J Indian Soc Remote Sens DOI 10.1007/s12524-016-0551-z
SHORT NOTE
Statistical Retrieval of Ozone and Meteorological Parameters Using SHADOZ Observations and Radiative Transfer Model Shuchita Srivastava 1 & P. K. Thapliyal 2 & M. V. Shukla 2 & J. S. H. Bisht 3 & D. Mitra 1
Received: 29 June 2015 / Accepted: 5 January 2016 # Indian Society of Remote Sensing 2016
Abstract A statistical retrieval algorithm has been developed using SHADOZ ozonesonde observations and radiative transfer model simulations for the retrieval of vertical profiles of temperature, water vapor and ozone. Retrieved profiles of ozone and meteorological parameters are compared with corresponding in-situ ozonesonde observations and IASI observations to test accuracy of the retrieval algorithm. The standard deviation error in the temperature profile, estimated using retrieved and ozonesonde observed profiles, is found to be in the range of 0.7–2.7° K. The percentage root mean square error (RMSE) estimated using retrieved and actual profiles of ozone and water vapor are found to be in the range of 5– 30 % and 10–30 %, respectively. The standard deviation error in temperature (2 to 3.5° K) and RMSE in water vapor (~35– 55 %) estimated using actual IASI observations (retrieved and observed) is relatively higher than the retrieval errors from the simulated radiances. The RMSE in the retrieval of total column ozone is also estimated using simulated and actual IASI radiances. The RMSE for simulated data is found to be 1.9 ± 1.4 %. The RMSE in the total column ozone estimated for actual IASI observations is found to be 3.8 ± 3.2 %. Inclusion of the zenith angle as a predictor in the regression coefficients has improved retrieval error of atmospheric parameters. The standard deviation error for temperature is improved by 0.2–0.3° K and RMSE for ozone and water vapor is
* Shuchita Srivastava
[email protected];
[email protected]
1
Marine and Atmospheric Sciences Department, Indian Institute of Remote Sensing, Dehradun 248001, India
2
Space Applications Centre, Ahmedabad, India
3
Physical Research Laboratory, Ahmedabad, India
improved by 2–4 % and 4–7 % respectively in different atmospheric regions. Keywords Ozone . Radiative transfer model . Statistical retrieval . SHADOZ
Introduction Tropospheric ozone plays an important role in the chemistry and radiation budget of the Earth’s atmosphere. It produces highly reactive OH radical which defines the lifetime of several trace gases in the troposphere. Thus it controls the oxidation capacity of the troposphere. Also, lower tropospheric ozone is a pernicious pollutant having detrimental impacts on human health and crop production (Ellingsen et al. 2008). In the upper troposphere, ozone is a potential greenhouse gas and can have a pronounced impact on global warming and climate change (Gauss et al. 2006). In the stratosphere, about 90 % of total atmospheric ozone is present which protects biota on the earth from potential damage by absorbing the harmful solar ultraviolet radiation. Knowledge of vertical profiles of ozone along with meteorological parameters like temperature and water vapour has become an essential part of numerical modelling for weather and climate predictions. The requirement of frequent measurements of these variables on global scale is fulfilled by satellite based instruments. Many satellite based instruments, such as Tropospheric Emission spectrometer (TES), Atmospheric infrared sounder (AIRS), Infrared Atmospheric Sounding Interferometer (IASI) etc. have been providing these measurements over last many years using infrared remote sensing technique. The AIRS onboard EOS-Aqua satellite and IASI onboard MetOp satellite are hyperspectral sounders with 2378 and 8461 infrared channels, respectively. The hyperspectral
J Indian Soc Remote Sens
sounders achieve spectral resolutions closer to the widths of the atmospheric absorption lines and provide better vertical resolution of atmospheric variables. The IASI offers the best vertical resolution of ozone and meteorological parameters in the troposphere/stratosphere (Hilton et al. 2012). Satellite instruments measure radiances emitted or reflected from Earth. These data are used to estimate underlying atmospheric variables (Herring and King 2000). The process of inferring these characteristics from observed radiances is called inverse-problem of atmospheric sounding (Kidder and VonderHaar 1995). Due to the large volume of data being collected by the IASI, compromises are required between retrieval accuracy and computational complexity. IASI makes observation in 8461 channels and that makes it very complex in terms of computational time as well as using this large volume of data in physical retrieval, because it employs use of radiative transfer model computations in real-time. There is another complexity in using this large volume of data due to large redundancy in observations from IASI channels that introduces a singularity in the observations. A computationally efficient method for determining the distribution of atmospheric ozone, temperature and moisture from satellite sounding measurements uses previously determined statistical relationships between observed or modeled radiances with corresponding atmospheric profiles. In the present research work, statistical retrieval algorithm is developed for the retrieval of ozone, temperature and water vapor contents over tropical region using SHADOZ ozonesonde observations and RTTOV simulated IASI radiances. The goodness of retrieval has been investigated by comparing the IASI real time observations with retrieved atmospheric parameters. The main objective of this work is to improve the ozone retrieval by using the best quality ozone profile measurements (SHADOZ) in the retrieval procedure as a training data. This can serve as a baseline algorithm to further improve the retrieval quality.
