Abstract: Interactions involving land use and transport have been increasingly studied by researchers who deal with dynamic phenomena, in favor of global sustainability. One method that represents this phenomenon consists of the Land Use Transport Interaction – LUTI model, which provides predictions of these interactions and identifies causal relationships for policy formulation. LUTI models identify the patterns that best relate land use to transport, projecting the evolution of the urban area. The goal of this research is to obtain a calibrated dynamic model capable of simulating scenarios, based on the interaction of land use and transport, in the Metropolitan Region of São Paulo. The application of the NDBI index demonstrated a high capacity to discretize natural and artificial materials, resulting in excellent separability, generating negative values on surfaces formed by vegetation and positive values for built-up areas. This result indicated that the NDBI index is the most appropriate choice for use in the band difference method, when used in the change detection algorithm, highlighting the type, extent and location of the change for each pixel in this time series. The generated product consists of a change detection output raster file, which represents the difference in pixels between scenes, the type or intensity of the change. The changes that represent expansion and densification was be subjected to statistical testing using the M index to test the separability capacity of the NDBI index in the discretization of classes of changes. The scars of the changes was be used as references for collecting and training pixels, in Landsat reflectance scenes, for use in the Deep Learning supervised classification method (Figure 3a), consisting of three phases: 1st) Sample training, 2nd) Deep Learning model training and 3rd) Performance. As a result, a specific land use classification model will be obtained for the RMSP, .dlpk file (Depp Leanring Package). The raster classified by the .dlpk model and other transport parameters (accessibility, distance traveled, gradient between roads and transport zones) will be used as input parameters for the AC – TRAZ model, for calibration, optimizing the characteristics of the RMSP. As a future perspective, transition maps between the real and the simulated will be generated to support the formulation of policy and urban planning.

Keywords: Remote Sensing, Change detection, Deep learning, Dynamic systems, Public policy.