Water vapor study using MODIS and GPS data at 64 continuous GPS stations (2002–2017) in indian subcontinent

Journal of Atmospheric and Solar–Terrestrial Physics 196 (2019) 105138

Contents lists available at ScienceDirect

Journal of Atmospheric and Solar-Terrestrial Physics journal homepage: www.elsevier.com/locate/jastp

Research Paper

Water vapor study using MODIS and GPS data at 64 continuous GPS stations (2002–2017) in indian subcontinent Sridevi Jade a, *, T.S. Shrungeshwara a, Boddapati Anil b a b

CSIR-4PI, CSIR Fourth Paradigm Institute (Formerly CSIR-CMMACS), Wind Tunnel Road, Bangalore, 560 037, India Real Time Governance State Centre, Velagapudi, Andhra Pradesh, 522503, India

A R T I C L E I N F O

A B S T R A C T

Keywords: MODIS (moderate resolution imaging spectroradio-meter) GPS (global positioning system) PWV (precipitable water vapor) Spatial variability function

Precipitable Water Vapor (PWV) is estimated using MODIS (Moderate Resolution Imaging Spectro-radio-meter) Terra level 3 data with daily resolution i. e MOD08_D3 at 64 cGPS (continuous Global Positioning System) stations spatially spread over Indian subcontinent between geodetic latitude 5� to 35� N and geodetic longitude of 70� to 96� E. MODIS-PWV is compared with GPS-PWV estimated at these cGPS stations to check the validity of water vapor retrieved from MODIS data in Indian subcontinent. Correlation coefficient (R2) between daily values of MODIS and GPS water vapor is above 0.9 with RMSE (root mean square error) of 2–5 mm for 22 cGPS in peninsular India, above 0.9 with RMSE of 3–6 mm for 5 cGPS in northeast India and above 0.8 with RMSE of 1–9 mm for 26 cGPS in Himalayas. PWV time series at all the cGPS stations indicated distinct seasonal cycle for both MODIS and GPS PWV with high RMSE (~6 mm) in wet months and low RMSE (~3 mm) during dry months. Taking advantage of broad spatial spread of stations and long span of data, model for spatial variability of GPSPWV for Indian subcontinent is proposed. Inter-annual and seasonal variability of GPS-PWV is discussed in detail for peninsular India, northeast India and Himalayas.

1. Introduction PWV is the term used for total integrated water vapor in the air column between the surface and the atmosphere in the zenith direction. Water vapor has high spatio-temporal variation especially in tropics with distinct zonal and vertical distribution. PWV is very sensitive near lower troposphere due to the presence of Atmospheric Boundary Layer (ABL) which is covered by a thermal inversion layer. The air below the inversion layer is moist and cloudy whereas the air above the inversion layer is cloud free and dry. Water vapor is dominant green house gas of the atmosphere and study of its spatio-temporal variation is very important to understand the distribution of clouds, rainfall, atmospheric storm systems, chemical reactions that occur in atmosphere and earth’s energy balance. Vertical and horizontal distribution of water vapor in the atmosphere is measured using balloon-borne radiosonde in­ struments, upward looking ground based water vapor radiometers, satellite remote sensing and GPS. In the recent years, dedicated meteo­ rological satellite sensors such as MODIS on Terra, Atmospheric Infrared Sounder (AIRS) on Aqua and several others give water vapor estimation at high spatial and temporal resolution (Gao et al., 1990; Kaufman and

Gao, 1992; Gao and Kaufman, 2003; King et al., 1992, 2003; Parkinson, 2003). MODIS 1� � 1� global grid data are routinely used for atmo­ spheric, biological, monsoon and climate related studies and assimilated in global circulation models. Water vapor products of MODIS are important constituents of global climate studies. GPS signal propagation through atmosphere is effected due to ionosphere and troposphere refraction which in turn is used to estimate total electron content (Jade and Shrungeshwara, 2018) and PWV in the atmosphere. GPS PWV estimated from the tropospheric delay suffered by GPS signals from the GPS satellite to the receiver on the ground (Bevis et al., 1992, 1994) is widely used due to its all weather coverage, high temporal resolution, low cost, easy installation and operation. MODIS PWV was validated in several regions globally using GPS PWV estimates (Li et al., 2003, 2005; Liu et al., 2006; Lu et al., 2011; Thomas et al., 2011; Bennouna et al., 2013; Chang et al., 2015; Vaquero-Martinez et al., 2017; Wang et al., 2017; Gurubuz and Jin, 2017). For the first time in Indian subcontinent, GPS based atmospheric water vapor studies were initiated in 2004 (Jade et al., 2005; Jade and Vijayan, 2008) and GPS PWV is estimated at several cGPS (continuous Global Positioning System) stations using both observed and low

* Corresponding author. E-mail addresses: [email protected] (S. Jade), [email protected] (T.S. Shrungeshwara), [email protected] (B. Anil). https://doi.org/10.1016/j.jastp.2019.105138 Received 11 March 2019; Received in revised form 16 August 2019; Accepted 30 August 2019 Available online 1 October 2019 1364-6826/© 2019 Elsevier Ltd. All rights reserved.

S. Jade et al.

Journal of Atmospheric and Solar-Terrestrial Physics 196 (2019) 105138

resolution 2.5� � 2.5� grid NCEP (National Centers for Environmental Prediction) derived meteorological parameters for a period of 4 years (2001–2004). Detailed comparison of PWV, surface pressure and tem­ perature values using observed and NCEP data was given. Spatial and seasonal variability of water vapor in Indian subcontinent were dis­ cussed based on these estimates. Kumar et al. (2009) gave comparison of MODIS and GPS PWV for two cGPS stations in Indo-Gangetic Plain for a period of one year. Validation of PWV extracted from MODIS Terra, Atmospheric Infrared Sounder (AIRS), Aerosol Robotic Network (AER­ ONET), NCEP was carried out using GPS PWV data (Prasad and Singh, 2009) at three (Bangalore, Hyderabad and Kanpur) cGPS stations. They concluded that MODIS near Infrared column product shows higher correlation of about 91% with GPS for these three regions with a sys­ tematic bias during wet and dry months which can be corrected as these are used for assimilation in to numerical weather prediction and global climate models. Using limited GPS PWV data at Varanasi and Kanpur continuous stations during 2007–2008 (Kumar et al., 2013), gave a high correlation (0.98) with daily MODIS near infrared water vapor with clear column. Seasonal correlation between the two products varies from 0.84 to 0.97 for limited data with no systematic seasonal bias. Joshi et al. (2013) gave comparison between MODIS NIR clear column and

GPS PWV for three years (2009–2011) at Almora GPS station located in Central Himalaya at altitude of 1259 m msl. Annual correlation is about 91% and seasonal correlation varies between 62 and 87% with sys­ tematic bias. Daily PWV values for the three year period give bias of 4.53 mm and RMSE of 4.06 with high RMSE and bias during monsoon season. MODIS PWV estimates at a high altitude (~4325 m) cold desert Hanle site located in trans-Himalaya are validated (Ningombam et al., 2016) using 8 years GPS data during 2005–2012. It was found that MODIS data is underestimating PWV during summer season due to dry terrain. Inter-annual, seasonal and diurnal variability of water vapor was discussed in detail for the Hanle region along with the correlation of PWV with rainfall and temperature. Spatial and temporal variability of GPS PWV for northeast India (Barman, 2016; Barman et al., 2017) during 2004–2012 was studied in detail to understand the effect of several processes on the water vapor variability. Correlation between rainfall and PWV was given for the northeast region which receives the highest rainfall in Indian subcontinent. Further (Ningombam et al., 2018), evaluated existing eleven empirical models to estimate water vapor for high altitude Hanle site using the 8 years of GPS data. In addition, moist parameters such as water vapor scale height, dew point temperature and water vapor pressure were examined in depth for the

Fig. 1. cGPS sites used for the water vapor study in Indian subcontinent with 1–13 years of data during 2002–2017. Average PWV at each site is denoted by column. GPS sites with collocated meteorological sensors are given in bold letters. 2