Dataset Utilized Infrared Atmospheric Sounding Interferometer (IASI) The infrared Atmospheric Sounding Interferometer (IASI) is an infrared fourier transform interferometer onboard EUMETSAT’s Meteorological Operation (MetOp-A) Satellite (Hilton et al. 2012), launched in October 2006. The satellite is in a polar sun-synchronous orbit with local time of 0930 h and 2130 h in descending and ascending nodes, respectively. It measures the radiance emitted from the earth in 8461 channels covering the spectral interval 645–2760 cm−1 (3.62–15.5 μm). IASI provides vertical distribution of meteorological parameters (temperature, humidity) and vertical distribution/total column content of various trace gases
(ozone, carbon monoxide, carbon dioxide, methane, sulphur dioxide, nitrous oxide, nitric acid etc). In the present work, statistical retrieval algorithm is developed for the retrieval of vertical profiles of temperature, water vapor and ozone using IASI radiances. Shadoz Ozonesonde and Radiosonde Observations SHADOZ (Southern Hemispheric ADditional OZonesonde) observation program consists of ozonsonde and radiosonde observations over eleven locations in the tropical region (Thompson et al. 2007). Each ozonesonde consists of a teflon pump, an ozone sensing electrochemical concentration cell (ECC) and an electronic interface board. The ECC sensor is made up of teflon cathode and anode chambers containing the platinum electrodes immersed in potassium iodide solutions of different concentrations. This sensor relies on the oxidation reaction of O3 with potassium iodide in solution and yields a precision better than ±3–5 % and an accuracy of about ±5– 10 % up to 30 km altitude (Smit et al. 2007). Radiosonde, based on resistance/capacitance based sensors, provides the measurement of temperature, pressure and relative humidity (Vaisala 1989). These sonde provides vertical profile of O3 and meteorological parameters from surface to about 30 to 35 km with vertical resolution of about 10 m. Fig. 1 depicts the geographical locations of SHADOZ sites and Table 1 shows the details of observational locations with total number of ozonesonde profiles available over these sites.
Methodology Radiative Transfer Model (RTTOV-10) Radiative Transfer for TIROS Operational Vertical Sounder (RTTOV) is a fast radiative transfer model for passive infrared and microwave satellite radiometers, spectrometers and interferometers (Matricardi 2009). It is an extensive, well validated, model for simulating satellite radiances using coefficient datasets for the channels/frequency regions of all important earth observing satellite instruments/sensors (Eyre and Woolf 1988). The forward radiative transfer model RTTOV (version 10) calculates the radiance corresponding to known atmospheric profiles of temperature and absorbing gaseous species. In a simplified way, the radiance Y can be written as: Y ¼ F ½X ðT ; q; O3 ; T s ; εÞ
ð1Þ
Where F is the forward model, X is the function of vertical distribution of temperature (T), water vapor (q) and ozone (O3) surface skin temperature (Ts) and emissivity (ε).
