swath width | 295 m |

Swath-to-swath overlap | Approximately 30% |

Pulse frequency | 71 kHz |

Radiata pine stand attributes of 10 plots at age 9 years with biomass carbon and LAI (pre-biomass) data, and 36 plots at age 9 (post-biomass) and 13 years with modelled carbon estimates, and computed periodic mean annual increments (PMAI) in stem volume and carbon stocks per plot over that period at Puruki Forest.

| |||||||||
---|---|---|---|---|---|---|---|---|---|

Age 9 years | Age 9 years | Age 13 years | |||||||

Variable | Mean | Min | Max | Mean | Min | Max | Mean | Min | Max |

MTH (m) | 13.0 | 11.6 | 14.6 | 13.4 | 11.0 | 16.0 | 20.1 | 17.2 | 22.5 |

BA (m^{2}/ha) |
32.4 | 21.0 | 48.9 | 25.6 | 12.5 | 46.2 | 45.1 | 25.0 | 73.3 |

LAI | 12.7 | 8.8 | 19.1 | - | - | - | - | - | - |

Needle retention | 2.02 | 0.97 | 2.64 | 1.95 | 0.97 | 2.64 | 1.87 | 1.34 | 2.75 |

Prune height (m) | 4.1 | 3.6 | 4.6 | 4.5 | 3.5 | 5.7 | 4.5 | 3.5 | 5.7 |

Vol_{ib} (m^{3}/ha) |
158 | 96 | 260 | 126 | 61.6 | 253 | 315 | 173 | 538 |

Trees/ha | 846 | 678 | 977 | 704 | 424 | 933 | 685 | 412 | 911 |

Under Ht (m) | 1.8 | 1.1 | 3.0 | 1.6 | 0.9 | 3.0 | 3.4 | 0.9 | 8.6 |

Under Cover (%) | 67 | 40 | 82 | 66 | 36 | 84 | 42 | 9 | 90 |

AG Live C (t/ha) | 39.1 | 24.2 | 58.9 | 34.5 | 17.6 | 71.4 | 83.0 | 46.1 | 138.7 |

BG Live C (t/ha) | - | - | - | 9.3 | 4.8 | 18.1 | 18.7 | 10.3 | 31.3 |

Dead wood C (t/ha) | - | - | - | 1.3 | 0.1 | 4.7 | 1.3 | 0.04 | 3.8 |

Litter C (t/ha) | - | - | - | 9.3 | 3.7 | 19.5 | 11.6 | 5.5 | 21.9 |

Total C (t/ha) | - | - | - | 55.4 | 26.2 | 109.5 | 114.6 | 62.0 | 194.3 |

PMAI Vol_{ib} (m^{3}/ha/yr) |
47.2 | 27.8 | 73.1 | ||||||

PMAI AG Live C (t/ha/yr) | 12.1 | 7.2 | 18.1 | ||||||

PMAI Total C (t/ha/yr) | - | - | - | - | - | - | 14.9 | 9.0 | 23.0 |

Least squares periodic mean annual increment (PMAI) in volume (model R^{2} = 90%), above ground live sequestration (model R^{2} = 88%), and total carbon sequestration (Model R^{2} = 87%) in relation to genetic type and aspect, after allowing for stocking differences at Puruki Forest. PMAI was calculated for 36 plots re-measured at stand ages 9 (2006) and 13 years (2010). Plots had been classified as predominantly North- or South-facing, or alternatively were classified as mixed aspect if the plot included a balance of North and South-facing aspects.

| ||||
---|---|---|---|---|

Volume PMAI^{3}/ha/year) |
AG Live C PMAI |
Total C PMAI | ||

Genetic type | GF30 | 58.3 (52.9–63.8) | 14.2 (12.8–15.7) | 17.6 (15.6–19.9) |

GF7 cuttings | 37.8 (30.6–45.0) | 10.1 (8.2–12.1) | 12.2 (9.6–14.8) | |

GF7 cuttings | 47.7 (39.4–56.0) | 12.2 (9.9–14.5) | 15.0 (12.0–18.0) | |

GF30/PC mix | ||||

HD | 53.8 (48.0–59.7) | 14.0 (12.4–15.6) | 17.5 (15.4–19.6) | |

GF2 | 45.8 (35.9–55.8) | 11.9 (9.2–14.6) | 14.8 (11.2–18.4) | |

PC | 49.1 (43.0–55.3) | 12.8 (11.1–14.4) | 16.0 (13.8–18.3) | |

Aspect | North-sunny | 55.6 (50.6–60.7) | 14.4 (13.0–15.7) | 18.1 (16.2–19.9) |

Mixed aspect | 48.3 (44.1–52.6) | 12.4 (11.3–13.6) | 15.3 (13.7–16.8) | |

South-shady | 42.3 (39.3–45.3) | 10.8 (10.0–11.6) | 13.3 (12.2–14.3) |

Mean and range of various LiDAR metrics (percentile heights, mean height, and dispersion statistics) based on 36 radiata pine plots measured at age 9 (2006) and re-measured at age 13 years (2010) at Puruki Forest. Metrics in 2006 were based on all returns and cover based on first returns with 5 m/5 m as height cut-offs, or first returns and all returns with 0.5 m/0.5 m set as ground/canopy height cut-offs. Metrics in 2010 were based on all returns with 0.5 m/0.5 m set as ground/canopy height cut-offs.

