Intensive human activities produce large amounts of anthropogenic heat (AH) released into the atmosphere. A large amount of AH emissions will increase the urban heat island effect. Ichinose et al. found that the AH emissions increased the heat island intensity by 1–1.5 °C in Tokyo [1
]. Fan and Sailor showed that AH emissions in Philadelphia can make the heat island intensity increase 2–3 °C in a winter night [2
]. The increase of temperature will have a series of effects on surface energy conversion and the atmospheric boundary layer [3
], and further affect local air quality [5
]. Young et al. investigated the effects of anthropogenic heat on ozone air quality during the summer in the Seoul metropolitan area. The release of anthropogenic heat was found to increase O3
concentration by 3.8 ppb in the urban area [5
]. Hence, accurately estimating anthropogenic heat is of great significance for the urban heat island, surface energy conversion, and climatic and environmental researches.
Anthropogenic heat flux (AHF) refers to the amount of anthropogenic heat emission (AHE) generated per unit time and unit area [6
]. Many researchers have carried out studies on the estimation of AHF, which generally includes three types of methods: (1) energy balance equation; (2) building energy modeling; (3) the energy-consumption inventory approach [7
]. The estimation accuracy of the energy balance equation approach is highly dependent on the estimation accuracy of the parameters in the model, including sensible heat, latent heat, and storage heat [10
]. Furthermore, an additional issue of the energy balance equation approach is that it relies on flux measurements that are not available for most locations. The building energy modeling, as its name suggests, only considers the heat emissions from buildings [12
The energy-consumption inventory approach is employed to estimate AHF based on socio-economic data and various types of energy-consumption data, which takes into account the heat released from industry, building, and transportation [6
]. The energy consumption from different emission sources are converted into anthropogenic heat, respectively, through a certain energy conversion experience coefficient, thus achieving an AHF estimation [1
]. Since the socio-economic data and energy consumption data are based on administrative division units, the energy-consumption inventory approach can only obtain the mean value of AHF on the spatial scale represented by the statistical data, which cannot show the fine spatial pattern of AHF in the unit. What is more, it is time-consuming and laborious to obtain the complete and detailed socio-economic data and energy-consumption data for large-scale AHF estimations. Thus, it is essential and meaningful to collect the appropriate grid data and comprehensively use the multi-source remote sensing data to construct the AHF estimation model to obtain the gridded AHF data quickly and efficiently, which can better express the spatial difference of AHF inside the municipal administrative divisions.
Nighttime light (NTL) data have a unique advantage in monitoring urbanization and human activities. The brightness of nighttime light can well reflect the spatial distribution of economic activities and energy consumption [20
]. There is a high linear correlation between the brightness of nighttime light and AH emission [23
]. Yang et al. [27
] estimated the AHF in China from 1992 to 2010 using the defense meteorological satellite program’s operational linescan system (DMSP/OLS) NTL data. Moreover, Ma et al. [28
] obtained the gridded AHF data in Zhejiang province in China based on DMSP/OLS NTL data. However, DMSP/OLS NTL data presented a serious pixel saturation phenomenon in the urban center, which to some extent led to an underestimation of the AHF in the area. Dong et al. [14
] developed a top-down method for estimating global anthropogenic heat emission based on the global radiance-calibrated DMSP/OLS NTL data, which can effectively avoid the serious pixel saturation phenomenon and LandScan 2013 global population database. The AHF results with a high resolution of 1 km were acquired in 2013. However, the characteristics of anthropogenic heat emission may be different due to the differences in regional development levels and climatic conditions. There are some limitations in the same consideration of AHF emission characteristics of all regions in the large-scale AHF estimation. Therefore, it is necessary and meaningful to carry out research on the method of AHF estimation.
In addition, some researchers have analyzed the availability of some indexes in estimating AHF to obtain more refined AHF data. For example, the AHF estimation model was established using human settlement index (HSI) [29
] to realize the estimation of gridded AHF [28
]. HSI is constructed according to the strong negative correlation between vegetation index and surface impervious data [29
]. Although it has achieved good results in the current applications of small- and medium-scale AHF estimation, its promotion and application in large-scale areas still needs further analysis and discussion. Moreover, Zhang et al. [32
] used the vegetation index to correct the nighttime light index and proposed the vegetation adjusted NTL urban index (VANUI), which uses vegetation signals to reduce NTL data saturation and increase the spatial variability of nighttime lightness values. Chen et al. characterized the spatiotemporal dynamics of anthropogenic heat flux in Beijing–Tianjin–Hebei region in China between 1995 and 2015 using the VANUI index [33
]. The applicability of VANUI in estimating the gridded AHF for a large area requires further analysis and comparison.