S. Jade et al.

Journal of Atmospheric and Solar-Terrestrial Physics 196 (2019) 105138

cold desert high altitude environment. In this study, we use cGPS data from 64 sites spatially spread over the Indian subcontinent between geodetic latitude 5� to 35� N and geodetic longitude of 70 to 96� E to validate the daily and seasonal MODIS PWV estimates during 2002–2017 with data span of 1–13 years. For the first time MODIS level 3 daily water vapor is validated using cGPS data of 64 sites with long data span than any earlier study and thus gives an op­ portunity for in-depth analysis. Previous studies on GPS water vapor variability were limited to few stations, region specific and short span of data. Spatial, seasonal and inter annual variability of GPS derived water vapor over the Indian subcontinent is discussed in detail during the 15 year period.

the ERA-Interim data set and ΔZ is the difference of geo-potential heights of two consecutive pressure levels. GPS-PWV at each site is calculated at 2 h interval from ZTD as detailed above using observed meteorological parameters (if available) and at 6 h interval using interpolated surface pressure and temperature values. Daily average GPS PWV is calculated from average of PWV values for each day. Several studies specific to Indian Subcontinent (Jade et al., 2005; Jade and Vijayan, 2008; Ningombam et al., 2016; Barman et al., 2017) indicate that the accuracy of GPS PWV estimates depend on the uncertainty in estimating ZTD, accuracy of interpolated surface pressure and temper­ ature values, weighted mean temperature of the atmosphere which is required to estimate PWV from ZWD.

2. Data and methods

2.2. MODIS data and analysis

2.1. GPS data and analysis

MODIS is mounted on-board the Earth Observing System (EOS) polar orbiting terra satellite which passes over equator twice in each day giving global scale coverage every 1–2 days. MODIS has moderate spatial resolution (10 � 10), wide swath (2330 km) and large spectral range (0.4–14.2 μm). PWV is derived using an algorithm (Kaufman and Gao, 1992; Gao and Kaufmann, 2003) based on observations of water vapor attenuation of near-IR solar radiation reflected by surfaces. Techniques using ratio of three water absorption channels (0.905, 0.936, 0.940 μm) and two atmospheric window channels (0.865, 1.24 μm) are used in the algorithm to determine atmospheric water vapor transmittance after removing surface reflection variation with wavelengths. Water vapor column is derived from transmittance based on radiative transfer theory calculations and using look up table pro­ cedures. MODIS-PWV derived using the algorithm detailed above has an error bar of 5–10% as reported by Gao and Kaufmann (2003). MODIS level 2 pixel based near-IR water vapor products and daily, 8-day and monthly level 3 near-IR water vapor products at 1� � 1� global grid are now routinely produced and archived. For the present study, MODIS Terra level 3 near-IR clear column daily surface water vapor products (MOD08_D3) version 5.1 is used (https://ladsweb.modaps.eosdis.nasa. gov/search/). MODIS level 3 daily surface water vapor products are used for study of annual and seasonal variation of water vapor over regional and global scales and hence chosen for the present study. MODIS PWV for each GPS station is obtained by horizontally interpo­ lation of the surface PWV values of four surrounding (1� � 1� ) MODIS grid points as given below

Data span and details of GPS data with 30 s interval of about 64 cGPS stations (Fig. 1) used for PWV estimation is given in Table 1. Zenith Total Delay (ZTD) suffered by GPS signals passing through atmosphere is obtained by phase processing of GPS data using GAMIT/GLOBK version 10.4 developed at MIT (Herring et al., 2010a, b). ZTD is estimated at 2 h interval using GPS observations with elevation cut-off angle of 10� to avoid multipath. ZTD is a the sum of Zenith Hydrostatic Delay (ZHD) and Zenith Wet Delay (ZWD) and is given as (1)

ZTD ¼ ZHD þ ZWD

Based on the Elgered et al. (1991) model, Zenith Hydrostatic delay is expressed as a function of surface pressure (Ps), latitude (λ) and ellip­ soidal height (h) as given below ZHD ¼

ð1

ð2:2779 � 0:0024ÞPs 0:00266 cosð2λÞ 0:00028 h Þ

(2)

ZWD is estimated as (ZTD-ZHD) and GPS-PWV is derived from ZWD (Askne and Nordius, 1987) as given below �

PWV ¼ Π � ZWD ​ ; ​ П ¼

ρ Rv

106

k3 Tm

�

(3)

þ k2’

where ρ is the density of liquid water, Rv is specific gas constant of water vapor, k’2and k3 are the atmospheric refraction constants and Tm is the weighted mean temperature of the atmosphere. Empirical relations to derive Tm from surface temperature is given by several earlier studies (Davis et al., 1985; Bevis et al., 1992, 1994; Mendes et al., 2000; Solbrig, 2000; Schüler et al., 2001). All these mean atmospheric temperature (Tm) models are evaluated for Indian subcontinent (Jade et al., 2005) and concluded that all the models agree within �1 mm deviation. We have used the empirical relation Tm ¼ 70.2 þ 0.72 T given by Bevis et al. (1992) for sites with collocated meteorological sensors measuring sur­ face temperature and pressure. П is the ratio of PWV/ZWD and is roughly about 0.15 (Bevis et al., 1994) with 20% variation depending on site location, height, weather and seasons. Since all the cGPS sites are not equipped with collocated meteoro­ logical sensors, we use surface temperature and pressure values inter­ polated to site location for all the cGPS sites using the interpolation scheme detailed by Schüler (2001) from ERA-Interim data set with high resolution of 0.125� � 0.125� with 37 vertical levels. For ERA-Interim data set, mean atmospheric temperature (Tm) is calculated from the vertical profiles (Davis et al., 1985) as given below Pn Pvj R Pv dZ j¼1 Tj ΔZj T Tm ¼ R P v (4) � Pn Pvj dZ 2 T j¼1 T2 ΔZj

PWV ¼

X wj PWVj ¼ w1 PWV1 þ w2 PWV2 þ w3 PWV3 þ w4 PWV4

(5)

Horizontal weights at each grid point (j ¼ 1,2,3,4) are calculated as wj ¼

wj w1 þ w2 þ w3 þ w4

(6)

The weighting coefficient wj for each grid points is defined as a function of the spherical distance between the respective grid points and � C position of the antenna given as wj ¼ Rωj . Where, R is the mean radius of the Earth, ω is the spherical distance between the respective grid points and position of the antenna (λ, φ), C is the weighting power (1 � C � 2). The spherical distance ω is calculated using spherical trig­ onometry as given below. � � � � cos ωj ¼ sin λj ⋅sinðλÞ þ cos λj ⋅cosðλÞ⋅cos φj φ (7) MODIS-PWV for each site is calculated as detailed above using the daily surface water vapor values extracted from MOD08_D3 (1� � 1� ) grid data. 3. Results and discussions

j

We have chosen cGPS sites with minimum 1 year of data for PWV estimation (Fig. 1, Table 1). Eight GPS sites (IISC, BAN2, IAOH, RSCL, GHTU, TZPR, BOMP and LHAZ) located in peninsular India, Ladakh,

where, Pv is partial water vapor pressure (in hPa), T is the atmospheric temperature (in K). The summation is done for each pressure levels of 3

S. Jade et al.

Journal of Atmospheric and Solar-Terrestrial Physics 196 (2019) 105138

Table 1 Details of cGPS sites used for the water vapor study along with data span. * denotes the cGPS sites with collocated meteorological sensors. Bias, RMSE and correlation coefficient R2 between daily value of MODIS and GPS PWV along with seasonal Bias, RMSE for the cGPS sites during 2002–2017 is given in the table. Geodetic Longitude (o E)

MSL Height (meters)

Peninsular Indian cGPS stations Delhi DELH 28.5 Kharagpur IITK 26.5 Lucknow LUCK 26.9 Lucknow LCKI 26.9 Lucknow LCK2 26.9 Varanasi BHUP 25.3 Khavda KHAV 23.9 Radhanpur RADP 23.8 Bela Temple BELP 23.9 Mount Abu MABU 24.7

