Comparative assessment of twelve methods for projecting forest aboveground biomass at the Howland Research Forest, Maine, USA. Companion analysis code and outputs for Weiskittel et al. (2026), Forestry.
This repository contains the R analysis code and SLURM submission scripts used to compare twelve aboveground biomass (AGB) projection methods at the Howland Research Forest:
- NAIP imagery combined with Acadian yield curves
- Area based LiDAR (Chapman Richards and Schumacher)
- LiDAR combined with Acadian yield curves
- ITC segmentation combined with strata averaging
- ITC combined with FVS NE
- ITC combined with FVS ACD
- Terrestrial laser scanning combined with Acadian projection
- Sample based field inventory
- Multi attribute yield curve matching (Tompalski et al. 2016, 2018)
- Wall to wall directional 3D point cloud projection (Kohek et al. 2022) with ALS decimated to 25 pts m⁻²
- Hybrid A: 2017 ITC anchor with default Kohek expansion coefficients and stratified pixel calibration
- Hybrid B: ITC + TLS calibration, with TLS fit H D allometry and locally calibrated Kohek M0 and M1 coefficients
Methods 1 through 8 are the original analysis (see Weiskittel et al. 2026 main text and supplementary v3). Methods 9 through 12 are added in the v4.5 revision and are the primary focus of the code in this repository.
| # | Method | Cost ($/ac) | RMSE 10 yr (%) | 95% CI | R² |
|---|---|---|---|---|---|
| 6 | ITC + FVS-ACD (best baseline) | 0.88 | 10.0 | [6.1, 14.3] | — |
| 9 | Curve matching | 1.50 | 35.9 | [35.7, 36.2] | −0.31 |
| 10 | Pure Kohek (holdout) | 8.20 | 26.4 | [25.9, 26.8] | 0.25 |
| 11 | Hybrid A (ITC 2017, MELiTE anchor) | 2.65 | 24.7 | [24.5, 25.0] | 0.20 |
| 12 | Hybrid B (ITC + TLS, MELiTE anchor) | 3.45 | 25.2 | [25.0, 25.5] | 0.17 |
Method 11 is the best of the four new methods. None of them beats the existing ITC + FVS ACD baseline. The directional 3D family adds individual tree resolution that area based methods cannot provide; that is its operational value.
howland-agb-12method/
├── R/ # Core analysis scripts
│ ├── methods_01_08/ # Original 8-method analysis (v4.4 baseline)
│ │ ├── README.md # Method-to-script mapping for Methods 1-8
│ │ ├── 01_howland_agb_yield_curves.R # Methods 2 and 3 (yield curves)
│ │ ├── 02_howland_lamb_plot_matching.R # Methods 4 and 7 (kNN, TLS)
│ │ ├── 03_howland_individual_tree_efi_fvs.R # ITC pipeline shared by Methods 4-6
│ │ ├── 04_howland_naip_projection_validation.R # Method 1 (NAIP)
│ │ ├── 05_howland_fvsne_projection_comparison.R # Methods 5 and 6 (FVS-NE, FVS-ACD)
│ │ ├── 06_recalibrate_with_2025als.R # Method 2 4-point recalibration
│ │ ├── 07_calibrate_fvsacd.R # Method 6 calibration
│ │ ├── 08_pixel_validation.R # Method 8 (field) and pixel comparison
│ │ ├── 09_decision_framework.R # Cost-accuracy quadrant framework
│ │ ├── 10_pixel_method_comparison.R # Wall-to-wall pixel comparison
│ │ ├── generate_manuscript_figures.R # Figures 1-13 (v4.4 main)
│ │ └── reviewer_response_analyses.R # First-round reviewer response analyses
│ ├── method09_curve_matching.R # Method 9 (Tompalski 2016, 2018)
│ ├── method10_directional_kohek.R # Method 10 (Kohek 2022, decimated ALS)
│ ├── method11_hybrid_itc_anchor.R # Method 11 (ITC 2017 anchor)
│ ├── method12_hybrid_itc_tls.R # Method 12 (ITC + TLS calibration)
│ ├── refinements_stratified_holdout_bootstrap.R # Post hoc refinements (M9-M12)
│ └── methods_11_12_ornl_sensitivity.R # Sensitivity test with ORNL 2017 anchor
├── figures/
│ ├── main_figures.R # Figs 3, 4, 5, 6 (manuscript)
│ └── supplemental_figures.R # Figs S5, S6, S7, S14 (supplemental)
├── slurm/ # SLURM submit scripts for OSC Cardinal
├── docs/
│ ├── DATA.md # Data sources and access
│ ├── METHODS.md # Method by method technical summary
│ ├── REPRODUCE.md # Step by step reproduction guide
│ ├── MANUSCRIPT_SECTIONS.md # Prose source for sections 2.9 to 2.12
│ ├── ORIGINAL_METHODS_MEMO.md # Background memo proposing Methods 9 to 12
│ ├── TECHNICAL_REVIEW.md # Pre-submission peer-style review record
│ └── QC_ASSESSMENT.md # v4.5 comprehensive QC pass
├── outputs/
│ └── sample_outputs/ # Small validation CSVs (tracked)
│ ├── README.md # Index of sample CSVs
│ ├── method09_validation.csv
│ ├── method10_validation.csv
│ ├── method11_validation.csv
│ ├── method12_validation.csv
│ ├── refined_method_stats.csv # Final headline numbers
│ ├── refinements_summary.csv # Bootstrap CIs
│ ├── methods_11_12_ornl_validation.csv
│ ├── methods_11_12_ornl_fits.csv
│ ├── Table_method_summary_12methods.csv
│ ├── TableS_boot_methods9_12.csv
│ └── TableS_stratified_fits.csv
├── CHANGELOG.md
├── CITATION.cff
├── LICENSE
└── README.md
The analysis was developed and run on the Ohio Supercomputer Center (OSC) Cardinal cluster (RHEL 9, SLURM scheduler) under project allocation PUOM0008.
