Comparison of GNSS and INSAT-3D sounder retrieved precipitable water vapour and validation with the GPS Sonde data over Indian Subcontinent

Precipitable water vapour (PWV) plays a key role in the atmospheric processes from climate change to micrometeorology. Its distribution and quantity are critical for the description of state and evaluation of the atmosphere in NWP model. Lack of precise and continuous water vapour data is one of the major error sources in short term forecast of precipitation. The task of accurately measuring atmospheric water vapour is challenging. Conventional in situ measurements of atmospheric water vapour is provided by GPS Sonde humidity sensors profile twice a day at 0000 and 1200 UTC mainly from limited land regions. In recent years India Meteorological Department (IMD) is computing PWV from 19 channel sounder of INSAT-3D in three layers 1000-900 hPa, 900-700 hPa and 700-300 hPa and total PWV in the vertical column of atmosphere stretching from surface to about 100 hPa under cloud free condition. These data most commonly were validated using spatially and temporally collocated GPS Sonde measurements. In this paper, INSAT-3D satellite retrieved PWV data are validated with column integrated PWV estimates from a network of ground based Global Navigation Satellite System (GNSS) over Indian subcontinent. The PWV retrieved by INSAT-3D sounder platform is very promising, being in a good agreement with the GNSS data recorded over India for the period June, 2017 to May, 2018. The root-mean-square (rms) differences of 5.4 to 7.1 mm, bias of -4.7 to +2.1 mm and correlations coefficient of 0.79 to 0.92 was observed between INSAT-3D and GNSS PWV. The correlations coefficient between GPS Sonde and GNSS derived PWV ranges from 0.85 to 0.98.


Introduction
Variability in precipitable water vapour (PWV) directly influences the formation of clouds, atmospheric water transport, energy conversion and its transport over large distances through hydrological cycle. It is meaningful to study spatial and temporal variability of water vapour in the atmosphere. To quantify the amount of water vapour in the atmosphere, the PWV is defined as the height (mm) of the water if all of the water substance contained in a vertical column of unit horizontal cross section were condensed into liquid. It is also referred to as the total column water vapour. There are two main methods to study the atmospheric PWV; ground based and space based. The important ground based observation methods are GPS radiosonde, sun photometry, ground based Global Navigation Satellite System (GNSS) and microwave radiometry while space based method include COSMIC radio occultation, moderate-resolution imaging spectroradiometer (MODIS), SCIAMACHY, atmospheric infrared sounder (AIRS), INSAT-3D sounder etc. These methods of obtaining PWV information allowed substantial progress in understanding of the processes of water vapour occurring in the lower atmosphere. The radiosonde is considered to be a useful reference tool for atmospheric sounding. The conventional atmosphere sensing techniques such as GPS radiosonde, microwave radiometer and sun photometry have limitations in acquiring the continuous transformation of atmospheric PWV due to their low, inhomogeneous spatial and temporal resolution and high operational costs. The advantages of ground-based GNSS retrievals over conventional PWV retrieval methods are that it provides high precision and high temporal resolution data under all weather conditions at very economical cost (Nilsson and Gradinarsky, 2006;Jin et al., 2007). Space-based satellite sensors provide even wider coverage including the world ocean than regional ground-based networks. Infrared radiances are strongly affected by clouds and because of this satellite retrieval of PWV requires cloud-free conditions. However, the satellite retrieved PWV observations must be inter-compared against reliable measurements in order to assure their quality.