J Indian Soc Remote Sens Fig. 1 SHADOZ sites
Statistical Retrieval Algorithm
statistical retrieval method requires regression coefficient matrix (RC) of size K x J such that
In statistical retrieval approach, a statistical relationship is developed between large atmospheric dataset (temperature, humidity, ozone etc.) and corresponding observed radiances by satellite. This relationship is then applied to other observed radiances to retrieve the profiles of these atmospheric parameters. The large atmospheric dataset utilized for the development of algorithm is known as training dataset. Suppose there are N sounding pairs in the training dataset, L (j × 1) is the column vector containing j radiances observed by satellite based instrument, X is the (K ×1) vector, paired with L, which contains atmospheric variables at K different pressure levels in the atmosphere. Finally, let Lavg and Xavg represent average column vectors for N atmospheric soundings. Then
dX ¼ RC ðdLÞ
Table 1
ð2Þ
Where, dX = X-Xavg and dL = L-Lavg. Finally, RC matrix can be estimated as RC ¼ dX :dLT : dL:dLT
−1
ð3Þ
Once RC matrix is known, atmospheric parameters can be calculated using observed radiances from Eq.(2). Statistical Retrieval Approach Using Training/Testing Dataset To develop statistical retrieval algorithm, training/testing dataset has been compiled using SHADOZ ozonsonde and radiosonde observations. A very strict quality check has been
Shadoz locations and number of ozonesonde profiles
Locations
Latitude (o)
Longitude (o)
Watukosek Java Suva, Fiji Pago Pago, Am. Samoa
−7.50 −18.13 −14.23
112.6 178.4 −170.56
Irene, South Africa La Reunion, France Kuala Lumpur, Malaysia Costa Rica Hilo, Hawaii Paramaribo Surinam Nairobi, Kenya Ha Noi, Vietnam
−20.90 −21.06 2.73 −9.98 19.43 5.81 −1.27 21.01
28.22 55.48 101.7 −84.21 −155.04 −55.21 36.80 105.80
Elevation (m) 50 6 77.0 1524 24 17 899 11 7 1795 7
No. of ozonesonde 308 324 519 274 479 336 330 582 511 653 145
J Indian Soc Remote Sens
applied on soundings and 35 % profiles are removed from the further analysis (Profiles having bad sounding data, profiles having ozone values not available for pressure less than 13 hPa etc.). Total 2730 ozonesonde and radiosonde profiles are used in the present study. These profiles are randomly subdivided into two categories – 1) Training dataset: No of profiles =2451 (90 % of total ozonesonde observations) 2) Testing dataset: No of profiles =279 (10 % of total ozonesonde observations) The vertical profiles of temperature, water vapor mixing ratio and ozone are interpolated to RTTOV standard pressure levels (1100 hPa to 0.005 hPa). The RTTOV model is run for each atmospheric profile and set of radiances and brightness temperatures are generated at selected 300 channels (Collard 2007). The training dataset and corresponding set of brightness temperatures are used for the estimation of regression coefficient matrix following statistical regression approach. The RC coefficient matrix is applied on IASI simulated radiances for the retrieval of atmospheric profiles of temperature, pressure and ozone corresponding to testing dataset. In addition, these atmospheric profiles are Fig. 2 Variation of retrieved temperature with respect to observed temperature in different atmospheric regions. Coefficient of determination R-square, standard deviation error (SDE) and bias in data are also shown
retrieved using IASI real time observed radiances and compared with IASI observed atmospheric profiles to check the goodness of retrieval algorithm.
