AGB v2.0.0

Metadata and map accuracy reporting for new and improved AGB/AGC/BGC modeling workflows contained within the operational_agb GitHub repo.
Author

Lucas Johnson

Published

July 18, 2024

Notable changes

  • First look at operational_agb!

  • Landsat collection 2

  • NSVB allometrics

  • SVM dropped from component algorithms

  • Additional LiDAR collections incorporated

Model metadata

  • n: 1650

  • p: 40

  • predictors: MSR, NBR, NDVI, SR, TCA, TCB, TCG, TCW, delta_MSR, delta_NBR, delta_NDVI, delta_SR, delta_TCA, delta_TCB, delta_TCG, delta_TCW, mag, dur, preval, rate, dsnr, twi, slope, TPI, TRI, roughness, flowdir, elevation, chm_simard, water_dist_raw, ppt, soltotal, tmax, tmin, lcpri, lcsec, ecozone, aspect, wetland, yod

AGB

Model formulas:

  • Direct: -4.99022608455678 + (agb_lightgbm_1_0012 * 0.521526351159874) + (agb_lightgbm_1_0193 * 0.027595177386714) + (agb_lightgbm_1_0405 * 0.312972671109803) + (agb_ranger_1_0804 * 0.171725567374107)
  • Indirect: -5.69796202468447 + (agb_pred_lightgbm_1_0638 * 1.00992085970713) + (agb_pred_ranger_1_0360 * 0.02165864323083)
  • Ensemble: (Intercept) = -0.212134790402864, direct = 0.526322450797974, indirect = 0.448435584718672

Map accuracy

Table 1: Map accuracy results (vs FIA) for select scales. RMSE, MAE, ME in \(\operatorname{Mg\ ha}^{-1}\). Scale = distance between hexagon centroids in km; PPH = plots per hexagon; n = number of comparison units (plots or hexagons).
Scale n PPH Model MAE % MAE RMSE % RMSE ME R2 dr
plot:pixel 456 Direct 47.60 34.24 59.98 43.14 -6.74 0.34 0.59
Indirect 50.35 36.22 61.89 44.52 -19.08 0.30 0.56
Ensemble 46.51 33.46 57.89 41.64 -8.39 0.39 0.60
20 km 269 1.7 Direct 41.07 28.80 51.91 36.41 -3.21 0.39 0.60
Indirect 45.96 32.24 56.54 39.66 -18.45 0.27 0.55
Ensemble 40.87 28.66 51.18 35.90 -6.15 0.40 0.60
30 km 164 2.78 Direct 33.08 23.67 41.76 29.87 -5.73 0.43 0.61
Indirect 40.16 28.73 49.20 35.20 -22.85 0.21 0.52
Ensemble 33.79 24.17 41.71 29.84 -9.52 0.43 0.60
50 km 71 6.42 Direct 22.92 16.38 29.82 21.31 -3.32 0.67 0.70
Indirect 30.14 21.54 38.37 27.42 -21.02 0.45 0.61
Ensemble 23.66 16.91 30.33 21.67 -7.43 0.66 0.69
Figure 1: Comparison of mapped AGB to FIA estimated AGB across selected scales represented by distances between hexagon centroids (plot:pixel, 20 km, 30 km, and 50 km). Geometric mean functional relationship (GMFR) trend line shown with dashed (orange) line, and 1:1 line shown with solid (red) line.
Figure 2: Map accuracy results (vs FIA) across multiple scales represented by distances between hexagon centroids. Blue lines are fit to data using cubic splines with three knots.
Figure 3: Empirical cumulative distribution function comparisons. Landsat-modeled predictions vs FIA plot-level estimates across model types.

AGC

Model formulas:

  • AGC frac: 0.48
  • Direct: -2.83609724778451 + (agc_lightgbm_1_0055 * 0.490175153844668) + (agc_lightgbm_1_0208 * 0.336618898141192) + (agc_lightgbm_1_0221 * 0.057765503001157) + (agc_ranger_1_0506 * 0.156535677432887)
  • Indirect: -2.46410323096981 + (agc_pred_lightgbm_1_0532 * 0.572767865958672) + (agc_pred_lightgbm_1_0638 * 0.413063967883088) + (agc_pred_ranger_1_0586 * 0.0418299846148557)
  • Ensemble: (Intercept) = 0.253798509091623, direct = 0.573772495933346, indirect = 0.403572593249995

