Comparison of the tropical tropopause in ECMWF reanalyses and GPS-RO satellite observation

(1)

Comparison of the tropical tropopause in ECMWF reanalyses and GPS-RO satellite observation

Johanne Mehren

Thesis submitted for the degree of Master of Science in Meteorology

60 Credits

Department of Geoscience

Faculty of Mathematics and Natural Sciences UNIVERSITY OF OSLO

June 14, 2020

(2)

c 2020 Johanne Mehren

Comparison of the tropical tropopause in ECMWF reanalyses and GPS-RO satellite observation

This work is published digitally through DUO – Digitale Utgivelser ved UiO

http://www.duo.uio.no/Printed: Reprosentralen, University of Oslo

(3)

Abstract

The tropical tropopause layer (TTL) acts as the gateway for water vapor and other chemical tracers to the stratosphere, and is important for the chemical composition and radiative balance. In this thesis, two reanalyses from the European Center for Medium-Range Weather Forecasting (ECMWF), ERA5 and ERA-Interim, are evaluated and compared to GPS Radio Occultation (GPS-RO), as an extension and deepening of the SPARC Reanalysis Intercom- parison Project (S-RIP) report, Chapter 8 (under revision) and Tegtmeier et al., 2020. The temperature and height of two tropical tropopause definitions are studied from 2007-2018, the cold point tropopause (CPT) and the lapse rate tropopause (LRT). The aim of the thesis is to compare the newly published ERA5 to its predecessor, ERA-Interim, by using satellite data as a reference point, and evaluate if seasonal and interannual variations are well represented in reanalyses when compared to GPS-RO. The updates in ERA5 includes highly improved temporal and spatial resolution, new model cycle and assimilation of more observations.

The climatological tropical annual mean temperature bias is positive and the height bias is negative for both reanalyses relative to GPS-RO. ERA5 has the smallest differences for both tropopause definitions, 0.2 K for CPT and 0.1 K for LRT, likely due to high vertical resolution, representing small changes in the temperature profiles accurately. The larger differences in ERA-Interim, 0.4 K for CPT and 0.7 K for LRT, maximises over central Africa and east Pacific.

The seasonal cycle is well represented in ERA5, while the tropopause height difference varies considerably throughout the year in ERA-Interim. Areas with increased CPT temperature biases are central Africa, the maritime continent and the east Pacific, which maximise during March and April for ERA5 (+0.4 K) and during January and February for ERA- Interim (+0.8 K). The effect of the El Niño Southern Oscillation (ENSO) is studied and the difference between ERA5 and GPS-RO is not enhanced during ENSO events, while the LRT temperature bias in ERA-Interim shows a small increase over the maritime continent during La Niña events. The shift in location of the lowest temperatures and highest altitudes during boreal summer to the area of the Asian summer monsoon region is captured by both reanalyses. ERA5 and ERA-Interim have enhanced CPT temperature biases south of the Indian

(4)

peninsula, and an additional LRT temperature and height maximum is found over the Ara- bian peninsula in ERA-Interim.

ERA5 produces tropical tropopause temperatures and heights in close agreement to GPS- RO, and improves the representation of the tropopause compared to its predecessor, ERA- Interim.

ii

(5)

1 Introduction

The tropical tropopause layer (TTL) holds properties of both the static stable stratosphere and the well-mixed troposphere and acts as a gateway for air transported to the stratosphere.

The entrainment of water vapor (WV) into the stratosphere is important for the chemical composition and is regulated by the low temperature in the upper boundary of the TTL, as WV undergoes freeze-drying here (Brewer, 1949). Air entering the stratosphere through the TTL can have an impact on the chemistry in this layer by contributing to ozone budget and change the radiative balance in the troposphere and the stratosphere (Fueglistaler et al., 2009). Two tropopause definitions are used in this thesis, 1) the cold point tropopause (CPT), which is the altitude at which the minimum temperature is found in the vertical temperature profile (Highwood and Hoskins, 1998) and 2) the lapse rate tropopause (LRT) , defined as the lowest level at which the lapse rate decreases to 2^◦C/km or less (WMO, 1957).

The upper TTL is affected by both annual and interannual variations. The tropopause has been documented in several studies to be coldest and highest during boreal winter (Seidel et al., 2001, Fueglistaler et al., 2009), mainly due to the Brewer-Dobson Circulation (BDC), which transports air from the tropical troposphere, through the TTL and to the stratosphere, and sinks at higher latitudes. The lowest CPT and LRT temperatures are usually found in areas of deep convection, like the maritime continent, central Africa and South America (Gettelman, Salby, et al., 2002). Interannual variability includes the El Niño Southern Os- cillation (ENSO), where changes in the equatorial and east Pacific sea surface temperature changes the Walker circulation. Previous studies have linked the different phases of ENSO to significant changes in convection patterns, having a strong influence on the location of the lowest tropopause temperature and WV transport through the TTL (Gettelman, Randel, et al., 2001). During boreal summer, the Asian monsoon dominates the tropical circulation, shifting the convection northwest from the west Pacific, and the TTL has the lowest temperatures and highest altitudes in this region (Devasthale and Fueglistaler, 2010).

