Elsevier

Atmospheric Research

Volume 181, 15 November 2016, Pages 124-132
Atmospheric Research

Retrieval of atmospheric properties with radiometric measurements using neural network

https://doi.org/10.1016/j.atmosres.2016.05.011Get rights and content

Highlights

  • Microwave radiometer measures brightness temperatures (BT) at two frequency bands.

  • Retrieval techniques are used to obtain temperature and humidity profiles from BT.

  • Three retrieval techniques are used to get reliable atmospheric profiles from BT.

  • Results are compared with radiosonde data to validate the efficacy of techniques.

  • Back Propagation neural network provides the best matching with minimal errors.

Abstract

Microwave radiometer is an effective instrument to monitor the atmosphere continuously in different weather conditions. It measures brightness temperatures at different frequency bands which are subjected to standard retrieval methods to obtain real time profiles of various atmospheric parameters such as temperature and humidity. But the retrieval techniques used by radiometer have to be adaptive to changing weather condition and location. In the present study, three retrieval techniques have been used to obtain the temperature and relative humidity profiles from brightness temperatures, namely; piecewise linear regression, feed forward neural network and neural back propagation network. The simulated results are compared with radiosonde observations using correlation analysis and error distribution. The analysis reveals that neural network with back propagation is the most accurate technique amongst the three retrieval methods utilized in this study.

Introduction

Atmospheric disturbances play a crucial role in promoting vertical circulations by transporting heat and moisture from the boundary layer to top of the atmosphere which leads to the growth of convective phenomena causing immense damage to life and property. As a result, remote sensing of temperature and humidity profiles is important in various walks of life. At present, many techniques are being used to measure the atmospheric properties of the lower atmosphere, namely, radar, radio-sounding or satellite imagery. Radiosondes launched from regional weather services measure and record the atmospheric profiles of temperature, humidity and wind-speed with a series of instability indices. However, radiosonde data are available only twice a day, which may not be adequate to reveal a rapidly varying state of the atmosphere. On the other hand, remote sensing of temperature and humidity from satellite imagery produces better spatial coverage with a poor temporal resolution.

Ground-based microwave radiometers have the advantage of continuously monitoring the atmosphere up to heights of 10 km with a better temporal resolution and can cover the temporal and spatial gaps of synoptic networks by radiosonde and satellite measurements. In addition, these passive instruments perform in all weather conditions (Xu et al., 2014, Xu et al., 2015). In the last few decades, profiling of various atmospheric parameters by MWRs has found numerous applications in atmospheric studies. As a result of latest technical improvements, radiometers can be used for profiling both temperature and humidity simultaneously (Solheim et al., 1998, Güldner and Spänkuch, 2001, Cadeddu et al., 2013, Renju et al., 2015, Harikishan et al., 2014). An additional advantage of MWR is the high accuracy of measurement of integrated liquid water (Westwater, 1978, Peter and Kämpfer, 1992) and measurement of the liquid water profile (Politovich et al., 1995, Solheim et al., 1998, Ware et al., 2003, Crewell et al., 2009, Ebell et al., 2010, Calheiros and Machado, 2014, Campos et al., 2014, Serke et al., 2014). This instrument measures the radiation intensity at a number of frequency channels in the microwave spectrum that are dominated by atmospheric water vapor and molecular oxygen emissions (Rose and Czekala, 2003, Knupp et al., 2009, Cadeddu et al., 2013, Wulfmeyer et al., 2015). It measures the radiated emission in the form of brightness temperatures and converts them to humidity, liquid and temperature profiles by a suitable transformation to a number of heights, providing a continuous measurement of different parameters. All weather operations provided by ground based microwave radiometers have made them useful in weather prediction and analysis. In the recent years, many attempts have been carried out to predict extreme weather events using microwave radiometers with a reasonable accuracy (Won et al., 2009, Madhulatha et al., 2013, Maitra et al., 2014, Cimini et al., 2015, Chakraborty et al., 2014, Chakraborty and Maitra, 2016, Illingworth et al., 2015).

