Anthropogenic Heat Flux Estimation Based on Luojia 1-01 New Nighttime Light Data: A Case Study of Jiangsu Province, China

: With the rapid process of urbanization, anthropogenic heat generated by human activities has become an important factor that drives the changes in urban climate and regional environmental quality. The nighttime light (NTL) data can aptly reﬂect the spatial distribution of social-economic activities and energy consumption, and quantitatively estimate the anthropogenic heat ﬂux (AHF) distribution. However, the commonly used DMSP / OLS and Suomi-NPP / VIIRS NTL data are restricted by their coarse spatial resolution and, therefore, cannot exhibit the spatial details of AHF at city scale. The 130 m high-resolution NTL data obtained by Luojia 1-01 satellite launched in June 2018 shows a promise to solve this problem. In this paper, the gridded AHF spatial estimation is achieved with a resolution of 130 m using Luojia 1-01 NTL data based on three indexes, NTL nor (Normalized Nighttime Light Data), HSI (Human Settlement Index), and VANUI (Vegetation Adjusted NTL Urban Index). We chose Jiangsu, a fast-developing province in China, as an example to determine the best AHF estimation model among the three indexes. The AHF of 96 county-level cities of the province was ﬁrst calculated using energy-consumption statistics data and then correlated with the corresponding data of three indexes. The results show that based on a 5-fold cross-validation approach, the VANUI power estimation model achieves the highest R 2 of 0.8444 along with the smallest RMSE of 4.8277 W · m − 2 and therefore has the highest accuracy among the three indexes. According to the VANUI power estimation model, the annual mean AHF of Jiangsu in 2018 was 2.91 W · m − 2 . Of the 96 cities, Suzhou has the highest annual mean AHF of 7.41 W · m − 2 , followed by Wuxi, Nanjing, Changzhou and Zhenjiang, with the annual mean of 3.80–5.97 W · m − 2 , while the ﬁgures of Suqian, Yancheng, Lianyungang, and Huaian, the cities in northern Jiangsu, are relatively low, ranging from 1.41 to 1.59 W · m − 2 . This study has shown that the AHF estimation model developed by Luojia 1-01 NTL data can achieve higher accuracy at city-scale and discriminate the spatial detail of AHF e ﬀ ectively.


Introduction
With the rapid development of urbanization, urban built-up areas have expanded dramatically. A number of natural landscapes have been replaced with various artificial surface, which causes sensible heat increases and latent heat decreases, resulting in an urban heat island (UHI) [1,2]. At the same time, the number of urban populations has continued to surge. According to the statistics from the performance and accuracy of different AHF estimation models are analyzed and compared, and the best model is determined based on a 5-fold cross-validation approach. Finally, the AHF mapping with a high resolution of 130 m is realized and the spatial distribution characteristics of AHF in urban areas are analyzed based on the AHF estimation result.

Study Area
The study area is Jiangsu Province, located in China's southeastern coastal region (30°45′N-35°20′N, 116°18′E-121°57′E). Jiangsu belongs to a subtropical and warm temperature climate transition zone with the average annual temperature around 13-16℃, the average annual rainfall between 150-400 mm. The regional gross domestic product (GDP) of Jiangsu increased from 4196.2 billion yuan in 2010 to 9259.5 billion yuan in 2018, with a growth rate of 121%, and the GDP per capita increased from 53,525 yuan to 115,168 yuan, ranking first among all the provinces in China. The total population increased from 78.7 million people to 80.5 million people. Moreover, the energy consumption also increased significantly, from 257.7 million tons of standard coal in 2010 to 314.3 million tons of standard coal in 2018 [28]. Jiangsu Province is divided into 13 prefecture-level cities (Nanjing, Wuxi, Xuzhou, Changzhou, Suzhou, Nantong, Liangyungang, Huaian, Yancheng, Yangzhou, Zhenjiang, Taizhou, and Suqian) and 96 county-level cities. The study area covers around 100,831 km 2 , some small coastal islands and reefs are not included in this study ( Figure 1).   [29]. It can be seen from Figure 2 that the northern and central of Jiangsu are predominated by cropland, and the south has a large area of continuous impervious surface. Statistics show that the area is dominated by cropland, making up 66.11% of the area, followed by impervious surface (16.81%), water (11.72%), and forest (4.12%).   [29]. It can be seen from Figure 2 that the northern and central of Jiangsu are predominated by cropland, and the south has a large area of continuous impervious surface. Statistics show that the area is dominated by cropland, making up 66.11% of the area, followed by impervious surface (16.81%), water (11.72%), and forest (4.12%). The development difference between north and south regions in Jiangsu is obvious. The cities in the south have faster economic growth and larger population inflow than those in the north. Therefore, this study can reveal the regional unbalanced development and the spatial distribution characteristics of AHF in Jiangsu, which is of great significance for the study of anthropogenic heat emissions.