This study aims to realize the fine gridded AHF estimation and mapping with a high resolution of 500 m in China in the year 2016. Considering the different characteristics of anthropogenic heat emission caused by the differences in regional development levels, this study adopts the partition modeling method for the first time to construct an AHF estimation model of sub-regions for better estimating the AHF in China. First, the socio-economic data and energy consumption data are used to estimate the AHF of the municipal (state, district, league) administrative regions in China based on the energy-consumption inventory approach. Next, taking the multi-source remote sensing data as independent variables, such as Suomi-NPP/VIIRS (Suomi national polar-orbiting partnership visible infrared imaging radiometer suite) NTL data which has a larger brightness range and can effectively overcome the pixel saturation phenomenon of DMSP/OLS data and MODIS (moderate resolution imaging spectroradiometer) data, and the AHF values obtained by the energy-consumption inventory approach as dependent variables, the gridded AHF estimation models are constructed in eight sub-regions respectively. Then, the performance and accuracy of different estimation models are analyzed and compared, and the optimal set of models are determined as the estimation scheme to achieve the AHF estimation in China. Finally, the refined AHF mapping with a resolution of 500 m was realized in China based on the model.
2. Data and Study Area Division
Socio-economic data and energy consumption data were derived from statistical yearbooks issued by the national bureau of statistics and local bureau of statistics [34
]. Although data of Tibet autonomous region was not collected, and the statistical indicators of data in Hong Kong, Macao, and Taiwan were slightly different, a total of 308 prefecture-level (state, district, league) socio-economic data and energy consumption data were collected. Meanwhile, the data inspection and preprocessing were carried out. For the missing data, the data was allocated according to the same type of index, or estimated by the linear regression method to ensure the integrity and reliability of the data used for modeling.
Multi-temporal MODIS NDVI (normalized difference vegetation index) (MOD13A1) data were downloaded from the United States Geological Survey (USGS) [35
]. The MOD13A1 product is 16-day synthetic data with a spatial resolution of 500 m. Data from April to October in 2016 were collected. A total of 28 scene images were covered in the whole country, which were re-projected from sinusoidal projection to Albers projection, and preprocessed successively by mosaicing and clipping. Then, quality control was performed according to the QC (quality control) subset of the MOD13A1 product to obtain applicable and high-quality data.
Suomi-NPP/VIIRS (national polar-orbiting partnership visible infrared imaging radiometer suite) NTL data were available at the National Oceanic and Atmospheric Administration (NOAA) [36
]. The Suomi-NPP/VIIRS annual synthetic NTL data in 2016 were obtained, which have been subjected to stray light, moonlight removal processing, and background noise suppression processing. The Suomi-NPP/VIIRS NTL data were finally re-projected to the Albers projection and the resample was completed based on the nearest neighbor method.
Due to the obvious regional differences in population distribution and economic development level in China, there are also large differences in the regional anthropogenic heat emissions. In order to better carry out the AHF estimation study in China, the whole study area was divided into eight sub-regions [37
]. The AHF estimation was carried out in each sub-region separately. The division of the sub-regions is shown in Figure 1
5.1. Applicability of HSI for Large-Scale AHF Estimation
HSI has achieved good results in the estimation of medium-scale and small-scale AHF [28
], but its application in estimating large-scale AHF still needs to be analyzed and discussed. In this study, the HSI and VANUI are constructed using NTL data and NDVI data, and the mean values of indexes in a prefecture-level city are calculated in order to establish the functional relationship with the mean AHF (as shown in Table 4
), so as to further analyze the applicability of indexes in large-scale AHF estimation.
It can be seen that there is a good fitting relationship between HSI and AHF in the ECR with a fitting R2 of 0.83. While the fitting coefficient is low in other sub-regions, especially in the MYER and NWR, the fitting R2 value is less than 0.1.
The reason might be that since the HSI is built based on the strong negative correlation between vegetation index and nighttime light brightness to express the characteristics of human activities. Generally, areas with a low NDVI value have more human activities and higher nighttime light brightness. However, many regions in China, especially in the northwest, have low vegetation coverage, sparse population, and also low nighttime light intensity. So the HSI usually cannot reflect the human activities in these areas. Therefore, HSI has obvious regional limitations in the application of large-scale AHF estimations.