The concept of GNSS meteorology was first introduced by Bevis et al. (1992) and PWV data were estimated from Global Positioning System (GPS) observations. As the basis of GNSS meteorology, the electromagnetic signal from a GNSS satellite suffers a delay as it passes across the atmosphere, partially as result of the presence of water vapour. The zenith total delay can be converted into PWV using surface pressure and temperature measurements (Bevis et al., 1994). Remarkable progress has been achieved in recent decades and the accuracy of the GNSS-derived PWV has been proved to be better than 1.5 mm using independent GNSS observations (Duan et al., 1996;Tregoning et al., 1998). Several studies have shown GNSS derived PWV distribution as a useful tool for monitoring severe weather events (Brenot et al., 2006(Brenot et al., , 2013Barindelli et al., 2018;Bonafoni and Biondi, 2016;Calori et al., 2016). A number of researchers have studied the variation in atmospheric water vapour in relation with the onset of southwest monsoon and rainfall. Rao et al. (2013) have studied the variations in refractivity during southwest Monsoon using GPS radio occultation. They have reported an appreciable increase in the refractivity difference between 600 and 100 hPa well before the arrival of Southwest Monsoon coupled with a dip at the time of onset in conformity with COSMIC GPS RO observations. Puviarasan et al. (2015) observed that the time series of PWV estimated from GPS increases to more than 60 mm when Southwest monsoon approaches and sharply decreases on withdrawal at the same location. He also recommended, GPS PWV data is more reliable tool for declaring monsoon arrival and withdrawal than using the satellite (INSAT) derived moisture, particularly over the land region. A dense network of continuously operating ground based GNSS receiver if deployed can accurately provide the progress as well as the activeness of Southwest monsoon on real time. Simon et al., 1994, have studied changes in western Indian Ocean moisture in the middle troposphere, upper troposphere and scale height of water vapour during the period of Monsoon Onset over Kerala (MOK) and the moisture and scale height of water vapour increased about 8 days prior to the onset of Southwest Monsoon over Kerala. Simon et al. (2006) observed favourable conditions leading to the MOK using Tropical Rainfall Measuring Mission TRMM/TMI derived total water vapour build up over the Western Arabian Sea by about two and half weeks. Singh et al. (2004) observed that amount of PWV over the Arabian Sea and the Indian Ocean is well correlated with onset day of Southwest monsoon over the Indian sub-continent. The maximum value of PWV in the 10-15° N latitude zone is to appear 7-12 days prior to the monsoon onset at Kerala coast. Jade & Vijayan (2008) observed that precise measurements of water vapour over the Himalayas and Tibetan plateau are helpful to investigate the land atmosphere interactions and their effect on the Asian monsoon circulation in terms of driving forces and feedback mechanisms.
The descriptions of retrieval techniques for the three instruments (GNSS, GPS Sonde and satellite) PWV are given in the present work. The objective of the paper is to study the accuracy of the GNSS-derived PWV dataset by comparison with that from GPS Sonde and INSAT-3D sounder retrieved PWV using data for the period June, 2017 to May, 2018.