Satellite Data
The NTL data used in this study is Luojia 1-01. We acquired all the GEC system geometry corrected NTL products covering Jiangsu in 2018, downloaded from the High-Resolution Earth Observation System of the Hubei Data and Application Center (http://www.hbeos.org.cn).
Since some AHF algorithms used in this article need the help of vegetation index, we further selected the Landsat 8 images in 2018. The product level of Landsat 8 images is Level 1T, downloaded from the United States Geological Survey (USGS) (https://earthexplorer.usgs.gov/). The acquisition time of Landsat 8 images had to be selected in summer from July to August as far as possible, because the images of the optimum growth of vegetation throughout the year in 2018 needed to be utilized in this study. However, there are many cloudy and rainy days in summer of subtropical area like Jiangsu, so a large number of Landsat 8 images are missing. The images from June to October were used as supplements, because Jiangsu is dominated by evergreen vegetation, and the growth trends of vegetation from June to October are relatively similar.

Remote Sensing Data Processing
Although the system geometry was corrected before being downloaded, Luojia 1-01 images still had slight geo-referencing errors by about 150-400 m, when referring to the Landsat 8 images. In order to ensure the accuracy of overlay analysis, a map-to-map geometric correction was performed. Taking Landsat 8 as the reference image, Luojia 1-01 was registered with Landsat 8.
Jiangsu is covered by multiple scene images of Luojia 1-01 images on different acquisition dates. In order to ensure the continuity of the spatial pixels, and eliminate the lighting noise existing in the NTL data, the overlapping areas were averaged for the brightness of light, and if the pixels in an image were 0, then this position was represented by 0 [30]. Then, the DN value image of NTL data of Jiangsu was obtained by mosaicking and clipping the processed images.
We detected a few outliers in the Luojia 1-01 NTL mosaicking image. The outliers are probably caused by the stable lights from fires of oils or gas wells located in those areas [31]. In order to correct The development difference between north and south regions in Jiangsu is obvious. The cities in the south have faster economic growth and larger population inflow than those in the north. Therefore, this study can reveal the regional unbalanced development and the spatial distribution characteristics of AHF in Jiangsu, which is of great significance for the study of anthropogenic heat emissions.

Satellite Data
The NTL data used in this study is Luojia 1-01. We acquired all the GEC system geometry corrected NTL products covering Jiangsu in 2018, downloaded from the High-Resolution Earth Observation System of the Hubei Data and Application Center (http://www.hbeos.org.cn).
Since some AHF algorithms used in this article need the help of vegetation index, we further selected the Landsat 8 images in 2018. The product level of Landsat 8 images is Level 1T, downloaded from the United States Geological Survey (USGS) (https://earthexplorer.usgs.gov/). The acquisition time of Landsat 8 images had to be selected in summer from July to August as far as possible, because the images of the optimum growth of vegetation throughout the year in 2018 needed to be utilized in this study. However, there are many cloudy and rainy days in summer of subtropical area like Jiangsu, so a large number of Landsat 8 images are missing. The images from June to October were used as supplements, because Jiangsu is dominated by evergreen vegetation, and the growth trends of vegetation from June to October are relatively similar.

