Variable: fNRec Ranger result Call: ranger(formulaString.lst[[j]], data = dfs, importance = "impurity", write.forest = TRUE, mtry = t.mrfX$bestTune$mtry, num.trees = 500) Type: Regression Number of trees: 500 Sample size: 34 Number of independent variables: 377 Mtry: 4 Target node size: 5 Variable importance mode: impurity OOB prediction error: 185.6971 R squared: -0.2992777 OOB RMSE: 13.627 Variable importance: [,1] rElevIndex 27.98242 Mg_M_agg30cm_AF_1km 26.04698 MAXENV3 25.85884 ECN_M_agg30cm_AF_1km 24.45150 NMSD3avg 22.62124 P.B_M_agg30cm_AF_1km 22.41333 NCluster_7_AF_1km 22.10676 SNDPPT_M_agg30cm_AF_1km 21.49865 B13CHE3 21.36394 Temperature 21.35071 NIRL00 19.67176 M43WSALT 19.53567 VBFMRG5 19.45228 Water_balance 19.23583 PRSCHE3 19.12990 eXtreme Gradient Boosting 34 samples 377 predictors No pre-processing Resampling: Cross-Validated (3 fold, repeated 1 times) Summary of sample sizes: 23, 22, 23 Resampling results across tuning parameters: eta max_depth nrounds RMSE Rsquared 0.3 2 50 15.18234 0.08977676 0.3 2 100 15.18353 0.08982070 0.3 2 150 15.18353 0.08982070 0.3 3 50 15.00116 0.09320908 0.3 3 100 15.00141 0.09322081 0.3 3 150 15.00141 0.09322081 0.3 4 50 15.04206 0.07410893 0.3 4 100 15.04201 0.07410726 0.3 4 150 15.04201 0.07410726 0.3 5 50 15.08852 0.08231704 0.3 5 100 15.08872 0.08232851 0.3 5 150 15.08872 0.08232851 0.3 6 50 15.00554 0.08478895 0.3 6 100 15.00585 0.08480235 0.3 6 150 15.00585 0.08480235 0.3 7 50 15.08022 0.07913931 0.3 7 100 15.08042 0.07915654 0.3 7 150 15.08042 0.07915654 0.3 8 50 15.08262 0.06599365 0.3 8 100 15.08264 0.06598472 0.3 8 150 15.08264 0.06598471 0.4 2 50 14.88445 0.10228126 0.4 2 100 14.88443 0.10228024 0.4 2 150 14.88443 0.10228024 0.4 3 50 14.79328 0.07890818 0.4 3 100 14.79327 0.07890814 0.4 3 150 14.79327 0.07890814 0.4 4 50 14.85352 0.06262032 0.4 4 100 14.85351 0.06262031 0.4 4 150 14.85351 0.06262031 0.4 5 50 14.94794 0.06252836 0.4 5 100 14.94794 0.06252836 0.4 5 150 14.94794 0.06252836 0.4 6 50 14.83880 0.06482174 0.4 6 100 14.83879 0.06482248 0.4 6 150 14.83879 0.06482248 0.4 7 50 14.93287 0.05801256 0.4 7 100 14.93287 0.05801255 0.4 7 150 14.93287 0.05801255 0.4 8 50 14.74715 0.08292705 0.4 8 100 14.74715 0.08292706 0.4 8 150 14.74715 0.08292706 0.5 2 50 14.87274 0.06001605 0.5 2 100 14.87274 0.06001606 0.5 2 150 14.87274 0.06001606 0.5 3 50 14.93540 0.06061883 0.5 3 100 14.93540 0.06061883 0.5 3 150 14.93540 0.06061883 0.5 4 50 14.80131 0.06340168 0.5 4 100 14.80131 0.06340168 0.5 4 150 14.80131 0.06340168 0.5 5 50 14.89984 0.06619110 0.5 5 100 14.89984 0.06619107 0.5 5 150 14.89984 0.06619107 0.5 6 50 14.90376 0.05953289 0.5 6 100 14.90376 0.05953289 0.5 6 150 14.90376 0.05953289 0.5 7 50 15.08949 0.09287150 0.5 7 100 15.08949 0.09287150 0.5 7 150 15.08949 0.09287150 0.5 8 50 14.93150 0.06656511 0.5 8 100 14.93150 0.06656510 0.5 8 150 14.93150 0.06656510 Tuning parameter 'gamma' was held constant at a value of 0 Tuning parameter 'colsample_bytree' was held constant at a value of 0.8 Tuning parameter 'min_child_weight' was held constant at a value of 1 RMSE was used to select the optimal model using the smallest value. The final values used for