5.2. Comparison of Goodness of Fit between VANUI and NTLnor and AHF
NTL data has been widely used in monitoring human residential dynamics and human activities [22
]. It is highly correlated with the urban population and socio-economic variables. It has been effectively used to estimate urban populations, GDP (Gross Domestic Product), built-up areas, and other urbanization variables [20
]. The results in this study further verified that nighttime light brightness can reflect the AH emission well. There is a good correlation between NTLnor
and AHF, indicating that NTLnor
can be effectively used for large-scale AHF estimation.
In addition, correlation analysis is conducted on the AHF and indexes in the prefecture-level city. The fitting R2
values of the VANUI in sub-regions are between 0.63 and 0.94, and the fitting R2
values of NTLnor
are between 0.69 and 0.95, as shown in Table 4
, indicating that there are significant correlations between AHF and the VANUI and NTLnor
The fitting results show that the VANUI is applicable to a large-scale AHF estimation. At the same time, the goodness of fit of the VANUI in MYER, MYAR, SCR, and NER are higher than that of NTLnor. By comparing the correlation between the VANUI, NTLnor and AHF, the effectiveness of the VANUI in improving the goodness of fit with AHF is discussed and confirmed, thus improving the estimation results of NTLnor. The results are obtained based on China as the research area. The AHF estimation scheme is also applicable to other areas where data from the study area is accessible.
5.3. Comparison of AHF Estimation Results of the Three Indexes
The AHF estimation results of the three indexes in Beijing, Shanghai, and Guangdong are shown in Figure 8
. By comparing the spatial morphology of the estimation results, the characteristics and differences of the estimation results of different indexes can be well recognized.
Through the comparison, it is found that the estimation results of the VANUI and NTLnor are relatively consistent on the whole. Moreover, the spatial pattern of AHF in the urban center can be well reflected, and the network distribution of road within the city can be clearly seen from the results, indicating that the results can effectively express the regional characteristics of AHF within the city.
The estimation results of the HSI have obvious saturation phenomenon in the urban center, which cannot reflect the high-value AHF distribution, and have poor performance on the regional characteristics of AHF within the city. It may be that the distribution characteristics and combination mode of factors, such as vegetation cover and human activities in the study area, influence the estimation results of the HSI.
The refined AHF mapping in China in the year 2016 is achieved in this study. First of all, the energy-consumption inventory approach is used to estimate the AHE and AHF values of the municipal (state, district, league) administrative regions. Next, the gridded AHF partition estimation models are constructed based on multi-source remote sensing data and the AHF data of the administrative division units. The performance and accuracy of different estimation models constructed based on multiple indexes are analyzed and compared in partitions. The set of optimal model is determined as the AHF partition estimation scheme. Finally, the refined AHF mapping with a resolution of 500 m was realized in China based on the model. The conclusions are as follows:
The estimated results of AHE and AHF values based on the energy-consumption inventory approach show that the AHE and AHF of provinces in China are quite different. The provinces with high AHE are Shandong, Guangdong, Jiangsu, and Hebei. Among them, the AHE in Shandong province is the highest, accounting for 10.17% of the country’s total AHE. The AHF values are compared in provinces. AHF in Shanghai is the highest, reaching 12.56 W·m−2, which is much higher than other provincial AHF values, followed by Tianjin, Beijing, and Jiangsu. The AHF values are 5.92 W·m−2, 3.35 W·m−2, and 3.10 W·m−2, respectively.
There are significant correlations between AHF and VANUI, NTLnor. Among them, the fitting R2 value of the VANUI and AHF in sub-regions is between 0.63 and 0.94. The fitting R2 value of NTLnor and AHF is between 0.69 and 0.95, and both of them are applicable to estimating large-scale AHF in China. The distribution characteristics and combination mode of vegetation cover, human activities, and other factors in the study area may have great influence on the estimation results of the HSI. The HSI has obvious regional limitations in the application of large-scale AHF estimation.
Compared with other AHF products, it is found that the AHF results of this study have higher spatial heterogeneity and can better characterize the spatial distribution pattern of AHF in a region. There are certain advantages to obtain fine AHF data by using the AHF partition estimation scheme, which can better realize AHF estimation according to the characteristics of AH emissions within sub-regions.
The AHF partition estimation scheme proposed in this study realizes the gridded AHF estimation in China. The refined AHF data with a resolution of 500 m obtained in this study can provide a support for the urban heat island and climatic and environmental researches. Since the statistical data are based on the administrative division unit, the model proposed at present cannot achieve the partition study of different AHF components, which still needs further study.