Data and methodology
India Meteorological Department (IMD) established a network of 25 ground based Global Navigation Satellite System (GNSS) receivers collocated with meteorological sensors for continuous monitoring of tropospheric total integrated precipitable water vapour (PWV) for weather events monitoring (Fig. 1). GNSS data were processed using the Trimble Pivot Platform (TPP) software to estimate PWV at every 15 minutes interval. The space and  (Niell, 1996) were used to map the signal path delay from zenith direction to other elevation angels.

GNSS Meteorology
The fundamental observable of GNSS is the signal propagation time from a satellite to a receiver. Multiplying the propagation time by the speed of light in vacuum gives the pseudo range between the satellite and the receiver. The pseudo range is mix of various errors such as clock errors both on receiver and satellite, atmospheric error, multipath error etc. For geodetic applications higher accuracy (millimetre level), carrier phase measurements are necessary because the phase can be measured to 1% of the wavelength of the carrier signal. A carrier phase measurement in units of length can be expressed as: where, Θ is the carrier phase observable, D is the true distance between the satellite and the receiver; c is the speed of light in vacuum, T r and T s are the receiver and satellite clock offsets which can be eliminated by double where, (2), 99.9% of ionospheric delay is calculated accurately and can be eliminated.
ΔL, the total delay is sum of zenith hydrostatic delay and zenith wet delay given by (Bevis et al., 1992;Bevis et al., 1994;Davis et al., 1985) in the neutral atmosphere given by: The zenith hydrostatic delay (ZHD) in mm can be estimated by measuring surface pressure P s (in hPa) as follows: where, f (ϕ, H) accounts for the variation in gravitational acceleration with latitude ϕ and the height H of the surface above the ellipsoid (in kilometres). The ZWD is obtained by subtracting the ZHD from ZTD, the PWV (mm) estimates were then derived by scaling the ZWD with the multiplication factor Π is given by: where, ρ is the density of liquid water, R v is the specific gas constant for water vapour, k 3 and k 2 ' are physical constant. The water-vapour weighted mean temperature T m of the atmosphere is defined and approximated as (Wang et al., 2005): where P v and T represent vapour pressure and absolute temperature along the zenith path, respectively and the integral intervals are from the earth surface to the atmospheric top. T m can be obtained from radiosonde vertical profile or from NWP analysis. Assuming a linear relation with surface temperature it is also possible to approximate T m from station surface temperature (T s ) as follows: T m = 55.8 + 0.77T S Error due to T m : Water vapour weighted vertically averaged mean temperature of the atmosphere T m is an important parameter of the relationship between PWV and the zenith wet delay as the accuracy of GPS estimates of PWV is directly related to the accuracy of T m . The watervapour weighted mean temperature of the atmosphere is defined and approximated as follows: Since k 2 /k 3 is small (~5.9 × 10 -5 K -1 ), the relative error of PWV approximately equals to that of T m (Bevis et al., 1994). On the basis of above equation, for T m values 240 K to 300 K, the 1% and 2% accuracies in PWV require errors in T m less than 2.74 K and 5.48 K on average respectively (Wang et al., 2005).

Mapping functions :
In GNSS data processing, the slant path delay is converted to the equivalent ZTD (sum of the ZHD and the ZWD) using hydrostatic and wet Mapping Functions (Niell, 1996) where, θ is the elevation angle seen from the ground antenna to the satellite.
The remaining term n λ and ε in equation (1) is integer ambiguity and sum of other unmodelled errors, e.g., signal multipath, antenna phase centre variations and radome effects can be estimated and removed from the long term observations.

Precipitable water from GPS Sonde
Precipitable water vapour content based on GPS Sonde observation can be estimated using the formula of Curry and Webster (1999).
where, 'g' is the average gravity acceleration of tropospheric air column, ρ w is the density of atmospheric water vapour, ∆p (hPa) is the difference in pressure between layers and q i is the discrete specific humidity measured from sonde observation. Three types of GPS sonde are being used in IMD (i) DFM-09 GPS sonde (ii) CF-06A GPS sonde (iii) RSG20A GPS sonde (Table 1). Several factors including dry bias, time lag, calibration error and sensor drift may introduce errors in the PWV estimation from radiosonde data (Miloshevich et al., 2009). It is assumed that measured pressure, temperature and dew-point depression were obtained along a strictly vertical ascent (Niell et al., 2001), although observational evidence suggests that radiosonde ascents are rarely  (6) 3.67-4.59 10 × 10 km MWIR(5) 6.38-11.33 10 × 10 km LWIR (7) 11.66-14.85 10 × 10 km strictly vertical because winds are generally present at all altitudes. Usually, radiosondes are expected to produce PWV with an uncertainty of a few millimeters, which is considered to be the accuracy standard of PWV for meteorologists (Niell et al., 2001).

