Skip to main content

Table 2 Comparison of the different models for the modeling of SOC stocks at various soil depths on the Qinghai Plateau

From: Uncertainties of soil organic carbon stock estimation caused by paleoclimate and human footprint on the Qinghai Plateau

Soil depth (cm)

Type

Model

R2

CC

RMSE

MAPE

NMSE

0–30

Model_Ori

RF

0.520

0.654

3.331

0.727

0.491

GBM

0.523

0.697

3.296

0.722

0.481

SVM

0.410

0.582

3.671

0.770

0.597

Model_PC

RF

0.526

0.669

3.297

0.700

0.481

GBM

0.553

0.719

3.189

0.653

0.450

SVM

0.449

0.614

3.550

0.735

0.558

Model_H

RF

0.523

0.664

3.311

0.706

0.489

GBM

0.527

0.700

3.285

0.704

0.478

SVM

0.451

0.615

3.544

0.768

0.556

Basic data

MC_NitrDepAll (Wet deposition of inorganic nitrogen), MC_LAI (Leaf area index), MC_PAR (Photosynthetically active radiation), V_AGBC (Aboveground biomass carbon), V_NDVI (Normalized differential vegetation index), V_FAPAR (Fraction of absorbed photosynthetically active radiation), V_NPP (Net primary productivity), S_MicroCN (C:N ratio of soil microbial biomass)

30–50

Model_Ori

RF

0.583

0.696

2.142

2.040

0.435

GBM

0.563

0.723

2.150

2.175

0.439

SVM

0.505

0.691

2.300

2.328

0.502

Model_PC

RF

0.611

0.726

2.062

2.069

0.403

GBM

0.588

0.744

2.086

2.219

0.413

SVM

0.538

0.712

2.215

2.428

0.466

Model_H

RF

0.582

0.699

2.139

2.048

0.434

GBM

0.582

0.736

2.100

2.213

0.419

SVM

0.508

0.681

2.282

2.392

0.494

Basic data

MC_Wind10 (10 m wind speed), MC_Tem (Modern temperature), MC_Pre (Modern precipitation), V_EVI (Enhanced vegetation index), V_GPP (Gross primary productivity), V_NPP, MC_NitrDepAll, MC_PAR, V_LAI, V_NDVI, V_FAPAR

50–100

Model_Ori

RF

0.655

0.761

3.730

0.919

0.360

GBM

0.682

0.794

3.547

0.879

0.326

SVM

0.637

0.776

4.758

0.847

0.366

Model_PC

RF

0.670

0.777

3.637

0.887

0.343

GBM

0.693

0.810

3.462

0.847

0.311

SVM

0.654

0.783

3.669

0.887

0.349

Model_H

RF

0.648

0.757

3.761

0.882

0.367

GBM

0.692

0.802

3.487

0.858

0.315

SVM

0.576

0.715

4.078

0.930

0.431

Basic data

MC_Wind10, MC_NitrDepAll, MC_PAR, V_EVI, V_GPP, V_NPP, V_FAPAR, S_MicroCN, S_MicroSMC (Soil microbial biomass carbon)

100–200

Model_Ori

RF

0.768

0.764

8.283

1.556

0.319

GBM

0.692

0.736

8.819

1.644

0.361

SVM

0.745

0.846

7.444

1.714

0.257

Model_PC

RF

0.778

0.790

7.926

1.647

0.292

GBM

0.775

0.806

7.734

1.673

0.278

SVM

0.876

0.935

5.184

1.700

0.125

Model_H

RF

0.809

0.777

8.017

1.563

0.299

GBM

0.794

0.806

7.662

1.644

0.273

SVM

0.928

0.961

3.964

1.763

0.073

Basic data

T_Slope (Slope), V_SIF (Sun-Induced Chlorophyll Fluorescence), V_NPP, MC_Surrunoff (Surface runoff), MC_Wind10, MC_NitrDepAll, MC_PAR

Paleoclimate factors

PC_Pre_LGM (Paleo-precipitation in the last glacial maximum), PC_Tem_LGM (Paleo-temperature in the last glacial maximum), PC_Pre_MidH (Paleo-precipitation in the mid-Holocene), PC_Tem_MidH (Paleo-temperature in the mid-Holocene)

Human footprint factors

H_Population (Population density), H_HumanFp (Human footprint)

  1. SOC represents soil organic carbon; Model_Ori represents SOC stock estimated without considering the paleoclimate or the human footprint factors; Model_PC represents SOC stock estimated considering the paleoclimate factors; Model_H represents SOC stock estimated considering human footprint factors. RF, GBM and SVM represent the random forest model, the gradient boosting machine model and the support vector machine, respectively. R2, CC, RMSE, MAPE and NMSE indicate the coefficient of determination, Lin’s concordance correlation coefficient, root mean square error, mean absolute percentage error and normalized mean square error, respectively. The selected variables were obtained by integrating the recursive feature elimination (RFE), Boruta, fscaret and mlr algorithms