TABLE 2.

Best fit models of seed weight (g), based on 65 Pinus strobiformis stands in Mexico and United States.

Method of variable selection	Machine learning algorithm	Independent variables	RMSE	MAE	R2
ROC	lm	GSP, Pseudotsuga menziesii, MAT, SMRPB, P. arizonica, J. deppeana	0.039	0.033	0.798
KW	brnn	GSP, MAT, Pseudotsuga menziesii, P. strobiformis, P. arizonica, SMRPB	0.040	0.031	0.780
KW	rf	GSP, MAT, Pseudotsuga menziesii, P. strobiformis, P. arizonica, SMRPB	0.041	0.033	0.752
PLS	avNNet	DD0, Pseudotsuga menziesii, P. arizonica, Arctostaphylos pungens, Arbutus xalapensis, SMRPB	0.043	0.037	0.769
PLS	nnet	DD0, Pseudotsuga menziesii, P. arizonica, Arctostaphylos pungens, Arbutus xalapensis, SMRPB	0.048	0.039	0.738
PLS	mlpWeightDecay	DD0, Pseudotsuga menziesii, P. arizonica, Arctostaphylos pungens, Arbutus xalapensis, SMRPB	0.075	0.061	0.468

PLS = Partial Least Squares, ROC = Receiver Operating Characteristic, KW = Kruskal Wallis, rf = Random Forest, brnn = Bayesian Regularized Neural Networks, lm = linear regression, mlpWeightDecay = Multi-Layer Perceptron, nnet = Neural Network, avNNet = Model Averaged Neural Network, RMSE = Root-mean-square error, MAE = Mean Absolute Error, R² = R squared, GSP = Growing season precipitation, April to September, Pseudotsuga menziesii = frequency of occurrence of Pseudotsuga menziesii in the neighborhood, SMRPB = Summer precipitation balance: (jul+aug+sep)/(apr+may+jun), Juniperus deppeana = frequency of occurrence of Juniperus deppeana in the neighborhood, P. arizonica = frequency of occurrence of P. arizonica in the neighborhood, Arbutus xalapensis = frequency of occurrence of Arbutus xalapensis in the neighborhood, Arctostaphylos pungens = frequency of occurrence of Arctostaphylos pungens in the neighborhood, MAT = Mean annual temperature (degrees C), P. strobiformis = frequency of occurrence of P. strobiformis in the neighborhood, DD0 = Degree-days < 0 degrees C (based on mean monthly temperature).