Assessment of Urban Green Cover and its Types using Threshold-based Approach

Gaikadi, S.¹ and Vasantha Kumar, S.²*

School of Civil Engineering, Vellore Institute of Technology (VIT), Vellore – 632014, Tamilnadu, India

*Corresponding Author

Abstract

Providing adequate green cover in the form of trees, parks and gardens is a nature-based solution which every city must adopt to address the urbanization and its impacts, as it helps to reduce the temperature, maintain the ecology and biodiversity, provide recreational facilities for the city inhabitants and many more. Accurate mapping and monitoring of Urban Green Spaces (UGS) through satellite images is essential to check whether a city has adequate green cover or not. Studies on UGS have mostly used Landsat and Sentinel-2A satellite images for extracting the green covered areas in cities. Though it is advantageous to use these freely available images, however the accuracy of extraction is a major concern as the images are of >10m spatial resolution. Distinguishing the various green cover types, such as trees, shrubs, cropland, etc. is not an easy task in low and medium resolution satellite images. To overcome these drawbacks, one has to use high (1-10m) and very-high resolution (<1m) images for extracting the UGS accurately, however, purchasing such ‘cm’ resolution satellite images is a costly affair. In order to balance between the cost and resolution, the best and viable option is to use the Google earth (GE) imagery and the same was attempted in the present study by proposing a simple threshold-based approach to extract the UGS and its types from GE imagery. To illustrate the proposed method, 15 sites from the city of Vellore, India were taken and the latest cloud free GE imagery was then downloaded, projected and carefully checked for its range of pixel values in different types of UGS classes. The threshold values of each UGS type were then used to reclassify the original GE image and finally the accuracy assessment was performed on the reclassified UGS map. The results were promising as the overall accuracy (OA) was > 90% at 12 sites and only 3 out of 15 sites exhibits OA in the range of 75-86%. In order to assess the performance of the proposed approach at other locations, 5 cities across the globe [2 from India (Bangalore and Hyderabad) and 3 from abroad (Barcelona, Kuala Lumpur, New York)] were taken and the results revealed that the proposed method performs well in extracting the UGS and its types as the OA was found to be above 90% at all the 5 cities. The proposed approach can be used by civic authorities not only to check whether the city has adequate green cover or not but also can monitor whether UGS has increased or decreased over the years using the historical GE images.

Keywords: Accuracy Assessment, Google Earth Images, Threshold Value, Urban Green Space, Urbanisation

1. Introduction

Urbanisation is the process through which people move from rural areas to cities due to increased work opportunities. In order to obtain better housing environment, better facilities for recreation and better health care facilities and social life too, an increasing number of people are being forced to relocate to cities and towns from rural areas. For a developing nation like India, urbanization is unavoidable; however, unplanned, rapid, and unsystematic urban growth may have adverse effects on a variety of environmental and ecological components, particularly on land and water. The problem is more critical as 68% of the world's population, up from 55% today, is anticipated to reside in cities in 2050 [1]. Nearly 90% of this population growth would occur in Africa and Asia. Due to the country's apparent rapid urbanization, the urban landscape appears to have undergone significant changes in terms of land use and land cover, placing a substantial strain on a number of natural resources [2].

One of the major effects of urbanization is loss of beautiful green cover in a city apart from other ill effects like water scarcity, polluted air, traffic jams, etc. Loss of green cover plays a major role as it directly deals with ecology and biodiversity.

In order to maintain the city with good environmental conditions, one need to adapt the pace of growing green within the empty spaces both indoor and outdoor [3]. Many cities across the world especially in developed nations have already realized the advantages of green spaces and thus took initiatives not only to increase the forest cover but also the green cover within cities, popularly called as urban green spaces (UGS).

Globally, the need for more UGS is a high priority on city policy agenda and it includes trees, grass, shrubs, or other types of vegetation within urban areas. The sustainable growth encourages active lifestyle and protect biodiversity. UGS also reduces the consequences of heat islands on a broader scale. The contributions they have made to lowering carbon emissions and enhancing public health have also received attention in recent years [4]. UGS promotes biodiversity, lowers summer city temperatures and reduces urban flood risk. UGS advances enhanced wellness, increased social cohesion, and ensures wellbeing of citizens. Considering the benefits associated with UGS, many countries in the world have formulated guidelines on how much green space should be available in a city on per capita or per hectare basis. For example, World Health Organization (WHO) has suggested a minimum of 9 m²/person of UGS and in European Union, the publicly accessible UGS in core cities is found to be 18.2 m²/inhabitant on an average, which is twice the recommended value [5]. In India, the suggested UGS per capita is 10-12 m²[6]. In Tamil Nadu state of India, the UGS targeted is 20 to 40% of city’s geographical area or in other words, one-third of the city should be covered by green [7]. In order to check, whether the existing UGS available per capita or per hectare is adequate or as per standards, it is essential to perform an assessment of UGS in a city/town so that if there is any deficiency, steps can be taken by the city authorities to increase the same. In recent years, the advancements in remote sensing and GIS and availability of open source data such as Google earth made UGS calculation easier than before.

