Abstract
This paper explores the application of emerging machine learning methods from image super resolution (SR) to the task of statistical downscaling. We specifically focus on convolutional neural network–based generative adversarial networks (GANs). Our GANs are conditioned on low-resolution (LR) inputs to generate high-resolution (HR) surface winds emulating Weather Research and Forecasting (WRF) Model simulations over North America. Unlike traditional SR models, where LR inputs are idealized coarsened versions of the HR images, WRF emulation involves using nonidealized LR and HR pairs, resulting in shared-scale mismatches due to internal variability. Our study builds upon current SR-based statistical downscaling by experimenting with a novel frequency-separation (FS) approach from the computer vision field. To assess the skill of SR models, we carefully select evaluation metrics and focus on performance measures based on spatial power spectra. Our analyses reveal how GAN configurations influence spatial structures in the generated fields, particularly biases in spatial variability spectra. Using power spectra to evaluate the FS experiments reveals that successful applications of FS in computer vision do not translate to climate fields. However, the FS experiments demonstrate the sensitivity of power spectra to a commonly used GAN-based SR objective function, which helps interpret and understand its role in determining spatial structures. This result motivates the development of a novel partial frequency-separation scheme as a promising configuration option. We also quantify the influence on GAN performance of nonidealized LR fields resulting from internal variability. Furthermore, we conduct a spectrum-based feature-importance experiment, allowing us to explore the dependence of the spatial structure of generated fields on different physically relevant LR covariates.
Significance Statement
We use artificial intelligence algorithms to mimic wind patterns from high-resolution climate models, offering a faster alternative to running these models directly. Unlike many similar approaches, we use datasets that acknowledge the essentially stochastic nature of the downscaling problem. Drawing inspiration from computer vision studies, we design several experiments to explore how different configurations impact our results. We find evaluation methods based on spatial frequencies in the climate fields to be quite effective at understanding how algorithms behave. Our results provide valuable insights into and interpretations of the methods for future research in this field.
Abstract
This paper explores the application of emerging machine learning methods from image super resolution (SR) to the task of statistical downscaling. We specifically focus on convolutional neural network–based generative adversarial networks (GANs). Our GANs are conditioned on low-resolution (LR) inputs to generate high-resolution (HR) surface winds emulating Weather Research and Forecasting (WRF) Model simulations over North America. Unlike traditional SR models, where LR inputs are idealized coarsened versions of the HR images, WRF emulation involves using nonidealized LR and HR pairs, resulting in shared-scale mismatches due to internal variability. Our study builds upon current SR-based statistical downscaling by experimenting with a novel frequency-separation (FS) approach from the computer vision field. To assess the skill of SR models, we carefully select evaluation metrics and focus on performance measures based on spatial power spectra. Our analyses reveal how GAN configurations influence spatial structures in the generated fields, particularly biases in spatial variability spectra. Using power spectra to evaluate the FS experiments reveals that successful applications of FS in computer vision do not translate to climate fields. However, the FS experiments demonstrate the sensitivity of power spectra to a commonly used GAN-based SR objective function, which helps interpret and understand its role in determining spatial structures. This result motivates the development of a novel partial frequency-separation scheme as a promising configuration option. We also quantify the influence on GAN performance of nonidealized LR fields resulting from internal variability. Furthermore, we conduct a spectrum-based feature-importance experiment, allowing us to explore the dependence of the spatial structure of generated fields on different physically relevant LR covariates.
Significance Statement
We use artificial intelligence algorithms to mimic wind patterns from high-resolution climate models, offering a faster alternative to running these models directly. Unlike many similar approaches, we use datasets that acknowledge the essentially stochastic nature of the downscaling problem. Drawing inspiration from computer vision studies, we design several experiments to explore how different configurations impact our results. We find evaluation methods based on spatial frequencies in the climate fields to be quite effective at understanding how algorithms behave. Our results provide valuable insights into and interpretations of the methods for future research in this field.
