Yunnan Province is a typical low-latitude plateau region in Southwest China, the northern part of which encompasses high mountains belonging to the southern extension of the Qinghai–Tibet Plateau (Figure 1). Mountains, plateaus, and basins account for 84%, 10%, and 6% of the total area of the province, respectively (Ma et al., 2021). The special geographic location and unique geomorphic environment of Yunnan Province produce obvious regional differences in climate with prominent three-dimensional climatic characteristics. The northern high mountains have a barrier effect on precipitation, and the deep valleys in the west provide a North–South corridor for the movement of warm humid air from the ocean (Wu et al., 2012). The average/maximum/minimum temperature, average precipitation, and relative humidity in Yunnan Province from 1961 to 2020 were 16.5°C (102 stations)/23.1°C (102 stations)/11.9°C (102 stations), 1,105 mm (105 stations), and 74.3% (102 stations), respectively.

This study used two data time series obtained from the Yunnan Meteorological Bureau (http://yn.cma.gov.cn/) for analysis of shallow (5, 10, 15, and 20 cm) and deep (40–320 cm) soil temperature (Table 1). The first dataset comprised measurements recorded at 08:00, 14:00, and 20:00 Beijing Time (owing to the prevalence of errors in data recorded at 02:00, they are not considered here) from 1 January 1961 to 28 February 2021; the second dataset included each hour observations from 1 January 2004 to 28 February 2021. The quality of the data is strictly controlled, i.e., station data are discarded if there are missing/erroneous values for more than 5 days continuously in each month for the first dataset and for more than 1 day for the second dataset. Overall, there were few stations with data series that failed quality control. Station migration is the main factor causing sudden change in soil temperature data series. To further guarantee data quality, the following conditions were adopted when selecting the stations for inclusion in the study: 1) Stations with migration distance of no greater than 20 km and vertical height displacement of no greater than 100 m during 1961–2020 were retained, as long as the station underwent no more than two relocations. 2) Missing soil temperature data were replaced by applying linear regression to data of different elevations and from neighboring stations. Shuttle Radar Topography Mission data with 30-m resolution were derived from the Geospatial Data Cloud for China (https://www.gscloud.cn/).

The annual and seasonal soil temperatures averaged over the period of 2004–2020 across Yunnan Province are illustrated in Figures 2A–E. It can be seen that the annual and seasonal shallow soil temperatures change considerably, whereas the annual average deep soil temperatures (40–320 cm) change little; the hourly maximum temperature is 23.9°C (15:00) at 5-cm depth and 19.6°C at 160- and 320-cm depth (00:00–23:00). With increase of the soil layer, the hourly average soil temperature in spring and summer decreases. In spring, the highest soil temperature is 26.0°C (15:00) at 5-cm depth and the lowest soil temperature is 18.0°C (00:00–01:00) at 5-cm depth; in summer, the highest soil temperature is 28.3°C (15:00) at 5-cm depth and the lowest soil temperature is 19.9°C (00:00–23:00) at 320-cm depth. The average hourly soil temperature in autumn and winter increases with increase of the soil layer (the hourly soil temperature is the same for each layer and at each hour at depths of 80–320 cm). In autumn, the highest temperature is 23.7°C (15:00) at 5-cm depth and the lowest temperature is 21.1°C (00:00–23:00) at 320-cm depth. In winter, the highest temperature is 19.3°C (00:00–04:00) at 5-cm depth and the lowest temperature is 10.2°C at 320-cm depth (00:00–23:00).

