Spatiotemporal Built-up Land Density Mapping Using Various Spectral Indices in Landsat-7 ETM+ and Landsat-8 OLI/TIRS (Case Study: Surakarta City)

Spectral indices variations support for rapid and accurate extracting information such as built-up density. However, the exact determination of spectral waves for built-up density extraction is lacking. This study explains and compares the capabilities of 5 variations of spectral indices in spatiotemporal built-up density mapping using Landsat-7 ETM+ and Landsat-8 OLI/TIRS in Surakarta City on 2002 and 2015. The spectral indices variations used are 3 mid-infrared (MIR) based indices such as the Normalized Difference Built-up Index (NDBI), Urban Index (UI) and Built-up and 2 visible based indices such as VrNIR-BI (visible red) and VgNIR-BI (visible green). Linear regression statistics between ground value samples from Google Earth image in 2002 and 2015 and spectral indices for determining built-up land density. Ground value used amounted to 27 samples for model and 7 samples for accuracy test. The classification of built-up density mapping is divided into 9 classes: unclassified, 0-12.5%, 12.5-25%, 25-37.5%, 37.5-50%, 50-62.5%, 62.5-75%, 75-87.5% and 87.5-100 %. Accuracy of built-up land density mapping in 2002 and 2015 using VrNIR-BI (81,823% and 73.235%), VgNIR-BI (78.934% and 69.028%), NDBI (34.870% and 74.365%), UI (43.273% and 64.398%) and Built-up (59.755% and 72.664%). Based all spectral indices, Surakarta City on 2000-2015 has increased of built-up land density. VgNIR-BI has better capabilities for built-up land density mapping on Landsat-7 ETM + and Landsat-8 OLI/TIRS.


Introduction
The current trend in present time is the acceleration of urbanization and the expansion of urban areas that make the earth's surface turn into an impermeable layer, which is built-up land [22]. Built-up land is an area that is filled and surrounded by building [19]. The growth of built-up land in urban areas has resulted in increased built-up land density [17]. Increased built-up land density, so it is necessary for the supervision of urban development [23]. Supervision on urban development, especially built-up land density can be done through remote sensing [2]. Remote sensing has advantages such as faster, accurate, and has direct access to the earth's surface [26]. Remote sensing can also be used spatiotemporal to observe changes in built-up land density in urban areas [7]. The study of urban areas in remote sensing can be used as a means of decision-making such as urban development planning, disaster management, and urban environmental management [9] [21]. in Central Java (the density of Central Java is only 1,038 people/km 2 ), so that the development of the built-up land is quickly developed. Surakarta City with an area of 44 km 2 , is bordered by Karanganyar and Boyolali regencies in the north, Karanganyar and Sukoharjo regencies in the east and west, and Sukoharjo regency in the south. According to population, Surakarta City is also the third largest city in the southern part of Java after Bandung and Malang [14]. The east side of the city is passed by the Bengawan Solo river.
In this research, the image used is Landsat-7 ETM+ and Landsat-8 OLI L1T (Standard Terrain Correction) image. This image can be found on path 119 and row 65 of the World Reference System (WRS). The Landsat-7 ETM+ image was acquired on May 13, 2002 and contains 8 bands, while Landsat-8 OLI is acquired on October 13, 2015 and contains 11 bands. The bands of Landsat-7 ETM+ image are stored as 8-bit digital numbers, while Landsat-8 OLI images are stored as 16-bit digital numbers [20].