Observations provide present day and historical records of the weather and climate, but are limited in space and time. Reanalyses solve this problem by using a forecast model as a first guess of the atmospheric state in the past, combined with the assimilation of obser-

(8)

vations available, which corrects the forecast, and provides a consistent understanding of atmospheric components for the last decades. The benefit of using reanalyses is the fixed forecast model, i.e. no changes due to updates in the model and produces analyses of the atmosphere based on available observations on fine horizontal and vertical scales (Fujiwara et al., 2017). Reanalyses have been used to evaluate Chemistry Climate Models (CCM) , and work as a guidance of how global models are able to simulate changes in the future due to climate change (Gettelman, Birner, et al., 2009). In this thesis, two reanalyses products will be used, both from the European Center for Medium-Range Weather Forecasting (ECMWF), the ERA-Interim and the newest reanalysis, ERA5. When evaluating two reanalyses from the same research institute, any differences between them can be due to updates in the model cycle, implementation of schemes, assimilation method and usage of observations.

Since the launch of the first Global Positioning System Radio Occulation (GPS-RO) satellite mission in 1995, several missions have followed. The technique "provides accurate high- vertical resolution of temperature in the stratosphere and the troposphere" (Anthes, 2011). In contrast to conventional data, such as radiosondes, GPS-RO has a global coverage. GPS-RO data works as a reference point for the comparison between ERA5 and ERA-Interim.

This thesis is an extension of Chapter 8 of the Stratosphere-troposphere Processes and their role in Climate (SPARC) Reanalysis Intercomparison Project (S-RIP) report (under revision), which focuses on the TTL, and Tegtmeier et al., 2020, a part of the S-RIP. The S-RIP aims to provide information about differences among current reanalyses products and their causes, providing guidance to reanalysis data users (Fujiwara et al., 2017). In Tegtmeier et al., 2020, eight reanalysis data sets are evaluated and compared to GPS-RO data for the period 1980-2010, but the focus period is from 2002-2010 to include ERA5 data. This thesis includes a comparison of the period from 2007 to 2018, and in addition addresses seasonal and interannual variations in the TTL.

The aim of this thesis is to address the following questions:

1. What are the mean differences between GPS-RO data and ECMF reanalyses for the TTL,

2. Where and when are the largest differences,

3. Can the difference be linked to the seasonal and interannual changes of the TTL and do the questions above provide an answer to

2

(9)

CHAPTER 1. INTRODUCTION

4. Is the representation of the tropical tropopause improved in ERA5 relative to ERA- Interim

Background of the tropopause and its characteristics are presented in 2. Information about the reanalyses and GPS-RO data, and how the TTL definitions are calculated is provided in Chapter 3. The results are shown and discussed in Chapter 4, followed by a short summary and outlook in Chapter 5.

(10)

(11)

2 Background

This chapter provides relevant background for the thesis. The structure of the atmosphere is described in Section 2.1, followed by an explanation of the Brewer-Dobson Circulation in Section 2.2. In Section 2.3, the TTL and the two tropopause definitions are described.

Seasonal variations are discussed in Section 2.4. Section (2.5 and 2.6, present how the ENSO and Asian summer monsoon have an impact on the TTL. The final section is a presentation of S-RIP and results in Tegtmeier et al., 2020.

2.1 General structure of the Atmosphere

The atmosphere is divided into four layers based on the vertical temperature structure, with the transition layer between them marked as apause. In the troposphere, the solar radiation is absorbed by the surface. The temperature decreases strongly with height (statically un- stable), as seen in Figure 2.1, due to less absorption in the air above. This changes at the tropopause, and in the stratosphere, the temperature increases with height (statically stable) due to absorption of ultraviolet radiation by ozone (Wallace and Hobbs, 2006). (Wallace and Hobbs, 2006)

The troposphere is characterized by strong vertical mixing of air due to the meridional tem-

Figure 2.1: The vertical temperature profile of the atmosphere. From Marshall and Plumb, 2008.

(12)

2.2. THE BREWER-DOBSON CIRCULATION

perature gradient, while the inversion layer in the stratosphere suppresses this. Thus, the tropopause distinguishes the two different layers and different regimes.

2.2 The Brewer-Dobson Circulation

The theory behind a global mass circulation, transporting tropospheric air into the stratosphere in the tropics, and moving poleward with descend in middle and high latitudes, was presented over fifty years ago, and is usually referred to as the Brewer-Dobson Circulation (BDC) (Butchart, 2014). Dobson et al., 1925, noted that only such a circulation could explain the high concentrations of stratospheric ozone at high latitudes and low concentrations at equator, given that ozone production occurs at the stratospheric equator. A few years later Brewer, 1949, linked the observed upper air water vapor measurements to a slow ascend- ing motions at the equator into the stratosphere, and descent at the poles. As explained by Holton et al., 1995, the transport of air is due to a "extratropical pump" or in some cases referred to as "Rossby-wave pump", which is driven by upward propagating wave dissipa- tion and a following wave drag in the troposphere. The wave drag works as a pump, sucking up air in the tropic troposphere, into the stratosphere and down to higher latitudes. This have an impact on the radiative equilibrium- air pushed upwards (downwards) in the tropical (high latitudes) stratosphere is cooled (warmed) adiabatically, causing the temperatures to be above (below) radiative equilibrium. The wave induced forces are strongest in winter, and the circulation acts as a single-cell transport in the winter hemisphere which affects the local convection. Stronger land-sea contrasts in the Northern Hemisphere (NH) compared to the Southern Hemisphere (SH), induces stronger wave forcing for the NH winter.