Remote Sensing Data Processing
Although the system geometry was corrected before being downloaded, Luojia 1-01 images still had slight geo-referencing errors by about 150-400 m, when referring to the Landsat 8 images. In order to ensure the accuracy of overlay analysis, a map-to-map geometric correction was performed. Taking Landsat 8 as the reference image, Luojia 1-01 was registered with Landsat 8.
Jiangsu is covered by multiple scene images of Luojia 1-01 images on different acquisition dates. In order to ensure the continuity of the spatial pixels, and eliminate the lighting noise existing in the NTL data, the overlapping areas were averaged for the brightness of light, and if the pixels in an image were 0, then this position was represented by 0 [30]. Then, the DN value image of NTL data of Jiangsu was obtained by mosaicking and clipping the processed images.
We detected a few outliers in the Luojia 1-01 NTL mosaicking image. The outliers are probably caused by the stable lights from fires of oils or gas wells located in those areas [31]. In order to correct the outliers, the processing procedure proposed by Shi et al. [31] was performed. Since Nanjing, Wuxi, and Suzhou are the three most developed cities in Jiangsu, the pixel values of the other areas should not exceed those cities theoretically. According to statistics, the highest DN value in Nanjing, Wuxi, and Suzhou is 793,457, which was used as a threshold to correct the outliers. Each pixel whose DN value was larger than 793,457 in NTL data was assigned as a new value. The new value was the maximal DN value within the pixel's immediate eight neighbors.
The images of Landsat 8 OLI were radiometrically calibrated using the equations from the Landsat 8 Data Users Handbook [32]. This converted the DN of the raw images to at-satellite reflectance.
The Luojia 1-01 NTL and Landsat 8 images were stitched and projected to the Albers Equal Area Conic Projection, respectively. We resampled the spatial resolution of Luojia 1-01 to 30 m, consistent with Landsat 8 (Figure 3).
Remote Sens. 2020, 12, x FOR PEER REVIEW 5 of 17 the outliers, the processing procedure proposed by Shi et al. [31] was performed. Since Nanjing, Wuxi, and Suzhou are the three most developed cities in Jiangsu, the pixel values of the other areas should not exceed those cities theoretically. According to statistics, the highest DN value in Nanjing, Wuxi, and Suzhou is 793,457, which was used as a threshold to correct the outliers. Each pixel whose DN value was larger than 793,457 in NTL data was assigned as a new value. The new value was the maximal DN value within the pixel's immediate eight neighbors. The images of Landsat 8 OLI were radiometrically calibrated using the equations from the Landsat 8 Data Users Handbook [32]. This converted the DN of the raw images to at-satellite reflectance.
The Luojia 1-01 NTL and Landsat 8 images were stitched and projected to the Albers Equal Area Conic Projection, respectively. We resampled the spatial resolution of Luojia 1-01 to 30 m, consistent with Landsat 8 (Figure 3).

Calculation of the AHF Spatial Estimation Indexes
In this study, Normalized nighttime light data (NTLnor) [12], Human Settlement Index (HSI) [21], and Vegetation Adjusted NTL Urban Index (VANUI) [24] were employed for AHF spatial gridding estimation. In previous studies, these three indexes were calculated based on DMSP/OLS and Suomi-NPP/VIIRS NTL data. HSI and VANUI also need to integrate the brightness of NTL with MODIS vegetation index products. However, at present, there is little literature on construction of AHF estimation index based on Luojia 1-01 NTL data. Therefore, this paper attempted to transplant the three indexes to Luojia 1-01 data and combined with the Normalized Difference Vegetation Index (NDVI) [33] from Landsat 8 images for AHF estimation. The equations of NTLnor, VANUI and HSI were expressed as:

Calculation of the AHF Spatial Estimation Indexes
In this study, Normalized nighttime light data (NTL nor ) [12], Human Settlement Index (HSI) [21], and Vegetation Adjusted NTL Urban Index (VANUI) [24] were employed for AHF spatial gridding estimation. In previous studies, these three indexes were calculated based on DMSP/OLS and Suomi-NPP/VIIRS NTL data. HSI and VANUI also need to integrate the brightness of NTL with MODIS vegetation index products. However, at present, there is little literature on construction of AHF estimation index based on Luojia 1-01 NTL data. Therefore, this paper attempted to transplant the three indexes to Luojia 1-01 data and combined with the Normalized Difference Vegetation Index (NDVI) [33] from Landsat 8 images for AHF estimation. The equations of NTL nor , VANUI and HSI were expressed as: Remote Sens. 2020, 12, 3707 6 of 17 where NTL max and NTL min are the maximum and minimum values in NTL image, respectively; NDVI max is the maximum value in the multitemporal composite NDVI data in 2018 [21]. The NDVI max is expressed as: where NDVI 1 , NDVI 2 , . . . , NDVI n are the multitemporal NDVI images from Landsat 8 in 2018. Through visual interpretation of Google Earth satellite imagery, the pure vegetation pixels were extracted from NDVI max , with a threshold of 0.70. Each pixel whose value is larger than or equal to 0.70 in NDVI max was simply defined as pure vegetation. Furthermore, the pixels with the brightness of NTL greater than 0 were extracted from the pure vegetation. The NTL min was obtained by counting the average brightness value of these pixels in the NTL image [34][35][36].

