Landsat Ensemble v1.0.0 Map Accuracy

Map accuracy/agreement assessment following the Riemann et. al. framework.

Description

Map accuracy assessment for the v1.0.0 model-weighted-ensemble Landsat model trained on predictions from the v0.0.99 landsat-fia predictions and the v0.0.99 landsat-lidar predictions(). FIA data inventoried between 2013 and 2019 were considered for this assessment.

Model Document

Riemann et. al.

Branch used to produce this document: lj_landsat-ensemble-1.0.0

FIA plot inclusion criteria

• All subplots must be sampled.
• Plots must be completely forested.
• Annual assessment vs pooled assessment:
  • For an annual assessment: only FIA plots sampled in the designated year were considered, and compared to the modeled AGB surface for said year.
  • For a pooled assessment: FIA plots inventoried in all years (2013-2019) were aggregated within each hex. Modeled values were extracted from each annual surface to match the inventory year for a particular FIA plot, but aggregated within each hex just the same.

Masking

LCMAP Collection 1.2 primary classification layers were used to mask out water, developed, and barren areas i.e. “!class %in% c(1, 5, 8)” pixels from the modeled AGB surfaces. Annual AGB surfaces were matched to the LMCAP layer from the same year.

AOA masking was also implemented, and results are shown both with and without AOA masking.

NOTE: Masked AGB pixels are set to NA, and excluded from the analysis.

Masked LCMAP classes:

• 1: Developed
• 5: Water
• 8: Barren

Agreement Statistics

Table 0: Plot:Pixel Pooled x LCPRI

LCPRI	n	Mean FIA	%RMSE	RMSE	% MAE	MAE	MBE	R\(^2\)
Cropland	32	25.38	151.27	38.39	126.71	32.16	23.73	-0.29
Grass/Shrub	34	32.35	115.38	37.33	96.12	31.09	25.58	0.26
Tree cover	1823	142.47	36.92	52.59	28.81	41.05	-1.51	0.35
Wetland	126	112.11	48.44	54.31	39.10	43.84	-5.55	0.36

Table 1: Plot:Pixel Pooled x LCPRI - AOA masked

LCPRI	n	Mean FIA	%RMSE	RMSE	% MAE	MAE	MBE	R\(^2\)
Cropland	30	25.32	143.50	36.34	119.58	30.28	20.29	-0.09
Grass/Shrub	25	39.15	88.08	34.49	75.62	29.61	22.86	0.50
Tree cover	1826	142.09	37.10	52.71	29.09	41.33	-0.19	0.36
Wetland	123	113.70	48.94	55.64	39.43	44.83	-4.52	0.35

Table 2: Pooled

Dist	n	PPH	Mean FIA	%RMSE	RMSE	%MAE	MAE	MBE	R\(^2\)
Plot:pixel	2015		136.85	38.21	52.29	29.90	40.92	-0.90	0.41
10	1053	1.91	137.41	32.92	45.23	25.21	34.64	-1.55	0.46
25	260	7.75	134.17	23.58	31.64	16.93	22.72	0.31	0.56
50	79	25.51	135.29	16.36	22.14	11.92	16.12	-0.38	0.68
100	27	74.63	125.98	14.23	17.93	7.88	9.93	3.20	0.72

Table 3: Pooled - AOA Masked

Dist	n	PPH	Mean FIA	%RMSE	RMSE	%MAE	MAE	MBE	R\(^2\)
Plot:pixel	2004		137.31	38.23	52.50	30.03	41.23	0.14	0.41
10	1051	1.91	137.73	32.96	45.40	25.26	34.79	-0.52	0.45
25	260	7.71	134.49	23.94	32.20	17.15	23.06	1.27	0.54
50	79	25.37	135.56	17.01	23.06	12.27	16.63	0.46	0.66
100	27	74.22	126.25	16.15	20.39	8.40	10.61	4.76	0.64

Continuous Metrics (Pooled, AOA Masked)

Scatter Plots (Pooled, AOA Masked)

Menlove and Healey (2019, AOA Masked)

66.1% of 236 estimates within FIA 95% CI. 7 estimates filtered (\(\geq\) 425 Mg ha\(^{-1}\)).