Arid-semiarid ecosystems constitute the largest biome globally. They cover about 41% of the Earth's surface, presenting low rainfall and high temperatures. The use of remote sensing can contribute to the assessment and characterization of this type of vegetation, enabling the estimation of forest parameters of interest. The objective of this study was to evaluate the efficiency of using drone-borne optical and unmanned laser scanning (ULS) data to estimate the aboveground biomass of six non-timber commercial forest species in arid-semiarid ecosystems, using machine learning algorithms. The efficiency of the type of remote sensing data and algorithms utilized, the stepwise regression (R-Step), random forests (RF), and support vector machine (SVM), was evaluated through a comparison with field estimates of total aboveground biomass (AGB(t)) and species-specific aboveground biomass (AGB(s)). The estimated AGB(t) was higher in La Sauceda given its state of conservation (A(1): 5.169 Mg ha(-1)) and lower in Ejido Hipolito (A(2): 2.339 Mg ha(-1)) due to forest management activity. The optical variables showed a higher correlation with AGB(t) (r = 0.42) compared to the ULS metrics (r = 0.17). The SVM algorithm, fed with a combination of ULS and optical (RGB and multispectral images) data, demonstrated greater precision when estimating AGB(t) in both areas (R-2 = 0.62, RMSE = 1.98 Mg ha(-1) in A(1); R-2 = 0.71, RMSE = 1.30 Mg ha(-1) in A(2)), as compared with the R-Step and RF. The accuracy increased when estimating AGB(s) (A(1): R-2 > 0.60 and RMSE < 1.433; A(2): R-2 > 0.54 and RMSE < 0.872). The results demonstrate the potential of the data captured with drone-mounted remote sensors for the assessment of non-timber forest resources in arid-semiarid ecosystems.