2.3 The tropical tropopause

As described in Section 2.1, the tropopause distinguishes the troposphere from the stratosphere, and the location of the tropopause varies meridionally, zonally and with seasons.

The tropical tropopause layer (TTL) is presented in Section 2.3.1, along with dynamical and chemical description. Two definitions of the tropical tropopause will be discussed in Section 2.3.2 and Section 2.3.3.

2.3.1 TTL

The tropical tropopause TTL was first introduced as a concept in the 1990s, as a transition layer holding properties of both the troposphere and the stratosphere. The lower boundary is usually at the main convective outflow, at about 12 km, and the upper boundary is the cold

6

(13)

CHAPTER 2. BACKGROUND

point tropopause at 17 km.

The TTL is the gateway for tropospheric chemical components to the stratosphere, such as water vapor, as this is a region of strong convection. Water vapor (WV) in the TTL is important in the radiation budget and in the chemical composition of the stratosphere. It is therefore important to study the processes controlling WV in the TTL. Stratospheric WV is mainly controlled by the slow ascent of the BDC, encountering cold tropopause temperatures and is dehydrated through freeze-drying, thus making a strong connection between cold TTL temperatures and a dry stratosphere (Brewer, 1949). In addition, deep convection can penetrate through the TTL, and directly inject WV in the lower stratosphere (Fueglistaler et al., 2009).

Today, the BDC is seen as the main reason behind seasonally and annually averaged mass exchange across the TTL (Butchart, 2014). Yulaeva et al., 1994, explained how the annual temperature cycle of the TTL could be linked to the BDC by investigating the temperature in the tropics and the extratropics, which are out of phase. A dynamical explanation is given- stronger circulation due to increased wave induced forcing in the NH winter requires stronger diabatic heating in the tropics, compensated with cooling in the extratropics.

Along equator, northeast and southeast trade winds converge, forming a band of convergence, the intertropical convergence zone (ITCZ). The location of the ITCZ changes with the seasonal variations of wind patterns, like the Asian monsoon (Wallace and Hobbs, 2006,

Figure 2.2: Illustration of the TTL. The black thick line denotes the tropopause. From Montzka et al., 2010

(14)

2.3. THE TROPICAL TROPOPAUSE

pp.15-16). Large-scale convection forms deep convective clouds, which have an impact on the radiation budget, the static stability and WV budget. Deep convective clouds are mainly 10-15 km high, transporting air from the boundary layer, as illustrated in Figure 2.2 by the main convective outflow. Also shown is convection reaching the TTL, often called over- shooting convection, and penetrating the stratosphere, which rarely occurs (Fueglistaler et al., 2009). In the TTL, clouds are mainly optical thin, such as cirrus clouds (Ci in Figure 2.2), due to the limitation of WV here (Fueglistaler et al., 2009).

Ozone is primarily produced in the stratosphere but can also be photochemically produced the troposphere. In the TTL, the abundance of ozone is controlled by trosphere-stratosphere exchange and chemical reactions (Fueglistaler et al., 2009). Ozone can have a radiative effect on the TTL. Indeed, there is a correlation between colder CPT temperatures and small tropopause ozone values (Gettelman, Birner, et al., 2009).

2.3.2 Cold point tropopause

The cold point tropopause (CPT) is defined as the level where the vertical temperature profile reaches its minimum (Highwood and Hoskins, 1998), and is located in the upper TTL (Figure 2.2). The CPT annual mean temperature is lowest at equator, with values of 192 K.

Above the CPT, the stratospheric inversion begins. The CPT is of great interest as it marks the coldest temperature air encounters when transported through the TTL and works as a boundary condition for water vapor entering the stratosphere (Mote et al., 1996). The annual CPT height is higher off equator due to zonally variable features, like an increase in height in JJA due to the Asian monsoon (Fueglistaler et al., 2009).

2.3.3 Lapse rate tropopause

By WMO, the tropopause is defined as the lowest level at which the lapse rate decreases to 2^◦ C/km or less, provided that the average lapse rate between this level and all higher levels within 2 km do not exceed 2 ^◦ C/km (WMO, 1957). The LRT is located below the CPT, just above the level of clear-sky-zero radiative heating in Figure 2.2, and has a higher temperature and lower height compared to CPT. Convection usually does not influence the tropopause temperature above the LRT and rarely penetrates into the stratosphere. It also marks the shift in static stability, as the stratospheric temperature is controlled by radiative heating. Unlike CPT, the LRT is a global definition. The LRT height is highest at equator and decreases sharply close to the jet stream due to increased static stability (Seidel et al.,

8

(15)

2001).