77.1 80.2 80.9 81.0 81.0 83.0 69.8 71.6 70.8 72.8

284.2 143.6 153.0 118.4 118.5 93.6 103.4 31.0 42.6 1679.6

532 429 1170 964 989 423 535 549 646 384

2.5/4.2 2.8/5.2 2.8/4.6 2.1/3.9 2.1/3.9 0.3/2.7 1.8/4.9 2.6/4.2 1.1/8.3 12.0/15.1

0.97 0.94 0.96 0.96 0.96 0.97 0.94 0.95 0.78 0.91

Dharoi Gandhinagar Udaipur Bhopal Jabalpur Dhanbad Durgapur Bhubaneswar Bombay Pune Hyderabad Bangalore* Bangalore* Kodaikanal

24.0 23.2 24.6 23.2 23.1 23.8 23.5 20.3 19.1 18.6 17.4 13.0 13.0 10.2

72.8 72.7 73.7 77.4 79.9 86.4 87.3 85.8 72.9 73.9 78.6 77.6 77.5 77.5

197.4 72.6 576.9 495.0 404.4 244.2 114.8 49.5 64.4 566.6 518.7 929.9 918.0 2339.4

467 608 330 705 327 433 569 1675 364 558 4333 4962 2294 2986

2.2/3.7 2.2/3.9 1.3/2.6 2.5/3.7 1.6/2.7 1.1/3.3 1.3/3.8 1.1/3.8 0.3/3.4 1.9/3.4 2.1/3.4 2.8/4.2 3.3/4.7 11.8/15.9

0.96 0.94 0.98 0.97 0.98 0.94 0.95 0.91 0.93 0.95 0.95 0.95 0.96 0.74

Trivandrum TVRM Colombo SGOC Maldives MALD Portblair PBR2 North East cGPS stations Tezpur* TZPR Guwahati* GHTU Shillong CSOS

8.4 6.9 4.2 11.6

77.0 79.9 73.5 92.7

79.5 19.1 5.5 39.1

663 628 912 1422

0.8/2.9 2.7/5.4 0.9/4.5 3.0/6.3

0.77 0.72 0.29 0.71

26.6 26.2 25.6

92.8 91.7 91.9

130.1 62.1 1599.6

3278 2753 1065

3.1/6.2 1.8/4.4 9.2/12.2

0.88 0.92 0.90

Shillong

25.6

91.9

1538.1

502

8.3/11.4

0.90

1.7/3.4 1.4/3.1 9.8/ 10.2 9.2/9.5

Lumami LUMA Imphal IMPH Aizwal AIZW Himalayan cGPS stations Panamik* PAN2 Leh* RSCL Hanle IAOH Naddi NADI Kothi KOT1 Jhiri GBKL Bhatwari BHTW Dehradun DEHR Dehradun WIH2 Nagoli GBSN Almora GBPK Nainital GBNL

26.2 24.7 23.7

94.5 93.9 92.7

956.4 812.4 820.0

2128 1658 1037

2.5/5.0 1.4/3.3 1.4/4.4

0.90 0.95 0.91

2.6/3.7 1.7/2.7 1.0/2.7

34.7 34.1 32.8 32.2 32.3 31.8 30.8 30.3 30.3 30.2 29.6 29.4

77.6 77.6 79.0 76.3 77.2 77.2 78.6 78.1 78.0 78.7 79.6 79.4

3260.8 3338.6 4325.0 1910.3 2544.6 1184.9 1863.0 692.2 666.3 1223.3 1306.1 2069.0

668 1666 2478 583 333 696 318 989 1301 1308 3385 1378

2.0/2.7 1.9/2.9 0.5/1.5 3.7/5.3 2.2/3.7 7.4/9.4 5.0/8.0 1.9/3.5 1.4/3.5 5.5/7.6 1.1/3.4 9.3/12.2

0.89 0.88 0.86 0.91 0.84 0.91 0.82 0.95 0.93 0.87 0.92 0.90

1.5/1.6 0.8/1.2 0.4/1.1 2.7/3.3 – 3.7/4.0 1.4/1.8 2.2/3.1 2.0/2.9 6.6/7.0 0.7/1.7 7.2/7.6

Dhangadhi Darchula Bhimchula

DNGD DRCL BMCL

28.8 29.7 28.7

80.6 80.5 81.7

189.0 2027.3 2292.3

426 536 608

4.2/6.5 2.0/5.0 6.5/8.6

0.94 0.82 0.88

3.1/3.5 0.1/1.4 4.8/5.2

Jumla Dolpa Ghorahi Jomsom Koldana

JMLA DLPA GRHI JMSM KLDN

29.3 29.0 28.0 28.8 27.8

82.2 82.8 82.5 83.7 83.6

2442.0 2569.5 785.3 3473.9 1831.5

680 681 680 680 681

0.2/2.5 2.3/4.5 1.9/4.2 0.2/4.0 12.0/16.4

0.92 0.87 0.93 0.66 0.91

Sarangkot Rumjartar Ramite

SRNK RMJT RMTE

28.3 27.3 27.0

83.9 86.6 86.6

1685.1 1387.7 2126.8

462 373 625

0.8/3.5 1.4/5.0 9.1/11.4

0.87 0.81 0.88

1.1/1.8 0.4/1.1 1.6/2.7 1.4/1.8 10.4/ 11.1 0.5/1.8 0.2/1.4 8.7/8.9

Station

Site code

DHAR ISRR UDAI BHOP JBPR DHAN DURG BHUB IITB PUNE HYDE IISC BAN2 KODI

SHIL

Geodetic Latitude (o N)

GPS Data (days)

GPS – MODIS Daily Bias/RMSE (mm)

GPS – MODIS Seasonal (mm) R2

Bias/ RMSE Winter 2.7/3.4 2.2/3.4 2.7/3.4 2.2/3.3 2.2/3.3 0.5/1.8 3.5/4.2 3.6/4.2 3.0/7.7 – 3.1/3.7 3.3/3.9 1.2/2.0 3.8/4.2 2.5/3.1 1.7/2.9 2.3/2.8 3.7/4.8 2.0/2.9 3.1/3.7 3.1/3.9 3.9/4.5 4.5/5.0 18.2/ 18.9 2.4/4.3 2.9/5.6 4.0/9.0 4.7/6.7

Bias/ RMSE Spring

Bias/ RMSE Summer

Bias/ RMSE Autumn

1.9/3.0 1.9/2.9 2.5/4.0 1.8/3.4 1.8/3.4 0.8/2.8 – 3.8/4.4 4.8/7.4 12.1/ 13.4 3.1/3.9 2.6/4.3 1.2/2.5 2.7/3.3 2.1/3.0 0.6/3.6 0.4/3.8 0.3/4.2 0.6/4.1 2.6/3.9 3.1/4.2 4.6/5.4 5.6/6.3 19.4/ 20.0 3.2/6.3 6.3/8.1 – 6.2/9.1

7.4/8.2 6.1/7.5 6.4/8.4 5.5/7.4 5.5/7.5 4.1/6.2 3.0/7.0 – 3.9/12.1 –

6.6/7.2 4.3/7.3 5.3/6.3 4.2/5.3 4.2/5.3 2.4/3.9 2.8/5.8 3.7/5.1 0.6/9.2 16.7/ 18.0 3.7/4.7 4.3/5.7 2.1/3.4 4.3/5.3 2.8/4.0 – 4.0/5.7 3.0/5.1 – 1.9/3.4 4.1/5.2 5.6/6.4 6.2/6.8 22.1/ 22.9 – 4.7/6.9 5.5/10.8 9.5/11.8