The trends of change in annual and seasonal hourly soil temperatures during 2004–2020 display a large range at depths of 5–20 cm, whereas the amplitude of the significant trend of change in hourly temperature in deep soil layers (40–320 cm) is largely constant and significant (except for individual hours in winter). Specifically, the trend of change annually and in spring is most consistent, whereas the hourly trend of warming in summer is more significant than that in other seasons (Figure 3). Meanwhile, annually, the slope of the trend ranges from −0.61°C to 0.84°C/year at 5-cm depth, −0.32°C–0.92°C/year at 10-cm depth, −0.17°C–0.36°C/year at 15-cm depth, and −0.10°C to 0.23°C/year at 20-cm depth. In spring, the slope of the trend ranges from −0.64°C to 0.92°C/year at 5-cm depth, −0.32°C–0.59°C/year at 10-cm depth, −0.16°C–0.40°C/year at 15-cm depth, and −0.08 to 0.25°C/year at 20-cm depth. In summer, the slope of the trend ranges from −0.57 to 0.76°C/year at 5-cm depth, −0.32°C–0.50°C/year at 10-cm depth, −0.18°C–0.33°C/year at 15-cm depth, and −0.10°C to 0.21°C/year at 20-cm depth. In autumn, the slope of the trend ranges from −0.56°C to 0.72°C/year at 5-cm depth, −0.33°C–0.48°C/year at 10-cm depth, −0.17°C–0.33°C/year at 15-cm depth, and −0.10°C to 0.20°C/year at 20-cm depth. In winter, the slope of the trend ranges from −0.66°C to 0.89°C/year at 5-cm depth, −0.35°C–0.53°C/year at 10-cm depth, −0.18°C–0.35°C/year at 15-cm depth, and −010°C to 0.22°C/year at 20-cm depth for each hour. Additionally, annually and seasonally, the trend is upward at 02:00–11:00 (significant) and downward at both 00:00–01:00 (significant) and 12:00–23:00 (insignificant) at 5-cm depth. The trend is upward at 04:00–13:00 (significant, except in autumn) and downward at both 00:00–02:00 (significant) and 15:00–23:00 (insignificant, annually and in spring; significant, in summer, autumn, and winter) at 10-cm depth. The trend is upward at 05:00–15:00 (significant, annually and in spring and summer; 06:00–15:00 significant, in autumn and winter) and downward at both 00:00–03:00 (significant) and 16:00–23:00 (insignificant, annually and in spring; significant, in summer, autumn, and winter) at 15-cm depth. The trend is upward at 06:00–17:00 (significant, annually and in spring; 07:00–16:00 significant, in summer and autumn; 08:00–16:00 significant, in winter) and downward at both 00:00–05:00 (00:00–02:00 significant, annually and seasonally) and 18:00–23:00 (20:00–23:00 significant, annually and in winter; 19:00–23:00 significant, in summer and autumn) at 20-cm depth. Furthermore, temperature change at depths of 0–20 cm is mainly dominated by a significant upward trend at the majority of stations, and mainly concentrated during 03:00–11:00, 04:00–13:00, 06:00–14:00, and 08:00–15:00 at depths of 5, 10, 15, and 20 cm, respectively, whereas deep soil temperature is mainly dominated by a significant downward trend, although it is evident only at a relatively small percentage of stations annually and in spring. In summer, autumn, and winter, the significant upward trend in shallow soil temperature is mainly concentrated during 02:00–17:00 for average timescales and the majority of stations, while the significant downward trend is mainly concentrated during 13:00–23:00. The percentage of stations with significant upward and downward trends in deep soil temperature is lower than that for shallow layers both annually and seasonally (Figure 4).

Figure 5 presents the temporal distributions of soil temperature at depths of 5–320 cm on the monthly scale during 2004–2020, which show marked variation. The range of fluctuation of shallow soil temperature (depth: 5–20 cm) is large during February–November, especially at depths of 5–10 cm. Conversely, the deep soil temperature (depth: 40–320 cm) is almost the same every hour without fluctuation, but it also shows different characteristics. The deep soil temperature increases with increase of the soil layer during January–February and November–December, especially during January–February. During April–September, the deep soil temperature decreases with increase of the soil layer. Meanwhile, we note that the trend of monthly soil temperature change is inconsistent with the average soil temperature (Figure 6). The magnitude of the rate of change of the shallow soil temperature (depth: 5–10 cm) fluctuates greatly, while the hourly rate of change of the deep soil temperature varies little, showing a significant trend of increase, especially during April–June and in August and December. Overall, the significant trend of change of hourly soil temperature in January, May–August, and November–December is more obvious than that in other months, whereas the significant trend of change in February is weaker. It can be seen from Figure 7 that the percentage of stations with a significant uptrend or downtrend in soil temperature in shallow layers is higher than that of stations with a significant uptrend or downtrend in deep soil temperature. Shallow soil temperature exhibits an upward trend from February–May, mainly concentrated during 02:00–15:00. In other months, the temperature in each soil layer exhibits a significant uptrend at a higher percentage of stations than in the previous period, and then a significant downtrend at a higher percentage of stations in the subsequent period. Furthermore, the deep soil temperature shows significant increase and decrease in the percentage of stations with little or no significant decrease from January–December in all soil layers.

Considering the 1-km DEM, the 08:00, 14:00, and 20:00 soil temperatures at depths of 5–320 cm averaged over 2004–2020 were interpolated using six types of machine learning (the optimal model, ensemble weights (%), and R ² are detailed in Supplementary Tables S1–S3. Except for a few values of R ² < 0.8, the other values of >0.8 indicated that the interpolation results were ideal and could be used for spatial interpolation). The spatial variation of soil temperature at depths of 5–320 cm is characterized (only the map of the spatial distribution at depth of 5 cm is shown) by lower values in northwestern and northeastern parts of Yunnan Province and higher values in the south, especially in the dry and hot valley areas (Supplementary Figure S1).