2.4 Seasonal variations

Several studies have investigated the climatology of the tropopause, using different observations, such as radiosondes (Seidel et al., 2001) and satellite data (Kim and Son, 2012).

As mentioned in Section 2.3.1, the annual variations in tropopause temperatures are mainly controlled by the BDC seasonality. The BDC is strongest in boreal winter and weakest during boreal summer. This can be seen in the climatological seasonal cycles of CPT and LRT variables, with the lowest temperatures during boreal winter and highest temperature during boreal summer. The lowest temperatures (188 K) are located over the west Pacific and maritime continent during boreal winter, while during boreal summer, the minimum (192 K) shifts west to the area of the Asian summer monsoon (Seidel et al., 2001). The annual mean pattern of CPT matches the boreal winter pattern, indicating that CPT climatology is strongly affected by the boreal winter (Kim and Son, 2012).

The CPT and LRT height have a reversed seasonal cycle, with the highest altitudes during boreal winter and lowest during boreal summer. The highest LRT height (16.8 km) is located over the north equatorial Africa during most of the year, with one notable variation during November, when the minimum LRT height is located over tropical Pacific. The CPT height has the similar pattern as the LRT with some differences, the maximum height is located over central America from December to March and over Africa for other months (Seidel et al., 2001).

The lowest tropopause temperatures are usually found in regions of deep convection, which also change location during the year. The ITCZ exhibits a seasonal cycle due to solar radiation, moving north during boreal summer and south during boreal winter (Wallace and Hobbs, 2006). The convective distribution in a recent study by Kim, Randel, et al., 2018, is shown in Figure 2.3. The shading marks the frequency distribution of deep convection above 15 km, while the red crosses marks deep convection above 17 km. During DJF, the convection is centered at the west Pacific and maritime continent, with two secondary maxima over South America and central Africa. In JJA, the convection center moves northwest to the area of the Asian monsoon. The convection patterns in Figure 2.3 are shown in other studies and linked to cold CPT temperatures, coinciding over the areas of deep convection (Hosking

(16)

2.5. INTERANNUAL AND SEASONAL VARIABILITY

Figure 2.3: Tropical convection distribution using CloudSat. The shading shows the frequency distribution of convection above 15 km (number of 15 km convection / total number of observations in a grid box) as percentage, and the red crosses mark cloud top height above 17 km. AM = Asian monsoon. WP = west Pacific. EQ = Equator. From Kim, Randel, et al., 2018

et al., 2010, Gettelman, Salby, et al., 2002). Kim and Son, 2012 argued that the outgo- ing longwave radiation (OLR) values are located slightly off the coldest TTL temperatures, indicating that other processes must be present to explain the seasonality.

2.5 Interannual and seasonal variability

The TTL also exhibits interannual variability associated with large scale dynamics and was studied by Krüger et al., 2008a. The Quasi Biennial Oscillation (QBO), a quasi-periodic change between easterly and westerly zonal winds in the tropical stratosphere, was found to cool the upper TTL during the easterly phase, while the opposite was discovered during the westerly phase. Volcanic eruptions and changes in the solar cycle also had an impact on the upper TTL temperature. The El Niño Southern Oscillation (ENSO) signal showed cold and dry tropopause during La Niña, while El Niño years were warm and less dry. This thesis will focus on the El Niño Southern Oscillation (ENSO) impact on the TTL, as the time period covered (2007-2018) includes several ENSO years with varying intensity. Section

10

(17)

2.5.1 explains the ENSO and its impact on TTL, El Niño and La Niña, followed by how ENSO events are detected (Section 2.5.2)

As discussed in Section 2.4, the Asian summer monsoon has a significant impact on the seasonal cycle of the TTL, and will be explained in Section 2.6.

2.5.1 ENSO

The ENSO is described by Wallace and Hobbs, 2006.

ENSO is a coupled ocean-atmosphere dynamic, represented by the pattern of positive sea surface temperature anomalies in the equatorial Pacific (El Niño) and the response in the atmospheric circulation (Southern Oscillation). El Niño and La Niña events represents the extremes in the ENSO cycle and the phase between them is called neutral conditions. The average (no El Niño or La Niña) winds at equator are easterly, which produces a west- ward wind stress and transports warm water. The corresponding atmospheric circulation, the Walker circulation, is characterized by strong convection of moist air over the eastern Pacific, and subsiding dry air over the west pacific. The Walker circulation reinforces the easterlies.

Changes in SST in the east and central Pacific can give rise to El Niño or La Niña.

El Niño events: During El Niño, the easterlies at equator are weakened, the wind stress is reduced and SST anomalies are warmer normal in the east and central Pacific. This in turn changes the Walker circulation, with strong convection shifting east, over the equatorial dry zone (Wallace and Hobbs, 2006, pp. 431-437). El Niño events are often referred to as warm events, and their influence on the convective pattern depends their strengths, i.e, how the SST anomalies are changing over time.

