An Automated Model to Evaluate Landscape Patches with Analysis of the Neighborhood Relations ^†

Abstract

The landscape should be analyzed in segments to understand its texture, structure, function, and changes. These segments can be used to evaluate landscape structure and for function analysis. In this context, the most important segments which form the landscape are landscape patches. Analysis and understanding of the landscape structure and ecological progress needs measurement of the landscape patches and evaluation. Therefore, the neighborhood ratio between the patches should be known. In this study, we propose an automated method, which is based on Python language, to compute this ratio with consideration of neighborhood degrees between the patches. The test site was Mugla-Koycegiz, a town in Turkey, where there is a huge population of Sweetgum (Liquidambar orientalis) trees, and the town is important for shoreline tourism. Urban area, water surface, agricultural areas, marsh, and forest classes were defined. Sentinel 2A multispectral satellite image was used and the Random Forest classification method applied. The derived patches were produced from the classification, and then converted to the vector form. All vector boundaries were converted to point features with 10 m intervals. The ratio of the number of points neighboring the specific class to all points along the boundary was computed automatically with developed script. Three different patches were analyzed, and the results are reported.

1. Introduction

Landscape should be analyzed in segments to understand its texture, structure, function, and changes. These segments are used to make evaluations and comments for landscape structure and function analysis [1,2]. In this context, the most important segments which form the landscape are landscape patches. Analysis and understanding of the landscape structure and ecological progress requires measurement of the landscape patches [3,4,5,6] and the derived data should be evaluated in the context of landscape ecology. For evaluation of landscape, several metrics are used [1,4,5,7].

Metrics provide any information regarding the components of the mosaic, the distribution of the components in the mosaic and their spatial status, the proportional state between each landscape type or the shape of the landscape elements, and support to interpret the landscape quality [1,4,8]. Landscape metrics can be classified according to patch, class, and landscape scale. Some of these metrics determine composition and some do configuration. Ecological processes are strongly connected to composition and configuration of landscapes independently and interactively [5,9].

Therefore, the metrics used in the evaluation of landscapes are generally classified and measured as area metric, edge metrics, shape metrics, core metrics, contrast metrics, aggregation metrics, subdivision metrics, isolation metrics. and diversity metrics [5]. This study focuses on the edge measurements. The Edge Density metric and the mean patch edge are an important indicator of habitat quality [1]. The patch edges constitute the neighboring areas of the vast zones defined as ecoton, where the areas exist which have the most intense interrelationships between different living things [1].

In one region, an increase in the edge density of a particular habitat area indicates that the edge effect is increased [10,11]. As the edge density increases, this is disadvantageous for the inner species in that habitat patch [12]. Low edge density is interpreted as high habitat quality [8]. The edge length and edge density metrics show an interpretation of the quality of the relevant landscape patch, but do not indicate the direction of other land uses adjacent to the landscape patch. That is, it does not measure how positive or negative the agricultural area or settlement adjacent to a landscape patch affects the landscape. In this context, the neighborhood ratio between the patches is measured in other land-uses adjacent to landscape patches.

This study proposes an automated algorithm that can be used to show what other land uses adjacent to a landscape patch affect the percentage of the landscape patch along the boundaries. In the context of the study, the relationship between the proposed model and landscape patches can be evaluated and the effects of environmental land uses on the relevant patch can be interpreted.

2. Materials and Methods

The study area is the town center of Koycegiz, Mugla province of Turkey. The reason for selecting as the test site was the hosting of endemic sweetgum (Liquidambar orientalis) tree populations while being under pressure of environmental land-uses. The test area has the center coordinates of area 36° 58′17′′ N and 28 ° 41′20′′ E, located in the south-west of Turkey (Figure 1).

In the study, three sweetgum population areas were selected from the current land uses, including settlements, agriculture, water surface, swamp, and stream. Patch 1 has an area of 0, 86 km², Patch 2 1, 56 km^2, and Patch 3 0, 47 km². The land uses adjacent to the three landscape patches are of different size and different quality. In the study, open access Sentinel 2A satellite imagery with 10 m spatial resolution was used as an input data. Random Forest (RF) and Maximum Likelihood Classification (MLC) methods were used for the classification of satellite images, and the highest accuracy value was determined with the RF method by using the (Confusion Matrix) algorithm at the end of the classification. The classified image obtained by this classification technique was used for each patch.

