1. Introduction
Weather information is indispensable to daily life. Understanding when and where people access weather information could have significant social benefits, such as aiding the development of user-oriented weather service systems and improving the efficiency of information dissemination. Previous studies demonstrated the importance of determining people’s use and perception of weather information. Some focused on behavior or attitudes from a general view. For example, Lazo et al. (2009) found that weather information was acquired most frequently in the morning, and early and late evening, to remain informed about the weather and aid the planning of daily activities and clothing choices. Others pay more attention during particular adverse weather. There is evidence showing that adverse weather may draw people’s attention to weather information. For example, people were found to use weather forecasts actively during hurricanes (Bostrom et al. 2018; Demuth et al. 2018). Li et al. (2021) found that weather application usage increased during a hailstorm in the megacities of Beijing and Tianjin in China (see the online supplemental material). The higher demand for weather information during hazardous weather perhaps is due to the desire to avoid related hazards. Although these studies demonstrated the relevance of weather conditions to the demand for weather information, further research with larger datasets and more comprehensive analyses is needed to improve our overall understanding of the effects of weather conditions on the demand for weather information.
The most commonly used method for studying the impact of weather on human behavior is the self-report survey in which participants are asked directly about their behaviors and attitudes. In recent years, massive amounts of data from weather application users have been collected automatically; data obtained through these applications could allow for more objective analysis of the demand for weather information. The ubiquity of “smartphones” has allowed for a wide range of data to be continuously collected, including measurements of the ambient atmosphere (Mass and Madaus 2014; Overeem et al. 2013) and mobile application usage records (Li et al. 2021). Additionally, smartphones also provide weather information that supplements that provided by television and radio (Lazo et al. 2009; Hickey 2015; Phan et al. 2018). Mobile applications can provide weather data instantaneously, and the time and location of data acquisition can be recorded for hundreds of millions of users. Such data are valuable for studies of the impact of weather on the demand for weather information. However, to the best of our knowledge, such data have rarely been used in studies on the demand for weather services.
In this study, smartphone data pertaining to the Moji Weather application were analyzed to explore the relationship between the demand for weather information and weather conditions. Moji Weather is a popular mobile weather application in many countries, especially China. Around 700 million total downloads and 60 million active users per day were recently reported for Moji Weather (Moji 2021a,b; J. Zhuang, Moji Co., Ltd., 2022, personal communication). From the usage data, we devised a demand index to measure variation in the demand for weather information in different regions. The demand index data were compared with hourly observations of precipitation, wind speed, temperature, and visibility to assess the relationship between demand for weather information and weather conditions by region. Such relationships could be used to tailor the types and amount of weather information according to the needs of weather application users.
The contribution of weather warnings to the demand for weather information was also examined. The results provide useful information for future studies aiming to evaluate and design weather warning systems.
This paper is organized as follows. Section 2 introduces the data and analysis methods. Section 3 presents the results of the analyses. Section 4 provides the conclusions and a final discussion of the limitations of the current work and possible future work.
2. Data and methods
a. Data description
In this study, we obtained smartphone data, meteorological surface observation data over China, and weather warning data for two Chinese megacities during 2017–18.
The smartphone data were collected and provided by Moji Co., Ltd. Because of data loss, 8602 of 8760 h of data were included for 2017, whereas a complete dataset (8760 h) was obtained for 2018. Approximately 6 billion Moji application usage records were generated within China during the study period. All records were binned into grid cells with a resolution of 0.5° × 0.5° (Fig. 1), for a total of 3415 grid cells. Data density was higher in eastern China than western China and was particularly high in the megacities in eastern China. The grid cells for Shanghai contained the most Moji usage records, exceeding 15 000 requests per hour on average. At the national scale, the average usage amount was 97 requests per hour per grid cell.
Distribution of log10-transformed Moji application weather data usage (color shading) and meteorological stations (black dots) across China during 2017–18.
Citation: Weather, Climate, and Society 14, 3; 10.1175/WCAS-D-21-0155.1
Hourly surface air temperature, precipitation, wind speed, and visibility observations from 2415 stations were obtained from the Chinese Meteorological Administration (CMA) after an initial quality control. Only stations with <20% missing or invalid data during the study period were selected, resulting in a final total of 1977 valid stations (Fig. 1).