Abstract
This study describes an automated analysis of real-time tropical cyclone (TC) aircraft reconnaissance observations to estimate TC surface winds. The wind analysis uses an iterative, objective, data-weighted analysis approach with different smoothing constraints in the radial and azimuthal directions. Smoothing constraints penalize the data misfit when the solutions deviate from smoothed analyses and extend the aircraft information into areas not directly observed. The analysis composites observations following storm motion taken within five hours prior and three hours after analysis time and makes use of prescribed methods to move observations to a Common Flight Level (CFL; 700-hPa) for analysis and reduce reconnaissance observations to the surface. Comparing analyses to several observed and simulated wind fields shows that analyses fit the observations while extending observational information to poorly observed regions. However, resulting analyses tend toward greater symmetry as observational coverage decreases, and show sensitivity to the first guess information in unobserved radii. Analyses produce reasonable and useful estimates of operationally important characteristics of the wind field. But, due to the radial and azimuthal smoothing and the under-sampling of typical aircraft reconnaissance flights, wind maxima are underestimated, and the radii of maximum wind are slightly overestimated. Varying observational coverage using model-based synthetic aircraft observations, these analyses improve as observational coverage increases, and for a typical observational pattern (two transects through the storm) the root-mean-square error deviation is < 10 kt (< 5 m s−1).
Abstract
This study describes an automated analysis of real-time tropical cyclone (TC) aircraft reconnaissance observations to estimate TC surface winds. The wind analysis uses an iterative, objective, data-weighted analysis approach with different smoothing constraints in the radial and azimuthal directions. Smoothing constraints penalize the data misfit when the solutions deviate from smoothed analyses and extend the aircraft information into areas not directly observed. The analysis composites observations following storm motion taken within five hours prior and three hours after analysis time and makes use of prescribed methods to move observations to a Common Flight Level (CFL; 700-hPa) for analysis and reduce reconnaissance observations to the surface. Comparing analyses to several observed and simulated wind fields shows that analyses fit the observations while extending observational information to poorly observed regions. However, resulting analyses tend toward greater symmetry as observational coverage decreases, and show sensitivity to the first guess information in unobserved radii. Analyses produce reasonable and useful estimates of operationally important characteristics of the wind field. But, due to the radial and azimuthal smoothing and the under-sampling of typical aircraft reconnaissance flights, wind maxima are underestimated, and the radii of maximum wind are slightly overestimated. Varying observational coverage using model-based synthetic aircraft observations, these analyses improve as observational coverage increases, and for a typical observational pattern (two transects through the storm) the root-mean-square error deviation is < 10 kt (< 5 m s−1).
Abstract
As a follow-on to a previous study that examined the tilt and precession evolution of tropical cyclones (TCs) in a critical shear regime, this study examines the processes leading to the subsequent divergent evolutions in tilt and intensity. The control experiment fails to resume its precession and reintensify, while the perturbed experiments with enhanced upper-level inner-core vorticity resume the precession after a precession hiatus period. In the control experiment, a mesoscale negative absolute vorticity region forms at the upper levels due to tilting in strong downtilt convection. This upper-level, negative-vorticity region is inertially unstable, causing the inward acceleration of upper-level radial inflow. This upper-level inflow subsequently becomes negatively buoyant due to diabatic cooling and descends, bringing midlevel, low equivalent potential temperature (θE ) air into the inner-core TC boundary layer, significantly disrupting the low-level TC circulation. Consequently, the disrupted TC vortex in the control is unable to recover. The upper-level negative vorticity region is absent in the perturbed experiments due to weaker downtilt convection, preventing the emergence of the disruptive inner-core downdraft. The weaker downtilt convection is caused by several factors. First, a stronger circulation aloft advects hydrometeors farther downwind, resulting in greater separation of the cooling-driven downdraft from the convective updraft region, and thus weaker dynamically forced lifting at low levels. Second, the mean θE of the low-level air feeding downtilt convection is smaller. Third, there is stronger and deeper adiabatic descent uptilt, causing more low-θE air diluting the downtilt updraft region. These results show how the full vortex structure is important to diverging TC evolutions in moderately sheared environments.
Abstract
As a follow-on to a previous study that examined the tilt and precession evolution of tropical cyclones (TCs) in a critical shear regime, this study examines the processes leading to the subsequent divergent evolutions in tilt and intensity. The control experiment fails to resume its precession and reintensify, while the perturbed experiments with enhanced upper-level inner-core vorticity resume the precession after a precession hiatus period. In the control experiment, a mesoscale negative absolute vorticity region forms at the upper levels due to tilting in strong downtilt convection. This upper-level, negative-vorticity region is inertially unstable, causing the inward acceleration of upper-level radial inflow. This upper-level inflow subsequently becomes negatively buoyant due to diabatic cooling and descends, bringing midlevel, low equivalent potential temperature (θE ) air into the inner-core TC boundary layer, significantly disrupting the low-level TC circulation. Consequently, the disrupted TC vortex in the control is unable to recover. The upper-level negative vorticity region is absent in the perturbed experiments due to weaker downtilt convection, preventing the emergence of the disruptive inner-core downdraft. The weaker downtilt convection is caused by several factors. First, a stronger circulation aloft advects hydrometeors farther downwind, resulting in greater separation of the cooling-driven downdraft from the convective updraft region, and thus weaker dynamically forced lifting at low levels. Second, the mean θE of the low-level air feeding downtilt convection is smaller. Third, there is stronger and deeper adiabatic descent uptilt, causing more low-θE air diluting the downtilt updraft region. These results show how the full vortex structure is important to diverging TC evolutions in moderately sheared environments.
