Long-term continuous observations of the horizontal inhomogeneity in lower-atmospheric water vapor concentration using A-SKY/MAX-DOAS

We conducted long-term (2017–2022) continuous observations of water vapor concentration in the lower atmosphere (0–1 km) over Tsukuba and Chiba, Japan, using the multi-axis differential optical absorption spectroscopy (MAX-DOAS) technique within the framework of the international Air Quality and Sky Research Remote Sensing Network (A-SKY). The accuracy of MAX-DOAS-derived lower-atmospheric water vapor concentration was validated against radiosonde measurements in Tsukuba (sample size = 1203), yielding a strong correlation (R = 0.971), thereby confirming the reliability of this method. Additionally, we demonstrated the capability of a four-azimuth-viewing MAX-DOAS system, comprising four A-SKY/MAX-DOAS instruments, to capture horizontal inhomogeneities in lower-atmospheric water vapor over Chiba. Analysis of the four directional data sets revealed a correlation between these inhomogeneities and atmospheric instability (increase in inhomogeneities during atmospheric instability). Under stable atmospheric conditions, the correlation coefficient between any two azimuths exceeded 0.95, whereas in unstable conditions, it decreased below 0.95 in all directions, occasionally dropping below 0.90. To further investigate this relationship, we identified 15 cases of significant horizontal inhomogeneity under unstable conditions, 10 of which coincided with the presence of a stationary front north of Chiba. Local analysis (LA) by the Japan Meteorological Agency indicated an inflow of warm, moist air toward the front, likely contributing to both the observed inhomogeneity and atmospheric instability. Additionally, while these inhomogeneities were captured by A-SKY/MAX-DOAS, they were not adequately detected in the lower-atmospheric LA, despite being within the error range of A-SKY/MAX-DOAS. This highlights the critical role of A-SKY/MAX-DOAS in monitoring lower-atmospheric inhomogeneities that conventional LA has underestimated.

Understanding atmospheric water vapor dynamics is essential for improving weather forecasts and numerical prediction models. In particular, accurately representing lower-atmospheric water vapor is crucial for enhancing the predictive accuracy of numerical models for mesoscale convective systems (MCSs; Houze 2004; Kato et al. 2003), which frequently lead to heavy rainfall. During intense precipitation events, such as the formation of Senjo-Kousuitai—a band-shaped heavy rainfall system extending 50–300 km in length and 20–50 km in width, driven by successive convective cell development (Kato 2020)—a continuous inflow of warm, moist air occurs below 1-km altitude toward the frontal region (Kato and Goda 2001; Kato, 2006). Numerical simulations have indicated significant horizontal inhomogeneity in lower-atmospheric water vapor during these events, underscoring the need for precise observational data to improve model performance.

Various methods have been employed to observe atmospheric water vapor. Key techniques include radiosondes, geostationary meteorological satellites (Bessho et al. 2016), the global positioning system (GPS; Shoji et al. 2004), microwave radiometers (MWRs), and water vapor Light Detection and Ranging (LiDAR). Radiosondes, MWRs, and water vapor LiDAR are particularly effective in obtaining vertical profiles in the planetary boundary layer (PBL). Radiosondes, the most widely used methods, provide direct atmospheric measurements but offer limited spatial coverage, with only 16 observation sites across Japan (Aerological Observatory 2021). MWRs primarily measure total precipitable water vapor (Ware et al. 2003), and recent advancements in deep learning, such as neural networks, have enabled the retrieval of vertical water vapor profiles from MWR data (Yan et al. 2020). Additionally, a one-dimensional variational technique has been applied to assimilate MWR observations into numerical weather prediction models, enhancing vertical profile accuracy (Araki et al. 2015). Water vapor LiDAR techniques, including Raman LiDAR (Melfi et al. 1969) and differential absorption LiDAR (Wulfmeyer and Bösenberg 1998), also provide vertical profiles (Kato et al. 2024; Yoshida et al. 2022).

