Abstract: Quantitative remote sensing technique is at the forefront of using remote sensing big data for ore prospecting. To explore the effects of number of samples on quantitative remote sensing modeling, we used multiple regression analysis method to perform remote sensing modeling and inversion experiments with varying number of surface bauxite samples. In this study, surface samples were first analyzed to obtain Al2O3 and SiO2 content information. Then, from the Landsat 8 remote sensing data, spectral reflectance of 1 to 7 spectral bands were read according to sample position. Next, multiple regression analysis on Al2O3 and SiO2 contents and corresponding reflectance of the 17 spectral bands were carried out using the SPSS software to establish a remote sensing quantitative inversion model according to the Al2O3 and SiO2 contents. In order to obtain the best inversion model, we randomly selected 6 batches of different number of samples for the modeling experiments, using about 2/3 of samples for modeling and the rest for model testing. The experimental results showed that as number of samples increased, the coefficient of determination (R2) first rose rapidly and then slowly declined; whereas the root mean square error (RMSE) behaved oppositely. At number of samples 50, maximum R2 and minimum RMSE were reached, showing an overall skewed data distribution. Finally, the remote sensing image was used to invert the established model at number of samples 50. The inversion results were in good agreement with experimentally measured Al2O3 and SiO2 contents in the study area, confirming that the modeling method is reliable and its application can be further expanded.

Key words: quantitative remote sensing, Madagascar, laterite type bauxite, multiple regression analysis, modeling, inversion