Landsat AGB 0.0.3: Big Tent

Landsat models built using 5 land cover types. Using 19 LiDAR coverages. 2021-07-03

Evaluation Results

Last iteration: 2021-06-29 - Landsat AGB 0.0.2: Full Coverage Foundation

Change Summary

• Using LCMAP primary land classes 2, 3, 4, 6, and 7
  • Excluding 1 (developed), 5 (water), and 8 (barren)

	RF (ranger)	GBM (LightGBM)	Ensemble (model weighted)	Ensemble (RMSE weighted)
RMSE	45.680	44.019	43.666	43.920
RMSE %	31.750	30.595	30.350	30.526
MBE	0.618	0.427	0.385	0.521
R2	0.703	0.721	0.725	0.724

AGB Distribution

summary(bind_rows(training, testing)$agb_mgha)

     Min.   1st Qu.    Median      Mean   3rd Qu.      Max. 
  0.00425  72.39922 143.66240 144.07367 216.35905 306.56705 

Validation Results

RMSE	Min	Median	Max
Rf	45.836	46.515	47.209
Lgb	43.919	44.745	45.395
Ensemble	43.891	44.621	45.321

“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	0.681	0.693	0.702
lgb	0.702	0.713	0.723
ensemble	0.704	0.715	0.725

Metadata

Ensembles

• RMSE-weighted model weights:

      lgb        rf 
0.5104045 0.4895955 

• Linear model weights:

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

Residuals:
     Min       1Q   Median       3Q      Max 
-213.711  -26.660   -0.104   27.125  281.211 

Coefficients:
                   Estimate Std. Error t value Pr(>|t|)    
(Intercept)      -1.875e+00  1.935e-01  -9.689   <2e-16 ***
rf_pred           3.003e-01  3.053e-03  98.367   <2e-16 ***
lgb_pred          6.759e-01  2.931e-03 230.592   <2e-16 ***
rf_pred:lgb_pred  2.060e-04  1.006e-05  20.478   <2e-16 ***
---
Signif. codes:  0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1

Residual standard error: 44.38 on 921496 degrees of freedom
Multiple R-squared:  0.7169,    Adjusted R-squared:  0.7169 
F-statistic: 7.78e+05 on 3 and 921496 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 \(p\)

• 43899 observations
  • 30719 training
  • 13180 testing
• 18 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

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] "variance"

$replace
[1] FALSE

$formula
agb_mgha ~ .

LGB:

$learning_rate
[1] 0.1

$nrounds
[1] 1500

$num_leaves
[1] 13

$max_depth
[1] -1

$extra_trees
[1] FALSE

$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] 0.5

$lambda_l2
[1] 6

$force_col_wise
[1] TRUE