| ||||||||||||
---|---|---|---|---|---|---|---|---|---|---|---|---|

Mean | Min. | Max. | Mean | Min. | Max. | Mean | Min. | Max. | Mean | Min. | Max. | |

Cover (%) | 64 | 38 | 93 | 86 | 69 | 99 | 75 | 61 | 94 | 91 | 85 | 97 |

P30 (m) | 7.0 | 6.1 | 8.1 | 5.3 | 2.9 | 8.2 | 4.2 | 2.2 | 7.3 | 9.7 | 5.9 | 13.4 |

P50 (m) | 7.8 | 6.8 | 9.1 | 7.0 | 4.6 | 9.2 | 6.3 | 3.7 | 8.5 | 12.1 | 8.9 | 15.2 |

P70 (m) | 8.8 | 7.5 | 10.3 | 8.3 | 6.4 | 10.4 | 7.9 | 5.9 | 9.8 | 14.0 | 11.2 | 16.8 |

P95 (m) | 10.8 | 9.0 | 12.6 | 10.5 | 8.7 | 12.6 | 10.3 | 8.6 | 12.4 | 17.2 | 14.8 | 19.7 |

Mean height (m) | 7.9 | 6.9 | 9.2 | 6.7 | 4.7 | 9.1 | 5.9 | 4.3 | 8.0 | 11.4 | 8.6 | 14.2 |

Sd (m) | 1.6 | 1.1 | 1.9 | 2.6 | 2.0 | 3.5 | 3.0 | 2.4 | 3.6 | 4.2 | 3.2 | 5.5 |

Cv (%) | 20 | 17 | 23 | 41 | 26 | 58 | 51 | 36 | 68 | 37 | 28 | 49 |

Regression coefficient p-values and R^{2} between cumulative LAI from the top of the canopy to 6 m above ground (LAI_6) and LiDAR metrics for a 9 year old pine forest based on all returns for P30 and cover based on first returns. Ground cut-off/cover height thresholds were selected to reflect the influence of various understorey tiers and the pine canopy on percentile height and cover metrics.

^{2} (%) | ||||
---|---|---|---|---|

| ||||

P30 | Cover (%) | |||

0.5 m/0.5 m | Log(LAI_6) | 0.0002 | 0.0368 | 88 |

1 m/1 m | Log(LAI_6) | 0.0006 | 0.0403 | 85 |

1 m/3 m | Log(LAI_6) | 0.0124 | 0.1317 | 80 |

5 m/5 m | Log(LAI_6) | 0.6031 | 0.0896 | 79 |

Multiple regression coefficients between radiata pine biomass carbon (t/ha of crown, stem, and crown plus stem) and LiDAR percentile height (P30) and cover (%) based on all returns. The regressions are based on 10 plots with measured and modelled biomass carbon (0.5 m ground and canopy cut-off heights). The model R^{2}, RMSE, and p-values are given for each model.

| ||||||||||
---|---|---|---|---|---|---|---|---|---|---|

R^{2} (%) |
RMSE | p-value | R^{2} (%) |
RMSE | p-value | R^{2} (%) |
RMSE | p-value | ||

Measured biomass | Regression | 81 | 1.69 | 0.0028 | 88 | 3.15 | 0.0005 | 93 | 3.40 | 0.0001 |

Estimate (s.e.) | t-value | p-value | Estimate (s.e.) | t-value | p-value | Estimate (s.e.) | t-value | p-value | ||

Intercept | −7.6 (6.8) | −1.1 | 0.300 | −38.4 (12.6) | −3.0 | 0.019 | −46.0 (13.6) | −3.4 | 0.012 | |

P30 | 2.2 (0.40) | 5.5 | 0.001 | 5.4 (0.75) | 7.2 | 0.0002 | 7.6 (0.81) | 9.3 | <0.0001 | |

Cover(%) | 0.10 (0.072) | 1.4 | 0.21 | 0.52 (0.13) | 3.9 | 0.006 | 0.62 (0.14) | 4.3 | 0.004 | |

| ||||||||||

R^{2} (%) |
RMSE | p-value | R^{2} (%) |
RMSE | p-value | R^{2} (%) |
RMSE | p-value | ||

| ||||||||||

Modelled biomass | Regression | 75 | 1.89 | 0.0075 | 89 | 3.60 | 0.0005 | 88 | 5.00 | 0.0007 |

Estimate (s.e.) | t-value | p-value | Estimate (s.e.) | t-value | p-value | Estimate (s.e.) | t-value | p-value | ||

Intercept | −13.0 (7.6) | −1.7 | 0.13 | −47.5 (14.5) | −3.3 | 0.013 | −60.7 (20.1) | −3.0 | 0.019 | |