Landsat-7 ETM+ and Landsat-8 OLI/TIRS data pre-processing
Satellite imagery data should be preprocessing first before used to find the indices value. The aim of preprocessing data to make pixel value in ideal condition so that it can be used for both visual and mathematical analysis [4]. Before performing image processing into spectral indices, Landsat-7 ETM+ and Landsat-8 OLI/TIRS images should be performed radiometric calibration and atmospheric correction. The raw image value calibrated pixel value or digital number (DN) of the multispectral sensor must be converted into Top of Atmosphere (TOA) reflectance.
The TOA spectral radiance of multispectral bands of Landsat-7 ETM+ and Landsat-8 OLI imageries calculated using the formula. In this study, we used the radiometric calibration module available in the Environment for Visualizing Images (ENVI Ver. 5.2, Exelis Visual Information Solutions, Boulder, CO, USA) software to derive the TOA spectral radiance. The pixel value is processed into a TOA spectral radiance, that is: Where is the spectral radiance at the sensor's aperture in watts/(meter 2 *srad*µm); the Greyscale and Brescale are respectively, the rescaled gain and rescaled bias contained in the level 1 product header or ancillary data record both expressed in watts/(meter 2 *srad*µm); the QCAL is the quantized calibrated pixel value in DN; the LMIN is the spectral radiance that is scale to QCALMAX; the QCALMIN is the Where Lλ is the TOA spectral radiance (Watts/(m 2 *srad*µm)), ML is the band-specific multiplication factor rescaling from metadata, AL is the band-specific additive rescaling factor from metadata, and Qcal is the value of standard pixel products quantized and calibrated (DN) [24]. From TOA spectral radiance, then converted to TOA reflected value.
Where = TOA reflectance, = angle of sun elevation, and = angle of sun zenith. TOA reflectance value converted to surface reflectance value.
The image processed into these spectral indices required atmospheric correction to the level of atsensor reflectance. Atmospheric correction of Landsat-8 OLI/TIRS and Landsat-7 ETM+ image using histogram adjustment. The histogram evaluated is a histogram of reflected values on the sensor in the form of fractional numbers [6]. The method used in histogram adjustment is dark subtraction [28].
Radiometric corrected satellite imagery is done geometric correction process to improve geometric position accuracy and reduce distortion of earth curvature effect [6]. The first thing to do is to determine the GCP (Ground Control Points) and then done the process of geometric correction. This control point is an object visible to the image as well as visible on the reference map used in geometric correction [11]. This control point can be an object visible to the image as well as seen on the reference map used in geometric correction, such as crossing between rivers and road or crossroads and some other objects clearly visible in the image and reference map [18]. The reference map in this study is a map of the Geospatial Information Agency with a projection of the WGS 1984 Transverse Mercator 49M with a UTM grid. The points taken as much as five points for each image with RMS less than 0.5.

Spectral indices
According to [15] UI can provide actual information related to urban conditions from satellite imagery. The spectral UI indice uses near infrared and middle infrared channels such as SWIR2. The UI spectral indice is used to describe the inverse relationship between the brightness levels of urban areas in near infrared and middle infrared [15]. NDBI has an advantage over a UI that has a unique spectral response from built-up land and other types of land cover [28]. UI and NDBI are the spectral indices developed and used to explore in observation the difference between the spectral response of built-up land in near infrared and middle infrared. The difference between NDBI and UI is NDBI using SWIR1, while UI uses SWIR2. Spectral indice based on visible band to see the condition of built-up land are VrNIR-BI and VgNIR-BI. The visible band spectral indice looks to use a combination of red and green channels with near infrared. The atnospherically corrected reflectance values and brightness temperatures are derived from the built-up indices.

Accuracy assessment
To measure the accuracy and capability of the five spectral indices (VrNIR-BI, VgNIR-BI, NDBI, UI and Built-up) for mapping the developed built-up land density using the standard error of estimate. Standard error of estimate is a measure of the number of regression model errors in predicting the value of Y [10]. Standard error of estimate has a mapping accuracy (percent) using some calculation of bottom, upper, minimum error (%) and maximum error (%) based on confidence level of 95%.