Map accuracy

Table 2: Map accuracy results (vs FIA) for select scales. RMSE, MAE, ME in \(\operatorname{Mg\ ha}^{-1}\). Scale = distance between hexagon centroids in km; PPH = plots per hexagon; n = number of comparison units (plots or hexagons).
Scale n PPH Model MAE % MAE RMSE % RMSE ME R2 dr
plot:pixel 456 Direct 23.06 34.27 29.16 43.34 -2.75 0.33 0.58
Indirect 24.28 36.08 29.88 44.41 -9.24 0.30 0.56
Ensemble 22.53 33.48 28.11 41.79 -4.04 0.38 0.59
Frac 22.54 33.50 28.07 41.71 -4.03 0.38 0.59
20 km 269 1.7 Direct 19.88 28.83 25.31 36.71 -1.13 0.38 0.60
Indirect 22.26 32.28 27.31 39.60 -8.95 0.27 0.55
Ensemble 19.84 28.76 24.84 36.02 -2.95 0.40 0.60
Frac 19.79 28.61 24.87 35.95 -3.05 0.39 0.59
30 km 164 2.78 Direct 16.14 23.89 20.44 30.24 -2.42 0.41 0.60
Indirect 19.35 28.63 23.64 34.98 -11.04 0.21 0.52
Ensemble 16.36 24.21 20.21 29.90 -4.57 0.42 0.60
Frac 16.41 24.19 20.47 30.17 -4.63 0.43 0.61
50 km 71 6.42 Direct 11.19 16.55 14.79 21.86 -1.28 0.65 0.70
Indirect 14.49 21.42 18.46 27.30 -10.11 0.45 0.61
Ensemble 11.58 17.13 14.83 21.93 -3.53 0.64 0.69
Frac 11.50 16.80 14.37 20.98 -2.96 0.70 0.71
Figure 4: Comparison of mapped AGC to FIA estimated AGC across selected scales represented by distances between hexagon centroids (plot:pixel, 20 km, 30 km, and 50 km). Geometric mean functional relationship (GMFR) trend line shown with dashed (orange) line, and 1:1 line shown with solid (red) line.
Figure 5: Map accuracy results (vs FIA) across multiple scales represented by distances between hexagon centroids. Blue lines are fit to data using cubic splines with three knots.
Figure 6: Empirical cumulative distribution function comparisons. Landsat-modeled predictions vs FIA plot-level estimates across model types.

BGC

Model formulas:

  • BGC frac: 0.09
  • Direct: -0.255677426150621 + (bgc_lightgbm_1_0016 * 0.535481079690739) + (bgc_lightgbm_1_0193 * 0.289172398102501) + (bgc_lightgbm_1_0208 * 7.75996008716589e-05) + (bgc_lightgbm_1_0221 * 0.0723762744023807) + (bgc_ranger_1_0485 * 0.0684275806382955) + (bgc_ranger_1_0323 * 0.0288670884781443) + (bgc_ranger_1_0981 * 0.0247054847927647)
  • Indirect: -0.556198534559293 + (bgc_pred_lightgbm_1_0638 * 1.01025358044001) + (bgc_pred_ranger_1_0586 * 0.0221981745462116)
  • Ensemble: (Intercept) = -0.0876253226862473, direct = 0.567303178251749, indirect = 0.422982838552703

Map Accuracy

Table 3: Map accuracy results (vs FIA) for select scales. RMSE, MAE, ME in \(\operatorname{Mg\ ha}^{-1}\). Scale = distance between hexagon centroids in km; PPH = plots per hexagon; n = number of comparison units (plots or hexagons).
Scale n PPH Model MAE % MAE RMSE % RMSE ME R2 dr
plot:pixel 456 Direct 4.32 33.74 5.52 43.16 -0.40 0.34 0.59
Indirect 4.60 35.91 5.70 44.53 -1.80 0.30 0.56
Ensemble 4.23 33.08 5.32 41.53 -0.77 0.39 0.60
Frac 4.27 33.38 5.35 41.79 -0.80 0.38 0.59
20 km 269 1.7 Direct 3.77 28.84 4.82 36.82 -0.12 0.38 0.60
Indirect 4.29 33.23 5.29 40.90 -1.90 0.29 0.56
Ensemble 3.80 29.37 4.73 36.60 -0.70 0.43 0.61
Frac 3.80 28.89 4.82 36.69 -0.62 0.38 0.59
30 km 164 2.78 Direct 3.07 24.03 3.92 30.66 -0.35 0.40 0.60
Indirect 3.56 28.29 4.41 35.00 -2.37 0.19 0.53
Ensemble 3.05 24.18 3.71 29.48 -1.05 0.42 0.60
Frac 3.18 24.76 3.96 30.80 -0.96 0.41 0.60
50 km 71 6.42 Direct 2.03 15.98 2.63 20.68 -0.19 0.68 0.71
Indirect 2.78 22.04 3.43 27.18 -2.11 0.50 0.61
Ensemble 2.10 16.63 2.69 21.30 -0.76 0.69 0.70
Frac 2.14 16.56 2.65 20.58 -0.73 0.70 0.72
Figure 7: Comparison of mapped BGC to FIA estimated BGC across selected scales represented by distances between hexagon centroids (plot:pixel, 20 km, 30 km, and 50 km). Geometric mean functional relationship (GMFR) trend line shown with dashed (orange) line, and 1:1 line shown with solid (red) line.
Figure 8: Map accuracy results (vs FIA) across multiple scales represented by distances between hexagon centroids. Blue lines are fit to data using cubic splines with three knots.
Figure 9: Empirical cumulative distribution function comparisons. Landsat-modeled predictions vs FIA plot-level estimates across model types.