Landsat AGB 0.0.2: Full Coverage Foundation

The basic landsat AGB models that we’ll work on building up from. Using 19 LiDAR coverages. 2021-06-29

Evaluation Results

Last iteration: 2021-06-15 - Early Efforts

Change Summary

• Sampling from all 19 LiDAR coverages
  • Script attempts to sample 0.05% of all non-NA forested AGB pixels per LiDAR surface, broken out across 20 bins. Approximately 25,000 pixels in total.
  • Bins are equal-interval categories ranging from 0 to the maximum AGB value of all input coverages.
• Dropped SVM due to huge computational demand as number of samples increase.

	RF (ranger)	GBM (LightGBM)	Ensemble (model weighted)	Ensemble (RMSE weighted)
RMSE	51.180	46.296	46.232	47.153
RMSE %	35.230	31.868	31.825	32.459
MBE	0.395	0.272	0.234	0.331
R2	0.634	0.694	0.695	0.686

AGB Distribution

summary(bind_rows(training, testing)$agb_mgha)

     Min.   1st Qu.    Median      Mean   3rd Qu.      Max. 
  0.00583  73.64861 145.60648 146.02056 218.85216 306.56705 

Validation Results

RMSE	Min	Median	Max
Rf	50.348	51.371	52.150
Lgb	45.518	46.629	47.644
Ensemble	46.395	47.391	48.188

“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	0.617	0.632	0.650
lgb	0.674	0.690	0.705
ensemble	0.670	0.684	0.701

Metadata

Ensembles

• RMSE-weighted model weights:

      lgb        rf 
0.5236072 0.4763928 

• Linear model weights:

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

Residuals:
    Min      1Q  Median      3Q     Max 
-265.37  -28.53    0.90   28.92  238.90 

Coefficients:
                   Estimate Std. Error t value Pr(>|t|)    
(Intercept)      -1.822e+00  3.226e-01  -5.649 1.62e-08 ***
rf_pred           1.487e-01  3.911e-03  38.033  < 2e-16 ***
lgb_pred          8.481e-01  3.558e-03 238.385  < 2e-16 ***
rf_pred:lgb_pred  8.857e-05  1.554e-05   5.699 1.21e-08 ***
---
Signif. codes:  0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1

Residual standard error: 46.52 on 528096 degrees of freedom
Multiple R-squared:  0.6912,    Adjusted R-squared:  0.6912 
F-statistic: 3.939e+05 on 3 and 528096 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 \(p\)

• 25165 observations
  • 17606 training
  • 7559 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] 500

$mtry
[1] 5

$min.node.size
[1] 3

$sample.fraction
[1] 1

$splitrule
[1] "maxstat"

$replace
[1] FALSE

$formula
agb_mgha ~ .

LGB:

$learning_rate
[1] 0.1

$nrounds
[1] 1000

$num_leaves
[1] 18

$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] 3

$lambda_l1
[1] 5

$lambda_l2
[1] 3

$force_col_wise
[1] TRUE