Estimation of the AHF of Administrative Unit Based on Statistics Data
The energy-consumption inventory approach was employed to calculate the AHF of the cities in Jiangsu based on the socio-economic data and various types of energy-consumption data from the statistical yearbook [28]. After comprehensive analysis of the statistical indicators in Jiangsu Province statistical yearbook, the four parts of anthropogenic heat emissions sources, industry, transportation, buildings (including both commercial and residential buildings), and human metabolism were considered [8,12,22,23,[37][38][39][40]. The total AHF is the sum of the four parts. The equation is expressed as: The industry heat is mainly derived from various types of energy consumption, such as coal, oil, gas, electricity, etc. [12]. In addition, the actual energy consumption is converted into the standard coal. The total industry consumption of Jiangsu is calculated, then the consumption of each city is allotted based on the ratio of the secondary industry. This equation is expressed as: where E I is the energy consumption of per ton of standard coal (tce); C is the standard coal heat, which is 29,306 kJ·kg −1 according to the national energy conversion standard; A is the study area (km 2 ); and T is 1 year. The transportation heat is the fuel waste heat discharged from vehicle energy consumption. The equation is expressed as: where V is the sum of civil automobiles; D is the annual average driving distance per vehicle (km), which is assumed average 2.5 × 10 4 km per year; E is the combustion efficiency (L·km −1 ), which is 12.7 L per 100 km; ρ is combustion density (kg·L −1 ); NHC is the net heat combustion (kJ·g −1 ), which is 45 kJ·g −1 [22]; A and T are the same as in Equation (6). The buildings heat is the energy consumption from wholesale and retail trade, accommodation, catering industry, and living consumption, including commercial and residential buildings. The commercial buildings heat and residential buildings heat were allotted based on the ratio of the tertiary industry and the population in each city, respectively. The equation is expressed as: where E BC and E BR are the commercial buildings energy consumption (tce) and residential buildings energy consumption (tce), respectively; C, A, and T are the same as in Equation (6). The energy consumption from human metabolism is divided into two states, active (7:00~23:00) and sleeping (23:00~7:00), respectively, according to the previous studies [37,41]. The equation is expressed as: where s 1 , s 2 are the metabolic rate of active state (171 W per person) and sleeping state (70 W per person), respectively; t 1 , t 2 are the hours of active time and sleeping time, respectively; P is the population; A and T are the same as in Equation (6).

Construction and Verification of the AHF Spatial Estimation Models
As was already mentioned, there is a significant correlation between brightness of NTL and AHF [17], which could be used for constructing the AHF spatial estimation model by employing the three indexes, NTL nor , HSI and VANUI). According to Equations (5)-(9), the annual mean AHF based on statistics data (AHF sta ) of 96 county-level cities in Jiangsu Province can be calculated. Taking AHF sta as the dependent variable (y), and the indexes as independent variable (x), and then regression analysis was carried out to establish the gridded AHF spatial estimation models. By comparing the fitting degree (R 2 ) of regression models based on the different indexes, the AHF estimation models with higher R 2 were preliminary selected. Furthermore, we used a 5-fold cross-validation approach to evaluate the results of AHF estimation models. The 5-fold cross-validation approach can be described as follows: the dataset was split randomly into 5 subsets, 1 of which was selected as the validation subset and the remaining 4 subsets were used as the training subset. Finally, the root mean square error (RMSE) and R 2 between the estimated and statistics values were employed to assess the model's performance [42].