Results and Discussion Retrieved Atmospheric Profiles vs. Testing Ozonesonde Observations The profiles of atmospheric parameters are subdivided into four atmospheric slabs (950–500 hPa, 500–250 hPa, 25070 hPa and 70–10 hPa) to show the variation of retrieved data corresponding to actual data. Figure 2 shows the scatter plots of retrieved temperature as a function of actual in-situ ozonesonde observations. The scatter plots are of good quality with R-squares above 0.93. Hence most of the variability in simulated temperature can be explained with the in-situ temperature. In general, the values of slope are quite close to unity (0.95–0.99). The intercepts of the linear regression vary between 1 and 10 K and show overall positive tendency. The standard deviation error is small (0.76 to 2.34) and bias between retrieved and actual data (−0.007 to 0.002) is close to zero. This supports the overall good agreement of the simulated temperature
J Indian Soc Remote Sens
with the in-situ temperature and underpins the capability of RTTOV model. The variability of retrieved ozone is shown with respect to in-situ observations in Fig. 3. Ozone shows relatively large spread between retrieved and actual datasets as compared to temperature in different atmospheric regions. In the lower (950–500 hPa) and middle troposphere (500–250 hPa), the slope values 0.59 and 0.63 indicate that retrieved ozone does not follow ozone variation in testing datasets very nicely. The intercepts of the linear regression vary between 12.8 and 15.8 ppbv and show overall positive tendency. The standard deviation error and bias vary from 9 to12 ppbv and 0.014 to 0.182 ppbv respectively. In the upper troposphere and lower stratosphere (250–70-10 hPa), ozone retrieval is found to be better with slope values of linear regression (0.89 and 0.98) close to 1 and R2 value more than 0.85. Lower bias between retrieved and actual data (−0.19 and −6.65 ppbv) indicates less systematic error. Standard deviation error (43.3 ppbv at 250–70 hPa and 474 ppbv at 70–10 hPa) seems to be large but actually % error is lower in the upper troposphere and lower stratosphere (~10 % only) with respect to lower/middle troposphere (~30 %). The water vapor mixing ratio scatter plot is shown in Fig. 4. The retrieval of water vapor is found to be good with slope, intercept and R2 values of linear regression Fig. 3 Variation of retrieved ozone with respect to observed ozone in different atmospheric regions. Coefficient of determination R-square, standard deviation error (SDE) and bias in data are also shown
0.93, 0.38 g/kg and 0.84 respectively. The bias in retrieved data is 0.08 g/kg and standard deviation error is 1.71 g/kg. Water vapor scatter plot is shown only for pressure region of 950–500 hPa. Above this pressure level, water vapor mixing ratio is very small and can be neglected.
Error in Simulated Atmospheric Profiles The goodness of retrieval algorithm is tested by estimating errors in the vertical distribution of temperature, water vapor and ozone. The zenith angle of observation changes the path length by a factor of Secant of zenith angle, e.g., at 60° zenith angle the path length becomes twice of the path length at the nadir. This affects absorption by the water vapour and ozone. Beyond 60° the path lengths becomes very large and for most of the retrieval applications this is set as the threshold maximum limit. To study the effect of zenith angles, the regression coefficient matrix RC is generated in two modes 1) Considering zenith angle 0° 2) Considering six zenith angles 0, 36.8°, 48.8°, 55.1°, 60°, 63.6° at fixed path length intervals (secant of zenith angle)
J Indian Soc Remote Sens
Fig. 4 Variation of retrieved water vapor mixing ratio with respect to observed water vapor mixing ratio. Coefficient of determination R-square, standard deviation error (SDE) and bias in data are also shown
The retrieval of atmospheric profiles is made using RC matrix in both the cases and standard deviation error (temperature)/root mean square error (water vapor and ozone) are estimated at each pressure level. The vertical profiles of error for temperature, ozone and water vapor are shown in Fig. 5. The standard deviation error is small in temperature profile in the troposphere (0.7–1.3 K). This error increases (~2.7 K) at 100 hPa (near tropopause), where temperature inversion takes place. The retrieval with consideration of six zenith angles is found to be better particularly above 200 hPa. The standard deviation error is reduced by 0.2–0.3 K. As expected from scatter plot of retrieved and observed ozone values, RMSE is significant particularly in the troposphere. RMSE is maximum near earth surface (40 %) and
Fig. 5 Standard deviation error (K) in temperature retrieval and root mean square error (%) in ozone and water vapor retrieval using RTTOV simulated radiance
decreased to 25 % in the middle troposphere. The error again increased near tropopause to 30 % and further decreased to 7– 12 % for 50–10 hPa range. These results are consistent with those obtained by Dufour et al. (2012). The effect of zenith angle consideration is clearly evident in the troposphere and near 10 hPa pressure level. The error is reduced by 2–4 %. However, error increased slightly near the earth surface. RMSE in water vapor profile is estimated only from surface to 200 hPa range. The RMSE is 10 % near the earth surface and increases to 30 % at 750 hPa. RMSE slightly decreases with decreasing pressure from 750 hPa to 200 hPa. The effect of zenith angle consideration is very clearly marked in water vapor RMSE profile. The error has been improved by 4–7 % between 750 hPa and 200 hPa. RMSE is estimated in retrieved total column ozone and shown in Fig. 6. The average RMSE is found to be 1.9 ± 1.4 %. RMSE values are found to be less than 4 % for most of the profiles. However, in one exceptional case, its value reached at 10 %.