Previous studies on water vapor inhomogeneity have primarily relied on data from radiosondes (Akiyama 1979), GPS (Bar-Sever et al. 1998; Aonashi et al. 2000; Shoji 2013), and MWRs (Bar-Sever et al. 1998; Aonashi et al. 2000). Akiyama (1979) reported intensive radiosonde observations conducted from 0 UTC on July 8 to 0 UTC on July 12, 1968, at five sites: Fukuoka, Kagoshima, and Naze (radiosonde stations in Kyushu, Japan), as well as Fukue and the research vessel Ryofu Maru in the East China Sea. Radiosondes were launched every six hours, providing detailed structural insights into the Baiu front through high temporal and spatial resolution observations at the synoptic scale. However, this observation network was limited to a short period, restricting its ability to capture long-term spatiotemporal variations. Therefore, a continuous multi-year observation network with higher spatiotemporal resolution is needed to better understand the role of water vapor inhomogeneity in MCS dynamics.

Bar-Sever et al. (1998) estimated the horizontal gradient of tropospheric path delay at the mesoscale using a single GPS receiver and validated their results against MWR-derived gradients. Aonashi et al. (2000) demonstrated a strong linear correlation between mesoscale inhomogeneity observed with GPS and MWR. Shoji (2013) introduced two indices representing mesoscale water vapor anisotropy, derived from gradient and inhomogeneity components calculated from GPS delays, revealing a link between horizontal water vapor inhomogeneity and convective precipitation. However, these methods primarily analyzed vertically integrated water vapor and faced challenges in estimating lower-atmospheric water vapor inhomogeneity. Aonashi et al. (2000) derived a precipitable water content gradient (Ruffini et al. 1999) from GPS under the assumption of a constant vertical gradient of relative humidity, which does not always hold in the real atmosphere. Additionally, discrepancies between the GPS-derived gradient and the spatially averaged gradient from GPS Earth Observation Network raise concerns about the reliability of different water vapor gradient estimations.

To directly observe lower-atmospheric water vapor distribution, three-dimensional (3D) GPS tomography has been proposed (Seko et al. 2004). Using a dense network of GPS receivers, Seko et al. (2004) estimated the 3D water vapor distribution associated with thunderstorms. Their method requires smoothing procedures based on radar data and the establishment of a background field derived from radiosonde observations. Recently, studies such as Miranda and Mateus (2021) and Saxena and Dwivedi (2023) have demonstrated the potential of using GNSS slant delays to reconstruct 3D water vapor fields without relying on auxiliary data. However, as noted in Bender et al. (2011), observation sites are preferably spaced at intervals of less than approximately 20 km, and particularly in order to accurately capture the distribution of water vapor in the lower troposphere, a dense GNSS receiver network is required.

The multi-axis differential optical absorption spectroscopy (MAX-DOAS) technique, employed within the international Air Quality and Sky Research Remote Sensing Network (A-SKY) as A-SKY/MAX-DOAS, is a unique, low-cost, passive ultraviolet (UV)–visible spectroscopy method capable of uncrewed, automatic, continuous observations. By directing the viewing angles of A-SKY/MAX-DOAS instruments toward different horizontal directions, it is expected that horizontal inhomogeneity in lower-atmospheric water vapor around an A-SKY/MAX-DOAS observation site can be captured.

While GNSS techniques require a dense receiver network, our method simply requires a set of four A-SKY/MAX-DOAS instruments with different observation azimuths at a single observation site. Therefore, our method could serve as an alternative or a complementary approach to GNSS techniques. Regardless of the chosen approach, given the role of lower-atmospheric water vapor in heavy rainfall disasters, further analysis of horizontal inhomogeneity through long-term, continuous PBL observations remains essential.