6.4/7.8 3.7/5.5 13.4/ 14.2 12.6/ 13.4 1.9/4.6 0.4/3.4 0.5/6.2

9.6/12.5 5.1/8.4 19.6/ 20.6 – 6.2/9.2 4.0/5.8 4.5/7.5

3.2/6.9 1.8/5.0 16.6/ 17.6 16.1/ 16.4 5.5/7.3 3.7/5.0 4.7/5.5

2.3/2.6 1.7/2.1 0.5/1.2 2.8/3.7 2.2/2.8 7.2/7.5 3.4/3.8 2.2/3.4 1.4/2.9 6.6/7.2 1.7/2.7 9.3/ 10.1 4.5/6.1 2.3/3.2 5.4/6.2

3.4/4.0 3.2/4.1 0.8/2.1 6.6/8.3 4.5/6.1 13.7/14.6 11.9/13.6 3.2/5.6 2.6/6.0 9.0/11.5 2.8/6.3 19.7/ 21.1 7.7/9.8 4.6/8.0 10.4/ 12.4 1.9/4.3 6.5/8.1 3.2/6.7 2.8/6.4 28.7/ 29.9 – – 15.4/ 16.9

1.7/2.1 1.5/2.3 0.5/1.5 4.5/5.4 1.7/3.1 8.2/9.2 – 2.0/3.9 1.0/3.8 5.9/7.7 0.4/3.6 14.0/ 15.1 – 1.0/4.4 10.6/ 11.6 0.6/2.7 2.6/4.8 3.6/6.1 0.2/4.4 22.1/ 23.8 1.1/4.2 1.3/5.5 15.1/ 15.9

0.7/1.8 2.1/2.8 1.8/3.3 0.6/2.1 12.8/ 14.4 1.9/3.8 1.2/4.5 10.4/ 11.5

1.6/5.9 1.3/6.2 –

4.3/5.4 2.6/3.7 1.8/4.5 2.6/6.8 1.1/6.5 4.7/6.5 – 4.2/5.3 5.7/6.5 6.4/6.9 20.3/ 20.9 – 5.0/7.5 – 9.4/12.4

(continued on next page)

4

S. Jade et al.

Journal of Atmospheric and Solar-Terrestrial Physics 196 (2019) 105138

Table 1 (continued ) Station

Site code

Geodetic Latitude (o N)

Geodetic Longitude (o E)

MSL Height (meters)

GPS Data (days)

GPS – MODIS Daily

GPS – MODIS Seasonal (mm)

Bias/RMSE (mm)

R2

Bias/ RMSE Winter

Bias/ RMSE Spring

Bias/ RMSE Summer

Bias/ RMSE Autumn

1.8/7.0 4.5/7.5 4.1/4.5 5.8/7.7 7.7/ 10.8 –

4.8/5.8 0.8/3.4 2.1/2.9 5.8/7.3 8.8/ 10.2 –

Syangboche Panthang Lhasa Lhasa* Bomdilla*

SYBC GBSK LHAS LHAZ BOMP

27.8 27.4 29.7 29.7 27.3

86.7 88.6 91.1 91.1 92.4

3824.9 1999.0 3659.3 3659.3 2510.7

332 1737 736 5037 2416

3.0/5.2 1.4/3.9 1.8/2.7 3.2/5.5 4.5/6.7

0.65 0.79 0.97 0.61 0.78

3.4/3.6 0.4/2.0 0.3/0.6 1.1/1.6 4.9/5.4

3.3/4.5 2.8/5.1 1.7/2.1 0.9/3.4 5.3/6.9

Anini

ANIN

28.8

95.9

1717.5

336

2.8/4.5

0.87

–

–

Fig. 3. GPS and MODIS PWV for Northeast India cGPS sites with long continuous data span.

Fig. 2. GPS and MODIS PWV for Peninsular Indian cGPS sites with long continuous data span.

MODIS) PWV at all the sites is plotted on the MODIS Terra water vapor distribution for the period of 2002–2017 over the Indian subcontinent (Fig. 5). Seasonal GPS PWV is calculated for all the cGPS sites for Winter (December to February), Spring (March to May), Summer (June to

Northeast India, Arunachal Himalaya and south of Tibet have collocated meteorological sensors (Table 1, Fig. 1). PWV for these 8 cGPS sites is estimated using observed pressure and temperature recorded at these sites. GPS-PWV calculated at these eight sites using ERA –Interim meteorological data gave precise water vapor values with a bias of 0.006 to 0.8 mm and RMSE of 0.6–1.8 mm when compared to GPS-OBS water vapor. Previous studies in Indian subcontinent indicate reasonable correlation between GPS-OBS and GPS-PWV estimated using ERAInterim data set with low resolution 2.5� � 2.5� grid (Jade and Vijayan, 2008) and 0.75� � 0.75� grid (Barman et al., 2017). GPS-PWV for all the sites is derived using high resolution 0.125� � 0.125� grid ERA-Interim data set with 37 vertical levels and is used to validate MODIS-PWV values for the Indian subcontinent. Average value of GPS-PWV at all sites is plotted in Fig. 1 which clearly depicts the spatial variability of water vapor in Indian subcontinent with medium to high water vapor values in peninsular and northeast India and low water vapor values in high altitude Himalayas. 3.1. Validation of MODIS PWV We divide Indian subcontinent (Fig. 1, Table 1) in to three regions: Peninsular India with normal to high rainfall, Northeast India with high rainfall and Himalayas with scanty to less rainfall. For few sites with long continuous span of GPS data, time series of daily MODIS and GPS PWV is plotted in Fig. 2 for peninsular India, Fig. 3 for northeast India and Fig. 4 for Himalayas. Daily average value of GPS-PWV is compared with daily MODIS-PWV for all the sites (Table 1). Average bias (GPS-

Fig. 4. GPS and MODIS PWV for Himalayan region cGPS sites with long continuous data span. 5

S. Jade et al.

Journal of Atmospheric and Solar-Terrestrial Physics 196 (2019) 105138

Fig. 5. MODIS Terra 1� � 1� grid water vapor distribution during 2002–2017 for Indian subcontinent. Vertical bars at the cGPS sites denote the positive bias (black) and negative bias (red) between the daily GPS and MODIS PWV values.

August) and Autumn (September to November) with minimum of 60–90 days of data during each season during 2002–2017 and compared with MODIS data (Table 1).

(Fig. 5, Table 1). For coastal sites TVRM and IITB located in southern tip and central region of west coast and for cGPS sites located in islands (SGOC, MALD and PBR2) the bias is positive indicating underestimation of PWV by MODIS data. Northeast India: MODIS PWV (Table 1, Fig. 5) has correlation coef­ ficient (R2) above 0.9 for all the sites spatially (23.7–26.6� N; 91.6 to 95� E) covering the region. For 5 cGPS sites (TZPR, GHTU, LUMA, IMPH and AIZW) bias is 2.5 to 3 mm and RMSE is 3–6 mm. MODIS PWV has a high bias of about 8.5 mm and RMSE of 11.5 mm for two cGPS sites located on Shillong Plateau at an altitude of ~1600 m. Only for TZPR and GHTU sites located in close proximity of Brahmaputra river with low altitude of 130 m and 62 m, bias is positive, which indicates that MODIS data is underestimating the PWV for these sites. Rest of the sites with altitude ranging from 800 to 1600 m bias is negative indicating overestimation of PWV by MODIS. Himalayan cGPS sites: There are about 29 cGPS sites in 2500 km Himalayan arc spatially spread from Ladakh Himalaya in the west to Eastern Himalayan Syntaxis. MODIS PWV (Table 1, Fig. 5) estimated at 23 cGPS sites has correlation coefficient (R2) above 0.8 with bias of 4.5