Spectral Indices
Appearance of built-up land on each Landsat-7 ETM+ and Landsat-8 OLI/TIRS indices is shown in figure 3. The bright color on the image shows the existence of the built-up land. Surakarta city has the appearance of higher built-up land compared with the surrounding area. This is seen from the bright color that dominates the Surakarta city compared with the surrounding area. Visually, Landsat-7 ETM+ has a higher sensitivity to the existence of built-up land compared to Landsat-8 OLI/TIRS. In Landsat-7 ETM+ it is clear that the difference between the built-up land that looks lighter and not built-up land that looks darker. Meanwhile  The statistical value of built-up land indice image presented in table 2 is able to explain the appearance of the built-up land based on the image pixel value. There are four components in the image statistics that can explain the appearance of objects in the image, namely the minimum value, maximum, mean, and standard deviation.   Mean is the average pixel value in the image obtained from the total pixel value divided by the number of pixels in the image. When the pixel value in the image has a relatively large value, the resulting value will be greater. The indice with the greatest mean is the UI in Landsat-7 ETM+ that is 52,899, while other indices of Landsat-7 ETM+ and Landsat-8 OLI/TIRS have very small and negative values. The mean value can also explain the description of built-up land in the field. The greater mean value shows that the area of built-up land recorded in the image is getting bigger.
The standard deviation (Stdev) contained in the image statistics shows the pixel value spreads in the image. The results showed that the standard deviation of all indices in Landsat-7 ETM+ and Landsat-8 OLI/TIRS, except that the UI indice in Landsat-7 ETM+ has a greater value than the mean or mean. It shows that the distribution of values within the image has a very wide spread, meaning that the pixel value in the image is very diverse and indicates that built-up land is not only concentrated in a single place, but spread in study area. VgNIR-BI has the greatest R 2 value compared to other indice which is 0.5863 in Landsat 8 OLI/TIRS. Similarly, the VrNIR-BI indice has the greatest value in Landsat-7 ETM+ that is 0.4869. This suggests that the VgNIR-BI and VrNIR-BI indices have the strongest correlation between image and field values. Therefore, the two indices are able to explain the existence of built-up land density.

Regression between ground value and spectral indices
The value of R 2 on the UI and Built-up indices has the lowest value compared to the values in the VrNIR-BI, VgNIR-BI, and NDBI indices. The UI indice has the lowest R 2 value in Landsat 8 OLI/TIRS of 0.0588 and the Built-up indice has the lowest R 2 value at Landsat 7 ETM+ of 0.113. Low regression value states that the existence of objects in the field can not be explained by image value well. Therefore, the UI indice is not good to explain the existence of built-up land density.  It is seen that the pattern of dispersion of increased built-up land density in Surakarta City leads to the north, west and south while the eastern direction does not increase. This is because a river bengawan solo in the eastern Surakarta City that inhibits the spread of increased built-up land density.  Source: Processing data, 2017 Table 3    Based on the overall statistical data in table 4, the VgNIR-BI spectral indice has the best ability for built-up land density mapping in Surakarta City on 2002 and 2015 using Landsat-7 ETM+ and Landsat-8 OLI/TIRS, while NDBI has the worst ability for Landsat-7 ETM+ and UI for Landsat-8 OLI/TIRS. VgNIR-BI has a maximum error, minimum error, maximum accuracy and minimum accuracy respectively of 52.424%, 18.176%, 81.823% and 47.545% using Landsat-7 ETM+ and 56.020%, 26.765%, 73.235% and 43.980% using Landsat -8 OLI/TIRS. The combined ability of green and red band in Landsat-7 ETM+ and Landsat-8 OLI/TIRS is the best combination of band for spatiotemporal built-up land density mapping in Surakarta City. as the Vr-NIR-BI indice on Landsat-8 OLI. The VgNIR-BI and VrNIR-BI indices on Landsat-7 ETM+ and Landsat-8 OLI imagery are able to represent a development of built-up land is clearly. This is evident from the value of the development of the area of built-up land. The built-up land indice applied to Landsat-7 ETM+ and Landsat-8 OLI is not fully able to represent well on built-up land. The regression graph and accuracy test were performed to find out the best indice in mapping the density of the built-up land in Surakarta City. The VgNIR-BI indice on Landsat-8 OLI and the VrNIR-BI indice on Landsat-7 ETM+ has the greatest R 2 value, so both indices are able to explain the existence of built-up land is great. Based on the results of the accuracy test, the value of VgNIR-BI on Landsat-7 ETM+ and NDBI on Landsat-8 OLI has a high accuracy value. Therefore, the VgNIR-BI has the best ability to built-up land density mapping in Surakarta City in 2002 and 2015 using Landsat-7 ETM+ and Landsat-8 OLI from visual and statistic perspective.