A-SKY/MAX-DOAS has been widely used in atmospheric chemistry research (e.g., Irie et al. 2008, 2011, 2015, 2016, 2019, 2021; Hoque et al. 2018a, b, 2022; Damiani et al. 2021, 2022, 2024; Ohno et al. 2022). Irie et al. (2011) pioneered the retrieval of water vapor vertical profiles using MAX-DOAS, enabling long-term continuous observations of water vapor in the PBL. However, the accuracy of A-SKY/MAX-DOAS water vapor measurements in the PBL has yet to be validated, and its potential for detecting horizontal inhomogeneity in water vapor remains unexplored.

This study aimed to conduct long-term continuous observations of horizontal water vapor inhomogeneity in the PBL and to evaluate the effectiveness of A-SKY/MAX-DOAS for water vapor measurements. To achieve this, A-SKY/MAX-DOAS was used to investigate the relationship between horizontal inhomogeneity in lower-atmospheric water vapor (0–1 km altitude) and atmospheric instability. From 2017 to 2022, continuous and simultaneous A-SKY/MAX-DOAS observations were conducted in Tsukuba (36.06° N, 140.13° E, 35 m a.s.l.) and Chiba (35.63° N, 140.10° E, 21 m a.s.l.), Japan. In Chiba, the four-azimuth-viewing MAX-DOAS (4AZ-MAXDOAS) system was employed, directing A-SKY/MAX-DOAS lines of sight north, east, south, and west. Although this system was originally designed to enhance the spatial representativeness of observations (e.g., Irie et al. 2021), this study utilized it to capture horizontal water vapor inhomogeneity. By leveraging A-SKY/MAX-DOAS-derived horizontal inhomogeneity data, this study examined the long-term relationship between lower-atmospheric water vapor variability and atmospheric instability over a six-year period.

Continuous ground-based observations were conducted using the A-SKY/MAX-DOAS system (Irie et al. 2008, 2011, 2015, 2019, 2021) over a 6-year period (2017–2022) at the Meteorological Research Institute in Tsukuba and Chiba University in Chiba, Japan (Fig. 1). The A-SKY/MAX-DOAS system, previously employed in various studies, is described below. The MAX-DOAS method is based on the well-established DOAS technique and applies the Lambert–Beer law (Platt and Stutz 2008). This method utilizes sunlight as a light source, and trace gas concentrations are accurately derived by analyzing their characteristic absorption spectral structures in high-wavelength-resolution hyperspectral measurements. Although these measurements are influenced by Rayleigh and Mie scattering from aerosols, such effects are minimized by approximating low-frequency variations with polynomials. The A-SKY/MAX-DOAS system consists of an outdoor telescope unit (PREDE Co., Ltd.) and an indoor high-resolution UV–visible spectrometer (Maya2000Pro, Ocean Insight Inc.) with a 25-μm slit. This spectrometer features 2048 detector channels and a wavelength resolution of 0.2–0.4 nm. The system’s field of view is estimated to be less than 1° (Irie et al. 2008).

Measurements were conducted in Tsukuba at three off-axis elevation angles (2°, 4°, and 8°) and in Chiba at five off-axis elevation angles (2°, 3°, 4°, 6°, and 8°), respectively. A reference angle of 70° was used instead of 90° to minimize signal variation across all angles while maintaining a constant integration time. To ensure a focus on aerosols and trace gases within the PBL, spectral observations were limited to low off-axis elevation angles (< 10°), thereby reducing sensitivity to altitudes above 3 km (Irie et al. 2015).

As reported by Irie et al. (2015), the validity of aerosol vertical profiles obtained using A-SKY/MAX-DOAS was confirmed through comparisons with cavity ring-down spectroscopy, LiDAR, and sky radiometer observations. The high precision and quality control of A-SKY/MAX-DOAS observational methods and algorithms were further demonstrated in two international MAX-DOAS intercomparison campaigns: the Campaign of Nitrogen Dioxide Measuring Instruments (CINDI; Roscoe et al. 2010) and CINDI-2 (Kreher et al. 2020), both held in Cabauw, the Netherlands.