All grid cells in this study had a mean hourly smartphone usage value of ≥5 (nonlogarithmic scale) and contained at least one meteorological weather station. A total of 750 grid cells (1245 stations) were analyzed (shaded areas in Fig. 2).
The four study regions (color shading). Beijing and Shanghai are marked with red crosses.
Citation: Weather, Climate, and Society 14, 3; 10.1175/WCAS-D-21-0155.1
As shown in Fig. 2, the grid cells covered four regions: Northern China (north of 34°N, east of 95°E), Xinjiang (north of 34°N, west of 95°E), the Sichuan Basin (south of 34°N and north of 28.5°N, east of 95°E), and Southern China (the remaining region south of 34°N). There were 277, 7, 60, and 406 grid cells in Northern China, Xinjiang, the Sichuan Basin, and Southern China, respectively. The northern and southern regions were delineated according to the traditionally used line of demarcation along the Qinling Mountains and Huaihe River (Liu et al. 2020; Li et al. 2015). We also distinguished the Sichuan Basin from Southern China, and Xinjian from Northern China. The Sichuan Basin, which is located in the eastern valley of the Tibet Plateau and is surrounded by mountains, has a climate that differs from the rest of Southern China. Similarly, Xinjiang, which is located in the most northwestern region of China, has a climate that is distinct from the rest of Northern China (Peel et al. 2007).
Weather warning data for two megacities, that is, Beijing, and Shanghai, whose locations are shown in Fig. 2, were obtained from the CMA. Warnings were issued for persistent low temperature, high temperature, high winds, rain storms, and other hazardous weather or weather-related hazards. Weather warnings for these two cities were issued at the district level. During the 2-yr study period, 3365 and 2182 warnings were issued in Beijing and Shanghai, respectively. To match the smartphone and weather data, the warning data were binned to hours and grid cells. A grid cell was considered to be a “warning cell” if a warning was issued for any part of it. A weather warning was assumed to have a major influence on a given hour if it was issued within the first 54 min thereof; otherwise, it was assumed to have a major influence on the next hour. Note that 54 min is an arbitrary cutoff; using other cutoffs such as 30, 36, 42, or 48 min did not affect the conclusions presented in later sections. The average number of weather warnings per grid cell for Beijing and Shanghai during the study period was 249 and 235, respectively.
b. Definition of demand index
By defining demand index this way, the diurnal variations in the demand for weather information can be removed. The demand index quantifies the extent to which demand for weather information deviates from the average level. To investigate the impact of real-time weather on the demand for weather information, we matched weather observations to the demand index on an hourly basis for each grid cell. The effect of each weather element was examined by setting a series of thresholds. For example, we estimated the effect of hourly precipitation by categorizing it into several groups and calculating the corresponding demand indexes.
Because the low probability of severe or adverse weather, the average demand index of a single grid cell in such weather may be inaccurate due to the small sample sizes. To estimate the community demand for weather information pertaining to a specific weather condition, we determined the demand index distribution using data from all grid cells within a region experiencing that weather condition.
3. Results
a. Seasonal and diurnal variations
To investigate seasonal and diurnal variations in the demand for weather information in the four regions shown in Fig. 2, we normalized the average hourly Moji application usage amount to [0, 1] according to the maximum difference for each grid cell and calculated the mean and standard deviation of the normalized values for all grid cells within each region. Overall, the northern regions (i.e., Northern China and Xinjiang) and southern regions (i.e., Southern China and the Sichuan Basin) exhibited distinctly different seasonal variations (Fig. 3a), with the amount of weather data used in summer being relatively higher in the northern regions. The amount of weather data used peaked in spring and winter in Southern China, spring and autumn in the Sichuan Basin, spring and summer in Northern China, and late spring to autumn in Xinjiang. These differences were not surprising because the weather conditions in the four regions exhibit different patterns of seasonal variation, which clearly drove the variations in user behavior.
(a) Seasonal and (b) diurnal variations from normalized average hourly Moji application usage in the four study regions in 2017–18. Each solid line represents the average normalized usage amount over all grid cells of a region. The upper and lower limits of an error bar represent the 25th and 75th percentiles of the normalized usage amount of all grid cells within the region.