Issues related to the accuracy of retrieval techniques used by microwave radiometers have been a matter of interest to researchers over the past few decades. In the past, many separate attempts have been made to retrieve the temperature and relative humidity profiles from brightness temperature (BT) measurements by a radiometer (C.-Mercader and Staelin, 1995, Solheim et al., 1998, Westwater, 1997, Ware et al., 2003, Lüdi et al., 2003, Matzler and Morland, 2009, Löhnert and Maier, 2012, Stähli et al., 2013, Westwater et al., 1999). According to Westwater et al. (2005) and Chan (2010), retrievals of thermodynamic profiles by MWR are done by two techniques, namely regression analysis and neural networks. The neural network technique is found to be a more reliable technique as it provides more realistic humidity profiles than that obtained with radiosondes, particularly in rainy conditions. According to Knupp et al. (2009) and Chan (2009), the retrieval of temperature and humidity profiles from MWR is done by neural network methods based on radiosonde data and a radiative transfer model. Ramesh et al. (2015) has used an adaptive neuro-fuzzy inference system (ANFIS) to retrieve profiles of temperature and humidity up to 10 km over the tropical station Gadanki (13.5 N, 79.2 E), India and have obtained considerably better results in temperature and humidity profiles. Ware et al. (2003) compared the radiometric, radiosonde and forecast soundings for evaluating the accuracy of radiometric temperature and water vapour soundings. Hewison (2007), Cimini et al., 2010, Cimini et al., 2011, Cimini et al., 2015 and Ishihara (2015) used a one-dimensional variational (1-DVAR) to retrieve atmospheric profiles of temperature and water content from brightness temperatures and compared the retrieved profiles with radiosonde observations. The statistical analysis revealed significant improvements in both the profiles compared to other retrieval techniques. Sánchez et al. (2013) has reported a bias in the temperature and humidity measurements obtained by MWR and applied some correction using linear adjustment technique, which has significantly improved the accuracy of temperature and humidity profiles. Again, some research attempts have also pointed at the existence of a wet bias below 5 km and a dry bias above it in the relative humidity profile (Chan, 2009, Xu et al., 2014). Radiosonde dry bias in moist, warm air and wet bias in dry, cold air (Turner et al., 2003, Miloshevich et al., 2004, Miloshevich et al., 2006, Miloshevich et al., 2009, Vömel et al., 2007, Yoneyama et al., 2008, Wang et al., 2013, Ansari et al., 2015) may contribute to these biases. The wet bias might be due to the presence of an elevated moist layer above the radiometer that is not encountered along the uncontrolled flight path of the radiosonde. Another possible reason might be the variable weather conditions generated due to water vapor and temperature variability in the lower troposphere. This variability may not be detected by the inversion technique used by the radiometer as it utilizes regression coefficients from a large number of radiosonde profiles which have an averaging effect. Apart from dry and wet biases, some other inaccuracies have also been reported by Löhnert and Maier (2012) and Paine et al. (2014). These types of errors are called open side-mount cryogenic calibration errors that appear as brightness temperature offsets in oxygen absorption bands and change after each absolute calibration due to (1) water condensation on the aluminium plate reflector connecting the cold load and the radiometer during LN2 calibration, and (2) oxygen condensation in liquid nitrogen (Paine et al., 2014). In order to prevent this, MWR radome blowers and heaters should be operated with maximum power. These types of errors can also be reduced by using pressurized, closed, top-mount cryogenic targets as in Radiometrics MP-3000 radiometers (Miacci et al., 2014) resulting in brightness temperature accuracies < 0.5 K. However, in the present case, water vapor condensation error is expected to be minimized since the radome blowers and heaters were in full power mode during the calibration process, and the maximum change in calibration temperature due to oxygen condensation is 2 K (Paine et al., 2014).

In the recent past, many other attempts have also been made to calibrate the radiosonde and radiometric measurements (Westwater, 1997, Miloshevich et al., 2001, Miloshevich et al., 2004, Miloshevich et al., 2006, Miloshevich et al., 2009, Löhnert and Maier, 2012, Turner et al., 2003, Mattioli et al., 2007, Ebell et al., 2010, Kottayil et al., 2012, Miacci et al., 2014, Chan and Lee, 2015). However, most of these attempts are reported for temperate locations and a very limited research has been reported on the retrieval accuracy obtained at a tropical location where the variability of atmospheric parameters are more prominent due to strong monsoon making the retrieval errors higher compared to that in temperate regions. Hence, a detailed analysis of retrieval errors is necessary for real time atmospheric monitoring by radiometers at tropical locations.

The main focus of the present study is to highlight the retrieval methodologies used by a ground based radiometer at Kolkata (22° 32′N, 88° 20′E), a tropical location. Three different techniques of retrieval have been employed, namely linear regression, feed forward neural networks and neural back propagation network. In each case the simulated results have been compared with those obtained from radiosonde data using correlation analysis, error distribution and error significance calculations. The analysis has shown that neural network with memory can give better agreement with radiosonde profiles compared to other retrieval techniques.