Data retrieval was performed using the Japanese MAX-DOAS profile retrieval algorithm, version 2 (JM2; Irie et al. 2008, 2011, 2015, 2019, 2021). The JM2 algorithm integrates high-precision onsite wavelength calibration, DOAS fitting, a radiative transfer model accounting for atmospheric sphericity, and an optimal estimation method to derive the vertical distribution of aerosols and trace gases in the lower troposphere. In radiative transfer calculations, surface influence on the box-air-mass factor (A_box) is incorporated using a lookup table (Irie et al. 2008), with A_box values computed via the Monte Carlo Atmospheric Radiative Transfer Simulator (Iwabuchi 2006). The analysis workflow of the JM2 algorithm is illustrated in Fig. 2. This algorithm enables the retrieval of water vapor vertical profiles with a spatial representativeness of approximately 10 km along the line of sight (Irie et al. 2011).

Irie et al. (2011) evaluated the validity of MAX-DOAS water vapor observations, estimating the uncertainty for a single measurement at 12% for random errors and 18% for systematic errors, with a total uncertainty of 18%. Systematic errors were assessed through additional retrievals assuming aerosol retrieval uncertainties of 30%, though this estimate may be conservative as not all error sources were considered. Cloud-related uncertainties were consistently addressed through a cloud screening process. A comprehensive description of the A-SKY/MAX-DOAS system, including instrumentation and the JM2 algorithm, can be found in the studies by Irie et al. (2008, 2011, 2015, 2019, 2021). While other studies have employed MAX-DOAS for water vapor observations (Wagner et al. 2013; Lin et al. 2020; Ren et al. 2021), A-SKY/MAX-DOAS uniquely restricts off-axis elevation angles to below 10° to minimize the influence of higher altitudes.

In Chiba, four A-SKY/MAX-DOAS instruments were deployed simultaneously, oriented toward the north (13°W), west (95°W), east (118°E), and south (175°E). While this configuration was originally designed to improve the spatial representativeness of observations around Chiba (e.g., Irie et al. 2021), this study leveraged it to assess horizontal inhomogeneity in water vapor. Data from all four azimuth directions were analyzed to investigate this variability.

Both volume mixing ratios (percentage by volume, %v), commonly used in atmospheric chemistry, and mass mixing ratios (g/kg), standard in meteorology, were examined. The unit "%v" represents the proportion of water vapor volume relative to the total volume of moist air.

3.1 Validation of A-SKY/MAX-DOAS

Figure 3 presents the time series of daily average water vapor concentrations at 0–1 km altitude from 2017 to 2022, as observed by A-SKY/MAX-DOAS in four directions at the Chiba site. Water vapor concentrations exhibited a clear seasonal variation, with higher values in summer and lower values in winter, consistent with previous findings (Trenberth et al. 2005). Additionally, trends in water vapor concentrations were largely consistent across all directions.

Figure 4 compares A-SKY/MAX-DOAS water vapor observations with radiosonde data from Tsukuba (Tateno) for the 0–1 km altitude range, validating the accuracy of A-SKY/MAX-DOAS measurements. Radiosonde observations were conducted twice daily at 9 and 21 Japan Standard Time (JST; JST = UTC + 9 h). For validation, we used the 9 JST data, which coincides with A-SKY/MAX-DOAS measurement times. The two observation points were approximately 250 m apart (Fig. 1). As shown in Fig. 4, A-SKY/MAX-DOAS and radiosonde data exhibited similar variations. Considering the total uncertainty in A-SKY/MAX-DOAS data (Irie et al. 2011) and the error bars, most measurements showed good agreement. Figure 5 further illustrates these correlations, with correlation coefficients (R) of 0.971 for water vapor concentration, 0.970 for the mass mixing ratio, and 0.967 for the number density, confirming the robustness of A-SKY/MAX-DOAS retrievals against radiosonde data.