the model were nrounds = 100, max_depth = 8, eta = 0.4, gamma = 0, colsample_bytree = 0.8 and min_child_weight = 1. RMSE: 14.747 R2: 0.083 XGBoost variable importance: Feature Gain Cover Frequency 1: B13CHE3 0.294847655 0.032661393 0.03649635 2: Mg_M_agg30cm_AF_1km 0.201506186 0.008675683 0.00729927 3: CLYPPT_M_agg30cm_AF_1km 0.181207936 0.041081909 0.03649635 4: Ca_M_agg30cm_AF_1km 0.117555826 0.039040572 0.03649635 5: M43WNALT 0.037679029 0.020158204 0.02189781 6: Zn_M_agg30cm_AF_1km 0.036500701 0.013013524 0.01459854 7: af_BDRICM_T__M_1km 0.032985689 0.013268691 0.02919708 8: SSI_NCluster_AF_1km 0.022626479 0.017351365 0.01459854 9: K_M_agg30cm_AF_1km 0.018950717 0.007910181 0.00729927 10: AAIavg_GYGA 0.012477020 0.015310028 0.04379562 11: BIO12ALT 0.011586778 0.260270477 0.21897810 12: M43BNALT 0.008341251 0.008165348 0.00729927 13: ENTENV3 0.007245472 0.089818831 0.08029197 14: P_M_agg30cm_AF_1km 0.005969332 0.008165348 0.00729927 15: NCluster_4_AF_1km 0.002923219 0.039550906 0.03649635 Ensemble validation RMSE: 13.972 R2: 0.041 -------------------------------------- Variable: fPRec Ranger result Call: ranger(formulaString.lst[[j]], data = dfs, importance = "impurity", write.forest = TRUE, mtry = t.mrfX$bestTune$mtry, num.trees = 500) Type: Regression Number of trees: 500 Sample size: 34 Number of independent variables: 377 Mtry: 20 Target node size: 5 Variable importance mode: impurity OOB prediction error: 16.70182 R squared: 0.2169388 OOB RMSE: 4.087 Variable importance: [,1] Fcover 21.529618 Fapar 18.440699 MANMCF5 18.380451 M13RB3A08 18.243483 Wdvi 17.372679 CHIRPSA 13.203840 PHIHOXagg0_30 12.883933 ENAX_M_agg30cm_AF_1km 11.591831 M13RB3ALT 11.544608 MAXENV3 11.112103 GAEZ_ET 10.988646 CMCF5avg 10.595683 Cassave_rainfed_intermed_baseline 10.032044 VW1MOD1avg 9.201425 ESMOD5avg 8.999925 eXtreme Gradient Boosting 34 samples 377 predictors No pre-processing Resampling: Cross-Validated (3 fold, repeated 1 times) Summary of sample sizes: 23, 23, 22 Resampling results across tuning parameters: eta max_depth nrounds RMSE Rsquared 0.3 2 50 3.836204 0.4023434 0.3 2 100 3.835927 0.4022958 0.3 2 150 3.835927 0.4022958 0.3 3 50 3.811407 0.3991708 0.3 3 100 3.811133 0.3991999 0.3 3 150 3.811133 0.3991999 0.3 4 50 3.807119 0.4065536 0.3 4 100 3.806860 0.4065695 0.3 4 150 3.806860 0.4065695 0.3 5 50 3.843540 0.3943848 0.3 5 100 3.843217 0.3944179 0.3 5 150 3.843217 0.3944179 0.3 6 50 3.792635 0.4096044 0.3 6 100 3.792402 0.4096262 0.3 6 150 3.792402 0.4096262 0.3 7 50 3.870011 0.3899893 0.3 7 100 3.869631 0.3900486 0.3 7 150 3.869631 0.3900486 0.3 8 50 3.762989 0.4073050 0.3 8 100 3.762728 0.4073227 0.3 8 150 3.762728 0.4073227 0.4 2 50 3.794338 0.3988201 0.4 2 100 3.794338 0.3988201 0.4 2 150 3.794338 0.3988201 0.4 3 50 3.726823 0.4137419 0.4 3 100 3.726823 0.4137419 0.4 3 150 3.726823 0.4137419 0.4 4 50 3.781841 0.4058668 0.4 4 100 3.781841 0.4058668 0.4 4 150 3.781841 0.4058668 0.4 5 50 3.862611 0.3913391 0.4 5 100 3.862610 0.3913391 0.4 5 150 3.862610 0.3913391 0.4 6 50 3.693929 0.4134035 0.4 6 100 3.693929 0.4134035 0.4 6 150 3.693929 0.4134035 0.4 7 50 3.694278 0.4179140 0.4 7 100 3.694278 0.4179140 