Precipitable water vapour from INSAT-3D sounder
Sounding system of the satellite provides vertical profiles of temperature (40 levels from surface to ~70 km), humidity (21 levels from surface to ~15 km) and integrated ozone from surface to top of the atmosphere. Vertical profiles of temperature and moisture can be derived from radiances in 19 channels, using the first guess from NWP data. INSAT-3D sounder channels brightness temperature values are averaged over a number of field of view (FOVs) prior to application of retrieval algorithm. Based on this, retrieval algorithm has option for retrieving the vertical profiles at 30 km (3 × 3 pixels) and 10 km resolution (each pixel). The Sounder has eighteen narrow spectral channels in shortwave infrared, middle infrared and long wave infrared regions and one channel in the visible region. The ground resolution at nadir is 10 × 10 km for all nineteen channels. Specifications of sounder channels are given in Table 2.
As INSAT-3D PWV is sensitive to the presence of clouds in the field of view (limitation of Infra-red sounder sensors), hence the PWV values collected under clear sky conditions were used in this study. Atmospheric profile retrieval algorithm for INSAT-3D Sounder is a two-step approach. The first step includes generation of accurate hybrid first guess profiles using combination of statistical regression retrieved profiles and model forecast profiles. The second step is nonlinear physical retrieval to improve the resulting first guess profile using Newtonian iterative method. The retrievals are performed using clear sky radiances measured by sounder within a 3 × 3 field of view (approximately 30 × 30 km resolution) over land for both day and night (INSAT-3D ATBD, 2015). The GNSS and INSAT-3D retrieved PWV values are matched at every hour. Four sets of regression coefficients are  where, "g" is the acceleration of gravity, p 1 = surface pressure, p 2 = top of atmosphere pressure (i.e., about 100 hPa beyond which water vapour amount is assumed to be in negligible). Unit of precipitable water is mm depth of equal amount of liquid water above a surface of one square meter. IMD is computing PWV from 19 channel sounder of INSAT-3D in three layers i.e., 1000-900 hPa, 900-700 hPa, 700-300 hPa and total PWV in the vertical column of atmosphere stretching from surface to about 100 hPa during cloud free condition. Monsoon, severe

Validation of PWV derived from ground based GNSS with GPS sonde
To evaluate the accuracy of the GNSS retrieved PWV across mainland India, the comparison are carried out with GPS sonde derived PWV using collocated 9 stations from June, 2017 to May, 2018. The GNSS and  (Table 3) is less than 4 mm except for Thiruvananthapuram (7.2 mm) and Bhubneshwar (4.94 mm). RMS differences between GPS Sonde and GNSS values of PWV are approximately less than 5 mm ( Table 3). The relatively larger PWV difference is observed at the coastal stations of India.  Table 4. The RMSE ranges from 5.41 to 7.14 mm. The bias between GNSS and INSAT-3D sounder values of PWV is less than 1 mm except for New Delhi (1.92 mm) and Jalpaiguri (2.14 mm). Fig. 5 shows Taylor diagram for different stations which provides standard deviation, correlation coefficient and RMS error between observed GNSS and

Distribution and variability of PWV
The PWV values depict strong seasonality; reaches maximum in monsoon season and minimum in the winter   May, 2018. The PWV value is largest in the coastal region of India due to sufficient moisture supply from the sea throughout the year. The annual mean values of PWV ranges from 43.4 to 54.7 mm at coastal sites. The PWV value is also high in north-east India. The lowest value of PWV of order less than 30 mm occurred over North India. The time series of the PWV values derived using GNSS and INSAT-3D sounder is given in Fig. 8. The annual variation of PWV based on seasons is clearly seen at all the stations with a peak during the monsoon season.

Conclusions
The present study has been carried out for GNSS stations of India Meteorological Department for the period of June, 2017 to May, 2018. The comparison of PWV retrieved from GNSS and GPS Sonde reveal the correlation coefficient varying from 0.98 at Nagpur and 0.85 at Raipur. The bias between the GPS Sonde and GNSS values of PWV is less than 4 mm except for Thiruvananthapuram and Bhubaneswar. The root mean square error between GPS Sonde and GNSS values of PWV are less than 5 mm at almost all the stations. The relatively larger PWV difference is observed at the coastal stations of India. The comparison of PWV retrieved from GNSS and INSAT-3D sounder exhibit a root mean square error ranging from 5.41 to 7.14 mm and a correlation above 0.79. The PWV distribution from GNSS stations across mainland India is also analysed. It was found that the largest PWV value occurs in the coastal region of India and in general PWV value shows a decreasing trend from south to north and increasing trend from west to east.
The comparison of PWV time series derived from GNSS with that from GPS sonde and INSAT-3D sounder data shows that the accuracy of the calculated PWV values is high. India is an ideal place for investigation of regional PWV responses to the climate change. The ground-based GNSS derived PWV will become an important independent data source in climate monitoring as the length of the time series grows. The GNSS-derived PWV value can be useful for investigation of weather forecasting and nowcasting in particular, precipitation forecasting. The simulation experiments should be carried out to assess the quality improvement of weather forecasts when GNSS derived PWV data are incorporated in models in order to fruitfully exploit the potential possibilities.