Landsat Thematic Mapper (TM) images acquired from the years 2001, 2007, and 2011 was used with Landsat Operational Land Imager (OLI) at a resolution of 30m for the years 2016 and 2017 in LULC classification. Table 1 represents the available Landsat imagery with a spectral band used in this study. All required satellite imagery for the study area was downloaded from the official site of USGS earth explorer. The city plan was obtained from the Sulaimaniyah municipality corporation as a shape file of study area.

In this study, the remote sensing software ENVI 5.3 was used for image processing. During the process, the imagery geometrically corrected through pre-processing calibration, which consist of atmospheric and topographic correction

[31] [32] . Image-processing techniques are statistical algorithms that change the visual appearance or geometric properties of the images, These corrections are required for selected multitemporal or multisensor Landsat images to improve visibility appearance and image quality and get more accurate results [9] [31] .

In Landsat Operational Land Imager (OLI) digital number values were converted to the surface reflectance with the fast line-of-sight atmospheric analysis of hypercubes (FLAASH) to remove atmospheric effects and create a surface reflectance image.

The pre-processed images were classified considering a supervised classification method. In this technique, the maximum likelihood was chosen which is based on Bayesian theory in estimating parameters of a probabilistic model [33] . Each pixel assigned for a class according to its probability. Mean vector and covariance metrics are the key component of maximum likelihood classification that can be retrieved from training data (signatures or region of interest ROI) [7] [14] . All data were used for this study were projected to the Universal Transverse Mercator (UTM) projection system with WGS 84 reference system. Figure 2 shows schematic workflow steps for land use land cover change LULC and accuracy assessment.

Landsat Thematic Mapper (TM) sensor was carried on Landsat 5 and images consist of six spectral bands with a spatial resolution of 30 meters for Bands 1 - 5 and 7 and one thermal band (Band 6). Landsat Operational Land Imager (OLI) and Thermal Infrared Sensor (TIRS) images consist of nine spectral bands with a spatial resolution of 30 meters for Bands 1 to 7 and Band 9 is useful for cirrus

cloud detection. The resolution for Band 8 (panchromatic) is 15 meters while thermal bands 10 and 11 resolutions are (100 * 30) meters [34] .

To create accuracy assessment 470 points selected in classified images with the congruent Landsat images 2001, 2007, 2011, 2016 and 2017 respectively, the classified images were compared with the reference images by means of the error matrix [19] [35] [27] [22] [36] .

Overall accuracy was calculated using Equation (1), while Kappa coefficient was calculated using equation (2) [18] [19] [20] .

O A = ( C / A ) ∗ 100 (1)

where;

OA: overall classification accuracy;

C: number of correct points;

A: total number of reference points.

K = N ∑ i = 1 r x i i − ∑ i = 1 r ( x i + ∗ x + 1 ) N 2 − ∑ i = 1 r ( x i + ∗ x + 1 ) (2)

where;

r: number of categories (Classes);

N: total number of point (observations pixels);

x_ii: number of points on row i and column i;

x_i+: marginal total of row i (from overall confusion/error matrix);

x_+i: marginal total of column (from overall confusion/error matrix).

Cohen’s Kappa coefficient (K) is a strong statistical method used to measure the agreement between predicted and observed categorizations of a dataset. This statistic method is useful, especially in predictive accuracy of classification [36] . Random sample points were used in post-classification for accuracy assessment, 470 random sample points are located by Arc-map software then all points labelled under their category type in the classification. The reference points were compared with the classification results at the same locations. The output point shape file or feature class that contains the random 470 points was computed in the confusion matrix to ensure that each point had a valid class asset values for the classified and ground-truth point [14] [20] .

The producer accuracy and user accuracy for each class calculated from the matrix as well as an overall kappa index of agreement. When Kappa coefficient is equal to (1.00) it means perfect agreement and if the value close to (0.00) it means poor agreement. Table 2 represents the rating criteria of Kappa statistics [33] [37] . User accuracy represents the probability that a pixel classified into a given category actually which represents that category on the ground [38] .

Four land cover classes identified from maximum likelihood classification. These classes are forest, Settlement area, soil, vegetation and as shown in Figure 3.

Classification results gave the LULC raster image of the selected area, it is possible to calculate the area for each class separately according to pixel geometry and statistical distribution of pixels through the study area. The change detection in different LULC categories in the Sulaimaniyah Governorate from 2001 to 2017 as illustrated in Table 3.

The accuracy report from the matrix of all classified images is tabulated in Table 4. An overall accuracy and kappa coefficient was obtained for the years 2001, 2001, 2011, 2016 and 2017 respectively. Various measures of accuracy assessment such as producer accuracy, user accuracy, overall accuracy and Kappa coefficient were carried out and are represented in Table 2. Kappa coefficients generally rated as substantial agreement for all classified.

Producers Accuracy (Omission Errors) The results from dividing the number of correctly classified pixels in each category on the major diagonal matrix by the number of reference pixels “known” to be of that category (the column total), This value represents how well reference pixels of the ground cover type are classified, While users Accuracy (Commission Error) computed by dividing the number of correctly classified pixels in each category by the total number of pixels that were classified in that category (the row total), This value represents the probability that a pixel classified into a given category actually represents that category on the ground.

While Kappa reflects the difference between actual agreement and the agreement expected by chance Kappa results means there is 80% better agreement than by chance alone, Different values of kappa and overall accuracy from 2001 to 2017 return to the choice of reference data, interaction between sensor and desired classification scheme and Error matrix is foundation of accuracy assessment.

PA: Producer Accuracy, UA: User Accuracy; OA: Overall Accuracy, K: Kappa coefficient.

During the 16 years period (2001 to 2017), settlement area increased by around 9%, which represent 51.80 km². On the other hand, bare soil decreased by 44.77% in area. Forest recorded minimum change according to deforestation in land use in the study area, while vegetation increased by over 37% in the area as represented in Figure 4.

Increased vegetation area refers to two factors which are settlement growth at the same duration and the difference in Landsat acquisition date, although all available images were selected from the same season while there is a difference in duration for first and last Landsat Image by around one month duration which is sufficient for vegetation growth.

Maximum likelihood classification result gave the raster image, Settlement area calculated according to the pixel geometry from Table 1 and statistical pixels distribution in attribute table through the study area.

Change in settlement area between the years 2001 to 2017 increased from 3.87% area in 2001 to rich 12.87% area in 2017 as shown in Figure 5.

Two rapid expansions in settlement area noted during evaluation of classified images between 2001 to 2017 and 2011 to 2016 respectively which are mainly caused by economic prosperity and population growth.

Change in settlement area returns to modification of land cover due to human processes while the change in land cover is the natural occurrence on the earth surface like Forest decreased in area from 2001 to 2017 according to Sulaimaniyah government plan to increase green area around the city but the consequences forest fire reduced the rate of forest development, as shown in Land cover variation tendency in Figure 6.

Bare soil decreased in area from 2001 to 2017 according to modification of land cover and Settlement expansion.

Vegetation and grassland development in the area after 2001 refer to economic prosperity and Human needs through population growth.