Citation: Weather, Climate, and Society 14, 3; 10.1175/WCAS-D-21-0155.1
Unlike the seasonal variations, the diurnal variations in the four regions were similar with respect to local solar time (LST). There was one major peak in the morning at 0700 LST and two secondary peaks in the early and late evening (Fig. 3b). The late evening peak occurred at 2100 LST in all four regions, while the early evening peak occurred at 1600 LST in Xinjiang, but at 1700–1800 LST in the other regions. According to Lazo et al. (2009), these peaks might be associated with people’s need for information for daily planning.
b. Demand index according to local extreme weather
Considering the high impact of extreme weather on the public and society, we first examined how the demand for weather information in different grid cells was affected by different types of local extreme weather, that is, by conditions within the top 0.5% of hourly precipitation and wind speed, the top and bottom 0.5% of temperature (representing hot and cold extremes, respectively), and the bottom 0.5% of visibility. The average responses to these extreme local weather conditions are plotted separately in Fig. 4. Among the five extreme local weather types, heavy precipitation had the highest demand index. High wind speed in most parts of the country; low visibility in Northern China and Southern China; extreme hot weather in Northern China, Xinjiang, and the Sichuan Basin; and extreme cold weather in Southern China also had positive demand indexes. Interestingly, Northern China and Xinjiang generally exhibited higher demand for weather information than Southern China in the case of local extreme hourly precipitation and extreme hot weather, while Southern China exhibited higher demand during extreme cold weather. Furthermore, Xinjiang and the Sichuan Basin were markedly different from other regions in the same latitude, with higher demand observed during extreme high wind and low visibility conditions. The Sichuan Basin was also unusual in terms of its higher demand under extreme precipitation and hot weather conditions relative to other parts of Southern China. These differences in the demand index among the four regions generally supported the manner in which the regions were distinguished, as shown in Fig. 2. The regional demand over all grid cells is shown in Fig. 4f, which validated our observations in Figs. 4a–e.
Grid maps of the demand for weather information under hourly extreme weather conditions, i.e., the top 0.5% for (a) precipitation, (b) wind intensity, and (c) temperature, and bottom 0.5% for (d) temperature and (e) low visibility. (f) Regional demand under extreme conditions (i.e., in Northern China, Xinjiang, Southern China, and the Sichuan Basin). The 10th, 50th, and 90th percentiles are indicated by the horizontal black lines. The 25th and 75th percentiles are indicated by the upper and lower edges of the boxes. Boxes are not shown for sample sizes <10.
Citation: Weather, Climate, and Society 14, 3; 10.1175/WCAS-D-21-0155.1
c. Influence of weather conditions on weather data usage
To further explore the effects of various weather conditions on people’s demand for weather information, we calculated the demand index over a range of values for each of the four weather elements. Hourly precipitation was examined at a scale of 5 mm, and wind intensity was classified using the Beaufort scale (Tropical Cyclone Data Center 2021); every two original consecutive wind level categories were collapsed into one category. Because of differences in maximum visibility among stations (Zhang et al. 2020), visibility larger than 10 km was set to 10 km, the modified values were examined according to a 2-km interval. We examined temperature with a 10°C interval.
Overall, we observed an obvious increase in the demand index in response to increasing precipitation and wind intensity in all regions; this trend was weaker for visibility and temperature, except for temperature above 40°C (Fig. 5). Demand began to rise even under light precipitation. For a given hourly precipitation range, the demand index was significantly higher in Northern China and Xinjiang than Southern China and the Sichuan Basin. While the demand levels in Northern China and Xinjiang were similar, that in Sichuan Basin was significantly higher relative to Southern China.
Boxplots of the demand index in response to (a) precipitation, (b) wind speed, (c) temperature, and (d) visibility. The 10th, 50th, and 90th percentiles of data demand are indicated by the horizontal black lines. The 25th and 75th percentiles are indicated by the upper and lower edges of the boxes. Boxes are not shown for sample sizes <10.