Figure 6 presents the correlation of daily average water vapor concentrations between each pair of the four directions at the Chiba site, totaling six pairs. As described earlier, A-SKY/MAX-DOAS observes water vapor concentrations over a spatial scale of approximately 10 km in each direction. The R values for all pairs were approximately 0.989, indicating strong consistency. However, slight directional variations in the distribution were observed. For example, water vapor concentrations in the north and west directions showed higher consistency, whereas pairs such as north–south and west–south exhibited lower consistency. The root-mean-square error (RMSE) was calculated for each directional pair, with all values around 0.1%v (∼0.8 g/kg). This suggests that horizontal inhomogeneity in water vapor cannot be overlooked. Prior studies have demonstrated that small variations in water vapor significantly impact precipitation forecasts. Kato et al. (2003) found that underestimating water vapor by 2 g/kg at 500-m altitude degraded forecast accuracy, while Yoshida et al. (2020) reported in their observing system simulation experiment that a 1 g/kg difference at 1 km altitude led to a 28% increase in predicted precipitation. Given these findings, the observed 0.8 g/kg variation underscores the importance of accounting for water vapor inhomogeneity using the 4AZ-MAXDOAS system.

3.2 Statistical analysis of water vapor inhomogeneity and atmospheric instability

The following section examines factors contributing to the horizontal inhomogeneity of water vapor at the Chiba site. Specifically, we analyze the relationship between horizontal lower-atmospheric water vapor inhomogeneity and atmospheric instability using A-SKY/MAX-DOAS observations in four directions from 2017 to 2022, combined with data from the Japan Meteorological Agency’s (JMA) MesoScale Model (MSM; Japan Meteorological Agency 2019). The MSM is a numerical weather prediction model with a 5-km grid spacing, providing forecasts eight times daily at 3-h intervals. Figure 7 explores the relationship between horizontal inhomogeneity in lower-atmospheric water vapor concentration, derived from A-SKY/MAX-DOAS observations at Chiba, and atmospheric instability, represented by the lifted index (LI), and, supplementarily, by the mixed-layer convective available potential energy (MLCAPE), both calculated from MSM. The LI is a convective parameter indicating atmospheric stability between the lower and mid-levels (Galway 1956). It is calculated by lifting an air parcel from the surface to the 500 hPa level and comparing the resulting parcel temperature to the ambient temperature at 500 hPa, with the value obtained by taking the temperature difference between them. The MLCAPE represents the buoyant energy available to an air parcel lifted from the mixed layer (May et al. 2022). MLCAPE was calculated based on potential temperatures, using a mixed-layer depth of 50 hPa, following previous studies (Lucas et al. 1994; Trier and Parsons 1995; Chuda and Niino 2005). A theoretical updraft speed was then estimated as \(\sqrt{2\text{MLCAPE}}\), assuming the complete conversion of this energy into vertical kinetic energy. As no upper-air observations were available near the Chiba site (35.63° N, 140.10° E), the LI and MLCAPE were derived from MSM at a nearby location (35.6° N, 140.125° E). The accuracy of this MSM-derived LI has been validated through comparison with radiosonde-derived LI at Tsukuba, showing a high correlation (R = 0.978) and an RMSE of 1.626.

In Fig. 7 (1), LI is plotted on the x-axis, while the correlation coefficient of water vapor concentrations between each directional pair in the 0–1 km altitude range is plotted on the y-axis. Only time periods without missing data for each pair were included in the correlation analysis. To minimize the impact of measurement errors, data points with relative errors exceeding 30% were excluded. Of the total 18,083 four-direction samples, 354 samples (1.96%) were removed. The results indicate that the correlation between water vapor concentrations worsens for all directional pairs when LI is below zero, signifying unstable atmospheric conditions and increased horizontal inhomogeneity of water vapor. This suggests that under unstable conditions, the spatial distribution of water vapor becomes more variable. Specifically, when the atmosphere was stable, the correlation coefficient exceeded 0.95 across all directional pairs. However, under unstable conditions, correlations dropped below 0.95 in all directions, with some cases showing values below 0.90. A similar trend was also observed from MLCAPE-derived updraft speed data instead of LI in Fig. 7 (2), reinforcing that the observed relationship does not depend on the specific choice of instability index.