Abstract
High-resolution profiles of vertical velocity obtained from two different surface-following autonomous platforms, Surface Wave Instrument Floats with Tracking (SWIFTs) and a Liquid Robotics SV3 Wave Glider, are used to compute dissipation rate profiles ϵ(z) between 0.5 and 5 m depth via the structure function method. The main contribution of this work is to update previous SWIFT methods to account for bias due to surface gravity waves, which are ubiquitous in the near-surface region. We present a technique where the data are prefiltered by removing profiles of wave orbital velocities obtained via empirical orthogonal function (EOF) analysis of the data prior to computing the structure function. Our analysis builds on previous work to remove wave bias in which analytic modifications are made to the structure function model. However, we find the analytic approach less able to resolve the strong vertical gradients in ϵ(z) near the surface. The strength of the EOF filtering technique is that it does not require any assumptions about the structure of nonturbulent shear, and does not add any additional degrees of freedom in the least squares fit to the model of the structure function. In comparison to the analytic method, ϵ(z) estimates obtained via empirical filtering have substantially reduced noise and a clearer dependence on near-surface wind speed.
Abstract
High-resolution profiles of vertical velocity obtained from two different surface-following autonomous platforms, Surface Wave Instrument Floats with Tracking (SWIFTs) and a Liquid Robotics SV3 Wave Glider, are used to compute dissipation rate profiles ϵ(z) between 0.5 and 5 m depth via the structure function method. The main contribution of this work is to update previous SWIFT methods to account for bias due to surface gravity waves, which are ubiquitous in the near-surface region. We present a technique where the data are prefiltered by removing profiles of wave orbital velocities obtained via empirical orthogonal function (EOF) analysis of the data prior to computing the structure function. Our analysis builds on previous work to remove wave bias in which analytic modifications are made to the structure function model. However, we find the analytic approach less able to resolve the strong vertical gradients in ϵ(z) near the surface. The strength of the EOF filtering technique is that it does not require any assumptions about the structure of nonturbulent shear, and does not add any additional degrees of freedom in the least squares fit to the model of the structure function. In comparison to the analytic method, ϵ(z) estimates obtained via empirical filtering have substantially reduced noise and a clearer dependence on near-surface wind speed.
Abstract
Modification of grasslands into irrigated and non-irrigated agriculture in the Great Plains results in significant impacts on weather and climate. However, there has been lack of observational data-based studies solely focused on impacts of irrigation on the PBL and convective conditions. The Great Plains Irrigation Experiment (GRAINEX) during the 2018 growing season collected data over irrigated and non-irrigated land uses over Nebraska to understand these impacts. Specifically, the objective was to determine whether the impacts of irrigation are sustained throughout the growing season.
The data analyzed include latent and sensible heat flux, air temperature, dew point temperature, equivalent temperature (moist enthalpy), PBL height, lifting condensation level (LCL), level of free convection (LFC), and PBL mixing ratio. Results show increased partitioning of energy into latent heat compared to sensible heat over irrigated areas while average maximum air was decreased and dewpoint temperature was increased from the early to peak growing season. Radiosonde data suggest reduced planetary boundary layer (PBL) heights at all launch sites from the early to peak growing season. However, reduction of PBL height was much greater over irrigated areas compared to non-irrigated croplands. Compared to the early growing period, LCL and LFC heights were also lower during the peak growing period over irrigated areas. Results note, for the first time, that the impacts of irrigation on PBL evolution and convective environment can be sustained throughout the growing season and regardless of background atmospheric conditions. These are important findings and applicable to other irrigated areas in the world.
Abstract
Modification of grasslands into irrigated and non-irrigated agriculture in the Great Plains results in significant impacts on weather and climate. However, there has been lack of observational data-based studies solely focused on impacts of irrigation on the PBL and convective conditions. The Great Plains Irrigation Experiment (GRAINEX) during the 2018 growing season collected data over irrigated and non-irrigated land uses over Nebraska to understand these impacts. Specifically, the objective was to determine whether the impacts of irrigation are sustained throughout the growing season.