Citation: Weather, Climate, and Society 14, 3; 10.1175/WCAS-D-21-0155.1
Our finding about the significant increase in demand in response to precipitation was in line with previous studies reporting that, among all of the elements of a weather forecast, people were most interested in precipitation (Lazo et al. 2009; Phan et al. 2018). This is intuitive because precipitation can often interfere with daily activities, necessitates the carrying of an umbrella, and reduces visibility (Gultepe 2008); this may cause road conditions to deteriorate (Jaroszweski and McNamara 2014) and lead to traffic jams.
Similar to precipitation, high wind speeds increased the demand for weather information in all regions. The demand index increased when the wind speed exceeded 5.5 m s−1. At wind speeds higher than 5.5 m s−1, Xinjiang had the highest demand, while the Sichuan Basin had a slightly higher demand than Northern and Southern China in the wind speed range of 5.5–10.7 m s−1. Southern China had a significantly higher response than Northern China when the wind speed exceeded 10.8 m s−1, possibly due to the landfall of tropical cyclones in Southern China.
Temperature had a small effect on the demand index, except for extremely high temperatures, that is, temperatures above 40°C. In Northern China, Xinjiang, and the Sichuan Basin, the demand for weather information significantly increased at temperatures above 40°C, with the demand in Xinjiang and Northern China being slightly higher than in the Sichuan Basin. In Southern China, the demand index was low in response to high temperatures but was slightly higher in the temperature range from −10° to 10°C than in the other regions. However, the demand was not significantly above the local average for Southern China.
Visibility has weaker effect on the demand index than other weather elements, indicating that the public had less interest in visibility fluctuations, perhaps because of its relatively small effect on daily activities in comparison with other weather elements. Further verification is required to confirm this hypothesis.
The different responses among these regions might be the result of local climate adaptation. According to the definition of sensory adaptation, sensitivity to a specific stimulus is reduced among people who are constantly exposed to it (Hewson and Tarrega 2017). Climate is one such stimulus to which people can adapt (Beall et al. 2012). People living in regions with frequent precipitation, high winds, and high or low temperature tend to be less sensitive to those conditions. Southern China experiences more wet and high-temperature days throughout the year than the other three regions (Yang et al. 2021), so people may be more adapted to them. Similarly, Xinjiang and the Sichuan Basin are less exposed to high wind speeds (Zhang and Wang 2020), and people therefore tend to pay more attention to windy conditions when they arise. High wind speeds are sometimes associated with tropical cyclones in Southern China, which might explain the greater attention paid by the public to wind speeds in that region. Apart from physiological and psychological adaptation, the fact that heaters and air conditioners are widely used in the northern and southern regions, respectively, could also have played a role in the different demand responses.
d. Determining high-demand weather criteria
Because the demand for weather information varies distinctly in different regions under different weather conditions, it is possible to provide locally customized weather services based on this knowledge. To quantify the relationship between the demand index and four types of weather conditions in the four regions, we performed a polynomial regression analysis. Prior to the regression analysis, we binned the data for each weather variable. Each variable was included in the regression model, and the demand for weather information was the dependent variable. The degree of the polynomial model was set to 3 for precipitation and wind, and 10 for temperature and visibility due to their more complex relationships. All regression analyses were conducted using the Python package scikit-learn (Pedregosa et al. 2011). The polynomial model showed a good fit to the data (Fig. 6).
Polynomial regression analysis showing the relationship between the weather data demand index and various weather conditions. Observed weather data were binned. The orange line represents the relationship between the mean observed weather data and mean observed demand index. The blue line is the regression curve. The orange shading indicates the 25th–75th percentile of the demand index.
Citation: Weather, Climate, and Society 14, 3; 10.1175/WCAS-D-21-0155.1
The 70th, 80th, 90th, 95th, and 99th percentiles of the demand indexes for the four regions (Table 1) were used as high-demand thresholds for matching with the regression curve equations, and to further characterize the impact of weather conditions on the demand for weather information. The weather criteria corresponding to these demand index percentiles are listed in Table 2. The precipitation and wind speed had low threshold values corresponding to the high-demand indexes (above the 70th percentile) in all regions. On the other hand, temperatures higher than 35°C drove increases in demand in all regions except Southern China. Low visibility was only weakly linked to a higher demand in Northern China and Southern China, whereas little attention was given to low visibility events in Xinjiang and the Sichuan Basin. These quantitative criteria for high-demand weather conditions in different regions are potentially useful for customizing and improving local weather services.