Existing studies on extracting UGS from satellite images have mostly used the Landsat and Sentinel-2A data [8] and [9]. Though it is advantageous to use these freely available images, however there are certain drawbacks. For example, accuracy of extraction is a major concern as the images are of ≥10m spatial resolution. Distinguishing the various green cover types, such as trees, shrubs, grassland, cropland, etc. is not an easy task in low and medium resolution satellite images. To overcome these drawbacks, one has to use high (1-10 m) and very-high resolution (<1 m) satellite images for extracting the UGS accurately [10] and [11].

However, purchasing such ‘cm’ resolution satellite images is a costly affair. So, the major limitations in existing studies are accuracy while classifying the low-resolution satellite images and huge cost of purchase if very high-resolution images of less than 1m are employed. In order to balance between the cost and resolution, the best and viable option is to use the Google earth (GE) images where one can clearly see the green cover and its types. There are many advantages in using GE images in remote sensing studies. The main advantage is, it is an open-source data and does not involve any purchase cost. Another advantage with GE is that it provides historical images since the year 2010. So, one can do change detection analysis with very high-resolution GE images, for example, how much UGS have increased or decreased over the years. Another major advantage with GE images is we can get latest imagery as GE often updates the satellite images especially for cities, thus one can get one image every two months on an average for cities like Chennai, Bengaluru, etc. The only drawback with GE image is we may not get the original multispectral bands i.e., Near Infrared (NIR), Red, Green, Blue with digital numbers. Hence the vegetation calculators such as normalized differenced vegetation index cannot be calculated on GE imagery to extract the green covered areas as there is no NIR or Red band available. One of the major gaps in existing studies is the GE imagery was used as a background image and onscreen manual digitisation was performed to prepare the UGS map.

However, it is a cumbersome and time-consuming task especially manual digitizing of features on a very high-resolution imagery [12] and [13]. In order to overcome the above said gap in existing studies and utilize the GE image more effectively, the present study proposed a threshold-based approach to extract the urban green cover and its types from GE image for the city of Vellore, India. The proposed approach is very easy to understand and apply and can be performed in any open-source GIS software also. The objectives of the present work are 1). To extract various UGS types, such as trees, shrubs and crop land from GE imagery using threshold-based approach and perform accuracy assessment 2). To perform universality checking of the proposed approach by extracting UGS in five cities across the globe and perform accuracy assessment. The scope of the study is limited to extraction of only green covered areas and does not include other landuse/landcover classes. All the landuse/landcover classes other than UGS are named as ‘Others’ while classification. Also, the scope is limited to only cities as focus is extracting only urban green spaces.

However, the study can be extended to extraction of forest cover too. The scope of the study is limited to use of only GE imagery and does not use any other satellite data for classification or accuracy assessment.

2. Materials and Methods

The study area chosen for the present work is in the city of Vellore which is the administrative headquarters of Vellore district in Tamil Nadu state of India as shown in Figure 1. Decades ago, it was just a municipality or a town with a population of only few lakhs but after it became a municipal corporation in 2008, the town has experienced massive urbanization and now it is one of the major cities in the state of Tamil Nadu. The major reasons for its tremendous urbanization are it has been selected for smart city mission by Govt. of India in 2015 and also it is one of the famous hubs for education in the country as Christian Medical College and Hospital (CMCH) and VIT University are located in Vellore. Also, the Vellore fort and Golden temple attracts huge number of tourists every year. In order to cater the housing requirements and commercial needs of the city, many new residential areas and shopping malls have come up in recent years. The urbanization has resulted in reduction in agriculture land over the years but still one can see agriculture lands on the Northern side of Palar River which bifurcates the city at its middle and runs from west to east as seen from Figure 1. A total of 15 sites within Vellore corporation were taken and their locations are shown in Figure 1. Google earth (GE) images of those 15 locations are showed in Figure 2. The 15 locations were decided based on the criteria that the selected location must have considerable amount of green cover when compared to other landuse/landcovers. As seen from Figure 2, all 15 locations are having significant green cover in terms of trees, shrubs, crop land, etc. It can also be seen from Figure 2 that each location has presence of other landuse/landcovers too such as built-up, water body, open land, etc. surrounding the green cover, so that we can cross check the performance of the proposed threshold-based approach in extracting the green cover alone accurately when compared to other land covers.