From the analysis in Fig. 7 (1), we identified 15 cases where the difference between the westward and southward directions exceeded 0.5%v under negative LI conditions, with the correlation between these two directions exhibiting the most significant decline. This threshold was determined based on the distribution shown in Fig. 8, where the frequency of samples with a difference greater than 0.5%v is discrete, indicating that such cases are relatively rare. In these cases (LI ≤ 0), the extracted samples accounted for 0.58% of all cases, with water vapor concentration differences of 0.5 ± 0.8%v (∼3.1 ± 5.1 g/kg), or more were observed between the west and south directions, indicating substantial horizontal inhomogeneity. These events, characterized by atmospheric instability and pronounced spatial variability, occurred between June and September. Additionally, in 10 of these 15 cases, a stationary front was present within 500 km north of the Chiba site, suggesting that A-SKY/MAX-DOAS captured warm, moist air inflows toward the frontal region. The statistical analysis identified 10 instances where a stationary front was positioned north of Chiba.

3.3 Analysis of representative cases with water vapor inhomogeneity

The following section presents representative cases observed in this study. Among them, the event with the largest water vapor concentration difference between the west and south directions, exhibiting the most pronounced horizontal inhomogeneity, was observed at 9 JST on September 8, 2018, and is selected as the representative case.

Figure 9 presents the time series of water vapor concentrations observed by A-SKY/MAX-DOAS on September 8, 2018, with 9 JST—the time of interest—highlighted by a rectangle. Even after accounting for the measurement uncertainty of a single A-SKY/MAX-DOAS observation (∼0.3%v or ∼3.1 g/kg; Irie et al. 2011) and RMSE between two directions (∼0.7%v or ∼4.4 g/kg), the observed spatial inhomogeneity in water vapor concentration at 9 JST reached 1.0 ± 0.8%v (∼6.3 ± 5.0 g/kg), indicating a substantial difference. To further quantify this variation, we calculated the mean and standard deviation of water vapor concentrations in the west and south directions from 8 to 11 JST, encompassing the period of maximum spatial inhomogeneity at 9 JST. The results indicate that the water vapor concentration in the west direction was 2.0 ± 0.2%v (∼12.7 ± 1.2 g/kg), while in the south direction, it was 2.9 ± 0.1%v (∼18.6 ± 0.6 g/kg), reinforcing the presence of significant spatial inhomogeneity during this period.