Error in Retrieved Atmospheric Parameters Using IASI Real-Time Observations The regression coefficients RC computed with zenith angle predictor is applied on IASI cloud-cleared radiances and retrieved products are compared with the IASI L2 products. The error in IASI retrieval is found to be higher with respect to retrieval made using in-situ ozonesonde observations (Fig. 7). The standard deviation error in temperature profile varies from 2.0–2.5 K from surface to about 130 hPa. The temperature error increases and becomes 3.5 K at 78 hPa and decreases further with decreasing pressure. Ozone RMSE is 40 % near surface which decreases to 20–30 % in the troposphere and shows a peak of 40 % at tropopause. Ozone RMSE decreases
J Indian Soc Remote Sens
Fig. 6 Root mean square error (%) in the total column ozone retrieved by RTTOV simulated radiances
Fig. 8 Root mean square error (%) in the total column ozone retrieved from IASI radiances
further with decreasing pressure (~10 %). The error in water vapor retrieval using IASI radiance is significantly higher with respect to retrieval using ozonesonde testing data set. The error is found to be 35–55 %. RMSE in total column ozone obtained from IASI observed radiances is shown in Fig. 8. Average RMSE is found to be 3.8 ± 3.2 %. RMSE values are found to be higher than error found in total column ozone retrieval using in-situ data by 4 % for most of the profiles. However, in some cases, RMSE is found in the range of 8–18 %. The cause of higher errors with actual IASI observations is due to the fact that regression coefficients were computed using simulated datasets whereas the actual observations may have biases w.r.t. radiative transfer model simulated brightness temperatures. To overcome this problem a bias correction procedure is required to make IASI observations consistent with the radiative transfer model simulations.
Summary and Conclusions
Fig. 7 Standard deviation error (K) in temperature retrieval and root mean square error (%) in ozone and water vapor retrieval using IASI radiances
Statistical retrieval technique is used for the retrieval of vertical profiles of temperature, water vapor and ozone using IASI simulated and observed radiances. Shadoz Ozonesonde observations are used as training/testing datasets for the calculation of regression coefficient matrix RC and assessment of the retrieved profiles of ozone and meteorological parameters. The salient features of retrieval using IASI simulated and observed radiances are given below. 1) The vertical variation of standard deviation error in temperature, estimated using simulated and observed profiles using testing ozonesonde data, is found to be in the range of 0.7–2.7 K. The standard deviation error estimated using IASI real time observations (retrieved and observed) is relatively higher (2 to 3.5 K) than retrieval error corresponding to the simulated observations.
J Indian Soc Remote Sens
2) The vertical profiles of RMSE estimated using simulated and observed profiles of ozone and water vapor (using testing ozonesonde data) is found to be in the range of 5–30 % and 10–30 % respectively. The RMSE profile estimated using IASI observations (retrieved and observed) is relatively higher particularly for water vapor for which RMSE reached more than 50 %. 3) Total column ozone RMSE is also estimated for simulated testing ozonesonde measurements and corresponding retrieved profiles. The simulated RMSE is found to be 1.9 ± 1.4 %. Total column ozone RMSE, estimated from actual IASI observations and corresponding retrieved profiles, is found to be 3.8 ± 3.2 %. 4) Consideration of zenith angle as predictor in regression coefficient matrix estimation has improved retrieval error of atmospheric parameters. Standard deviation error for temperature is improved by 0.2–0.3 K and RMSE for ozone and water vapor is improved by 2–4 % and 4– 7 % respectively in different atmospheric regions. In general, the accuracy of 1–2° K for temperature and 15– 30 % for water vapour are considered significant in the troposphere. Delcloo et al. (2011) reported accuracy of ozone in the range of 10–20 % in the stratosphere and 30 % in the troposphere for IASI instrument. The error estimated in the present study are found to be consistent with standard accuracies of these parameters. However, the differences in the retrieval statistics of simulated and real IASI dataset is due to the fact that radiative transfer model dependent bias correction procedure is not applied and the spectral surface emissivity in the real data is not prescribed. The retrieval algorithm needs improvement in terms of the inclusion of non-linear retrieval techniques and surface emissivity prescription.
