ABSTRACT
Agave lechuguilla Torr., of the family Agavaceae, is distributed from southwestern United States to southern Mexico and is one of the most representative species of arid and semiarid regions. Its fiber is extracted for multiple purposes. The objective of this study was to generate a robust model to predict dry fiber yield (Dfw) rapidly, simply, and inexpensively. We used a power model in its linear form and bioclimatic areas as dummy variables. Training, generation (80%) and validation (20%) of the model was performed using machine learning with the package 'caret' of R. Using canonical correlation analysis (CCA), we evaluated the relationship of Dwf to bioclimatic variables. The principal components analysis (PCA) generated two bioclimatic zones, each with different A. lechuguilla productivities. We evaluated 499 individuals in four states of Mexico. The crown diameter (Cd) of this species adequately predicts its fiber dry weight (R2 = 0.6327; p < 0.05). The intercept (ß0), slope [lnCd (ß1)], zone [(ß2)] and interaction [lnCd:Zona (ß3)] of the dummy model was statistically significant (p < 0.05), giving origin to an equation for each bioclimatic zone. The CCA indicates a positive correlation between minimum temperature of the coldest month (Bio 6) and Dwf (r = 0.84 and p < 0.05). In conclusion, because of the decrease in Bio 6 of more than 0.5°C by 2050, the species could be vulnerable to climate change, and A. lechuguilla fiber production could be affected gradually in the coming years.