3.1.1. Daily PWV Peninsular India: For 22 cGPS sites located in peninsular India (Table 1, Fig. 5), MODIS PWV has correlation coefficient (R2) above 0.9 with bias of 3.0 to 0.3 mm and RMSE of 2.5–5 mm. MODIS PWV for rest of the sites SGOC (bias: 2.7 mm, RMSE: 5.4 mm) located in the Srilankan islands, PBR2 (bias: 3 mm, RMSE: 6 mm) located on Andaman islands, coastal site TRVM (bias: 0.7 mm, RMSE: 3 mm), high altitude (2000 m) site KODI (bias: 11.8 mm, RMSE: 16 mm) and BELP (bias: 1.1 mm, RMSE: 8.3 mm) site in Gujarat have R2 of 0.7–0.8. Correlation coefficient at MALD (bias: 0.9 mm, RMSE: 4.5 mm) site located in Maldives islands south of peninsular India is very low (0.3) indicating that MODIS data fails to give a reliable estimate of water vapor for this site. High altitude (1600 m) site MABU located in Gujarat has bias of ~11.5 mm and RMSE of ~15 mm. For most of the peninsular India sites bias is negative indicating that MODIS is over estimating the PWV 6

S. Jade et al.

Journal of Atmospheric and Solar-Terrestrial Physics 196 (2019) 105138

to þ5 mm with RMSE value of 1.5 mm (IAOH) to 6.7 mm (BOMP). Rest of the sites have high value of bias ( 12 to 7 mm) and RMSE (7–16 mm). Himalayan sites have varying topography with altitude of sites ranging from 200 to 4300 m as it is highly undulated terrain. About 22 sites have positive bias indicating underestimation of PWV by MODIS data in the most of the Himalayan terrain. In Summary, daily PWV using MODIS data (Fig. 5) has correlation coefficient (R2) above 0.9 with low negative bias and RMSE for most of the cGPS sites located in peninsular and northeast India when compared to undulated Himalayan region (positive bias). In peninsular and northeast India, sites located in coastal areas (TRVM, SGOC), tiny islands (PBR2, MALD) and close vicinity of rivers (TZPR, GHTU), MODIS PWV has low correlation coefficient (R2), positive bias and high RMSE indicating underestimation of PWV by MODIS data. Similar trends were observed in Iberian Peninsula (Bennouna et al., 2013; Vaquero_Martinez et al., 2017), MODIS level 2 products and GPS give better agreement for continental sites when compared to coastal sites and MODIS products overestimate low water vapor values and underestimate high water vapor values. The moist sites have peak water vapor of up to 80 mm during monsoon (Figs. 1, 3 and 5), and it was recorded that the bias between GPS and MODIS PWV increases with increase in water vapor content (Li et al., 2003; Lu et al., 2011). This is caused due to un­ certainties associated with spectroscopic data base, calculation of at­ mospheric water vapor transmittance. Moreover, the four MODIS 1� � 1� grid points surrounding these sites (Figs. 1 and 5) are located on land, ocean and water bodies as compared to the rest of the sites for which MODIS grid points are located on land. For sites located in undulated Himalayan terrain and high altitude sites in peninsular (KODI, MABU), northeast India (CSOS, SHIL), surface heights approximately corresponding with MODIS PWV are horizontally interpolated to the site location (h) using SRTM (Shuttle Radar Topog­ raphy Mission, https://dds.cr.usgs.gov/srtm/version2_1/SRTM30/srt m30_documentation.pdf) C-Band data with 30 m horizontal resolution (C30) at 1� � 1� grid identical to MODIS grid using the following equation h ¼ w1 h1 þ w2 h2 þ w3 h3 þ w4 h4

which the site is located and water vapor variation. Sites with high negative bias indicate that the MODIS surface grid used for interpolation is at low altitudes (positive Δh) compared to the site elevation and vice versa. In summary, MODIS surface data is unable to estimate reliable PWV values for sites located in coastal regions, islands and in highly undulated terrains. 3.1.2. Seasonal PWV MODIS data depicts distinct seasonal cycle of PWV (Figs. 2–4, Table 1) for peninsular, northeast and Himalayan region with high bias and RMSE for all the sites during the summer and autumn seasons (June to November) during which Indian monsoon is active when compared to winter and spring seasons. Previous study at four sites in peninsular India (Prasad and Singh, 2009; Kumar et al., 2013), similar large sea­ sonal bias and RMSE was observed for MODIS data which is attributed to Indian monsoon (June–November). It was observed that several factors such as variation in temperature, rainfall, humidity and wind field causes the seasonal water vapor fluctuations at these sites. Similar sea­ sonal bias of MODIS data for Indian subcontinent was observed (Joshi et al., 2013; Ningomban et al., 2016) for a two Himalayan sites with ~1259 m and ~4500 m altitude. For Himalayan sites with altitude above 3000 m, seasonal bias is less compared to the rest of Himalayan sites which is due to low water vapor values attributed to dry environ­ ment in this region. For very high altitude sites such as Hanle, the inversion layer is very strong in cold winter months with very low water vapor (Fig. 4) and weak during summer months. Similar results were observed in southern Tibet (Lu et al., 2011) during monsoon season when the summer brings large amount of water vapor from tropical ocean to southern Tibet causing large water vapor variation for sites with altitude less than 3000 m. Large bias during summer and autumn seasons is due to large water vapor fluctuations at these sites which may be due to uncertainties associated with MODIS retrieval algorithm (Lu et al., 2011). Sources of errors in the MODIS near-IR PWV retrieval (Kaufman and Gao, 1992) are attributed to (i) 4–7% uncertainty in the spectral reflectance of the surface, (ii) 2% due to haze, (iii) 3–6% caused by sensor calibration, (iv) 2–3% accounts for pixel registration between the channels, (v) 1–2% caused by shift in channel location, (vi) 1% due to undetected clouds, (vii) 1% accounts for atmospheric temperature and moisture profiles and (viii) 0.7% accounts for mixed pixels when additional channels are used. In Iberian peninsula, it was found that the performance of MODIS level 2 products depends on seasonal water vapor cycle (Bennouna et al., 2013; Vaquero_Martinez et al., 2017) with overestimation of PWV in summer and winter and underestimation during spring and autumn. Thus MODIS data is unable to precisely de­ pict the seasonal variability of water vapor in Indian Subcontinent when compared to GPS PWV.

(8)

Horizontal weights at each grid point (j ¼ 1,2,3,4) are calculated using Eqs. (6) And (7). Height difference (Δh) between GPS height (HG) and surface height (HS) corresponding to MODIS grid is calculated as Δh ¼ HG

HS

(9)

Bias between GPS and MODIS PWV and the corresponding height difference Δh is plotted in Fig. 6. For high altitude sites in peninsular (KODI, MABU) and northeast India (CSOS, SHIL) where there is a sudden change of elevation, MODIS data has high negative bias (~9–12 mm), indicating over estimation of PWV as the MODIS grid points (Fig. 6) used for interpolation are on low altitudes i.e. positive Δh. Most of the Hi­ malayan cGPS sites have positive bias indicating underestimation of PWV by MODIS data as surface height is higher than the cGPS height i. e negative Δh. Also bias is low (~0.2–3 mm) for most sites as the four grid points are at similar heights and water vapor variation decreases with height. High positive bias of ~4–8 mm for four Himalayan sites (GBKL, DNGD, BHTW, GBSN) is due to highly undulated terrain in this region with the MODIS grid points located at either very low or high altitudes. Himalayan sites (BMCL, RMTE, GBNL, KLDN) indicated negative bias (~3–12 mm) as the MODIS grid points are located at low surface height thus overestimating the PWV. In Himalayas, MODIS surface data is unable to precisely capture the water vapor variability with height as the four grid points are located in undulated terrain with varying heights. For example for the highest altitude Hanle site MODIS give lowest values of bias and RMSE as the four grid points used for interpolation are located approximately at the same altitude. Also high altitude Hanle site has cold desert environment with very low water vapor of less than 20 mm (Fig. 4). The same does not apply to all the sites in Himalayas as the performance of MODIS PWV depends on undulations of the terrain in