P30 | 2.1 (0.45) | 4.6 | 0.002 | 6.3 (0.86) | 7.3 | 0.0002 | 8.4 (1.19) | 7.0 | 0.0002 | |

Cover (%) | 0.15 (0.08) | 1.9 | 0.10 | 0.58 (0.15) | 3.8 | 0.007 | 0.74 (0.21) | 3.5 | 0.011 |

Multiple regression relationship between modelled above ground (crown plus stem) carbon and LiDAR percentile height (P30) and cover (%), based on 36 plots in three trials measured in 2006 and 2010 at Puruki.

| ||||||
---|---|---|---|---|---|---|

R^{2} (%) |
RMSE | p-value | R^{2} (%) |
RMSE | p-value | |

Model | 81 | 5.95 | < 0.0001 | 53 | 16.35 | < 0.0001 |

Estimate (s.e.) | t-value | p-value | Estimate (s.e.) | t-value | p-value | |

Intercept | −45.8(10.2) | −4.5 | 0.001 | −115.0 (94.3) | −1.22 | 0.23 |

P30 | 7.52 (0.70) | 10.7 | < 0.0001 | 10.2 (1.73) | 5.90 | < 0.0001 |

Cover (%) | 0.67 (0.13) | 5.2 | < 0.0001 | 1.09 (0.94) | 1.16 | 0.25 |

Independent variables used to predict periodic mean annual increment (PMAI) in volume, above ground live carbon, and total carbon in 36 radiata pine plots included LiDAR (P95–P30), genetic type, aspect and stocking. Regression model R^{2} values apply to 2006 or 2010 LiDAR data.

Dependent Variable | Number of Independent Variables | Model | R^{2} (%) 2006 |
R^{2} (%) 2010 |
---|---|---|---|---|

PMAI in Volume | 1 | P95–P30 | 53 | 56 |

2 | P95–P30, Aspect | 73 | 75 | |

3 | P95–P30, Aspect, Genetic type | 90 | ||

4 | P95–P30, Aspect, Genetic type, Stocking | 93 | ||

PMAI in AG Live C | 1 | P95–P30 | 53 | 58 |

2 | P95–P30, Aspect | 75 | 77 | |

3 | P95–P30, Aspect, Genetic type | 88 | ||

4 | P95–P30, Aspect, Genetic type, Stocking | 91 | ||

PMAI in Total C | 1 | P95–P30 | 50 | 55 |

2 | P95–P30, Aspect | 75 | 77 | |

3 | P95–P30, Aspect, Genetic type | 88 | ||

4 | P95–P30, Aspect, Genetic type, Stocking | 91 |

Multiple regression model coefficients for estimating PMAI in stem volume increment and carbon sequestration from age 9 to 13 years, using 2010 LiDAR percentile heights (P95–P30), aspect, genetic type, and stocking per plot for 36 plots in Puruki Forest.

_{ib} |
||||||
---|---|---|---|---|---|---|

| ||||||

R^{2} (%) |
RMSE | p-value | R^{2} (%) |
RMSE | p-value | |

93 | 3.79 | < 0.0001 | 91 | 1.06 | < 0.0001 | |

Estimate (s.e.) | t-value | p-value | Estimate (s.e.) | t-value | p-value | |

Intercept | 44.3 (9.62) | 4.61 | 0.0001 | 11.8 (2.70) | 4.37 | 0.0002 |

P95–P30 | −3.005 (0.92) | −3.27 | 0.0030 | −0.755 (0.26) | −2.93 | 0.0069 |

Radiation = H | 6.865 (2.19) | 3.13 | 0.0042 | 1.83 (0.62) | 2.97 | 0.0063 |

= M | 0.0 | 0.0 | ||||

= L | −1.990 (2.07) | −2.07 | 0.3455 | −0.62 (0.58) | −1.07 | 0.2949 |

Genetic type = GF30 | 8.68 (2.31) | 3.76 | 0.0009 | 1.35 (0.65) | 2.09 | 0.0464 |

= HD | 4.73 (2.25) | 2.10 | 0.0451 | 1.22 (0.63) | 1.93 | 0.0649 |

= GF7 cuttings/PC/GF30 | 3.82 (4.06) | 0.942 | 0.3550 | 0.77 (1.14) | 0.674 | 0.5061 |

= PC | 0 | 0 | ||||

= GF2 | −3.70 (3.19) | −1.16 | 0.2559 | −0.97 (0.89) | −1.08 | 0.2887 |

= GF7 cuttings | −11.19 (2.39) | −4.48 | 0.0001 | −2.60 (0.67) | −3.87 | 0.0007 |

Stocking | 0.0345 (0.01) | 3.626 | 0.0012 | 0.008 (0.003) | 3.14 | 0.0045 |

We thank the Ministry for the Environment (MfE) for funding the measurement of permanent sample plots and biomass and leaf area measurements at Puruki. We would also like to thank field personnel who helped assess tree growth, health and leaf area at Puruki, in particular Doug Graham. Rod Brownlie provided accurate location data for the trial plots at Puruki. The LiDAR data and aerial photography over Puruki Forest were acquired by New Zealand Aerial Mapping. We also would like to thank John Novis, Ministry of Agriculture and Forestry, for providing information on the genetic quality of radiata pine from nursery surveys.