Section snippets

Experimental setup and data

A multi-frequency radiometer (RPG-HATPRO) operated at the Institute of Radio Physics and Electronics, University of Calcutta, Kolkata, India has been used to obtain the temperature and humidity profiles. It consists of two receiving sections (22.24–31.4 GHz and 51.3–58 GHz) along with a noise diode, a data acquisition system, rain sensor, GPS clock, pressure sensor, and a temperature sensor (Rose and Czekala, 2009). It measures the brightness temperatures at these frequency bands and converts

Methodology

The radiometer used in the present study measures the radiated and scattered energy from the surroundings in form of brightness temperatures at two frequency bands. The first band (22.24–31.4 GHz) is sensitive to water vapor and hence is used for humidity sensing while the second one, namely 51.26–58 GHz is sensitive to oxygen absorption and hence is used for temperature sensing. Temperature and humidity profiles of atmosphere are obtained from brightness temperatures using suitable retrieval

Comparative study of retrieval techniques

In the previous section, three retrieval techniques have been utilized. Now, these techniques are compared to investigate which retrieval technique has provided the highest level of matching with the radiosonde observations. A fourfold analysis is incorporated to find the best retrieval technique for this radiometer. Further details about the comparative studies have been elaborated in the following subsections.

Conclusions

A radiometer measures brightness temperatures at various frequencies and utilizes retrieval techniques to obtain atmospheric profiles from them. In this paper, three different techniques have been utilized to retrieve the atmospheric parameters (temperature and relative humidity) from radiometer derived brightness temperatures at various frequencies and in each case the simulated results have been compared with simultaneous radiosonde measurements. After a detailed analysis of all the

Acknowledgement

The financial support provided by ISRO (ISRO/RES/4/614/2014-15 dated 02.06.2014) under the projects (1) “Ku/Ka Band Channel Modelling for SATCOM Links over the Indian Region” and (2) “Space Science Promotion Scheme” are thankfully acknowledged.

References (58)

  • C.R. C.-Mercader et al.

    Passive microwave relative humidity retrievals using feedforward neural networks

    IEEE Trans. Geosci. Remote Sens.

    (1995)
  • M. Cadeddu et al.

    Atmospheric radiation measurement (ARM) program network of microwave radiometers instrumentation, retrievals

    Atmos. Meas. Tech.

    (2013)
  • P.W. Chan

    Performance and application of a multi-wavelength, ground based microwave radiometer in intense convective weather

    Meteorol. Z.

    (2009)
  • P.W. Chan

    The use of neural network retrieval for thermodynamic profiles of a ground-based microwave radiometer

  • D. Cimini et al.

    Thermodynamic atmospheric profiling during the 2010 Winter Olympics using groundbased microwave radiometry

    IEEE Trans. Geosci. Remote Sens.

    (2011)
  • D. Cimini et al.

    Forecast indices from ground-based microwave radiometer for operational meteorology

    Atmos. Meas. Tech.

    (2015)
  • D. Cimini et al.

    Temperature and humidity profiling in the Arctic using ground-based millimeter-wave radiometry and 1DVAR

    Trans. Geosci. Remote Sens.

    (2010)
  • S. Crewell et al.

    Can liquid water profiles be retrieved from passive microwave zenith observations?

    Geophys. Res. Lett.

    (2009)
  • J. Güldner et al.

    Remote sensing of the thermodynamic state of the atmospheric boundary layer by ground-based microwave radiometry

    J. Atmos. Ocean. Technol.

    (2001)
  • G. Harikishan et al.

    Macrophysical and microphysical properties of monsoon clouds over a rain shadow region in India from ground-based radiometric measurements

    J. Geophys. Res.

    (2014)
  • S. Haykin

    Neural Network—A Comprehensive Foundation

    (1994)
  • T. Hewison

    1D-VAR retrieval of temperature and humidity profiles from a ground-based microwave radiometer

    Trans. Geosci. Remote Sens.

    (2007)
  • A. Illingworth et al.

    Ground-based passive microwave profiling during dynamic weather conditions

    J. Atmos. Ocean. Technol.

    (2015)
  • H. Ishihara

    1DVAR Analysis, Japanese Cloud Seeding Experiments on Precipation Augmentation Meeting (JCSEPA)

    (2015)
  • K. Knupp et al.

    Ground-Based Passive Microwave Profiling during Dynamic Weather Conditions

    J. Atmos. Oceanic Technol.

    (2009)
  • A. Kottayil et al.

    On the importance of Vaisala RS92 radiosonde humidity corrections for a better agreement between measured and modeled satellite radiances

    J. Atmos. Ocean. Technol.

    (2012)
  • U. Löhnert et al.

    Operational profiling of temperature using ground-based microwave radiometry at Payerne: prospects and challenges

    Atmos. Meas. Tech.

    (2012)
  • A. Lüdi et al.

    The retrieval of temperature profiles with the ground based radiometer system ASMUWARA

  • A. Madhulatha et al.

    Nowcasting severe convective activity over southeast India using ground based microwave radiometer observations

    J. Geophys. Res.: Atmos.

    (2013)
  • Cited by (0)

    View full text