Accuracy Assessment of Land Use/Land Cover Indices for Al-Rusafa in Baghdad Governorate by Remote Sensing Technology and GIS

The management and planning of natural and artificial resources depend on accurately monitoring land cover changes. Land cover change mapping and monitoring used to require expensive field surveys. Remote sensing is cheaper and more practical for mapping land use and cover changes. The Tigris River divides the Iraqi capital, Baghdad, into two parts: Karkh and Rusafa. Al-Rusafa was selected as a study area for current research, which has had rapid population and urban growth in recent decades. The current research applies the support vector machine technique to supervised LU/LC maps’ classification into barren regions, water bodies, vegetation cover and built-up regions. Spectral indicators were calculated: Enhanced Vegetation Index, Modified Normalized Difference Water Index, Normalized Built-Up Area Index, Dry Bareness Index in addition to calculating the accuracy assessment and Kappa coefficient. Using the Landsat 9 satellite image, ArcGIS 10.8 and Envi5.3 software were used to analyze and evaluate the results and field points observed by GPS devices. The results showed that the SVM classification algorithm accurately revealed the categories of LU/LC, where the classification accuracy reached 95%, and that the arid lands covered most of the study area 848.864 km2 and water bodies 76.747 km2, the vegetation and the built-up regions 466.459 km2 and 439.077 km2, respectively. The spectral indices showed slightly different areas of barren lands (DBSI 752.589 km2, 93% accuracy), vegetation (EVI 423.651 km2, 96% accuracy), and water bodies (MNDWI 73.187 km2, 98% accuracy) and built-up areas (NBAI 501,731 km2, 90%accuracy). The Support Vector Machine method outperforms other classification methods, and the spectral indicators employed in this work are useful and dependable for extracting each LU/LC category. In conclusion, Landsat 9 satellite data can reliably and swiftly detect ground cover.


Introduction
Land use and land cover mapping are crucial, incorporating diverse factors to meet specific requirements.Terms for land usage and land cover are frequently combined.Consequently, it is crucial to define them precisely and cover, representing the Earth's surface's physical and biological cover, including urban areas, agricultural regions, marshes, (semi-)natural areas, and water bodies [1].However, land use indicates a territory's current and upcoming planned human activities, residential, industrial, agricultural, commercial, forest, and recreational [2].It expresses human activity developed for social, political, economic, and cultural.Using observations of Earth as a foundation in various disciplines.The climate changes and the urban extension have caused a significant loss in agricultural lands, Water bodies shrinkage, air quality decreases, and land surface temperature increases.Rapid changes in LU/LC include urban sprawl, degradation of land, and the transformation of agricultural land into a barren landscape or urban regions, particularly in developing nations.This type of shift impacts local and regional environments, ultimately impacting global climate and land cover changes [3].Therefore, evaluating land use and cover changes is critical for creating a sustainable land use strategy that has a vital role in the global ecosystem [4].
Data from satellite sensors have become an essential tool for researchers studying LU/LC change.Reduced survey times, latest map availability, lower costs, improved efficiency, pixel-based digital picture classification, and spectral details are only a few of the many vital contributions of remote sensing technology.They offer timely and thorough coverage of any particular topic and are also a great source of information,therefore allowing for the successful assessment of natural resources and the monitoring of changes in land use or cover [4].Supervised classification learning systems select training data for each manually defined class type by the interpreter.This enables them to categorise pixels according to their spectral characteristics (such as reflectance values or digital numbers) [5].
Previous research used a variety of programs, methods, and strategies to investigate the lu/lc changes in many cities worldwide.Pal and Mather [6] Two multi-class SVM experiments used ML and ANN classification.Multispectral (Landsat-7 ETM+) data reveal land cover in eastern England and central Spain.SVM outperforms ML and ANN classifiers for high-dimensional data and limited training datasets.Deilmai et al. [7] MLC and SVM classifiers evaluated multispectral data.Landsat Thematic Mapper in Johor, Malaysia, recognized forests, oil palm, urban areas, water, and rubber.SVM outclassed MLC, and SVMs reliably classify land cover.A comparison between maximum likelihood and support vector machines was made to classify land use in Hilla City, Babylon, Iraq [8].SVM accuracy is 94.48% with a 0.90 kappa coefficient.These values estimate and extract land cover/use better than Maximum Likelihood.This land use map classifier is the most accurate and consistent in the research area.[9] From 1988 to 2016, the SVM and ML classification algorithms were tested to classify land use in the Kathmandu Valley, Nepal.SVM outperforms MLC in classification accuracy.Calculation of NDVI, NDBI, and NDWI was done from 2001 and 2006 using Landsat ETM+ images [10].These indices were used to monitor Kufa and Abbasiyah.Vegetation, built-up areas, and aquatic bodies were classified as unsupervised, while Analyzing LU/LC changes was the primary goal.The techniques proved effective and reliable for LU/LC change detection.The study of the urban sprawl's temporal and spatial dimensions and its dynamic changes in Karachi, Pakistan [11].The study utilised Landsat imagery from 1991, 2000, and 2013 to illustrate the capabilities of remote sensing data and the effectiveness of the suggested methodologies in geographic research.[12] This study analysed Landsat-TM and Landsat 8 images from 1984 and 2015 to determine vegetation cover spectral indices in the Wadi Alarab basin.NDVI, RVI, SAVI, and EVI2 were extracted to find the best index for monitoring and detecting plant cover changes in the research area.EVI2 had the strongest statistical correlation when used to map vegetation distribution.R^2 (0.96) for four indices in use.[13] Analyzed the vegetation cover in Campo Belo do Sul in Brazil, by utilizing five vegetation indicators which Calculated NDVI, EVI, SAVI, LAl, and NDWI.No vegetation index best represented all study classes.Nonetheless, the majority of NDVI, SAVI, and EVI had suitable adjustments in most of them.
A supervised Maximum Likelihood Classification (MLC) method on Landsat 8 images used to identify land cover in Baghdad from 2015 to 2020 [14].The land cover classes changed during the investigation.Urban, vegetation, and water areas increased by 7.5%, 9.5%, and 1.5%, respectively, while bare soil dropped by 18.5%.This study aims to use Landsat 9 images to create LU/LC maps by supervised classification SVM method and divide them into four categories: Bare soil land, water bodies, vegetation cover and builtup area, and barren land.Additionally, calculating different spectral indicators, EVI, MNDWI, DBSI, NBAI, to evaluate the accuracy assessment of each index and to compare with the accuracy of classification.