The data analyzed include latent and sensible heat flux, air temperature, dew point temperature, equivalent temperature (moist enthalpy), PBL height, lifting condensation level (LCL), level of free convection (LFC), and PBL mixing ratio. Results show increased partitioning of energy into latent heat compared to sensible heat over irrigated areas while average maximum air was decreased and dewpoint temperature was increased from the early to peak growing season. Radiosonde data suggest reduced planetary boundary layer (PBL) heights at all launch sites from the early to peak growing season. However, reduction of PBL height was much greater over irrigated areas compared to non-irrigated croplands. Compared to the early growing period, LCL and LFC heights were also lower during the peak growing period over irrigated areas. Results note, for the first time, that the impacts of irrigation on PBL evolution and convective environment can be sustained throughout the growing season and regardless of background atmospheric conditions. These are important findings and applicable to other irrigated areas in the world.
Abstract
East African countries benefit economically from the largest freshwater lake in Africa: Lake Victoria (LV). Around 30 million people live along its coastline, and 5.4 million people subsist on its fishing industry. However, more than 1000 fishermen die annually by high-wave conditions often produced by severe convective wind phenomena, which marks this lake one of the deadliest places in the world for hazardous weather impacts. The World Meteorological Organization launched the 3-yr High Impact Weather Lake System (HIGHWAY) project, with the main objective to reduce loss of lives and economic goods in the lake basin and improve the resilience of the local communities. The project conducted a field campaign in 2019 aiming to provide forecasters with high-resolution observations and to study the storm life cycle over the lake basin. The research discussed here used the S-band polarimetric Tanzania radar from the field campaign to investigate the diurnal cycle of the convective mode over the lake. We classified the lake storms occurring during the two wet seasons into six different convective modes and present their diurnal evolution, organization, and main radar-based attributes, thereby extending the knowledge of convection on the lake. The result is the creation of a “convection catalog for Lake Victoria,” using the operational forecast lake sectors, and defining the exact times for the different timeslots resulting from the HIGHWAY project for the marine forecast. This will inform methods to improve the marine operational forecasts for Lake Victoria, and to provide the basis for new standard operation procedures (SOP) for severe weather surveillance and warning.
Significance Statement
In this work we use new radar data over Lake Victoria, Africa, to study convective mode organization and its diurnal cycle over the lake. This work is of particular importance due to the numerous hazardous weather events and related accidents on the lake, including capsized boats, plane crashes, floods, and hailstorms on the shore settlements, that are responsible for a high annual fatality toll. Results of our analyses provide updated information for operational marine forecasts using relevant time segments and sectors of the lake to improve nowcasting operations in Lake Victoria.
Abstract
East African countries benefit economically from the largest freshwater lake in Africa: Lake Victoria (LV). Around 30 million people live along its coastline, and 5.4 million people subsist on its fishing industry. However, more than 1000 fishermen die annually by high-wave conditions often produced by severe convective wind phenomena, which marks this lake one of the deadliest places in the world for hazardous weather impacts. The World Meteorological Organization launched the 3-yr High Impact Weather Lake System (HIGHWAY) project, with the main objective to reduce loss of lives and economic goods in the lake basin and improve the resilience of the local communities. The project conducted a field campaign in 2019 aiming to provide forecasters with high-resolution observations and to study the storm life cycle over the lake basin. The research discussed here used the S-band polarimetric Tanzania radar from the field campaign to investigate the diurnal cycle of the convective mode over the lake. We classified the lake storms occurring during the two wet seasons into six different convective modes and present their diurnal evolution, organization, and main radar-based attributes, thereby extending the knowledge of convection on the lake. The result is the creation of a “convection catalog for Lake Victoria,” using the operational forecast lake sectors, and defining the exact times for the different timeslots resulting from the HIGHWAY project for the marine forecast. This will inform methods to improve the marine operational forecasts for Lake Victoria, and to provide the basis for new standard operation procedures (SOP) for severe weather surveillance and warning.
Significance Statement
In this work we use new radar data over Lake Victoria, Africa, to study convective mode organization and its diurnal cycle over the lake. This work is of particular importance due to the numerous hazardous weather events and related accidents on the lake, including capsized boats, plane crashes, floods, and hailstorms on the shore settlements, that are responsible for a high annual fatality toll. Results of our analyses provide updated information for operational marine forecasts using relevant time segments and sectors of the lake to improve nowcasting operations in Lake Victoria.