A change in soil temperature might alter the physical properties of soil, which could then impart substantial impact on the physical, chemical, and biological processes occurring both above and below ground (Davidson and Janssens., 2006; Giardina et al., 2014; Hartley., 2014; Zhang et al., 2016; Li et al., 2022). Understanding the variations of soil temperature is of great importance for assessing energy exchange between the soil and the atmosphere in relation to climate change (Du et al., 2017; Shen et al., 2020; Liu et al., 2022; Oogathoo et al., 2022; Song et al., 2022). This study found a significant seasonal pattern contrary to the variation of average shallow and deep soil temperature, i.e., it decreases (increases) with increasing soil depth in spring and summer (autumn and winter) at 00:00–23:00. This finding is the same as that reported for Jiangsu Province, China (Shi and Chen, 2021), and eastern Australia (Knight et al., 2018) for daily soil temperature, showing that soil temperature is sensitive to short-term weather processes in terms of land–atmosphere exchange. In spring and summer, the land surface acts as an energy source and the deep soil acts as an energy receiver; in autumn and winter, their relations are reversed (Hu and Feng, 2003). Such reversed vertical energy processes result in the reversal of the vertical pattern of soil temperature (Shi and Chen, 2021). Previous research based on the daily soil temperature shallow and deep soil temperature revealed rising soil temperatures annually and seasonally in eastern Australia (Knight et al., 2018), metropolitan cities of Korea (Cheon et al., 2014), Croatia (Sviličić et al., 2016), the Tibetan Plateau, China (Zhu et al., 2018; Fang et al., 2019; Wang et al., 2020), Jiangsu Province (Shi et al., 2021) and China (Wang et al., 2018, 2021; Shi et al., 2021), and our results are consistent. Importantly, we found that the trend of change in shallow soil temperature (0–20 cm) fluctuates greatly, while the marked trend of change in deep soil temperature (40–320 cm) is almost the same at annual, seasonal, and monthly scales on average during the time series. However, the trend of change in annual, seasonal, and monthly hourly shallow soil temperature (0–20 cm) is obvious for most stations, whereas the corresponding changes in deep soil temperature (40–320 cm) are not, which means that deeper strata are probably affected less by meteorological fluctuations in Yunnan Province. Furthermore, the trend of spatiotemporal change in shallow soil temperature is very obvious. Therefore, it is necessary to consider the increase in soil respiration and the change in carbon sinks in aboveground and underground biomass when calculating the carbon balance of an ecosystem (Zhang et al., 2016), especially in Yunnan Province, which has rich biodiversity.

We also revealed the vertical variation of the average soil temperature with elevation. The average soil temperature decreases at the three specific times (08:00, 14:00, and 20:00), and the average soil temperatures are different at depths of 5 and 10–20 cm. Moreover, the average soil temperatures are higher at 14:00 than at 08:00 and 20:00 at 5-cm depth, whereas the values are higher at 20:00 than at 08:00 and 14:00 at depths of 10–20 cm. The increase in soil temperature in different soil layers with elevation is significant except at 5-cm depth, and rate of change is significant at 08:00 and 14:00 at depths of 15–20 cm; the change is significant at depth of 20 cm in all elevation zones, except at some elevations at 20:00. We first revealed the vertical gradient change in shallow soil temperature in low-latitude highland areas (Yunnan Province), which has not been reported before and might have some relevance for regional climate modeling and ecosystem modeling.

Different plant species have their own biological minimum during their growth cycle, germination, and emergence, which is very important for growth (Sviličić et al., 2016). If soil temperature drops below a certain threshold, the amounts of water and nutrients that roots can absorb decrease, which can lead to delay in certain developmental stages of plants (Porter and Gawith, 1999; Sviličić et al., 2016). Soil temperature accounted for approximately 90% of the change in net photosynthesis (Schwarz et al., 1997). Similarly, soil temperature is associated with limited underground plant (i.e., root) growth and development, which accounts for 50%–90% of all terrestrial plant growth (Ruess et al., 2003; Steinaker and Wilson, 2008). Therefore, soil temperature plays a key role in crop growth. Furthermore, a stable soil temperature above 12°C–14°C is suitable for sowing spring maize (Du et al., 2019). Meanwhile, Leeper et al. (2021) also highlighted that start of season estimates based on 5-cm soil temperature not only had lower measures of error than the traditional start of season based on 0-cm surface temperature, but also were a better match with the start of season based on NDVI for 41.3% (27.2%) of stations in years when 5-cm soil temperature (air temperature) measurements were available. In this study, we found that the start date has advanced, the end date has become delayed, duration has extended, and the critical accumulated soil temperature of ≥12 and 14°C at 5-cm depth has an upward trend, which are characteristics conducive to the emergence of crops and to the growth of strong seedlings and roots. For example, such changes can advance the date of sowing of early spring maize and support robust growth (Du et al., 2019). There is no doubt that the 5-cm soil temperature has undergone substantial change, and that plant phenology, vegetation shift, niche variety, and ecosystem stability are all likely to be impacted by soil warming, together with other ecological processes. This finding could support the government of Yunnan Province in adjusting the local agricultural structure and reasonably formulating appropriate sowing dates for corn, potatoes, and other crops. However, higher soil temperatures can be detrimental to some agricultural crops, causing many plant diseases and pests to become activated and microbiological activity to become reduced (Sviličić et al., 2016).