Study area
The study area is represented by Al-Rusafa in Baghdad Governorate (the capital of Iraq).Baghdad Governorate is situated in Central Iraq between (32° 48' 00'' -33° 45' N) latitudes and (43° 50' 00''-45° 00' 00'' E) longitudes Baghdad lies on both banks of the Tigris River, which divides it into Karkh (the western part) and Rusafa (the eastern part).The area has the most populous governorate in Iraq, with approximately 7 million inhabitants in 2019.
The research area is represented by the administrative boundaries of the municipalities of Baghdad for the side, Rusafa, as shown in Figure 1.Where several cities are on the Rusafa side to the east of the Tigris River (Municipalities of Al-Rusafa, Al-Shaab, Adhamiya, Al-Mada'in, Al-Husseiniya, New Baghdad and Al-Sader), these municipalities variate in size, population, and topography, where the area of Al-Rusafa side covers approximately (362) km2 and Tigris River covers about (11) km2 within the municipalities of Baghdad [15].

Data Used
One Landsat satellite image was downloaded from the website USGS (United States Geological Survey) website, with a 30-meter resolution Landsat-9 OLI covering the study area for 2023-07-14.On the other hand, several software programs were used in this study, such as ArcGIS 10.8 and Envi 5.3.This image used the World Geodetic System 1984 (WGS84) and the Universal Transverse Mercator projection (UTM) system, zone 38, path 168, row 037, and a cloud cover of 0%.Table 1.includes the bands in the image used in this study.

Image Pre-processing
After the images were downloaded, radiometric calibration and atmospheric correction measures were taken to enhance the image data by converting the radiation from the sensor into surface reflectance values in ENVI 5.3 software was the model Fast Line-of-Sight Atmospheric Analysis of Spectral Hypercubes (FLAASH) [16].

Supervised Classification
The choice of a classification technique is determined by factors such as the nature of the resources, the nature of the application, and the data availability.[17].Therefore, This study utilises the pixelbased supervised classification technique by applying a statistical learning-based classifier known as the non-parametric rule of support vector machines.The choice of this method is grounded on extensive research studies that have found that support vector machine (SVM) achieves high accuracy compared to other classification methods.
This method sorts the image into the training courses using a type of SVM classification based on training samples.In supervised classification, the user must select a region of interest that functions as a classifier and categorizes all image pixels [18].