La Niña events: La Niña events are characterized by colder SST anomalies than normal in the east and central Pacific, due to strengthened easterlies. The Walker circulation intensify, with increased convection over the west Pacific and enhanced subsidence over the east Pa- cific. Like El Niño, La Niña events varies in strength and time, and are named cold ENSO events.

Changes in mean surface temperature and winds, convection patterns and the vertical temperature structure due to warm and cold ENSO events have an imapct in TTL temperatures.

Gettelman, Salby, et al., 2002 investigated cloud fraction and linked the coldest tropopause temperatures to regions of deep convection during El Niño events. The shift of convection pattern is shown to have a significant impact on the water vapour in the upper troposphere.

(18)

2.5. INTERANNUAL AND SEASONAL VARIABILITY

Gettelman, Randel, et al., 2001 showed that the water vapor mixing ratio follows the pattern of the convective activity, with a maximum in the regions of strong convection and a minimum just above the convection. Thus, by investigating the temperature and height of the of the tropopause during ENSO provides important information about the representation in the reanalyses.

2.5.2 How ENSO is evaluated

There are numerous ways of describing the ENSO cycle, depending on time periods and areas in focus. This thesis will use National Oceanic and Atmospheric Administration (NOAA) Ocean Niño Index (ONI, https://origin.cpc.ncep.noaa.gov/products/analysis_

monitoring/ensostuff/ONI_v5.php), which describes the SST anomalies in the Niño 3.4 region (5^◦N- 5^◦S, 120^◦−170^◦W) with a 30-year base period. An ONI value≥0.5 is defined as an El Niño event, while ONI≤ −0.5 is defined as a La Niña event. ONI is calculated using 3-month running SST mean, and the base period is updated every 5th year to take the heating of the Central Pacific due to global warming into account (NOAA, 2020). In Table 2.1 are El Niño periods marked with red numbers and La Niña periods marked with blue

Figure 2.4: ONI 2006-2018. OND = October-November-December mean, NDJ = November-December-January mean, DJF = December-January-February mean. The blue and red background separates the ENSO intensity into weak (light colors), moderate (reg- ular colors) and strong (dark colors). Data fromhttps: // origin. cpc. ncep. noaa.

gov/ products/ analysis_ monitoring/ ensostuff/ ONI_ change. shtml

12

(19)

numbers, covering the period from 2006 to 2018. The inclusion of 2006 is to see if the start year of the analysis, 2007, might have been influenced by an ENSO event. The time series of different 3-month means ONI presented in Figure 2.4 illustrates the variation strength of El Niño and La Niña events. It also shows the small variation in ONI for a three month mean during boreal winter.

2.6 The Asian summer monsoon

The Asian summer monsoon (ASM) is a seasonal shift of wind pattern due to differences in heating of land and sea. In early boreal summer, the large continent heats fast, warm air rises and forms a low pressure at the surface. Over the ocean, the temperature is relatively cooler, and a high pressure exists here. The surface winds blow from high to low pressure, transporting moist air over warm land, which is forced upwards due to orography in south- western flank of the Himalayas and causes deep convective updrafts throughout the summer (Wallace and Hobbs, 2006). This is the lower tropospheric monsoon influence, but the upper troposphere also exhibits large dynamical changes. The Tibetan Plateau is a highly elevated area, which absorbs incoming solar radiation. This produces strong meridional temperature gradients over a sizeable part of the upper troposphere (Webster et al., 1998). A high pressure forms above, know as the Tibetan high or the monsoon anticyclone, which extends all the way through to the lower stratosphere (Krishnamurti and Bhalme, 1976). It is important to

Figure 2.5: Mean vertical velocities from ERA-Interim shown as filled contours. The black dashed line marks the poition of the anticyclone. Orange contours are OLR. Arrows are mean 850 hPa horizontal winds from ERA-Interim. From Nützel et al., 2016

(20)

2.6. THE ASIAN SUMMER MONSOON

point out that the Asian summer monsoon varies in strength from year to year and also during the season, with active and break periods (Krishnamurti and Bhalme, 1976).

During boreal summer months, the tropical tropopause is cold and high over the Asian summer monsoon area, and is associated with strong and deep convection (Kim and Son, 2012, Gettelman, Salby, et al., 2002). The area of the ASM is shown in Figure 2.5, where the dashed black line illustrates the monsoon anticyclone and the orange contours show mean OLR. The location of the deep convection is southeast of the monsoon anticyclone. Mean vertical velocity are indicated by filled contours, showing upward winds on the eastern side of the anticyclone and downward winds on the western side, over the Arabian Peninsula (Nützel et al., 2016). Changes in convection centers and upward motions have a significant impact on the dehydration of the stratosphere and transport through the TTL (Devasthale and Fueglistaler, 2010). By investigating the temperatures and height of the tropopause during boreal summer, it is of interest to see if there is a strong monsoon signal in the reanalyses and if the reanalyses captures the shift in location of the lowest temperatures in the GPS-RO data set.