Figure 10 presents the weather map and radar imagery for this case. A stationary front extended from the northeast to the southwest and was positioned approximately 150 km north of the Chiba site. Based on this, we hypothesize that the inflow of warm, moist air toward the stationary front increased the horizontal inhomogeneity of lower-atmospheric water vapor, leading to atmospheric instability. To verify this hypothesis, we analyzed the water vapor concentration field in greater detail using JMA’s local analysis (LA) data. The LA, with a 2-km grid spacing, represents the initial conditions for JMA’s Local Forecast Model (LFM; Japan Meteorological Agency 2019), which has the same resolution and provides 24 updates per day using Automated Meteorological Data Acquisition System and radar data for 10-h weather forecasts. Figure 11 displays the LA-derived distribution of water vapor concentration at an altitude corresponding to 950 hPa at 9 JST on September 8, 2018. The figure confirms the presence of warm, moist air inflow from the southwest toward the frontal region, suggesting that this inflow contributed to the pronounced horizontal inhomogeneity and atmospheric instability. A similar pattern of warm, moist air inflow toward the frontal region was observed in the other nine cases where a stationary front was identified north of the Chiba site, though variations in inflow strength and direction were noted. These findings indicate a consistent association between significant horizontal inhomogeneity in water vapor and atmospheric instability in such conditions. To assess whether the horizontal inhomogeneity detected by A-SKY/MAX-DOAS was similarly captured by the LA, we analyzed the LA data for all 15 identified cases. To ensure a fair comparison, we considered the spatial representativeness of A-SKY/MAX-DOAS, which is approximately 10 km. Following this, we extracted a corresponding region in the LA grid, marked by cross-symbols (×ばつ) in Fig. 11b, and computed the average water vapor concentration within this range. The maximum horizontal inhomogeneity between the west and south directions in the LA among the 15 cases was 0.2%v (∼1.2 g/kg), whereas for the specific case shown in Figs. 9, 10 and 11, the LA recorded an inhomogeneity of only 0.01%v (∼0.06 g/kg). While Fig. 11 indicates that the LA qualitatively captures horizontal inhomogeneity, its magnitude is significantly lower than that observed with A-SKY/MAX-DOAS. Given that A-SKY/MAX-DOAS recorded an inhomogeneity of approximately 1.0%v (∼6.3 g/kg) in Fig. 9, the LA appears to underestimate the extent of horizontal inhomogeneity. Even after accounting for measurement errors (∼0.3%v or ∼1.9 g/kg for a single A-SKY/MAX-DOAS direction and ∼0.4%v or ∼2.5 g/kg RMSE for two directions) and potential mismatches in observation ranges and directional coverage between LA and A-SKY/MAX-DOAS, the LA’s detection of horizontal inhomogeneity remains lower than expected. This suggests that while LA can identify spatial variations in water vapor, it may systematically underestimate their magnitude.

These findings highlight the importance of long-term, continuous observations of horizontal inhomogeneity in lower-atmospheric water vapor. Water vapor in the lower atmosphere plays a crucial role in precipitation formation, and localized variations can substantially impact the development of heavy rainfall. The statistical analysis of A-SKY/MAX-DOAS suggests that the inflow of warm, moist air toward stationary fronts enhances atmospheric instability near the front, increasing the likelihood of extreme rainfall events. However, even LA, the highest-resolution numerical weather prediction model operated by the JMA struggles to fully capture fine-scale inhomogeneity in lower-atmospheric water vapor. This underscores the value of A-SKY/MAX-DOAS, which can detect inhomogeneities that are underestimated in numerical models, making it an essential tool for atmospheric monitoring. For future research, expanding observational coverage is necessary to capture broader-scale inhomogeneities. Deploying four MAX-DOAS instruments as a single unit and extending observations in multiple spatial directions may enable the detection of large-scale water vapor variations. As a four-unit system has been confirmed to effectively measure water vapor inhomogeneity, the next step is to compare these results with MWR observations, as demonstrated by Aonashi et al. (2000). Unlike GPS tomography techniques, which require dense receiver networks of GNSS stations, this method offers a practical approach for widespread deployment. Ultimately, A-SKY/MAX-DOAS has the potential to enhance our understanding of the spatial structure of water vapor, improve data assimilation in numerical models, and contribute to early warning systems for heavy rainfall disasters.