Abstract
As one of the most prominent weather systems over the Indian subcontinent, the Indian summer monsoon low pressure systems (MLPSs) have been studied extensively over the past decades. However, the processes that govern the growth of the MLPSs are not well understood. To better understand these processes, we created an MLPS index using bandpass-filtered precipitation data. Lag regression maps and vertical cross sections are used to document the distribution of moisture, moist static energy (MSE), geopotential, and horizontal and vertical motions in these systems. It is shown that moisture governs the distribution of MSE and is in phase with precipitation, vertical motion, and geopotential during the MLPS cycle. Examination of the MSE budget reveals that longwave radiative heating maintains the MSE anomalies against dissipation from vertical MSE advection. These processes nearly cancel one another, and it is variations in horizontal MSE advection that are found to explain the growth and decay of the MSE anomalies. Horizontal MSE advection contributes to the growth of the MSE anomalies in MLPSs prior to the system attaining a maximum amplitude and contributes to decay thereafter. The horizontal MSE advection is largely due to meridional advection of mean state MSE by the anomalous winds, suggesting that the MSE anomalies undergo a moisture–vortex instability (MVI)-like growth. In contrast, perturbation kinetic energy (PKE) is generated through barotropic conversion. The structure, propagation, and energetics of the regressed MLPSs are consistent with both barotropic and moisture–vortex growth.
Abstract
As one of the most prominent weather systems over the Indian subcontinent, the Indian summer monsoon low pressure systems (MLPSs) have been studied extensively over the past decades. However, the processes that govern the growth of the MLPSs are not well understood. To better understand these processes, we created an MLPS index using bandpass-filtered precipitation data. Lag regression maps and vertical cross sections are used to document the distribution of moisture, moist static energy (MSE), geopotential, and horizontal and vertical motions in these systems. It is shown that moisture governs the distribution of MSE and is in phase with precipitation, vertical motion, and geopotential during the MLPS cycle. Examination of the MSE budget reveals that longwave radiative heating maintains the MSE anomalies against dissipation from vertical MSE advection. These processes nearly cancel one another, and it is variations in horizontal MSE advection that are found to explain the growth and decay of the MSE anomalies. Horizontal MSE advection contributes to the growth of the MSE anomalies in MLPSs prior to the system attaining a maximum amplitude and contributes to decay thereafter. The horizontal MSE advection is largely due to meridional advection of mean state MSE by the anomalous winds, suggesting that the MSE anomalies undergo a moisture–vortex instability (MVI)-like growth. In contrast, perturbation kinetic energy (PKE) is generated through barotropic conversion. The structure, propagation, and energetics of the regressed MLPSs are consistent with both barotropic and moisture–vortex growth.
Abstract
Spatial patterns of tropical cyclone tornadoes (TCTs), and their relationship to patterns of mesoscale predictors within U.S. landfalling tropical cyclones (LTCs) are investigated using multicase composites from 27 years of reanalysis data (1995–2021). For 72 cases of LTCs with wide-ranging TC intensities at landfall, daytime TCT frequency maxima are found in the northeast, right-front, and downshear-right quadrants when their composites are constructed in ground-relative, TC-heading relative, and environmental shear relative coordinates, respectively. TCT maxima are located near maxima of 10-m–700-hPa bulk wind difference (BWD), which are enhanced by the TC circulation. This proxy for bulk vertical shear in roughly the lowest 3 km is among the best predictors of maximum TCT frequency. Relative to other times, the position of maximum TCT frequency during the afternoon shifts ∼100 km outward from the LTC center toward larger MLCAPE values. Composites containing the strongest LTCs have the strongest maximum 10-m–700-hPa and 10-m–500-hPa BWDs (∼20 m s−1) with nearby maximum frequencies of TCTs. Corresponding composites containing weaker LTCs but still many TCTs, had bulk vertical shear values that were ∼20% smaller (∼16 m s−1). Additional composites of cases having similarly weak average LTC strength at landfall, but few or no TCTs, had both maximum bulk vertical shears that were an additional ∼20% lower (∼12 m s−1) and smaller MLCAPE. TCT environments occurring well inland are distinguished from others by having stronger westerly shear and a west–east-oriented baroclinic zone (i.e., north–south temperature gradient) that enhances mesoscale ascent and deep convection on the LTC’s east side.