14

(21)

Ocean Niño Index 2006-2018

DJF JFM FMA MAM AMJ MJJ JJA JAS ASO SON OND NDJ

2006 -0.79 -0.67 -0.47 -0.28 -0.05 0.04 0.12 0.27 0.48 0.71 0.90 0.95 2007 0.71 0.32 -0.03 -0.23 -0.29 -0.41 -0.54 -0.84 -1.13 -1.40 -1.54 -1.60 2008 -1.59 -1.42 -1.19 -0.92 -0.75 -0.54 -0.35 -0.26 -0.30 -0.41 -0.60 -0.73 2009 -0.80 -0.69 -0.52 -0.24 0.09 0.35 0.47 0.54 0.65 0.95 1.31 1.57 2010 1.55 1.31 0.94 0.44 -0.09 -0.59 -1.03 -1.38 -1.61 -1.70 -1.69 -1.59 2011 -1.37 -1.09 -0.83 -0.64 -0.47 -0.38 -0.46 -0.65 -0.89 -1.06 -1.14 -1.03 2012 -0.81 -0.62 -0.50 -0.38 -0.18 0.05 0.27 0.33 0.31 0.21 0.01 -0.21 2013 -0.38 -0.33 -0.24 -0.21 -0.27 -0.34 -0.38 -0.35 -0.31 -0.23 -0.22 -0.27 2014 -0.37 -0.36 -0.17 0.13 0.30 0.23 0.07 0.03 0.18 0.44 0.59 0.66 2015 0.60 0.56 0.62 0.79 1.02 1.25 1.54 1.83 2.11 2.37 2.53 2.64 2016 2.53 2.23 1.68 1.03 0.48 -0.00 -0.34 -0.57 -0.68 -0.74 -0.71 -0.56 2017 -0.29 -0.06 0.15 0.29 0.39 0.38 0.16 -0.14 -0.44 -0.70 -0.88 -0.97 2018 -0.87 -0.76 -0.60 -0.41 -0.13 0.06 0.11 0.20 0.43 0.70 0.85 0.81

Table 2.1: Ocean Niño Index, based on 3-month running average of SST anomalies. Red marks periods with ONI ≥ 0.5, defined as an El Niño event, and blue marks periods with ONI ≤ −0.5, which is defined as a La Niña event. Data from https: // origin. cpc. ncep. noaa. gov/ products/ analysis_ monitoring/

ensostuff/ ONI_ change. shtml

(22)

2.7. SPARC REANALYSIS INTERCOMPARISON PROJECT

2.7 SPARC Reanalysis Intercomparison Project

This thesis is an extension of Chapter 8, "The tropical tropopause layer" of the S-RIP report (under review) and of Tegtmeier et al., 2020. In Tegtmeier et al., 2020, eight reanalysis data sets are evaluated, including ERA-Interim and ERA5, for the period 1980-2010, but due to limtied available ERA5 data, it focuses on 2002-2010. Reanalyses are compared to GPS-RO (also called GNSS-RO).

The tropical mean (20^◦S - 20^◦N) CPT and LRT temperatures for 2002-2010 averaged show that ERA5 agrees best relative to GPS-RO, with a bias of+0.05K and -0.04 K, respectively.

ERA-Interim has the second best agreement,with a warm bias of 0.3K for CPT and 0.5 K for LRT. A time series of the tropical mean difference between reanalyses and GPS-RO is shown in Figure 2.6. ERA5 has a negative bias until late 2007, when COSMIC radio occultation data was assimilated, causing a jump in the temperature difference, also evident

Figure 2.6: Tropical mean (20^◦S−20^◦N, temperature differences between reanalysis and satellite data, 2002-2010. The cold point temperatue difference is marked with a think line, the lapse rate temperature difference is marked with a thin line. The dashed line shows when COSMIC data was assimilated (late 2007). From Tegtmeier et al., 2020

16

(23)

for ERA-Interim. The ERA-Interim difference shows more interannual variations, and is overall larger than for ERA5 (≈0.5 K). The two tropopause definitions differ in which have the largest bias.

In Figure 2.7, the latitude-longitude CPT temperature distribution of the GPS-RO data is illustrated, along with the difference between the reanalyses evaluated and GPS-RO data. It is evident that the lowest temperatures are found over the west Pacific and maritime continent, consistent with the background provided in this section. The differences show that ERA5 and ERA-Interim agrees well to GPS-RO, relative to the other reanalyses. ERA-Interim shows increased temperature bias over the central Africa and South America, but not over the western Pacific. Tegtmeier et al., 2020, linked this pattern to enhanced Kelvin wave activity, which is stronger over the areas of large temperature bias.

This thesis includes a comparison of the period from 2007 to 2018.

Figure 2.7: CPT temperature for GPS-RO data (a)), and the difference between reanalyses and GPS-RO (b-f)). Panel g) shows the difference between CFSR and GPS-RO for a period of high Kelwin wave activity. From Tegtmeier et al., 2020.