Demand index thresholds for defining high-demand weather conditions.
Quantitative criteria for defining high user demand for five types of weather data.
e. Demand index as an indicator of adverse weather and warnings
As can be seen from the previous section, some weather conditions, such as heavy precipitation, high wind, and high temperature, are often associated with a high demand for weather information, that is, there is a close linkage between the demand index and unfavorable weather with respect to human activities and psychology. It is natural to examine if the demand index can precisely indicate adverse of high-impact weather. Herein, adverse weather was defined as that in which one or more of the four weather elements corresponded to the top 10% of the average demand index values. The demand index and weather data for two megacities, Beijing and Shanghai, were used for the analysis. Weather warning data were also analyzed because weather warnings may also contribute to an increase in the demand index.
Because there was a degree of temporal overlap between the occurrence of adverse weather and weather warning issuance, we conducted an analysis for three cases: adverse weather only, weather warning only, and both. To quantify the correlation between the occurrence of adverse weather and the demand index, we calculated the proportion of the data samples in each case in which the demand index exceeded a given threshold. The proportions of high-demand events associated with the real-time occurrence of adverse weather only, weather warning issuance only, and both are shown in Fig. 7 over a range of demand index thresholds.
Proportion of high-demand events associated with the occurrence of adverse weather only, weather warning issuance only, and both according to the demand index threshold in (a) Beijing and (b) Shanghai.
Citation: Weather, Climate, and Society 14, 3; 10.1175/WCAS-D-21-0155.1
The proportion associated with the occurrence of adverse weather increased almost monotonically as the demand index increased from 0 to 12 for both cities (Fig. 7; sum of the blue and red areas). Some nonmonotonic increases may have been caused by random fluctuations, as the sample size decreased and the demand index increased. More important, a large proportion of high-demand events was associated with adverse weather, indicating that a high-demand index is a good indicator of adverse weather impacting on people’s daily lives.
The proportion of high-demand events associated only with a weather warning also increased monotonically as the demand index increased (Fig. 7, yellow area), which confirmed the correlation between weather warnings and the demand index. The results also indicated that some high-demand events were not associated with real-time adverse weather were linked to weather warnings. Including the impact of weather warnings, the proportion of high-demand events associated with adverse weather information increased for all thresholds in both cities, especially for the higher demand indexes, even reaching 100% in Shanghai at demand indexes higher than 12 (Fig. 7b).
4. Conclusions and discussion
In this study, smartphone (Moji Weather application) usage data from China during 2017–18 were analyzed alongside surface hourly observations of precipitation, wind speed, temperature, and visibility to explore the relationship between the demand for weather information and weather conditions. To investigate the regional dependence of the relationship, the study was conducted in four regions in China: Northern China, Xinjiang, the Sichuan Basin, and Southern China. Our analysis included diurnal and seasonal variations of the demand for weather information, the demand in response to local extreme weather, the influence of weather conditions on weather data requests, high-demand weather criteria (as perceived by the users), and the relative impacts of adverse weather and weather warnings on demand. To rule out any influence of daily human activities, the demand index focused on the impact of weather conditions.
The diurnal variations of demand for weather information found in this study were consistent with those reported previously (Lazo et al. 2009). People sought weather information more frequently in the morning, and early and late evening. These peaks match daily commute and sleep patterns, indicating an increase in demand in association with the planning of activities for the current day or next day. Our study also revealed significant seasonal variation in demand for weather information, as well as large differences in the patterns of variation among the four study regions. The seasonal variation, and its regional variability, suggested a dependence of the demand for weather information on weather conditions.
The demand for weather information was shown to increase obviously with hourly precipitation and wind intensity, but was not sensitive to temperature variations, except when the temperature approached 40°C. Among the weather elements investigated, precipitation was found to have the most significant effect on the demand index, consistent with Lazo et al. (2009) and Phan et al. (2018).
Our analysis also indicated that precipitation, high wind, and high temperature could induce distinctly different responses in different regions. In Northern China, there was higher demand in response to precipitation and high temperature, while in Southern China there was higher demand in response to low temperature. Relative to other regions, in Xinjiang and the Sichuan Basin demand was higher in response to high wind speeds and lower in response to low visibilities.