Abstract
Spatial patterns of tropical cyclone tornadoes (TCTs), and their relationship to patterns of mesoscale predictors within U.S. landfalling tropical cyclones (LTCs) are investigated using multicase composites from 27 years of reanalysis data (1995–2021). For 72 cases of LTCs with wide-ranging TC intensities at landfall, daytime TCT frequency maxima are found in the northeast, right-front, and downshear-right quadrants when their composites are constructed in ground-relative, TC-heading relative, and environmental shear relative coordinates, respectively. TCT maxima are located near maxima of 10-m–700-hPa bulk wind difference (BWD), which are enhanced by the TC circulation. This proxy for bulk vertical shear in roughly the lowest 3 km is among the best predictors of maximum TCT frequency. Relative to other times, the position of maximum TCT frequency during the afternoon shifts ∼100 km outward from the LTC center toward larger MLCAPE values. Composites containing the strongest LTCs have the strongest maximum 10-m–700-hPa and 10-m–500-hPa BWDs (∼20 m s−1) with nearby maximum frequencies of TCTs. Corresponding composites containing weaker LTCs but still many TCTs, had bulk vertical shear values that were ∼20% smaller (∼16 m s−1). Additional composites of cases having similarly weak average LTC strength at landfall, but few or no TCTs, had both maximum bulk vertical shears that were an additional ∼20% lower (∼12 m s−1) and smaller MLCAPE. TCT environments occurring well inland are distinguished from others by having stronger westerly shear and a west–east-oriented baroclinic zone (i.e., north–south temperature gradient) that enhances mesoscale ascent and deep convection on the LTC’s east side.
Abstract
The surface and air temperature gradient (T S00-T air) drives the development of the convective boundary layer and the occurrence of clouds and precipitation. However, its variability is still poorly understood due to the lack of high-quality observations. This study fills in this gap by investigating the diurnal to decadal variability in T S00-T air from 2002 to 2022 based on hourly observations collected at over 100 stations of the U.S. Climate Reference Network. It is found that T S00-T air reaches its maximum at noon with an average of 6.85°C over the Contiguous United States, which decreases to 4.28°C when the soil moisture exceeds 30%. The daily minimum of T S00-T air has an average of −2.08°C, which generally occurs in the early evening but is postponed as the cloud fraction decreases. Moreover, while existing studies have used the near-surface soil temperature, such as the 5-cm soil temperature (T S05), to calculate T S05-T air, we find that T S00-T air and T S05-T air have opposite diurnal cycles, and their amplitudes differed drastically. The daily minimum of T S00-T air has a significant decreasing trend (−0.50±0.007°C/decade) from 2002 to 2022 due to T air increasing at a higher rate than T S00 during the nighttime. The occurrence frequency of near surface stable condition (T S00-T air<0) increases significantly, and the frequency of unstable condition (T S00-T air>0) decreases notably throughout the year except for winter. When it is stable, the magnitude of T S00-T air tends to decrease while the T S00-T air tends to increase when it is unstable, which is consistent with the drying condition caused by precipitation deficit. This study provides the first observational evidence on how T S00-T air responds to warming.
Abstract
The surface and air temperature gradient (T S00-T air) drives the development of the convective boundary layer and the occurrence of clouds and precipitation. However, its variability is still poorly understood due to the lack of high-quality observations. This study fills in this gap by investigating the diurnal to decadal variability in T S00-T air from 2002 to 2022 based on hourly observations collected at over 100 stations of the U.S. Climate Reference Network. It is found that T S00-T air reaches its maximum at noon with an average of 6.85°C over the Contiguous United States, which decreases to 4.28°C when the soil moisture exceeds 30%. The daily minimum of T S00-T air has an average of −2.08°C, which generally occurs in the early evening but is postponed as the cloud fraction decreases. Moreover, while existing studies have used the near-surface soil temperature, such as the 5-cm soil temperature (T S05), to calculate T S05-T air, we find that T S00-T air and T S05-T air have opposite diurnal cycles, and their amplitudes differed drastically. The daily minimum of T S00-T air has a significant decreasing trend (−0.50±0.007°C/decade) from 2002 to 2022 due to T air increasing at a higher rate than T S00 during the nighttime. The occurrence frequency of near surface stable condition (T S00-T air<0) increases significantly, and the frequency of unstable condition (T S00-T air>0) decreases notably throughout the year except for winter. When it is stable, the magnitude of T S00-T air tends to decrease while the T S00-T air tends to increase when it is unstable, which is consistent with the drying condition caused by precipitation deficit. This study provides the first observational evidence on how T S00-T air responds to warming.