Determination and Verification of the Best AHF Spatial Estimation Models
In order to realize the gridded AHF mapping and take full advantages of the high spatial resolution of Luojia 1-01 NTL data, the regression analysis of three indexes (x) and AHF sta (y) were carried out by using linear, quadratic polynomial, exponential, logarithmic, and power functions, respectively. The results of AHF spatial estimation models based on different indexes and functions are shown in Table 1. It can be seen that all the correlations are statistically significant (p < 0.005). Overall, the R 2 of NTL nor and VANUI are higher than that of HSI. In term of functions, the power function has the highest R 2 in the three indexes. Among them, the R 2 of power function based on NTL nor and VANUI are greater than 0.83, which are much higher than 0.26 of HSI. Although the R 2 of VANUI power function model is 0.8444, which is slightly higher than 0.8370 of NTL nor , the difference between VANUI and NTL nor is not obvious. Figure 4 gives the 5-fold cross-validation results of AHF power function estimation models based on three indexes. It shows that among the three indexes, the strongest correlations between AHF sta and AHF estimation values obtained based on VANUI model of Jiangsu's 96 county-level cities, which yielded an R 2 of 0.8228 and RMSE of 4.8277 W·m −2 . Followed by the NTL nor model, which achieved a slightly low degree of agreement, with R 2 of 0.8119 and RMSE of 4.9782 W·m −2 . Meanwhile, the HSI model had an extremely low agreement, with the R 2 value being only 0.6274, and the RMSE was as high as 7.4805 W·m −2 . These results suggest that Luojia 1-01 NTL data can aptly estimate the AHF with the use of a proper index, such as VANUI or NTL nor . In this study, the estimation accuracy of VANUI is higher than that of NTL nor , so the VANUI power function model is selected as the best AHF spatial estimation model of Jiangsu in 2018. Although the R 2 of VANUI power function model is 0.8444, which is slightly higher than 0.8370 of NTLnor, the difference between VANUI and NTLnor is not obvious. Figure 4 gives the 5-fold crossvalidation results of AHF power function estimation models based on three indexes. It shows that among the three indexes, the strongest correlations between AHFsta and AHF estimation values obtained based on VANUI model of Jiangsu's 96 county-level cities, which yielded an R 2 of 0.8228 and RMSE of 4.8277 W·m −2 . Followed by the NTLnor model, which achieved a slightly low degree of agreement, with R 2 of 0.8119 and RMSE of 4.9782 W·m −2 . Meanwhile, the HSI model had an extremely low agreement, with the R 2 value being only 0.6274, and the RMSE was as high as 7.4805 W·m −2 . These results suggest that Luojia 1-01 NTL data can aptly estimate the AHF with the use of a proper index, such as VANUI or NTLnor. In this study, the estimation accuracy of VANUI is higher than that of NTLnor, so the VANUI power function model is selected as the best AHF spatial estimation model of Jiangsu in 2018.

Analysis the Gridded AHF Mapping Results
According to the best AHF spatial estimation model of VANUI power function (y = 696.2097

Analysis the Gridded AHF Mapping Results
According to the best AHF spatial estimation model of VANUI power function (y = 696.2097 x 0.9304 ), the gridded AHF mapping of Jiangsu in 2018 with a spatial resolution of 130 m (Figure 5a) and the distribution of annual mean AHF values in each county (Figure 5b) were completed. It can be seen that the distribution of AHF in Jiangsu shows a significant difference between the north and south. In northern Jiangsu, such as Liangyungang, Huaian, Xuzhou, Yancheng, and Suqian, the AHF values are relatively low. That is because there are large areas of continuous cropland and many scattered urban built-up areas ( Figure 2). Meanwhile, the AHF of cities in central Jiangsu, such as Yangzhou, Taizhou, and Nantong, have slightly higher AHF than the north. In contrast, in the south of Jiangsu, the scope of impervious surface in urban built-up areas has expanded dramatically and connected with each other (Figure 2), forming the high-value AHF areas in urban regions.
Remote Sens. 2020, 12, x FOR PEER REVIEW 9 of 17 the distribution of annual mean AHF values in each county (Figure 5b) were completed. It can be seen that the distribution of AHF in Jiangsu shows a significant difference between the north and south. In northern Jiangsu, such as Liangyungang, Huaian, Xuzhou, Yancheng, and Suqian, the AHF values are relatively low. That is because there are large areas of continuous cropland and many scattered urban built-up areas ( Figure 2). Meanwhile, the AHF of cities in central Jiangsu, such as Yangzhou, Taizhou, and Nantong, have slightly higher AHF than the north. In contrast, in the south of Jiangsu, the scope of impervious surface in urban built-up areas has expanded dramatically and connected with each other (Figure 2), forming the high-value AHF areas in urban regions. According to the statistics in Figure 5a, the annual mean AHF of Jiangsu in 2018 is 2.91 W·m −2 . Among the 13 prefecture-level cities, Suzhou contributes the highest AHF, which reaches 7.41 W·m -2 , followed by Wuxi, Nanjing, Changzhou, and Zhenjiang, with AHF range of 3.80-5.97 W·m −2 , while According to the statistics in Figure 5a, the annual mean AHF of Jiangsu in 2018 is 2.91 W·m −2 . Among the 13 prefecture-level cities, Suzhou contributes the highest AHF, which reaches 7.41 W·m −2 , followed by Wuxi, Nanjing, Changzhou, and Zhenjiang, with AHF range of 3.80-5.97 W·m −2 , while the figures of Suqian, Yancheng, Lianyungang, and Huaian, the cities in northern Jiangsu, are relatively low, ranging from 1.41 to 1.59 W·m −2 , only half of the whole province ( Figure 6).
Remote Sens. 2020, 12, x FOR PEER REVIEW 10 of 17 the figures of Suqian, Yancheng, Lianyungang, and Huaian, the cities in northern Jiangsu, are relatively low, ranging from 1.41 to 1.59 W·m −2 , only half of the whole province ( Figure 6). Statistics of the contribution rates of AHF grades are shown in Table 2. In Jiangsu, the contribution rate of the AHF grade of 0-5 W·m −2 is the highest, reaching 87.21%. These areas with low-value AHF are mainly croplands and water, as well as a few forests. And the contribution rate of AHF >5 W·m −2 is about 12.79%. Comparing the result between AHF (Figure 5a) and land cover (Figure 2), it can be seen that the spatial distribution of AHF >5 W·m −2 is consistent with an impervious surface. It indicates that the areas with frequent human activities can be well represented by the results of AHF.