(24)

(25)

3 Data and methods

In this chapter, the reanalyses (Section 3.1) and satellite data (Section 3.2) are presented, followed by a description the assimilation of GPS-RO in the reanalyses (Section 3.3). The final section (Section 3.4) of this chapter presents how the tropopause definitions are calculated, and an explanation of the ENSO and the Asian summer monsoon analyses.

3.1 Reanalyses

Reanalyses provide information about the previous states of the atmosphere by using a fixed forecast model as a first guess of the atmospheric state. This is then corrected by assimilation of available observations within the assimilation window. The reanalyses used in the thesis are presented below.

3.1.1 ERA-Interim

ERA-Interim is a reanalysis produced by ECMWF and was released in 2008. It is an upgrade from the previous reanalysis, ERA-40, with improvements in the data assimilation system which includes representation of the hydrological cycle and the stratospheric circulation (Dee et al., 2011). ERA-Interim is based on the integrated forecast system by ECMWF, cycle 31r2.

Temporal and spatial resolution are listed in Table 3.1. It is the first ECMWF reanalysis using the four-dimensional variational analysis system (4D-Var), illustrated in the left panel of Figure 3.1, providing the best fit between assimilated observations and the forecast model in the assimilation window (12 hours) (Fujiwara et al., 2017). A variational bias correction

ERA-Interim ERA5

Model cycle 31r2 (2006) 41r2(2016)

Spatial resolution

79 km globally 60 levels to 0.1 hPa

31 km globally, 62 km for EDA 137 levels to 0.01 hPa

Analysis output frequency Every 6th hour Hourly

Table 3.1: ERA5 and ERA-Interim, adapted from Hersbach, Rosnay, et al., 2018

(26)

3.1. REANALYSES

scheme is implemented in the data assimilation, which corrects or minimizes biases. The update in ERA-Inteim includes an automated scheme for correcting biases in the satellite radiance observations. If the radiance data was assimilation without a bias control, it could be evident as significant jumps in the reanalysis product (Dee et al., 2011). The cloud scheme in ERA-Interim benefits from an updated forecast model compared to ERA-40, including a parametrization allowing supersaturation with respect to ice in the cloud-free part of a grid box at temperatures lower than 250 K (Berrisford et al., 2011). This has an impact on relative humidity close to the tropopause, providing more realistic values in the tropics.

Other changes affecting the TTL are the upward transport of air through the tropopause, resulting in a better representation of stratospheric humidity, and improvements in Madden- Julian Oscillation, which have an impact on the Asian monsoon and ENSO (Dee et al., 2011).

3.1.2 ERA5

ERA5 is also produced by ECMWF, and is the replacement of ERA-Interim, with improvement in vertical and horizontal resolution, updates in the IFS model cycle and data assimilation development (see Table 3.1). The IFS cycle 41r2 was used in operational medium-range forecasting in 2016, an update which includes over ten years of research and development.

The assimilation of observations is done by a hybrid incremental 4-dimensional data assimilation (4D-Var) system, with an ensemble component of 10 independent members, estimat- ing background error for the 4D-Var. This Ensemble of Data Assimilation (EDA) is new for ERA5 (Hersbach, Rosnay, et al., 2018). The assimilation is illustrated in Figure 3.1 (right),

Figure 3.1: Illustration of 4D-Var (left) and 4D-Var EDA (right). From https:

// confluence. ecmwf. int/ display/ FUG/ 2. 5+ Model+ Data+ Assimilation%

2C+ 4D-Var, author: Timothy Hewson

20

(27)

CHAPTER 3. DATA AND METHODS

where the blue line represents the initial forecast by the model, which is adjusted by observations (purple lines) in the assimilation window (black horizontal line). The EDA is illustrated in green lines as a number of analyses started at the same time, with small perturbations from an initial analysis.

The ERA5 data used is based on 6-hourly sigma-hybrid model level data, as model levels provides a clear improvement when compared to standard pressure levels, see Tegtmeier et al., 2020, Figure 1 and 2. Full model levels in the reanalyses are converted to pressure levels using a standard pressure of 1013.25 hPa and Equation 3.1 (from Berrisford et al., 2011)

p_k= 1

2(p_k−1/2+p_k+1/2) (3.1)

wherep_k−1/2is the half model levels,

p_k−1/2=A_k−1/2+B_k+1/2p_s (3.2)

andA_k−1/2andB_k+1/2are coefficients specified for the reanalyses, i.e ERA5 with 137 model levels have different pressure levels than ERA-Interim, which has 60 model levels.

3.1.3 Difference between ERA-Interim and ERA5

Several improvements have been made in ERA5 compared to ERA-Interim. The most important improvements are listed below, documented in Hersbach et al.,2018.

• An improvement in horizontal and vertical resolution, plus a higher model lid (see Table 3.1).

• Improved temporal resolution. ERA5 produces hourly analyses fields, ERA-Interim produced 6-hourly analyses fields.

• Overall, ERA5 assimilates more conventional and satellite data, with an increased number of radiance data, microwave imagers, scatterometer wind and GPS-RO.