Required software:
- R 4.4.0
- gcc 12.3.0
- gdal 3.7.3, geos 3.12.0, proj 9.2.1
- AcadianGY v12.3.6 (sourced separately at
AcadianGY_12.3.6.r)
Required R packages:
- tidyverse 2.0.0
- terra 1.9.1
- sf 1.0.16
- lidR 4.1.2
- FNN 1.1.4
- future 1.34.0
- future.apply 1.11.0
- ggplot2 4.0.2
- patchwork 1.2.0
- ggrepel 0.9.5
- ggtext 0.1.2
- scales 1.3.0
The scripts are written for the Cardinal HPC environment. Adapt paths and module loads as needed for other clusters.
# 1. Clone
git clone https://github.com/holoros/howland-agb-12method.git
cd howland-agb-12method
# 2. Set environment variables to point at your data
export HRF_ROOT=/path/to/HRF
export OUTPUT_DIR=$HRF_ROOT/output
export FIELD_DIR=$HRF_ROOT/data/field_inventory
export TLS_DIR=$HRF_ROOT/data/tls_2025
export NORM_DIR=$HRF_ROOT/output/als_2025_processing
# 3. Submit Method 9 first (curve matching builds the AcadianGY library)
sbatch slurm/submit_method09_curve_matching.sh
# 4. Submit hybrid Methods 11 and 12 (fast: about 30 seconds each)
sbatch slurm/submit_method11_hybrid_a.sh
sbatch slurm/submit_method12_hybrid_b.sh
# 5. Run post hoc refinements (stratified calibration, holdout, bootstrap CIs)
sbatch slurm/submit_refinements.sh
# 6. Run Method 10 (decimation + ITC + Kohek; longer: about 40 minutes)
sbatch slurm/submit_method10_kohek.sh
# 7. Generate figures
sbatch slurm/submit_main_figures.sh
sbatch slurm/submit_supplemental_figures.shSee docs/REPRODUCE.md for the full reproduction sequence with expected outputs.
Raw data are not included in this repository because they are large and partially restricted. See docs/DATA.md for sources and access:
- 2012, 2014, 2015, 2017, 2021, 2023, 2025 G LiHT airborne LiDAR (NASA distributed via ORNL DAAC)
- 2025 MELiTE ABA AGB raster (validation reference)
- 2025 Phoenix RANGER ALS (~92 pts m⁻²)
- 2025 RIEGL VZ 400i terrestrial laser scan
- 56 ground inventory plots
- 2017 wall to wall ITC tree list (Howland_ITC_2017.gpkg, 208,632 trees)
If you use this code, please cite both this repository and the manuscript:
Weiskittel, A. R. (2026). howland-agb-12method: comparative assessment of
twelve AGB projection methods at Howland Research Forest [Software].
https://github.com/holoros/howland-agb-12method
Weiskittel, A. R., et al. (2026). Comparative assessment of twelve methods
for projecting forest aboveground biomass at the Howland Research Forest,
Maine, USA. Forestry, in press.
See CITATION.cff for machine readable citation metadata.
MIT (see LICENSE). Data products distributed under their respective licenses; see docs/DATA.md.
Aaron R. Weiskittel · University of Maine · aaron.weiskittel@maine.edu
Compute resources provided by the Ohio Supercomputer Center under allocation PUOM0008. ORNL DAAC G LiHT product hosting. NASA G LiHT mission for multi temporal acquisitions. Field inventory and TLS deployment supported by USDA Forest Service Northern Research Station.