Also, vegetation indices a) NDVI (Normalized Difference Vegetation Index), b) MNDWI (Modified Normalized Difference Water Index), c) NDBI (Normalized Difference Built-up Index) were calculated and added as bands in images to enhance information about vegetation, water and impervious surfaces. They were used together with RGB, NIR, SWIR bands of Landsat images for further processing.

The B2-B7 (RGB, NIR, SWIR) bands of Landsat images, along with NDVI, MNDWI and NDBI index bands were utilized in the machine learning algorithm.

To calculate the accuracy of the classification algorithm and the reliability of our result, the confusion matrix should be calculated. In the field of machine learning confusion matrix - also known as error matrix - is a special table that illustrates the performance of an algorithm (supervised learning). Each column of the table represents the cases in a predicted class while each row represents the cases in an actual class (or vice versa). Its name is derived from the fact that it makes it easy for us to observe if the system confuses one category with another.

From the confusion table we can calculate the accuracy based on the formula: $accuracy={(TP+TN)\over(TP+TN+FP+FN)}$, where TP = true positive, TN = true negative, FP = false positive, FN = false negative. Simply put, accuracy captures "the percentage of correct estimates." Kappa is another index that helps evaluate the performance between a set of classifiers. Kappa is designed to compare the performance of any classifier to the baseline of a "random classifier". In particular, $κ={(accuracy-random accuracy)\over(1-random accuracy)}$

In a similar way we calculated the mask for the areas that have nearby urban green spaces of 1 hectare (10000 square meters) and more, i.e. a minimum of 11 pixels (30mx30m=$900m^2$, 10000/900=11.1) of green space per 300 meter radius. Single intensity maps of greenery were used, where trees, bushes and grass have a value of 1.

The accuracy of the machine learning mechanism ranges as follows:

Summary of landcover class areas 2013

Summary of landcover class areas 2018

Summary of landcover class areas 2022

Regarding the index of accessibility to green spaces, it increased over the years by 7 percentage points, but there is a huge room for improvement. The following numbers were observed:

Although most of Athens geographically is relatively close to green spaces, this is not the case in the areas of the center, west and southwest of the center, which are the areas with the highest population density. A breath of fresh air for these areas are the few large parks such as the Tritsi park, the Veikou grove, the Peristeri grove and the green spaces in Lykavitto, Pnyka, Zappeio that have been marked in the image.

It is clear that no significant progress has been recorded in Athens around the issue of urban green spaces. Many squares even today are being "redeveloped" having cement as their main element, or at best having a low-level grass section. The metropolitan parks of Athens are few and most citizens only have access to smaller parks and squares.

Apart from the clear impact that greenery has on reducing temperature and increasing air quality, there are also many collateral "losses". Greater energy consumption and costs due to the temperature, respiratory problems, less exercise potential, reduced social cohesion and psychological tension.

In any case, emphasis should be placed on the most affected areas of the center and the western-south-western suburbs to reverse this alarming situation of an unsustainable city.