3.2. GPS – PWV variability Spatial and temporal variability of cGPS sites is discussed in subse­ quent sections using the daily average PWV values. 3.2.1. Spatial Spatial variability of water vapor depends upon surface temperature profile, wind flow pattern, precipitation, vertical humidity variation, distribution of land masses and water bodies. Average PWV of all the sites (Fig. 1) with minimum one year of GPS data evenly distributed across all the seasons of the year and mean sea level height of each site is plotted in Fig. 7. In peninsular India average PWV is 26–36 mm at most of the low altitude interior sites, 41–50 mm at coastal (IITB, BHUB, TRVM) and island sites (MALD, PBR2, SGOC), 13 mm at MABU (1680 m altitude) and 18 mm at KODI (2340 m altitude). For the seven sites in northeast India the average PWV varies from 22 mm at high altitude site (CSOS, SHIL) to 46 mm at TZPR site located in the close vicinity of Brahmaputra river. For Himalayan region highest average PWV of ~40 mm is recorded for DNGD site with very low altitude (190 m) 7

S. Jade et al.

Journal of Atmospheric and Solar-Terrestrial Physics 196 (2019) 105138

Fig. 6. Bias between GPS and MODIS PWV for Himalayan cGPS sites and high altitude sites in peninsular and northeast India along with difference in GPS and SRTM heights (Δh) using C30 data corresponding to MODIS 1� � 1� grid.

located in Nepal Himalaya to lowest value of ~5 mm at high altitude site (4325 m) located at Hanle (IAOH) in Ladakh Himalaya. High PWV value at sites located in coastal areas, islands and close to water bodies, is due to land-sea interactions, high moisture flux and water vapor trans­ mission. Our results indicate that there is no systematic variation of GPSPWV with respect to latitude and longitude of the site. Average PWV is inversely related to altitude of the site (Fig. 7) which is due to low temperature at high altitudes and decrease in the depth of water vapor column. Since ability of air to carry moisture is inversely proportional to temperature and density, water vapor decreases with increase in alti­ tude. Theoretical relation for the average PWV of Indian subcontinent and mean sea level height (h) (Fig. 7) follows a second degree expo­ nential fit with 95% confidence bounds and correlation coefficient of 0.86. Nonlinear relation of GPS PWV with altitude indicates that several factors such as moisture flux, wind pattern, temperature, precipitation, evaporation, site location etc. influence the water vapor variability with height. Mapping function П (Bevis et al., 1992, 1994) between PWV and ZWD is a function of Tm, the weighted mean temperature of the atmo­ sphere (Eq. (3)). Bevis et al. (1994) plotted the value of П as a function of Tm and found that the relationship is very nearly linear. To evaluate the uncertainty in parameter П, relative error in П was plotted as a function of relative error in Tm and concluded that the uncertainty in П closely approximates relative error in Tm when it exceeds 1%. Usually the relative error in the computation of Tm is more than 1% and hence the uncertainty in П is solely due to the error in Tm estimation. Value of П approximates to 0.15 � 20% variation depending on Tm estimation (Bevis et al., 1994). Tm is a function of partial pressure of water vapor and temperature (Eq. (4)) which vary with location, altitude, weather and seasons. П equal to PWV/ZWD ratio for the 64 cGPS sites is calcu­ lated using Eqs. (3) And (4) from 6hr interval ERA-Interim data set and

averaged over the total number of available days to study the spatial variability of П. For a span of 1–13 years in the Indian sub-continent (5–35� N; 70 to 96� E), PWV/ZWD ratio lies between 0.153 and 0.166 which is within �10% variation. Our previous study (Jade et al., 2008) spanning 4 years gave PWV/ZWD ratio of 0.148–0.163 for 24 cGPS sites in Indian subcontinent. П parameter varies exponentially with height and nonlinear with latitude for the Indian subcontinent. Taking advantage of large spatial and temporal coverage of the current data set with 64 cGPS sites, spatial variability of П for Indian subcontinent is refined as П¼

2 PWV ¼ fðλ; hÞ ¼ eaþðb:hÞþðc:λ Þ ZWD

(10)

where a ¼ 1.79; b ¼ 1.7 � 10 5 (1/m); c ¼ 9.64 � 10 6 (1/deg), stan­ dard error ¼ 0.347% h ¼ mean sea level height in m and λ ¼ latitude in degrees The above model fit has 95% confidence bounds with correlation coefficient R2 of 0.99 and the residuals of the fit are of the order of �0.25%. The above relation can be used to precisely estimate PWV for Indian subcontinent directly from ZWD without the need to compute the weighted mean temperature. 3.2.2. Seasonal PWV time series (Figs. 2–4) of all the sites indicate distinct seasonal variability with peak values in summer months (June to August) coin­ ciding with Indian monsoon and minimum values in winter (December to February). For seasonal variability study, daily PWV for seasons having 60–90 days of data for each season of all the available years are used. Highest PWV is observed in summer months and lowest PWV in winter months for all the Indian cGPS sites. Hence, Percentage Seasonal 8

S. Jade et al.

Journal of Atmospheric and Solar-Terrestrial Physics 196 (2019) 105138

Fig. 7. Average GPS-PWV estimated during 2002–2017 at all the cGPS sites along with the mean sea level height. Bold purple line denotes the theoretical relation fitted to the data which follows a second degree exponential with 95% confidence bounds and R2 0f 0.86.

Variability(PSV) at each cGPS site is defined as PSV ¼

PWVsummer PWVwinter � 100 PWVsummer þ PWVwinter

The above fit has 95% confidence bounds and correlation coefficient R2 of 0.73. Nonlinear relation between the seasonal variability and surface temperature variation shows that the seasonal PWV variation for tropics (Bock et al., 2007) is due to several factors such as surface temperature, rainfall, humidity, moisture transport, convection, evap­ oration etc.

(11)

For most of the cGPS sites in Indian subcontinent (Fig. 8) interseasonal variability is 50–80%. For coastal and island sites interseasonal variability is 4% at Maldives, 8% at SGOC, 17% at TRVM, 21% at PBR2. Since these are moist sites with moderate temperature difference between summer and winter, seasonal variability of PWV is relatively low. Peninsular sites (KODI, IISC, BAN2, HYDE, BHUB, PUNE) and moist sites TZPR, GHTU in northeast India record seasonal vari­ ability of 40–50% as these sites have moderate winter temperatures. Measured surface temperature using collocated meteorological sensors at cGPS sites: IISC (peninsular India), IAOH and LHAZ (Himalayas) is compared for a period of seven years (2009–2016) with the world weather online (WWO) data (https://www.worldweatheronline.com) of surface temperature. WWO monthly surface temperature data compares well with a correlation coefficient of 0.9 at IISC, 0.98 at IAOH and 0.95 at LHAZ with the observed value. Since observed surface temper­ ature data is not available at all the cGPS sites, we use WWO data to study the relation between surface temperature and PSV. Surface tem­ perature variation (Δt) for all the cGPS sites is calculated as difference of maximum average temperature in summer and minimum average tem­ perature in winter for the period of 2009–2017 and plotted in Fig. 8. Relation between the percentage seasonal variability (PSV) and surface temperature difference (Δt) follows second order power law and is given as PSV ¼ aΔtb þ c

3.2.3. Annual Year to year variation of PWV is predominantly related to precipi­ tation and surface temperature variation during each year. We have four cGPS sites (KODI, IISC, HYDE, BHUB) in Peninsular India, 4 sites (GHTU, TZPR, IMPH, LUMA) in northeast India and 4 sites (GBPK, GBSK, BOMD, LHAZ) in Himalayas (Figs. 1–4) with long span of data (3–13 years) for studying year to year PWV variability. For each year with minimum of 300 days of data, deviation of yearly average PWV from the total average of all available years for each cGPS site is plotted in Fig. 9. Interannual variability (IAV) is calculated as total deviation in mm at each cGPS site for all the available years. Percentage IAV (PIAV) for each cGPS site is given by PIAV ¼

IAV PWV

� 100

(13)

IAV in mm and PIAV values at all the twelve sites for the period spanning 3–13 years is plotted in Fig. 10. Average inter-annual PWV variation for all the peninsular sites located in south India is ~4–5 mm with inter-annual percentage variability of 15% at IISC with 13 years of data and 12% at HYDE with 11 years of data. BHUB site with 3 years of data has 9% inter annual variability from 2005 to 2010 and high altitude site KODI site with 4 years of data has 22% variability from 2004 to 2013. Northeast sites (TZPR, GHTU) show inter-annual PWV variation of ~4 mm with inter-annual percentage variability of ~10% at GHTU