In this study, we conducted long-term, continuous observations of lower-atmospheric water vapor concentrations in the 0–1 km range at Tsukuba and Chiba, Japan, from 2017 to 2022 using A-SKY/MAX-DOAS. The A-SKY/MAX-DOAS observations at Tsukuba exhibited an extremely high correlation with radiosonde measurements (R = 0.971), validating the reliability of our measurement method. We then analyzed the LI as an indicator of atmospheric stability and found that the horizontal distribution of water vapor became increasingly inhomogeneous under unstable atmospheric conditions. When atmospheric conditions were stable, the correlation coefficient between water vapor concentrations in different directions consistently exceeded 0.95. However, under unstable conditions, the correlation coefficient dropped below 0.95 in all pairs of different directions, with the values in some cases falling below 0.90. A detailed analysis of 15 cases with pronounced water vapor inhomogeneity and unstable atmospheric conditions revealed that in 10 cases, a stationary front was present north of the Chiba observation site. Further investigation using LA data from the LFM suggested that the inflow of warm, moist air from the southwest contributed to this inhomogeneity. However, when replicating the A-SKY/MAX-DOAS observation range in the LA data, we found that while the detected inhomogeneities were within the measurement margin of error, they were not fully captured by the lower-atmospheric LA. This finding underscores the importance of A-SKY/MAX-DOAS in detecting lower-atmospheric inhomogeneities that may be overlooked by numerical models. Building on these findings, future research will focus on deploying a four-unit MAX-DOAS system to capture broader-scale inhomogeneities, comparing measurements with MWRs, and ultimately improving data assimilation and early warning systems for heavy rainfall disasters.

The data are available upon request from the corresponding author (s.mizobuchi.chiba@gmail.com).

A-SKY:

Air Quality and Sky Research Remote Sensing Network

MAX-DOAS:

Multi-axis differential optical absorption spectroscopy

4AZ-MAXDOAS:

4-Different-azimuth-viewing multi-axis differential optical absorption spectroscopy

LiDAR:

Light Detection and Ranging

3D:

Three-dimensional

CINDI:

Cabauw Intercomparison campaign of Nitrogen Dioxide Measuring Instruments

JM2:

Japanese MAX-DOAS profile retrieval algorithm, version 2

JMA:

Japan Meteorological Agency

R :

Pearson correlation coefficient

n :

Sample size

RMSE:

Root-mean-square error

MCSs:

Mesoscale convective systems

UV:

Ultraviolet

GPS:

Global positioning system

MWRs:

Microwave radiometers

PBL:

Planetary boundary layer

JST:

Japan Standard Time

MLCAPE:

Mixed-layer convective available potential energy

MSM:

MesoScale Model

LI:

Lifted index

LA:

Local analysis

LFM:

Local Forecast Model

The authors are grateful to two anonymous referees and an editor for their instructive comments. The A-SKY/MAX-DOAS observations conducted in Tsukuba were supported by the Meteorological Research Institute of the Japan Meteorological Agency. Radiosonde data were obtained from the University of Wyoming’s database (https://weather.uwyo.edu/upperair/sounding.html). The mesoscale numerical weather prediction model GPV (MSM) data were collected and distributed by the Research Institute for Sustainable Humanosphere, Kyoto University (http://database.rish.kyoto-u.ac.jp/index-e.html). The authors would like to thank Enago (www.enago.jp) for the English language review.

This research was supported by the Environment Research and Technology Development Fund (JPMEERF20215005) of the Environmental Restoration and Conservation Agency of Japan, JSPS KAKENHI (JP22H03727 and JP22H05004), the JAXA 3rd research announcement on the Earth Observations (Grant No. 19RT000351).

Authors and Affiliations

1. Center for Environmental Remote Sensing, Chiba University, 1-33 Yayoicho, Inage-Ku, Chiba, 263-8522, Japan

   Shunya Mizobuchi & Hitoshi Irie

2. National Research Institute for Earth Science and Disaster Resilience (NIED), 3-1 Tennodai, Tsukuba City, Ibaraki, 305-0006, Japan

   Shingo Shimizu

Contributions

SM and HI designed the present study, performed the observation and analysis, and wrote the paper with support from all the authors. SS gave useful comments. All authors have read and approved the final manuscript.

Corresponding author

Correspondence to Shunya Mizobuchi.

Competing interests

The authors declare that they have no competing interests.

Publisher's Note

Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.

Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article's Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article's Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by/4.0/.

Reprints and permissions