Acknowledgments We thank S. K. Saha, S. P. S. Kushwaha, A. Senthil Kumar, Y. V. N. Krishna Murthy and A. S. Kiran Kumar for their encouragement and support. We are grateful to RTTOV team members for making radiative transfer model freely available. We are also thankful to SHADOZ team members for providing ozonesonde and radiosonde observations over various locations in the tropical region.
References Collard, A. D. (2007). Selection of IASI Channels for use in Numerical Weather Prediction. Quarterly Journal of the Royal Meteorological Society, 133, 1977–1991. Delcloo A, Daniel Hurtmans, Pierre-Fran¸cois Coheur, Cathy Clerbaux. Validation of IASI ozone profiles using balloon sounding data. 2011 EUMETSAT meteorological satellite conference, Sep 2011, Oslo, Norway. Dufour, G., Eremenko, M., Griesfeller, A., Barret, B., Le Flochmoen, E., Clerbaux, C., Hadji-Lazaro, J., Coheur, P. F., & Hurtmans, D. (2012). Validation of Three Different Scientific Ozone Products Retrieved from IASI Spectra Using Ozonesondes. Atmospheric Measurement Techniques, 5, 611–630. Ellingsen, K., Gauss, M., Van Dingenen, R., Dentener, F. J., Emberson, L., Fiore, A. M., et al. (2008). Global Ozone and air Quality: A Multi-Model Assessment Ofrisks to Human Health and Crops. Atmospheric Chemistry and Physics Discussions, 8, 2163–e2223. Eyre, J. R., & Woolf, H. M. (1988). Transmittance of Atmospheric Gases in the Microwave Region: A Fast Model. Applied Optics, 27, 3244– 3249. Gauss, M., Myhre, G., Isaksen, I. S. A., Grewe, V., Pitari, G., Wild, O., et al. (2006). Radiative Forcing Since Preindustrial Times due to Ozone Changes in the Troposphere and the Lower Stratosphere. Atmospheric Chemistry and Physics, 6, 575–599. Herring, D.D. & King, M.D. (2000). Space-based observations of the Earth. In Murdin, P. (Ed.), Encyclopedia of Astronomy and Astrophysics. Institute of Physics Publishing, pp. 2959-62. Hilton, F., Armante, R., August, T., Barnet, C., Bouchard, A., Camy-Peyret, C., et al. (2012). Hyperspectral Earth Observation from IASI: five years of accomplishments. American Meteorological Society, 93, 347–370. Kidder, S. Q., & von der Haar, T. H. (1995). Satellite meteorology-an introduction. Academic Press. Matricardi, M. (2009). Technical Note: An Assessment of the Accuracy of the RTTOV Fast Radiative Transfer Model Using IASI Data. Atmospheric Chemistry and Physics, 9, 6899–6913. Smit, H. G. J., Straeter, W., Johnson, B. J., Oltmans, S. J., Davies, J., Tarasick, D. W., et al. (2007). Assessment of the Performance of ECCozonesondes Under Quasi Flight Conditions in the Environmental Simulationchamber: Insights from the Juelich Ozone SondeIntercomparisonExperiment (JOSIE). Journal of Geophysical Research, 112, D19306. doi:10.1029/2006JD007308. Thompson, A.M., Witte, J.C., Smit, H.G.J., Oltmans, S.J., Johnson, B.J., Kirchhoff, V.W.J H., & Schmidlin, F.J. (2007). Southern Hemisphere Additional Ozonesondes (SHADOZ) 1998-2004 tropical ozone climatology: 3. Instrumentation, station-to-station variability, and evaluation with simulated flight profiles. Journal of Geophysical Research, 112, D03304. doi:10.1029/2005JD007042. Vaisala Inc. (1989), RS 80 Radiosondes. Upper-Air Systems product information. Reference No. R0422-2.Vaisala Inc., USA.