(12)

where a ¼ 587.5 � 30.92; b ¼ 1.346 � 0.071; c ¼ 72.12 � 3.80; 9

S. Jade et al.

Journal of Atmospheric and Solar-Terrestrial Physics 196 (2019) 105138

Fig. 8. Percentage seasonal variability plot with change in temperature for the cGPS sites with 60–90 days of data for each season during 2002–2017. Bold purple line denotes the empirical relation fitted to the data.

with 7 years of data and 9% at TZPR with 8 years of data. IMPH and LUMA sites with 3 years of data indicate inter-annual variation of ~ 1 mm with 4% variability at IMPH and about 3.8 mm variation with 11% variability at LUMA. GBPK cGPS site located in Northwest Himalaya with 7 years data (2003–2014) records inter-annual variation of ~4 mm with 18% variability and GBSK site located in Sikkim Himalaya with 3 years (2004–2008) show ~ 3 mm annual PWV variation with 12% variability. BOMP site located in Arunachal Himalaya with 6 years of data indicates inter annual variation of PWV of ~1.5 mm with 9% variability. Further north, Lhasa site located to the south of Tibet plateau with 14 years of data (2003–2014) records inter-annual variation of ~2 mm with 19% variability.

data overestimates PWV as MODIS grid points are at low altitudes. High negative bias for few Himalayan sites is due to MODIS grid points used for interpolation are located in undulated terrain with varying heights. Daily MODIS PWV data is unable to depict the seasonal variability of water vapor with high values of bias and RMSE during Indian monsoon months (June to November). Hence daily MODIS PWV data can only be used to get a first order estimate of water vapor in Indian subcontinent and should be used after necessary corrections for water vapor variability studies. 3. Average GPS-PWV of Indian subcontinent sites is inversely corre­ lated to the altitude of GPS site and theoretical relation fitted follows second degree exponential model with 95% confidence bounds. Spatial variability of П, mapping function for PWV and ZWD for the current data set fits with R2 of 0.99, standard error of 0.35% and residuals of �0.25% for the Indian subcontinent. Weighted mean temperature of the atmosphere at a GPS site is not required for water vapor estimation as this spatial variability function of П gives the water vapor directly from ZWD. 4. Seasonal PWV variation for tropical regions like Indian subcontinent is a complex process involving several factors such as humidity, temperature variation, rainfall, moisture fluxes, convection and evaporation. However, variation of surface temperature during the summer and winter season is directly correlated to inter-seasonal variability of water vapor and the empirical relation fitted follows a power law. 5. Inter annual variation of GPS-PWV is low (1–5 mm) for the 12 cGPS sites with data span of 3–13 years. Inter annual percentage vari­ ability of water vapor is 9–19% for Himalayan region, about 3–10% for northeast India and 9–22% for peninsular India. These results donot indicate any noticeable systematic trends in the inter-annual variability, hence long and continuous span of GPS data is required for long period trend analysis of water vapor.

4. Conclusions PWV study in Indian subcontinent using MODIS and GPS data at 64 cGPS stations spatially spread over the Indian subcontinent during 2002–2017 gives the following conclusions. 1. Water vapor estimates from GPS observations using observed (GPSOBS) and interpolated (GPS-PWV) pressure and temperature values give a very low bias and RMSE for the cGPS sites in Indian subcon­ tinent and hence GPS-PWV can be used for water variability studies. However for diurnal variability studies of water vapor we still need GPS-OBS as it gives water vapor values for <2 h interval whereas GPS-PWV values using ERA- Interim meteorological parameters are at 6 h interval. 2. For sites located in coastal areas, islands, close vicinity of rivers and high altitude region with varying topography (Himalayas), daily MODIS PWV product has high positive bias and RMSE indicating underestimation of PWV by MODIS data. For high altitude sites in peninsular and northeast with sudden change in topography MODIS 10

S. Jade et al.

Journal of Atmospheric and Solar-Terrestrial Physics 196 (2019) 105138

Fig. 9. Deviation of average PWV of each year from the total average PWV for all the available years for 12 cGPS sites with 3–13 years of data with minimum of 300 days in each year and equally spread over all the seasons.

Fig. 10. Inter-annual PWV variation in mm and percentage inter-annual variability for 12 cGPS sites with 3–13 years of data with minimum of 300 days in each year and equally spread over all the seasons.

11

S. Jade et al.

Journal of Atmospheric and Solar-Terrestrial Physics 196 (2019) 105138

Declaration of competing interest

operation and application of GPS. Intech. https://doi.org/10.5772/ intechopen74172. Joshi, S., Kumar, K., Pande, B., Pant, M.C., 2013. GPS-derived precipitable water vapour and its comparison with MODIS data for Almora, central Himalaya, India. Meteorol. Atmos. Phys. 120 (3–4), 177–187. https://doi.org/10.1007/s00703-013-0242-z. Kaufman, Y.J., Gao, B.C., 1992. Remote sensing of water vapor in the near IR from EOS/ MODIS. IEEE Trans. Geosci. Remote Sens. 30, 871–884. King, M.D., Kaufman, Y.J., Menzel, W.P., Tanre, D., 1992. Remote sensing of cloud, aerosol and water vapor properties from the Moderate resolution Imaging Spectrometer (MODIS). IEEE Trans. Geosci. Remote Sens. 30, 1–27. King, et al., 2003. Cloud and aerosol properties, precipitable water and profiles of temperature and water vapor from MODIS. Geosci. Remote Sens. IEEE Trans. 41, 442–458, 10.1.1109/TGRS.2002.808226. Kumar, S., Singh, A.K., Prasad, A.K., Singh, R.P., 2009. Annual variability of water vapor from GPS and MODIS data over the indo-gangetic plains. J. Indian Geophys. Union 13 (1), 17–23. Kumar, S., Singh, A.K., Prasad, Anup, K., Singh, R.P., 2013. Variability of GPS derived water vapor and comparison with MODIS data over the Indo-Gangetic plains. Phys. Chem. Earth 55 (57), 11–18. Li, Z., Muller, J., Cross, P., 2003. Comparison of precipitable water vapor derived from radiosonde, GPS and Moderate-Resolution Imaging Spectroradiometer measurements. J. Geophys. Res. 108 (D20), 4651. https://doi.org/10.1029/ 2003JD003372. Li, Z., Muller, J., Cross, P., Fielding, E.J., 2005. Interferometric synthetic aperture radar (InSAR) atmospheric correction: GPS, Moderate Resolution Imaging Spectroradiometer (MODIS) and InSAR integration. J. Geophys. Res. 110, B03410. https://doi.org/10.1029/2004/2004JB003446. Liu, J., Liang, H., Sun, Z., Zhou, X., 2006. Validation of the Moderate-Resolution Imaging Spectroradiometer precipitable water vapor product using measurements from GPS on the Tibetan Plateau. J. Geophys. Res. 111, D14103. https://doi.org/10.1029/ 2005JD007028. Lu, Ning, Qin, Jun, Yang, Kun, Gao, Yang, Xu, Xiangde, Koike, Toshio, 2011. On the use of GPS measurements for Moderate Resolution Imaging Spectrometer precipitable water vapor evaluation over southern Tibet. J. Geophys. Res. 116, D23117. https:// doi.org/10.1029/2011JD016160. Mendes, V.B., Prates, G., Santoa, L., Langely, R.B., 2000. An evaluation of accuracy of models for the the determination of the weighted mean temperature of the atmosphere. In: Proc. ION.2000 Natl. Tech. Meet, Anaheim, CA, USA, C4, pp. 433–438. http://w3.uagl.pt/~gprates/ION2000.pdf. Ningombam, S.S., Jade, S., Shrungeshwara, T.S., Song, H.-J., 2016. Validation of water vapor retrieval from moderate resolution imaging spectro-radiometer (MODIS) in near infrared channels using GPS data over IAO-hanle, in the trans-himalayan region. J. Atmos. Sol. Terr. Phys. 137 (January), 76–85. https://doi.org/10.1016/j. Jastp.2015.11.019. Elsevier. Ningombam, S.S., Jade, S., Shrungeshwara, T.S., 2018. Parameterization of water vapor using high-resolution GPS data and empirical models. J. Atmos. Sol. Terr. Phys. 168 (March), 58–69. https://doi.org/10.1016/j.jastp.2018.01.009. Parkinson, C.L., 2003. Aqua: an Earth_observing satellite mission to examine water and other climate variables. IEEE Trans. Geosci. Remote Sens. 41, 173–183. Prasad, A.K., Singh, R.P., 2009. Validation of MODIS terra, AIRS, NCEP/DOE AMIP-II reanalysis- 2, and AERONET sun photometer derived integrated precipitable water vapor using ground-based GPS receivers over India. J. Geophys. Res. 114 (D5), D05107. https://doi.org/10.1029/2008JD011230. Schueler, T., 2001. On Ground Based GPS Tropospheric Delay Estimation. Ph.D. Thesis. Univ. der Bundeswehr München, Neubiberg, Germany. Schueler, Torben, Posfay, Andrea, Hein, Guenter W., Biberger, Robert, 2001. A global analysis of the mean atmospheric temperature for GPS water vapor estimation. In: Proceedings of the 14th International Technical Meeting of Satellite Division of the Institute of Navigation (ION GPS 2001), Salt Lake City, UT, September 2001, pp. 2476–2489. Solbrig, P., 2000. Untersuchungen uber die Nutzung numerischer Wettermodelle zur Wasserdampfbestimmung mit Hilfe des Global Positioning Systems. Diploma Thesis. Institute of Geodesy and Navigation, University FAF Munich, Germany. Thomas, I.D., King, M.A., Clarke, P.J., Penna, N.T., 2011. Precipitable water vapor estimates from homogeneously reprocessed GPS data: an intertechnique comparison in Antartica. J. Geophys. Res. 116, D04107. https://doi.org/10.1029/ 2010JD013889. Vaquero-Martinez, J., Anton, M., Ortiz de Galisteo, J.P., Cachorro, V.E., Costa, M.J., Roman, R., Bennouna, Y.S., 2017. Validation of MODIS integrated water vapor product against reference GPS data at the Iberian Peninsula. Int. J. Appl. Earth Obs. Geoinf. 63, 214–221. https://doi.org/10.1016/j.jag.2017.07.008. Wang, Y., Su, H., Jiang, J., Livesey, N., Santee, M., Froidevaux, L., et al., 2017. The linkage between stratosphere water vapor and surface temperature in an observation-constrained coupled general circulation model. Clim. Dyn. 48 (7), 2671–2683.