Air temperature and precipitation are the main factors that determine the variation in soil temperature (Helama et al., 2011; Yang et al., 2018; Zhang et al., 2021; Zhao et al., 2021). However, large-scale atmospheric or boundary conditions might contribute to the change in nighttime soil surface temperatures (Wang et al., 2018). We found that soil temperature has significant negative correlation with elevation and latitude at 00:00–23:00 during 2004–2020 and at 08:00, 14:00, and 20:00 annually and seasonally during 1961–2020. Conversely, Zhu et al. (2018), Cuo et al. (2013), and You et al. (2010) all argued that the correlation between the uptrend in soil temperature and station elevation is non-significant. Albergel et al. (2015) found that incorporating orography data in a forecast system had strong impact on the outcomes of the performance evaluation of soil temperature forecasts. This seems contradictory and the discrepancy might be attributable to different dataset types, different methods, and different study periods (Pepin et al., 2015; Zhu et al., 2018), and more obvious spatial heterogeneity, all of which need further study. We found significant positive correlation between soil temperature and both surface temperature (AST0, AMaxST0, and AMinST0) and air temperature (AT, ATAMax, and ATAMin), and between soil temperature at the majority of stations and both WS and SD at 00:00–23:00 and at 08:00, 14:00, and 20:00, but with the opposite characteristics. Most stations have significant negative (positive) correlation with SD (WS) at 00:00–23:00 at depths of 5–320 cm, whereas most stations have significant positive (negative) correlation with SD (WS) at 08:00, 14:00, and 20:00 at depths of 5–20 cm on annual and seasonal timescales. Furthermore, the correlation between soil temperature and precipitation (TP, DP, and NP) and RH is weaker at 00:00–23:00 (except TP) and at 08:00, 14:00, and 20:00 on annual and seasonal timescales. Significant correlation with air temperature and snow cover was also found in China (Wang et al., 2018; Zhang et al., 2021); however, on the Tibetan Plateau, Fang et al. (2019) identified the central role of air temperature in soil warming, whilst also highlighting that the relationship between soil temperature and precipitation is complicated owing to the presence of frozen ground in the area. Higher precipitation also induced higher soil temperature (0–320 cm), while inhibiting the impact of the freeze–thaw process on soil temperature in summer during 1960–2014 for 66 stations. Wang et al. (2020) found that soil temperature (0–20 cm) variability was strongly correlated with the change in air temperature but weakly correlated with precipitation during 1965–2014 for 56 stations. It can be seen that, even in the same area, different results will be obtained by revealing those factors that affect the change in temperature of different soil layers. Therefore, it remains necessary to further explore the driving factors of temperature change in different soil layers in different regions using various methods.

One of the limitations of this study is the uncertainty regarding the influencing factors. For example, even for the same location (Tibetan Plateau), the results obtained can differ from those reported in the literature (Fang et al., 2019; Wang et al., 2020). Furthermore, we used six machine learning methods to interpolate the spatial distribution of hourly soil temperature, and the interpolation parameters are relatively ideal (please see Supplementary Material S1); however, other influencing factors such as snow cover and vegetation also need further consideration.

At the same time, more stations also need to further collect, so that the interpolation is more accurate. Additionally, Wang et al. (2018) indicated that the effects of urbanization on soil temperature are small, whereas Cheon et al. (2014) revealed that the soil temperature in metropolitan cities increases with the increase of anthropogenic urban heat. Therefore, in future research, many other variables such as land use, the urban heat island, snow cover, vegetation growth, the soil microbial community, and soil properties should also be considered. Meanwhile, field studies should be conducted to improve our understanding of the response of soil temperature to climate change (Wang et al., 2020).