Comparisons of Regression Tree Models For Sub-Pixel Imperviousness Estimation In a Gulf Coast City of Mississippi, USA
Coastal Sciences, Gulf Coast Research Laboratory
Ocean Science and Engineering
We studied the impact of shaded impervious surface area (ISA), atmospheric correction, and seasonal sensitivity, which have been generally ignored in previous studies, on ISA estimation at the sub-pixel scale using regression tree modelling. The study area is Pascagoula City on the Mississippi Gulf Coast, USA. Results showed that inclusion of shaded ISA as the response variable improved the model performance by reducing average error (AE) from 10.17 to 9.36%. Modelling with model-based atmospherically corrected imagery as predictors further reduced AE to 9.27%. The regression tree model using summer imagery as predictors (summer model) finally improved AE to 8.56%, compared with 9.28%, 9.50%, and 8.80% when using early spring, late spring, and autumn images as predictors, respectively; therefore the summer model was considered the optimal model. It was further applied to other seasonal images (i.e. early spring, late spring, and autumn images, as predictors) and the AE was 9.93%, 10.09%, and 9.12%, respectively, showing low seasonal sensitivity within this region. The findings in our study improved the modelling accuracy and expanded the scope of its future application in ISA estimation. © 2014 Taylor & Francis.
International Journal of Remote Sensing
(2014). Comparisons of Regression Tree Models For Sub-Pixel Imperviousness Estimation In a Gulf Coast City of Mississippi, USA. International Journal of Remote Sensing, 35(10), 3722-3740.
Available at: https://aquila.usm.edu/fac_pubs/17978