Validation of AHF Estimation Results
It is difficult to validate AHF estimation results according to field measured data because the sensible or latent heat caused by human activities or background conditions cannot be completely distinguished [8,22,43]. Therefore, the results are usually verified by comparing them with previous similar studies [44,45]. Due to the inconsistency in the data sources, modeling methods, and research scales, there is often no absolute comparability between the study results, but it can be used as reference data to verify the rationality of research results as a whole [12]. Xie et al. [46], Chen et al. [18], and Wang et al. [12] employed an energy-consumption inventory approach based on statistics data to estimate the spatial distribution of AHF in China. Part of the estimation results of AHF in Jiangsu of the three existing studies are compared with those obtained in this study. The detailed comparison is shown in Table 3. The AHF estimation result in this study is basically consistent with the three existing studies results, especially in the order of magnitude. Statistics of the contribution rates of AHF grades are shown in Table 2. In Jiangsu, the contribution rate of the AHF grade of 0-5 W·m −2 is the highest, reaching 87.21%. These areas with low-value AHF are mainly croplands and water, as well as a few forests. And the contribution rate of AHF >5 W·m −2 is about 12.79%. Comparing the result between AHF (Figure 5a) and land cover (Figure 2), it can be seen that the spatial distribution of AHF >5 W·m −2 is consistent with an impervious surface. It indicates that the areas with frequent human activities can be well represented by the results of AHF.

Validation of AHF Estimation Results
It is difficult to validate AHF estimation results according to field measured data because the sensible or latent heat caused by human activities or background conditions cannot be completely distinguished [8,22,43]. Therefore, the results are usually verified by comparing them with previous similar studies [44,45]. Due to the inconsistency in the data sources, modeling methods, and research scales, there is often no absolute comparability between the study results, but it can be used as reference data to verify the rationality of research results as a whole [12]. Xie et al. [46], Chen et al. [18], and Wang et al. [12] employed an energy-consumption inventory approach based on statistics data to estimate the spatial distribution of AHF in China. Part of the estimation results of AHF in Jiangsu of the three existing studies are compared with those obtained in this study. The detailed comparison is shown in Table 3. The AHF estimation result in this study is basically consistent with the three existing studies results, especially in the order of magnitude.  [12] For Suzhou, Wuxi, and Nanjing, the three cities that have the highest AHF estimation values in Jiangsu, as examples (Figure 7), the spatial rationality of AHF result can be verified by comparing the distribution of AHF in the urban spaces with the different characteristics of land use and land cover types. The figure also illustrates the differences in the spatial details of AHF achieved by Luojia 1-01 and Suomi-NPP/VIIRS, respectively.   [12] For Suzhou, Wuxi, and Nanjing, the three cities that have the highest AHF estimation values in Jiangsu, as examples (Figure 7), the spatial rationality of AHF result can be verified by comparing the distribution of AHF in the urban spaces with the different characteristics of land use and land cover types. The figure also illustrates the differences in the spatial details of AHF achieved by Luojia 1-01 and Suomi-NPP/VIIRS, respectively. It can be seen from Figure 7 that the gridded AHF estimation results with high resolution of 130 m based on Luojia 1-01 NTL data can effectively reveal the detailed spatial distribution of AHF inside the cities. The sharp and clear urban anthropogenic heat emissions networks are formed by the dense and orderly urban road systems. According to the different traffic flow and busy degree of the main roads, the AHF values of them vary slightly. The AHF values of urban roads are generally between 30 and 100 W·m −2 . AHF can also reach more than 100 W·m −2 at some road intersections. In addition to the network distribution, AHF also forms the massive agglomeration in some prosperous areas of the city center. The high-value AHF areas are mainly distributed in urban commercial areas and old towns, such as the commercial street around the Suzhou ancient town, Wuxi city center, Baile Square and Huiju city complex, and Nanjing Xinjiekou city center. In addition, the large municipal public facilities areas, such as railway stations, the grand theatre, the international conference and exhibition center, and the Olympic center also have high AHF. The results of this study show that the AHF in these high-value areas are mostly concentrated in the range of 50-150 W·m −2 , and in some extra high-value areas can exceed 200 W·m −2 . Furthermore, there are some median-value areas of AHF are between 15 and 50 W·m −2 ; those are mainly distributed in the urban residential districts. Meanwhile, the AHF of some villages and suburban residential areas is lower, with AHF ranging of 5-15 W·m −2 ( Table 4). The above comparative analysis shows that the gridded AHF estimation results in this study have a good presentation of AHF spatial details insider the city. Furthermore, the AHF results align with the different characteristics of urban land use and land cover types in spatial distribution. To summarize, the AHF estimation in this study obtained good AHF estimation results.

