Urban green spaces (UGSs) provide essential environmental services for the well-being of ecosystems and society. Due to the constant environmental, social, and economic transformations of cities, UGSs pose new challenges for management, particularly in fast-growing metropolitan areas. With technological advancement and the evolution of deep learning, it is possible to optimize the acquisition of UGS inventories through the detection of geometric patterns present in satellite imagery. This research evaluates two deep learning model techniques for semantic segmentation of UGS polygons with the use of different convolutional neural network encoders on the U-Net architecture and very high resolution (VHR) imagery to obtain updated information on UGS polygons at the metropolitan area level. The best model yielded a Dice coefficient of 0.57, IoU of 0.75, recall of 0.80, and kappa coefficient of 0.94 with an overall accuracy of 0.97, which reflects a reliable performance of the network in detecting patterns that make up the varied geometry of UGSs. A complete database of UGS polygons was quantified and categorized by types with location and delimited by municipality, allowing for the standardization of the information at the metropolitan level, which will be useful for comparative analysis with a homogenized and updated database. This is of particular interest to urban planners and UGS decision-makers.

Aerosol and meteorological remote sensing data could be used to assess the distribution of urban and regional fine particulate matter (PM_2.5), especially in locations where there are few or no ground-based observations, such as Latin America. The objective of this study is to evaluate the ability of Modern-Era Retrospective Analysis for Research and Application, version 2 (MERRA-2) aerosol components to represent PM_2.5 ground concentrations and to develop and validate an ensemble neural network (ENN) model that uses MERRA-2 aerosol and meteorology products to estimate the monthly average of PM_2.5 ground concentrations in the Monterrey Metropolitan Area (MMA), which is the main urban area in Northeastern Mexico (NEM). The project involves the application of the ENN model to a regional domain that includes not only the MMA but also other municipalities in NEM in the period from January 2010 to December 2014. Aerosol optical depth (AOD), temperature, relative humidity, dust PM_2.5, sea salt PM_2.5, black carbon (BC), organic carbon (OC), and sulfate (SO₄²⁻) reanalysis data were identified as factors that significantly influenced PM_2.5 concentrations. The ENN estimated a PM_2.5 monthly mean of 25.62 μg m⁻³ during the entire period. The results of the comparison between the ENN and ground measurements were as follows: correlation coefficient R ~ 0.90; root mean square error = 1.81 μg m⁻³; mean absolute error = 1.31 μg m⁻³. Overall, the PM_2.5 levels were higher in winter and spring. The highest PM_2.5 levels were located in the MMA, which is the major source of air pollution throughout this area. The estimated data indicated that PM_2.5 was not distributed uniformly throughout the region but varied both spatially and temporally. These results led to the conclusion that the magnitude of air pollution varies among seasons and regions, and it is correlated with meteorological factors. The methodology developed in this study could be used to identify new monitoring sites and address information gaps.

Classification of remotely sensed data with artificial neural networks is called neuro-classification, and this technique has shown great potential. The amount of data used for training a neural network affects the accuracy and efficiency of the neural network classifier. A neural network was trained separately with 5, 10, 15, and 20 per cent of image data from a Landsat Thematic Mapper scene, which was acquired 29 July 1987 for an agricultural region within Indiana, U.S.A. At a risk level of 5 per cent, the results showed that (a) classifiers NN-5% (neuro-classification with 5 per cent of the image data used for training), AW-10%, and AW-15% did not differ from one another, (b) classifiers AW-15% and AW-20% did not differ from each other, but (c) classifiers NN-5% and AW-10% differed from classifier AW-20%. The training rates were reduced by more than 10 seconds cycle^-1 as we increased the percentage of the image data for training a neural network. Approximately 5-10 per cent of the image data are needed to train a neural network classifier adequately to obtain satisfactory performance.