2.1 Data Collection and Processing

The satellite images from Google earth (GE) are the input data for the present work and hence the Google Earth Pro version was downloaded and installed on the computer. The GE was then checked for the availability of satellite imagery pertaining to the study area based on three criteria: (1). The images should be of recent one (2). The images should be free from clouds because even a small cloud cover may affect the results (3). The images should be from the same sensor. Sometimes the satellite images covering the study area may be from different sensors and in that case, half of the image would appear in one tone, texture and the other half would have a different tone, texture. It is not advisable to consider such imagery as it would adversely affect the classification results. In order to download the images from GE, an open-source GIS software, called ‘Smart GIS’ was used [14]. There is a tool called ‘Connect Google Earth’ in Smart GIS which can be used to visualize and preview the Google earth image for the study area chosen before it can be downloaded. The required zoom level was kept and then the latest GE images of all the 15 locations were downloaded.

Figure 2 shows the downloaded GE images of various locations within Vellore city in India. It can be seen that there is no cloud cover in any of the images. Also, one can clearly see the area covered by green. The GE images obtained from Smart GIS will usually be in geographic coordinate system, i.e., in decimal degrees. Such images cannot be used for measurement of area covered by green as they are in spherical coordinate system.

Hence the downloaded GE images were converted to projected coordinate system, i.e., rectangular or cartesian coordinate system in meters using Universal Transverse Mercator (UTM) projection in ArcGIS software. Once the input GE images are ready, the threshold-based approach can then be used to extract only the green covered areas and the same has been explained below.

2.2 Threshold-based Approach to Assess the Urban Green Cover

The present study proposed a threshold-based approach to extract the green cover from GE images and its step-by-step procedure is shown in Figure 3. The first step is selection of study area and identifying suitable locations for UGS extraction. The present study has selected 15 locations in the city of Vellore, India. The locations were carefully selected in such a way that the green cover is predominant at each location covering different types of UGS such as trees, shrubs, etc. Once the test locations are identified, the next step is to select the suitable GE image for analysis based on the three criteria as explained before, namely, the image should not have cloud cover, should be of recent one and must be from the same sensor. The Elshayal smart GIS software was then used to download the required GE images for the study locations selected. The downloaded GE images were then opened in ArcGIS software for projecting it into UTM projection. Before applying the threshold-based method, it is essential to first decide the number of UGS classes required. Based on visual inspection and field survey, it was found that there are three types of UGS classes that are present in the study area, namely, trees, shrubs and crop land. Trees are predominant at all the 15 locations when compared to other two classes. Once the UGS classes are decided, the threshold-based method can be applied, by carefully examining the range of pixel values for each UGS classes in each of the GE image of various locations.

Suppose if a GE image of a particular location consists of trees, shrubs and crop land, pixel ranges, i.e., minimum and maximum pixel values of each of these categories in that GE image are noted down. Likewise for all the other locations also, the range of values were noted down. Once the range is known, reclassification tool was then applied in ArcGIS software to prepare the UGS map with required number of classes. It is a trial-and-error procedure, i.e., identifying the pixel range and applying reclassification is repeated until the desired accuracy is achieved. For accuracy assessment, the random sampling method was adopted to get the testing points in the range of 20 to 40 which were randomly distributed over the resultant classified map using ‘Create Accuracy Assessment Points’ tool in ArcGIS software. Their ground truth information was obtained from the original/background GE image. Before running the ‘Compute Confusion Matrix’ tool of ArcGIS, it was ensured that the points are well distributed throughout the study area. The ground truth was then compared with the classified output to prepare the Overall accuracy (OA) and Kappa coefficient. Once the accuracy assessment results are acceptable, the universality checking of the proposed approach was then performed. It means whether the proposed approach would work well or not in other cities was verified by taking sample locations from different cities of the world. For the present work, 2 locations from Bangalore and Hyderabad cities in India and 3 locations from Barcelona, Kuala Lumpur and New York cities were taken. The threshold-based approach as explained above was then applied and finally OA and Kappa were arrived to check the performance of the proposed approach in extracting UGS from GE images.

Figure 3: Methodology flowchart

Figure 4: UGS Maps prepared using Threshold-based approach for 15 study locations in Vellore, India

3. Results and Discussion

3.1 Results of UGS Extraction from GE

The UGS map prepared for all the 15 locations in Vellore, India is shown in Figure 4. The range of values used for reclassification is shown in Table 1. It can be seen from Table 1 that the range of values for each UGS class is not same and also the values are not same for the same class in different locations. For example, in location 3, the pixel range of shrubs is 71 to 110 but whereas same shrubs exhibit pixel values of less than 35 in location 14. The results of UGS maps in Figure 4 revealed that the threshold-based approach performs well as it classifies correctly the UGS and its types and very well differentiates from other landuse/landcover classes. The results of accuracy assessment as shown in Table 2 revealed that the overall accuracy (OA) was > 90% at 12 sites and only 3 out of 15 sites exhibits OA in the range of 75-86%. The reason for less OA at three sites is, at some places, the water body is covered with algae and hence they are misclassified as green cover due to its greenish appearance in GE image though they are actually the waterbodies.

Table 1: Range of pixel values for various UGS classes for the 15 study locations from Vellore, India