Using 2-yr datasets of the demand for weather information and weather observations, we performed a polynomial regression analysis and identified regional high-demand criteria for various weather variables. The criteria varied greatly among the different regions. Although the criteria were proposed for the specific weather data used in this study, the approach adopted here could be applied to other datasets.
By taking the weather conditions corresponding to the top 10% of the demand index as adverse weather, we also calculated the correlations between both the occurrence of adverse weather and the demand index, and weather warnings and the demand index. Our results showed that a high demand for weather information was strongly related to adverse weather and moderately related to weather warnings. Adverse weather interferes with people’s daily activities, and it was therefore not surprising that it had a strong correlation with the demand index. Weather warnings can play an important role in informing the public about weather risks and the need to take protective measures.
Our study suggested that analyzing smartphone usage data could reveal valuable information about the public’s responses to various weather conditions, including extreme weather, as well as public perception of high-impact weather. The large spatial coverage of the data allowed us to analyze these responses in several regions with different climatic conditions. The high-demand weather conditions identified for the different regions could be applied for customizing weather services for these regions. The findings could help to optimize weather information communication methods, inform the design of user-oriented weather forecast evaluation systems, and further improve user satisfaction with weather services.
A major limitation of this study was its short study period. Although we included ∼16 000 h of data for each grid cell, the rarity of extreme weather events made it difficult to build a sufficiently complex model to capture fine-scale variations for all four weather variables, and therefore weather variables were analyzed individually. Moreover, the relationships between the demand for weather information and weather conditions may shift under changing climates; therefore, the conclusions of this study will require regular updating. Changes in these relationships over time would be an interesting topic for future research. This will require smartphone datasets to be preserved and expanded constantly.
One possible future research activity would be to further divide the four regions distinguished herein into smaller areas to obtain fine-scale information about the effect of weather conditions on the demand for weather information; this would require a longer data collection period than that used in this study. Applying the results from our study to weather forecast could provide predictions of demand index in advance, which would also generate significant social and commercial benefits. Applying the method used in this study to systematically explore the effects of weather warnings is another important research target.
Acknowledgments.
This study was supported by the National Natural Science Foundation of China Grants 42030607 and 42142011. We thank Moji Corporation for providing the smartphone information. The use of the smartphone data abides by the rules of China’s personal information protection law (China National People’s Congress 2021). The National Center for Atmospheric Research is sponsored by the U.S. National Science Foundation.
Data availability statement.
Data on hourly weather observations were obtained from the National Meteorological Information Center of the China Meteorological Administration. Because of confidentiality agreements, data of user demand for weather information can only be made available after request from Moji Co., Ltd.
REFERENCES
Beall, C. M., N. G. Jablonski, and A. T. Steegmann, 2012: Human adaptation to climate temperature, ultraviolet radiation, and altitude. Human Biology: An Evolutionary and Biocultural Perspective, John Wiley and Sons, 177–250.
Bostrom, A., R. Morss, J. K. Lazo, J. Demuth, and H. Lazrus, 2018: Eyeing the storm: How residents of coastal Florida see hurricane forecasts and warnings. Int. J. Disaster Risk Reduct., 30, 105–119, https://doi.org/10.1016/j.ijdrr.2018.02.027.
China National People’s Congress, 2021: Personal information protection law of the People’s Republic of China. Accessed 5 March 2022, http://www.npc.gov.cn/npc/c30834/202108/a8c4e3672c74491a80b53a172bb753fe.shtml.
Demuth, J. L., and Coauthors, 2018: “Sometimes da #beachlife ain’t always da wave”: Understanding people’s evolving hurricane risk communication, risk assessments, and responses using Twitter narratives. Wea. Climate Soc., 10, 537–560, https://doi.org/10.1175/WCAS-D-17-0126.1.
Gultepe, I., 2008: Measurements of light rain, drizzle and heavy fog. Precipitation: Advances in Measurement, Estimation and Prediction, Springer, 59–82.
Hewson, L., and A. Tarrega, 2017: Sensory adaptation. Time-Dependent Measures of Perception in Sensory Evaluation, J. Hort, Ed., Wiley-Blackwell, 67–87.
