Landsat:LiDAR AGB 0.0.7: Forced 0s

Hudak-method models using LiDAR surfaces with LCMAP classes 1, 5, 8 forced to 0. 2021-08-23

Evaluation Results

Last iteration: 2021-08-19 - Landsat AGB 0.0.6: LCMAP Stratification

Most comparable iteration: 2021-07-19 - Landsat AGB 0.0.5: High Class

Change Summary

• Forced LiDAR pixels with LCMAP classes 1, 5, 8 (developed, water, barren) to 0
• This iteration follows 0.0.1-0.0.5; it does not have much in common with 0.0.6 (which was a dead end)

	RF (ranger)	GBM (LightGBM)	Ensemble (model weighted)	Ensemble (RMSE weighted)
RMSE	47.459	45.927	45.453	46.060
RMSE %	33.433	32.354	32.020	32.448
MBE	-0.641	-0.736	-0.802	-0.689
R2	0.692	0.694	0.700	0.699

AGB Distribution

summary(bind_rows(training, testing)$agb_mgha)

     Min.   1st Qu.    Median      Mean   3rd Qu.      Max. 
  0.00022  70.41696 139.65191 141.55986 213.40259 306.56705 

Validation Results

RMSE	Min	Median	Max
Rf	47.775	48.464	49.245
Lgb	45.779	46.385	47.333
Ensemble	46.080	46.697	47.613

“Ensemble” here refers to the RMSE-weighted ensemble; the linear model ensemble is trained using the predictions from the validation stage (so doesn’t have any “test” data until we do the final evaluation).

R2	Min	Median	Max
rf	0.665	0.682	0.694
lgb	0.674	0.690	0.703
ensemble	0.676	0.693	0.705

Metadata

Ensembles

• RMSE-weighted model weights:

      lgb        rf 
0.5094451 0.4905549 

• Linear model weights:

Call:
lm(formula = agb_mgha ~ rf_pred * lgb_pred, data = pred_values)

Residuals:
    Min      1Q  Median      3Q     Max 
-210.03  -27.61   -0.28   29.03  216.35 

Coefficients:
                   Estimate Std. Error t value Pr(>|t|)    
(Intercept)      -4.894e+00  2.215e-01  -22.10   <2e-16 ***
rf_pred           3.767e-01  4.449e-03   84.68   <2e-16 ***
lgb_pred          5.374e-01  4.350e-03  123.52   <2e-16 ***
rf_pred:lgb_pred  7.120e-04  1.221e-05   58.31   <2e-16 ***
---
Signif. codes:  0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1

Residual standard error: 46.06 on 927796 degrees of freedom
Multiple R-squared:  0.6938,    Adjusted R-squared:  0.6938 
F-statistic: 7.007e+05 on 3 and 927796 DF,  p-value: < 2.2e-16

Coverages

• 18 coverages:
  • 3_county, allegany_steuben, cayuga_oswego, clinton_essex_franklin, columbia_rensselaer, erie_genesee_livingston, franklin_st_lawrence, fulton_saratoga_herkimer_franklin, great_lakes, hudson_hoosic, long_island, madison_otsego, north_east, oneida_subbasin, scoharie, southwest_b, southwest, warren_washington_essex

\(n\) and \(p\)

• 44196 observations
  • 30929 training
  • 13267 testing
• 29 predictors
  • tcb, tcw, tcg, nbr, mag, ysd, chm_simard, nys_precip, nys_tmax, nys_tmin, nys_aspect, nys_dem, nys_slope, nys_twi, delta_tcb, delta_tcw, delta_tcg, delta_nbr, lcpri2, lcpri6, lcpri4, lcpri3, lcsec3, lcsec4, lcsec1, lcsec2, lcsec5, lcsec8, lcsec6

Component Models

• Tuning used 5-fold CV
• Final hyperparameters:

Random forest:

$num.trees
[1] 1000

$mtry
[1] 4

$min.node.size
[1] 2

$sample.fraction
[1] 1

$splitrule
[1] "maxstat"

$replace
[1] FALSE

$formula
agb_mgha ~ .

LGB:

$learning_rate
[1] 0.1

$nrounds
[1] 100

$num_leaves
[1] 20

$max_depth
[1] -1

$extra_trees
[1] TRUE

$min_data_in_leaf
[1] 10

$bagging_fraction
[1] 0.9

$bagging_freq
[1] 1

$feature_fraction
[1] 0.9

$min_data_in_bin
[1] 13

$lambda_l1
[1] 0.1

$lambda_l2
[1] 2

$force_col_wise
[1] TRUE