Authors declare that there is no conflict of interest. Acknowledgements We sincerely thank the editor Prof. Dora Pancheva, associate editor and anonymous reviewers for their encouraging attitude, time and effort. We acknowledge that the anonymous reviewer’s comments were very useful in improving the quality of the manuscript. This is a CSIR-4PI ARiEES contribution. We acknowledge all the concerned scientific and technical personnel involved in operating and maintaining the cGPS stations. References Askne, J., Nordius, H., 1987. Estimation of tropospheric delay for microwaves from surface weather data. Radio Sci. 22 (3), 379–386. Barman, P., 2016. Spatial and Temporal Variation of Atmospheric Precipitable Water Vapor and Active Surface Deformation Studies in Northeast India Using GPS. Ph.D thesis. Tezpur University. http://shodhganga.inflibnet.ac.in/handle/10603/99800? mode¼full. Barman, P., Jade, S., Kumar, A., Jamir, W., 2017. Interannual, spatial, seasonal, and diurnal variability of precipitable water vapor over northeast India using GPS time series. Int. J. Remote Sens. 38, 391–411. Bennouna, Y.S., Torres, B., Cachorro, V.E., Ortiz de Galisteo, J.P., Toledano, C., 2013. The evaluation of the integrated water vapour annual cycle over the Iberian Peninsula from EOS-MODIS against different ground-based techniques. Q. J. R. Meteorol. Soc. 139, 1935–1956. https://doi.org/10.1002/qi.2080. Bevis, M., Businger, S., Herring, T.A., Rocken, C., Anthes, R., Ware, R., 1992. GPS meteorology: remote sensing of atmospheric water vapor using the global positioning system. J. Geophys. Res. 97 (D14), 15787–15801. https://doi.org/ 10.1029/92JD01517. Bevis, M., Businger, S., Chiswell, S., Herring, T.A., Anthes, R.A., Rocken, C., Ware, R.H., 1994. GPS meteorology: mapping zenith wet delays onto precipitable water. J. Appl. Meteorol. 33, 379–386. Bock, O., Guichard, F., Janicot, S., Lafore, J.P., Bouin, M.N., Sultan, B., 2007. Multiscale Analysis of precipitable water vapor over africa from GPS data and ECMWF analyses. Geophys. Res. Lett. 34 (9), L09705. https://doi.org/10.1029/2006GL028039. Chang, L., Gao, G., Jin, S., He, X., Xiao, R., Guo, L., 2015. Calibration and evaluation of precipitable water vapor from MODIS infrared observations at night. IEEE Trans. Geosci. Remote Sens. 53, 2612–2620. https://doi.org/10.1109/TGRS.2014.23630 89. Davis, J.L., Herring, T.A., Shapiro, I.I., Rogers, A.E.E., Elgered, G., 1985. Geodesy by interferometry: effects of atmospheric modelling errors on estimates of baseline length. Radio Sci. 20 (6), 1593–1607. Elgered, G.J.L., Davis, Herring, T.A., Shapiro, I.I., 1991. Geodesy by radio interferometry: water vapor radiometry for estimation of the wet delay. J. Geophys. Res. 96, 6541–6555. Gao, B.C., Goetz, A.F.H., 1990. Column atmospheric water vapor and vegetation liquid water retrievals from airborne imaging spectrometer data. J. Geophys. Res. B 95, 3549–3564. Gao, B.C., Kaufman, Y.J., 2003. Water vapor retrievals using Moderate Resolution Imaging Spectrometer (MODIS) near-infrared channels. J. Geophys. Res. 108, 4389. https://doi.org/10.1029/2002JD003023. Gokhan, Gurbuz, Jin, Shuanggen, 2017. Long-time variations of precipitable water vapour estimated from GPS, MODIS and radiosonde observations in Turkey. Int. J. Climatol. https://doi.org/10.1002/joc.5153. Herring, T.A., King, R.W., Mcclusky, S.C., 2010. Documentation of the GAMIT GPS Analysis Software Release 10.4. Department of Earth, and Planetary Sciences, Massachusetts Institute of Technology, Cambridge, MA. Herring, T.A., King, R.W., Mcclusky, S.C., 2010. GLOBK, Global Kalman Filter VLBI and GPS Analysis Program, Version 10.4. Department of Earth, and Planetary Sciences, Massachusetts Institute of Technology, Cambridge, MA. Jade, S., Vijayan, M.S.M., Gaur, V.K., Prabhu, T., Sahu, S., 2005. Estimates of precipitable water vapour from GPS data over the Indian subcontinent. J. Atmos. Sol. Terr. Phys. 67 (6), 623–635. https://doi.org/10.1016/j.jastp.2004.12.010. Jade, S., Vijayan, M.S.M., 2008. GPS-based atmospheric precipitable water vapor estimation using meteorological parameters interpolated from NCEP global reanalysis data. J. Geophys. Res. 113 (D3), 1–12. https://doi.org/10.1029/ 2007JD008758. Jade, S., Shrungeshwara, T.S., May 30th 2018. Ionosphere variability in low and midlatitudes of india using GPS-TEC estimates from 2002 to 2016, multifunctional

12