Hickey, W., 2015: Where people go to check the weather. FiveThirtyEight, https://fivethirtyeight.com/features/weather-forecast-news-app-habits/.
Jaroszweski, D., and T. McNamara, 2014: The influence of rainfall on road accidents in urban areas: A weather radar approach. Travel Behav. Soc., 1, 15–21, https://doi.org/10.1016/j.tbs.2013.10.005.
Lazo, J. K., R. E. Morss, and J. L. Demuth, 2009: 300 billion served: Sources, perceptions, uses, and values of weather forecasts. Bull. Amer. Meteor. Soc., 90, 785–798, https://doi.org/10.1175/2008BAMS2604.1.
Li, R., Q. Zhang, J. Sun, Y. Chen, L. Ding, and T. Wang, 2021: Smartphone pressure data: Quality control and impact on atmospheric analysis. Atmos. Meas. Tech., 14, 785–801, https://doi.org/10.5194/amt-14-785-2021.
Li, S., S. Yang, and X. Liu, 2015: Spatiotemporal variability of extreme precipitation in north and south of the Qinling-Huaihe region and influencing factors during 1960–2013. Prog. Geogr., 34, 354–363, https://doi.org/10.11820/dlkxjz.2015.03.010.
Liu, J., Q. Yang, J. Liu, Y. Zhang, X. Jiang, and Y. Yang, 2020: Study on the spatial differentiation of the populations on both sides of the “Qinling-Huaihe Line” in China. Sustainability, 12, 4545, https://doi.org/10.3390/su12114545.
Mass, C. F., and L. E. Madaus, 2014: Surface pressure observations from smartphones: A potential revolution for high-resolution weather prediction? Bull. Amer. Meteor. Soc., 95, 1343–1349, https://doi.org/10.1175/BAMS-D-13-00188.1.
Moji, 2021a: About Moji. Moji Co., Ltd., http://www.moji.com/about/.
Moji, 2021b: About Moji culture. Moji Co., Ltd., http://www.moji.com/about/culture/.
Overeem, A., J. C. R. Robinson, H. Leijnse, G. J. Steeneveld, B. K. P. Horn, and R. Uijlenhoet, 2013: Crowdsourcing urban air temperatures from smartphone battery temperatures. Geophys. Res. Lett., 40, 4081–4085, https://doi.org/10.1002/grl.50786.
Pedregosa, F., and Coauthors, 2011: Scikit-learn: Machine learning in Python. J. Mach. Learn. Res., 12, 2825–2830.
Peel, M. C., B. L. Finlayson, and T. A. McMahon, 2007: Updated world map of the Köppen-Geiger climate classification. Hydrol. Earth Syst. Sci., 11, 1633–1644, https://doi.org/10.5194/hess-11-1633-2007.
Phan, M. D., B. E. Montz, S. Curtis, and T. M. Rickenbach, 2018: Weather on the go an assessment of smartphone mobile weather application use among college students. Bull. Amer. Meteor. Soc., 99, 2245–2257, https://doi.org/10.1175/BAMS-D-18-0020.1.
Sisco, M. R., V. Bosetti, and E. U. Weber, 2017: When do extreme weather events generate attention to climate change? Climatic Change, 143, 227–241, https://doi.org/10.1007/s10584-017-1984-2.
Tropical Cyclone Data Center, 2021: The extended Beaufort wind scale. China Meteorological Administration, https://tcdata.typhoon.org.cn/en/zy_wind.html.
Yang, X., B. Zhou, Y. Xu, and Z. Han, 2021: CMIP6 evaluation and projection of temperature and precipitation over China. Adv. Atmos. Sci., 38, 817–830, https://doi.org/10.1007/s00376-021-0351-4.
Zhang, Y., L. Gao, L. Cao, Z. Yan, and Y. Wu, 2020: Decreasing atmospheric visibility associated with weakening winds from 1980 to 2017 over China. Atmos. Environ., 224, 117314, https://doi.org/10.1016/j.atmosenv.2020.117314.
Zhang, Z., and K. Wang, 2020: Stilling and recovery of the surface wind speed based on observation, reanalysis, and geostrophic wind theory over China from 1960 to 2017. J. Climate, 33, 3989–4008, https://doi.org/10.1175/JCLI-D-19-0281.1.