0.4 7 150 3.694278 0.4179140 0.4 8 50 3.688472 0.4110503 0.4 8 100 3.688472 0.4110503 0.4 8 150 3.688472 0.4110503 0.5 2 50 3.820601 0.4094829 0.5 2 100 3.820601 0.4094829 0.5 2 150 3.820601 0.4094829 0.5 3 50 3.954561 0.3876387 0.5 3 100 3.954561 0.3876387 0.5 3 150 3.954561 0.3876387 0.5 4 50 3.949525 0.3855733 0.5 4 100 3.949525 0.3855733 0.5 4 150 3.949525 0.3855733 0.5 5 50 4.043834 0.3688972 0.5 5 100 4.043834 0.3688972 0.5 5 150 4.043834 0.3688972 0.5 6 50 3.910200 0.3836654 0.5 6 100 3.910200 0.3836654 0.5 6 150 3.910200 0.3836654 0.5 7 50 3.940272 0.3907858 0.5 7 100 3.940272 0.3907858 0.5 7 150 3.940272 0.3907858 0.5 8 50 3.875181 0.3898764 0.5 8 100 3.875181 0.3898764 0.5 8 150 3.875181 0.3898764 Tuning parameter 'gamma' was held constant at a value of 0 Tuning parameter 'colsample_bytree' was held constant at a value of 0.8 Tuning parameter 'min_child_weight' was held constant at a value of 1 RMSE was used to select the optimal model using the smallest value. The final values used for the model were nrounds = 100, max_depth = 8, eta = 0.4, gamma = 0, colsample_bytree = 0.8 and min_child_weight = 1. RMSE: 3.688 R2: 0.411 XGBoost variable importance: Feature Gain Cover Frequency 1: Fapar 4.786555e-01 0.048868128 0.04123711 2: ENAX_M_agg30cm_AF_1km 2.833340e-01 0.012217032 0.01030928 3: AAIavg_GYGA 1.864183e-01 0.271649299 0.25773196 4: yFertilised_CassavaT2 2.869506e-02 0.022278117 0.02061856 5: NCluster_7_AF_1km 1.289611e-02 0.022996766 0.02061856 6: P.B_M_agg30cm_AF_1km 4.119178e-03 0.011498383 0.01030928 7: B02CHE3 2.022073e-03 0.054976644 0.05154639 8: af_agg_30cm_PWP__M_1km 1.336849e-03 0.032698527 0.04123711 9: B04CHE3 8.978421e-04 0.061085160 0.05154639 10: Wdvi 5.126154e-04 0.011498383 0.01030928 11: af_BDRICM_T__M_1km 2.969200e-04 0.008623787 0.02061856 12: ENTENV3 2.875018e-04 0.090549766 0.08247423 13: P.T_M_agg30cm_AF_1km 2.250121e-04 0.008983112 0.01030928 14: Al_M_agg30cm_AF_1km 1.265373e-04 0.013295005 0.02061856 15: NMSD3avg 7.183179e-05 0.057491915 0.05154639 Ensemble validation RMSE: 3.81 R2: 0.317 -------------------------------------- Variable: fKRec Ranger result Call: ranger(formulaString.lst[[j]], data = dfs, importance = "impurity", write.forest = TRUE, mtry = t.mrfX$bestTune$mtry, num.trees = 500) Type: Regression Number of trees: 500 Sample size: 34 Number of independent variables: 377 Mtry: 16 Target node size: 5 Variable importance mode: impurity OOB prediction error: 267.4312 R squared: 0.2613264 OOB RMSE: 16.353 Variable importance: [,1] Fapar 345.6508 M13RB1A08 247.2135 Slopeclassc3 231.9134 Fcover 230.8948 CRFVOL_M_agg30cm_AF_1km 229.4407 EVEENV3 226.4782 M43WVALT 214.2407 M13NDVIA08 197.3992 ENTENV3 194.3412 MAXENV3 187.9370 Cassave_actual_baseline 174.2224 CLYPPT_M_agg30cm_AF_1km 168.3625 BLDFIE_M_agg30cm_AF_1km 164.1207 P_M_agg30cm_AF_1km 163.9889 M43BVALT 156.8672 eXtreme Gradient Boosting 34 samples 377 predictors No pre-processing Resampling: Cross-Validated (3 fold, repeated 1 times) Summary of sample sizes: 22, 24, 22 Resampling results across tuning parameters: eta max_depth nrounds RMSE Rsquared 0.3 2 50 15.77765 0.4006731 0.3 2 100 15.77808 0.4006833 