Discussion
The AHF estimation index and its model are the key factors of AHF spatial estimation. In this study, the indexes of NTL nor , HSI, and VANUI were employed to estimate the AHF. We found that there are the significant correlations between NTL nor , VANUI. and AHF sta , especially in the power fitting equation, with R 2 about 0.8. However, the AHF estimation model based on HSI, the R 2 is as low as 0.26. Examining the result of HSI (Figure 8a), an unreasonable phenomenon could be found, that is, the value of water in HSI is much higher than other land covers, which directly leads to the lower fitting relationship between HSI and AHF sta . Analyzing the calculation formula of HSI in Equation (2), we found that in the extreme case, the NTL intensity of water areas should be close to 0, and the vegetation index (NDVI) with the range from −1 to 1, may be close to the minimum value, theoretically. Therefore, in this extreme situation, the HSI value in the water areas may present very high. Zhang et al. [24] also pointed out the drawbacks to the HSI. Since the relationship between NTL and NDVI follows a power law, when NTL has a maximum value, HSI will increase exponentially as NDVI approaches 0.
As such, when the water was removed from HSI (HSI-water) (Figure 8b), the fitting degree between HSI-water and AHF sta was greatly improved as expected ( Table 5). The R 2 is from 0.2641 of HSI to 0.7177 of HSI-water based on power function. The 5-fold cross-validation result of HSI-water AHF estimation model in Figure 9 illustrates that compared with HSI (R 2 is 0.8887 and RMSE is 6.8358 W·m −2 ), HSI-water based on power function has a significant improvement in correlations (R 2 is 0.8887) and accuracy (RMSE is 6.8358 W·m −2 ) between AHF estimation values and AHF sta values of 96 county-level cities in Jiangsu. However, if compared with VANUI (RMSE is 4.8277 W·m −2 ) or NTL nor (RMSE is 4.9782 W·m −2 ), there is still a certain accuracy gap in the result of HSI-water. Wang et al. [12] also found the same conclusion in the comparative study of these three indexes and pointed out that result of HSI have poor performance on the regional characteristics of AHF estimation.
Equation (2), we found that in the extreme case, the NTL intensity of water areas should be close to 0, and the vegetation index (NDVI) with the range from −1 to 1, may be close to the minimum value, theoretically. Therefore, in this extreme situation, the HSI value in the water areas may present very high. Zhang et al. [24] also pointed out the drawbacks to the HSI. Since the relationship between NTL and NDVI follows a power law, when NTL has a maximum value, HSI will increase exponentially as NDVI approaches 0.   As such, when the water was removed from HSI (HSI-water) (Figure 8b), the fitting degree between HSI-water and AHFsta was greatly improved as expected ( Table 5). The R 2 is from 0.2641 of HSI to 0.7177 of HSI-water based on power function. The 5-fold cross-validation result of HSI-water AHF estimation model in Figure 9 illustrates that compared with HSI (R 2 is 0.8887 and RMSE is 6.8358 W·m −2 ), HSI-water based on power function has a significant improvement in correlations (R 2 is 0.8887) and accuracy (RMSE is 6.8358 W·m −2 ) between AHF estimation values and AHFsta values of 96 county-level cities in Jiangsu. However, if compared with VANUI (RMSE is 4.8277 W·m −2 ) or NTLnor (RMSE is 4.9782 W·m −2 ), there is still a certain accuracy gap in the result of HSI-water. Wang et al. [12] also found the same conclusion in the comparative study of these three indexes and pointed out that result of HSI have poor performance on the regional characteristics of AHF estimation.  Over the past 20 years, DMSP/OLS and Suomi-NPP/VIIRS NTL data have been proven to be useful proxy measures for monitoring and analyzing human activities [27,47,48]. The coarse spatial resolution of these NTL data often means that the studies can only be done at large scales but will be restricted at city scales.
As can be seen from the results of AHF and the comparison between Luojia 1-01 and Suomi-NPP/VIIRS in the detailed images ( Figure 7) and their profiles of NTL data (Figure 10), Luojia 1-01 NTL images shows great advantages in capturing finer spatial details within urban areas. First, the main roads in urban areas can be presented more clearly by Luojia 1-01, while they are difficult to be reflected in Suomi-NPP/VIIRS. Moreover, the overall trends of the two NTL data profiles are basically similar, but the fluctuations in Luojia 1-01 are more frequently than those in Suomi-NPP/VIIRS. Moreover, the variation ranges of the peaks and valleys of Luojia 1-01 are larger than Suomi-NPP/VIIRS. These differences indicate that Luojia 1-01 NTL data contains more spatial information and has more capacities to reduce the saturation of NTL in urban areas. Over the past 20 years, DMSP/OLS and Suomi-NPP/VIIRS NTL data have been proven to be useful proxy measures for monitoring and analyzing human activities [27,47,48]. The coarse spatial resolution of these NTL data often means that the studies can only be done at large scales but will be restricted at city scales.
As can be seen from the results of AHF and the comparison between Luojia 1-01 and Suomi-NPP/VIIRS in the detailed images ( Figure 7) and their profiles of NTL data (Figure 10), Luojia 1-01 NTL images shows great advantages in capturing finer spatial details within urban areas. First, the main roads in urban areas can be presented more clearly by Luojia 1-01, while they are difficult to be reflected in Suomi-NPP/VIIRS. Moreover, the overall trends of the two NTL data profiles are basically similar, but the fluctuations in Luojia 1-01 are more frequently than those in Suomi-NPP/VIIRS. Moreover, the variation ranges of the peaks and valleys of Luojia 1-01 are larger than Suomi-NPP/VIIRS. These differences indicate that Luojia 1-01 NTL data contains more spatial information and has more capacities to reduce the saturation of NTL in urban areas.