The study has three steps; database generation, classification, and neighborhood calculation (Figure 2).

In the first stage, zoning plans, the environmental plan, and satellite images of the study area were obtained and transferred to GIS and a database was created. The zoning plan and the environmental layout plan were obtained from the related public institutions and the satellite images were obtained through open access while the boundaries were determined by taking into consideration the sweetgum of the sweetgum forests. Within this scope, three landscape patches consisting of sweetgum populations in the study area were determined and the study was focused on these patches.

In the second stage, two different classification techniques were used to perform the classification images on the satellite images, then the classification accuracy was calculated by using the Confusion Matrix algorithm and the raster image obtained by this method was transferred to the third stage after the RF method gave higher accuracy.

In the third stage, the derived classes from the raster were converted to the vector line data, and the lines were converted to the point features for every 10 meters. The closest neighbors of each point were counted in a table, and then for each boundary, the percentage of neighborhood was calculated.

3. Results

The study area is composed of six land uses, namely agriculture, settlement, sweetgum forests, swamp, Lake Ecosystem, and stream. All classes have different sized neighborhood relations. In order to determine the neighborhood relations of these classes, classification was applied using the RF algorithm. The classification result is shown in Figure 3. Each class is shown in a different color.

As a result of classification, neighboring edges were converted from raster to vector format. Then, the boundaries were converted to point features for every 10 meters (Figure 4.).

As seen above, the middle parts of the study area and the east are used as settlement areas. These settlements are adjacent to Patches 2 and 3. This neighborhood is an indication of the fact that the sweetgum forest patches are under pressure. The patch number 1 located in the west is adjacent to agriculture and the river. It can be said that the patch is under pressure in terms of its potential to be used as an agricultural area. The neighborhood ratios of the selected patches (1,2,3) can be found in

Table 1. Neighborhood ratios of the three patches (3, 2, 1).

Patch Nr.	Neighbor	NrPCN	TNrP	Ratio (%)
3	Marshland	178	4610	3,86
	Agriculture	4154	4610	90,11
	Settlement	278	4610	6,03
2	Marshland	980	8911	11,00
	Lake	937	8911	10,52
	Settlement	2359	8911	26,47
	Agriculture	4635	8911	52,01
1	Marshland	603	7376	8,18
	Stream	722	7376	9,79
	Agriculture	6051	7376	82,03

.

NrPCN is the total number of boundary points which are neighbors with the neighbor, and TNrP is the total number of points along the boundary of the patch. The sum of the neighborhood ratios for each patch is 100 as expected.

Findings from the study show that the biggest neighbors of the selected landscape patches are agricultural areas. It can be understood that sweetgum patches are adversely affected by these agricultural areas because it is known that especially in the agricultural areas of the region, one-year crops are grown, and pesticides are used to obtain high crop yields. Sweetgum patches adjacent to swamp areas can be said to be affected positively by eliminating water demand. Similarly, the sweetgum patches adjacent to the stream and lake ecosystem are thought to be positively affected. The settlement areas are expected to affect the sweetgum patches negatively as a result of intensive human pressure and the creation of new settlements. The ratios of these positive and negative effects within the context of neighborhood relations are shown in Table 1.

4. Conclusions

The data and theory of landscape patch quality generally support the decrease of the probability of local extinction as the patch area increases [13]. Furthermore, these data show that the reduction of the patch area and the increase in the number of parts reduce the patch quality [14]. Only the evaluation of the patch area and isolation were criticized for being too restrictive. Scientists interested in ecology have used many measures of landscape structure in order to determine habitat loss and patch quality, but these measures are not sufficiently good [15]. In this context, in the evaluation of patch quality, scientists emphasize the need to assess environmental variables and current land uses in addition to patch areas [16,17,18]. This study proposes a method based on neighborhood relations for the evaluation of landscape patch quality. An automatic model was developed to evaluate the positive or negative impact ratio of other land uses adjacent to the landscape patch. The developed model transforms the neighborhood relations of landscape patches into numerical values quickly and practically by using machine learning algorithms. Using these numerical values, the rate at which a landscape patch is affected by its neighbors can be understood and interpreted. Hence, this study proposes an automated method to assist ecology scientists in assessing landscape patch quality.