Assessment and simulation of land use and land cover change impacts on the land surface temperature of Chaoyang District in Beijing, China

Study area

Chaoyang District is located in the middle of metropolitan Beijing (39°55′ N −116°26′E, ∼2 to ∼116 m) (Fig. 2). This district has an area of 478 km² and a population of 3.5 million people between the 2nd and 5th ring roads. This district is the largest and most densely populated urban area of Beijing, with a density of 7,530 people/km². The mean annual precipitation is approximately 644 mm, and the mean annual temperature is approximately 13.5 °C. Spring and autumn are short, and the average summer and winter period last approximately 3 and 5 months, respectively. The municipality, as well as the Chinese National Government, spends almost 0.5 million USD per day on the development of this district. The district has jurisdiction over 22 sub-district offices and 20 area offices.

Data collection

The spatial images of LTM (Landsat 4–5) and OLI (Landsat 8) with a 30 m resolution from 1990 to 2018 were obtained from the USGS Earth Explorer for evaluating changes in LULC and LST. The entire Landsat scene cloud cover for the years 1990, 1997, 2004, 2011 and 2017 was approximately 3%–20%, but it was less than 1% in the study area (Table 1).

Computation of land use and land cover change (LULCC)

Supervised classification is a method that is a “probability algorithmic program” applied within the ESRI ArcGIS 10.5 for land use/cover classification. Maximum likelihood classification (MLC) is a primer supervised classification scheme used in remote sensing tactics for data-image information. It has minimal computational time among alternative supervised tactics, in which the pixels that should not be unclassified become classified. Ground verification was performed in uncertain areas through supported bottom clothing where misclassified areas were corrected by imposition and rearranging the samples in ESRI ArcGIS version 10.5. The point of reference was meant to assess the mapping accuracy. The three basic land use/cover types identified within the study area were (1) vegetation, (2) developed/built-up area, and (3) waterbodies (Fig. 4).

Retrieval of land surface temperature (LST)

Radiometrically corrected Landsat images with the thermal infrared band (Band-6) were used to derive the land surface temperature (LST). The procedure, which involves the conversion of a digital number (DN) to an at-satellite brightness temperature, pursues correction for atmospheric absorption, re-emission and surface emissivity that has equally been utilized (Amanollahi et al., 2016; Li et al., 2016; Snyder et al., 1998; Sobrino & Jiménez-Muñoz, 2014; Sobrino, Jiménez-Muñoz & Paolini, 2004) to convert the spectral radiance to the top of atmosphere (TOA) brightness temperature beneath the hypothesis of uniform emissivity by these equations. (1)${\displaystyle T_k=\frac{K_2}{ln\left(\frac{K_1}{L_{\lambda }}+1\right)}}$ where T_k is the at-satellite brightness temperature in Kelvin (K), L_λ is the spectral radiance (W/m2*sr*µm), K₂ is the calibration constant (K), and K₁ is the calibration constant (W/m2*sr*µm).

We can calculate L_sat by using the following equation: (2)${\displaystyle L_{sat}=\frac{\left(L_{max}-L_{min}\right)}{QCA{L}_{max}-QCA{L}_{min}}\times \left(DN-QCA{L}_{min}\right)+L_{min}}$ where DN is the digital pixel number such as Band 6, QCAL_max = 255 is the maximum quantized calibrated pixel value corresponding to L_max, QCAL_min = 0 is the minimum quantized calibrated pixel value corresponding to L_min, L_max = 17.04 (mW/ cm²sr⋅ µm) is the spectral at-sensor radiance that is scaled to QCAL_max, and L_min = 0 (mW/ cm²sr⋅ µm) is the spectral at-sensor radiance that is scaled to QCAL_min.

From the calculation in Eq. (2), the radiance (T_k)) is converted to surface temperature in Kelvin (K), which is then converted to Celsius (^oC) by using the following equation: (3)${\displaystyle T_C=T_k-273.15}$ where T_C is the temperature in Celsius (°C), and T_k is the temperature in Kelvin (K).

The land surface temperature (LST) was computed (Fig. 5) by using the following equation (Li et al., 2016; Sobrino, Jiménez-Muñoz & Paolini, 2004): (4)${\displaystyle LST=\frac{T_B}{1+\left(\frac{\lambda \cdot T_B}{\rho }\right)\cdot ln\left(\epsilon \right)}}$

Markov chain model

This model is based on the creation of Markov stochastic process systems for the prediction of a status being changed to another (Muller & Middleton, 1994). The Markov chain model is usually used to simulate the changes, dimensions, and trends of land use cover changes. Additionally, the produced probability transition matrixes were used to forecast and determine the possible situations of future land-use change and urban growth patterns and to study the simulation trends of land surface temperature (LST) (Al-sharif & Pradhan, 2014). The prediction of future LULCC was performed by using a land-use change modeller (LCM) in Terrset (Clark Labs TerrSet 18.31), and land-use changes were analysed and the situation was projected to 2025 by CA-Markov. The LST can be calculated based on the conditional probability formula by using the following equations:

(5)${\displaystyle S\left(t+1\right)=P_{ij}\times S\left(t\right)}$ (6)${\displaystyle P_{ij}=\left(\begin{array}{c}\hfill P_{11}\hfill \\ \hfill P_{21}\hfill \\ \hfill P_{n1}\hfill \end{array}\begin{array}{c}\hfill P_{12}\hfill \\ \hfill P_{22}\hfill \\ \hfill P_{n2}\hfill \end{array}\begin{array}{c}\hfill P_{1n}\hfill \\ \hfill P_{2n}\hfill \\ \hfill P_{n3}\hfill \end{array}\right)}$ Moreover, (7)${\displaystyle \left(0\le P_{ij}<1and\sum _{j=1}^N{P}_{ij}=1,\left(i,j=1,2,\dots \dots .n\right)\right)}$ where S(t) is the state of the system at time t; $S\left(t+1\right)$ is the state of the system at time $\left(t+1\right)$; P_ij is the matrix of transition probability in a state.

Accuracy assessment of LULCC

The user’s accuracy of the supervised classification (LULCC maps) and the Kappa coefficient were determined by using terrset IDRISI. The overall classification accuracy was 94.26%, 92.96%, 91.86%, 89.56% and 91.89% for the years 1990, 1997, 2004, 2011 and 2018, respectively (Table 4). Although the overall Kappa coefficient was above 0.8399, it showed strong agreement (Cleve et al., 2008; Xiao et al., 2007).

Simulation in LULCC and LST

The cellular automata (CA) and stochastic transition matrix of Markov chain models produced the LULCC and LST for the projected period of 2025. Approximately 12% of the negative change in green cover was estimated, which was approximately 164.92 km², and the estimated temperature fluctuation will be 32%, with an approximately 10.74 °C rise in the vegetation area. The built-up area will expand by ∼6% (283.04 km²), adding an ∼35% (12.66 °C) rise in temperature for the projected period of 2025 (Fig. 11). Water bodies will decrease to 5.98 km^2, which is approximately −79%. The accuracy of the maps of projected land use cover change in 2025 was classified by the Kappa coefficient value, which was above 0.97.