Conclusions
In this article, using the new generation NTL data of Luojia 1-01 with finer spatial resolution, supplemented by Landsat 8 vegetation index data, the AHF spatial estimation models were constructed, and the gridded AHF mapping with high spatial resolution of 130 m in Jiangsu Province in 2018 was achieved. The conclusions are summarized as follows: (1) AHF can be effectively estimated by using Luojia 1-01 NTL data. Among the three indexes used in this study, NTLnor, HSI (or HSI-water,) and VANUI, the VANUI has the most significant correlation with annual mean AHF based on statistics data (AHFsta) of county-level cities in Jiangsu, and AHF spatial estimation model based on its power function has the highest accuracy.

Conclusions
In this article, using the new generation NTL data of Luojia 1-01 with finer spatial resolution, supplemented by Landsat 8 vegetation index data, the AHF spatial estimation models were constructed, and the gridded AHF mapping with high spatial resolution of 130 m in Jiangsu Province in 2018 was achieved. The conclusions are summarized as follows: (1) AHF can be effectively estimated by using Luojia 1-01 NTL data. Among the three indexes used in this study, NTL nor , HSI (or HSI-water,) and VANUI, the VANUI has the most significant correlation with annual mean AHF based on statistics data (AHF sta ) of county-level cities in Jiangsu, and AHF spatial estimation model based on its power function has the highest accuracy.
(2) From the estimation results of AHF mapping, the annual mean AHF of Jiangsu in 2018 was 2.91 W·m −2 . In terms of the AHF spatial distribution, the higher AHF values are obviously concentrated in the cities in southern Jiangsu, such as Suzhou, Wuxi, Changzhou, and Nanjing, while the AHF in the northern cities was lower. The concentration of higher AHF is closely related to the level of regional economic development and population density. (3) Compared with Suomi-NPP/VIIRS, Luojia 1-01 NTL data with finer spatial resolution and has more potentials in distinguishing AHF in urban areas from different land use and land cover types. This ability to discriminate the spatial detail of AHF will contribution to the precise management of urban or regional anthropogenic heat emissions.
Anthropogenic heat emissions aggravate the temperature rise in the urban areas and increase the amplitude of the UHI. The reason is that anthropogenic heat not only affects temperature directly, in the form of AHF, it also affects it indirectly by altering storage and consequently the outgoing long-wave and sensible heat flux terms. Thus, the gridded AHF estimation results can be added to the urban surface-energy balance parametrization and regional scale climate modelling to better understand and model human-environment interactions in the future.