Spectral Indices
A spectral index is a mathematical equation used to measure an image's different spectral bands per pixel.It is calculated as a ratio of broadband spectral bands or as the normalized difference between two bands, and spectral indices are designed to emphasise pixels in a satellite images that represent the proportion or absence of a specific land-cover category.
Before index calculations, the initial digital number (DN) data are transformed to obtain the top of atmosphere planetary reflectance (ρλ).This conversion was accomplished by utilizing offset and gain values extracted from the Metadata file associated with the images [19], by using the following formula: M ρ is the band is multiplicatively rescalig, A ρ is the band additive rescaling, Q cal Pixel values that constitute the quantized and calibrated standard product (=DN) θ The local sun elevation angle is measured in degrees.
In the following subsections, the reflectance characteristic of some common land cover types is discussed:

Enhanced Vegetation Index (EVI)
The MODIS Land Discipline Group has suggested the Enhanced Vegetation Index (EVI).Both NDVI and EVI are vegetation indexes that have a worldwide coverage.Their purpose is to offer reliable geographical and temporal data about vegetation on a global scale [20].
The modified NDVI method utilises the blue band to correct the red band for the effects of atmospheric aerosol scattering.In regions with dense vegetation, the Enhanced Vegetation Index (EVI) is more sensitive.Additionally, it partially accounts for atmospheric factors and background noise originating from the canopy [21].
Where, NIR = near infrared band5, RED = band4, BLUE = band2.is Values of this index are in the range -1 to 1.

Modified Normalized Difference Water Index (MNDWI)
MNDWI is more suitable for water improvement in densely populated areas due to its ability to effectively reduce and eliminate the noise associated with built-up regions.
To enhance open water features, modify the NDWI such that it uses a SWIR1 band rather than an NIR band.It can correctly and quickly tell the difference between water and non-water features [22].
Where Green = band3, SWIR = band 6.The values of this index are in the range -1 to +1.

Dry Bareness Index (DBSI)
Many indicators could not accurately detect the bare soil category, as extracting the bare land category is difficult due to the similarity in spectral signatures between the built-up regions and the barren land the following equation an index was used, which was developed by [24] The DBSI readings can vary from -2 to +2.Greater values indicate more bare soil.

Results of Classification
Figure 2. illustrates the map of the supervised classification by the SVM method on 14 July 2023, as divided the study area into four classes: barren lands, agriculture land, built-up regions, and water bodies.
The map of LU/LC shows that most of the area is covered with bare soil and agricultural lands devoid of vegetation.

Result of Indices
Using LULC indices the original bands of satellite photos improved supervised classification accuracy, kappa, and overall accuracy.
Isolation (thresholding) was performed using ArcGIS software to classify LU/LC Classes, which means the maps resulting from calculating the indicators have been reclassified.
Figure 3. shows that the vegetation appears green, while the non-vegetative cover which include (built-up regions, barren land, and water bodies) is grey.
The vegetation cover was separated from other classifications.The EVI value between -1 and +1, where the vegetation value is higher than (0.2).   Regions with and without bare soil can be mapped using a suitable thershold for the class of bare soil.Areas with a DBSI value of 0.18 or higher were designated bare soil, whereas lower values were classified as other classes, as shown in Figure 5.

Accuracy Assessment
The assessment of post-classification accuracy is considered the most crucial component in validating the LU/LC maps produced by the models.The ability of a classifier to classify correctly a given set of samples is called accuracy.The data used to assess the approach's performance must be different from the data used to train the classifier The accuracy assessment process used a confused matrix on the land cover map listed in Table 2.

CONCLUSIONS
The main findings of this study summarized as follows.: • Extracting a land use map for the Rusafa side in Baghdad Utilising remote sensing technology and GIS, Landsat 9 images were used with a spatial resolution of 30 metres for the 2023 summer season.• Analysis results of Landsat-9 images through SVM method supervised classification showed the accuracy of the information for mapping LU/LC.Where the optimum, accurate image at overall classification accuracy is 95%, and the overall kappa is 0.93.• Some factors, including the choice of the best classification method, the suitable band, and the optimal threshold, may impact the effectiveness and accuracy of the categorization.
• The spectral indices' accuracy ranged ) 93-98)%.This means it is accurate and reliable in extracting LU/LC maps.

Figure 2 .
Figure 2. Map of LU/LC Classes Areas of Al-Rusafa Side by Using Supervised Classification.
The Kappa coefficient shows how well the validation data and the projected values match, The overall accuracy of a procedure can be determined by dividing the number of correctly evaluated samples by the total number of testing samples [ 26 ] .This stage was all about gathering testing sites for study areas.These sites are significant in determining the accuracy assessment for each classification algorithm and checking classification validation and kappa coefficient.40 points from field observation by GPS were used to calculate the accuracy of the classified image, and 160 Random points by stratified random sampling utilizing Google Earth Pro for inaccessible areas.

Table 2 .
Area of LU/LC Classes for Al-Rusafa Side and Accuracy of Classification.Kappa coefficient = 0.933In addition, calculated the accuracy assessment and kappa coefficient for each indicator.The results are in Table3.

Table 3 .
Accuracy and Area of Each Index.