0.3 2 150 15.77808 0.4006833 0.3 3 50 15.68199 0.4130488 0.3 3 100 15.68076 0.4131132 0.3 3 150 15.68076 0.4131132 0.3 4 50 16.27779 0.3604495 0.3 4 100 16.27691 0.3604985 0.3 4 150 16.27691 0.3604985 0.3 5 50 16.14101 0.3728676 0.3 5 100 16.13982 0.3729292 0.3 5 150 16.13982 0.3729292 0.3 6 50 16.11344 0.3711817 0.3 6 100 16.11239 0.3712441 0.3 6 150 16.11239 0.3712441 0.3 7 50 15.92993 0.3929483 0.3 7 100 15.92887 0.3930216 0.3 7 150 15.92887 0.3930216 0.3 8 50 16.19762 0.3726773 0.3 8 100 16.19654 0.3727517 0.3 8 150 16.19654 0.3727517 0.4 2 50 16.71978 0.3364876 0.4 2 100 16.71982 0.3364866 0.4 2 150 16.71982 0.3364866 0.4 3 50 15.96919 0.4341920 0.4 3 100 15.96919 0.4341920 0.4 3 150 15.96919 0.4341920 0.4 4 50 14.91439 0.4875095 0.4 4 100 14.91438 0.4875095 0.4 4 150 14.91438 0.4875095 0.4 5 50 14.79858 0.4648629 0.4 5 100 14.79857 0.4648629 0.4 5 150 14.79857 0.4648629 0.4 6 50 15.24246 0.4498177 0.4 6 100 15.24245 0.4498177 0.4 6 150 15.24245 0.4498177 0.4 7 50 16.08846 0.3952140 0.4 7 100 16.08845 0.3952140 0.4 7 150 16.08845 0.3952140 0.4 8 50 15.65491 0.3983262 0.4 8 100 15.65490 0.3983262 0.4 8 150 15.65490 0.3983262 0.5 2 50 16.66277 0.3627868 0.5 2 100 16.66277 0.3627868 0.5 2 150 16.66277 0.3627868 0.5 3 50 15.55958 0.4116431 0.5 3 100 15.55958 0.4116431 0.5 3 150 15.55958 0.4116431 0.5 4 50 14.97357 0.4269180 0.5 4 100 14.97357 0.4269180 0.5 4 150 14.97357 0.4269180 0.5 5 50 17.02573 0.3232432 0.5 5 100 17.02573 0.3232432 0.5 5 150 17.02573 0.3232432 0.5 6 50 16.33954 0.3677072 0.5 6 100 16.33954 0.3677072 0.5 6 150 16.33954 0.3677072 0.5 7 50 15.20759 0.4401347 0.5 7 100 15.20759 0.4401347 0.5 7 150 15.20759 0.4401347 0.5 8 50 16.14593 0.3727142 0.5 8 100 16.14593 0.3727142 0.5 8 150 16.14593 0.3727142 Tuning parameter 'gamma' was held constant at a value of 0 Tuning parameter 'colsample_bytree' was held constant at a value of 0.8 Tuning parameter 'min_child_weight' was held constant at a value of 1 RMSE was used to select the optimal model using the smallest value. The final values used for the model were nrounds = 100, max_depth = 5, eta = 0.4, gamma = 0, colsample_bytree = 0.8 and min_child_weight = 1. RMSE: 14.799 R2: 0.465 XGBoost variable importance: Feature Gain Cover Frequency 1: Fapar 0.6094941529 0.028210526 0.024539877 2: P_M_agg30cm_AF_1km 0.1226540481 0.026105263 0.024539877 3: Cassave_actual_baseline 0.1094813260 0.014315789 0.012269939 4: af_agg_ERZD_TAWCpF23mm__M_1km 0.0731046455 0.236210526 0.202453988 5: af_ERZD__M_1km 0.0272541187 0.057263158 0.049079755 6: B02CHE3 0.0163347521 0.008421053 0.018404908 7: BLDFIE_M_agg30cm_AF_1km 0.0110028000 0.055789474 0.055214724 8: M13RB1ALT 0.0103400401 0.006315789 0.006134969 9: CLYPPT_M_agg30cm_AF_1km 0.0081244281 0.013684211 0.018404908 10: RANENV3 0.0049015180 0.013052632 0.012269939 11: EACKCL_M_agg30cm_AF_1km 0.0027383643 0.005263158 0.006134969 12: B04CHE3 0.0011433607 0.008421053 0.012269939 13: M43WNALT 0.0010926781 0.026526316 0.024539877 14: M13RB1A01 0.0009766945 0.004421053 0.006134969 15: B14CHE3 0.0004615932 0.024631579 0.024539877 Ensemble validation RMSE: 15.064 R2: 0.382 --------------------------------------