• ERA5 differs from previous reanalyses from ECMWF with an uncertainty estimate based on a 10-member Ensemble of Data Assimilation (EDA), which estimates the background error for the 4D-Var and provides important information about the uncertainty in the analyses.

(28)

3.2. GPS-RO DATA

• Updated model cycle (from 31r2 to 41r2), which includes improvements in all components (land, ocean waves, atmosphere) relative to ERA-Interim and the assimilation methodology. Changes in model input, with forcing data suitable for climate reanalyses, which includes ozone, total solar irradience, greenhouse gases and stratospheric sulphate (Hersbach, Bell, et al., 2019).

• Cloud parametrization changes, which includes an improvement of the tropical cycle, and the diurnal cycle of convection.

• Improvements in the global balance of precipitation and evaporation. In the deep tropics, precipitation over land is better represented (Hennermann, 2020).

A study by Hoffmann et al., 2019 compared ERA5 and ERA-Interim for Lagrangian transport simulations for 2017. In the tropical tropopause region, ERA5 shows ERA-Interim reduced strong updrafts, and is linked to differences location of convection in the two reanalyses.

3.2 GPS-RO data

The use of Global Positioning System (GPS) radio occultation (RO) limb sounding technique started in the mid 1990ies with the GPS/Meteorology (GPS/MET) experiment (Anthes, 2011). The technique is illustrated in 3.2, showing the Low-Earh Orbiting (LEO) satellite receives signals from a GPS satellite, and as the LEO satellite is sets behind the Earth, the signal is bent, demonstrated by the tangent point.

The signal passes through the atmosphere, providing precise vertical profiles of the bending angle, which is used to compute the refractivityN. Equation 3.3 shows how the refractivity

Figure 3.2: Figure illustrating the GPS-RO technique. From Anthes, 2011.

22

(29)

is a function of temperature (T), pressure (p), water vapor pressure (e), electron density (n_e) and the frequency (f) of the GPS signal (Anthes, 2011).

N=77.6p

T +3.73×10⁵ e

T²−4.03×10⁷n_e

f² (3.3)

Combing refractivity profiles into a set of linear equations, the temperature in the lower stratosphere and upper troposphere can be derived from Equation 3.3. The temperature derived from GPS-RO provides data with random and systematic errors of 1 K and 0.2 K, respectively (Fujiwara et al., 2017).

Results from GPS/MET revealed how RO produced accurate high-vertical resolution soundings of the atmosphere (Anthes, 2011). Since then, several new missions have been released, with an increased number of soundings each day compared to GPS/MET, for example the COSMIC mission has about 2000 global sounding per day (Anthes, 2011).

3.2.1 GPS-RO data set

The GPS-RO data set used in this analysis is constructed from multiple satellite missions, listed in Table 3.2, and was provided by Robin Pilch Kedzierski¹, who downloaded it from the COSMIC Data Analysis and Archive Center (CDAAC,https://cdaac-www.cosmic.

ucar.edu/cdaac/products.html). All temperature profiles are atmospheric profiles with moisture information (wetPrf), interpolated to 100 meter vertical resolution. Re-processed data was used if available, otherwise post-processed data was included. The GPS-RO data is interpolated to reanalyses model levels using the barometric formula (Tegtmeier et al., 2020), and the cold point and lapse rate are identified for each profile, using the definitions from Section 2.3.2 and 2.3.3. The available data set for this thesis consist of daily data on a 5^◦×5^◦lat-lon grid between 30^◦S and 30^◦N from January 2007 to December 2018. Monthly means of daily means are calculated using the xarray package in Python, taking leap years into account, done by the author.

3.3 Assimilation of GPS-RO in reanalysis

The observations assimilated in reanalysis are usually grouped in two categories- conven- tional and satellite data. While conventional data, like radiosondes, aircraft measurements, are limited in space and time, satellite data are more continuous in space but still in-homogeneous

1GEOMAR Helmholtz Centre for Climate Modelling and Analyses

(30)

3.3. ASSIMILATION OF GPS-RO IN REANALYSIS

Mission Launched Ended

CHAMP July 15, 2000 September 19, 2010 GRACE March 17, 2002 October 17, 2017 COSMIC April 15, 2006 May 1, 2020 Metop-A October 19, 2006 ongoing Metop-B September 17, 2012 ongoing

SAC-C November 21, 2002 August 15, 2013 Terra-SAR-X June 15, 2007 ongoing

Table 3.2: Satellites missions used in GPS-RO data set

in time. Updates and availability of observations have strong impacts on the quality of the reanalysis that assimilate them (Fujiwara et al., 2017).

ERA5 and ERA-Interim assimilates both conventional and satellite data. As this thesis will compare these two reanalysis with GPS-RO observations, a description of how this product is assimilated is given below.

3.3.1 Assimilation of GPS-RO in ERA-Interim

A study done by Poli et al., 2010, investigates the assimilation of GPS-RO data from three missions (CHAMP, COSMIC and MetOp-A GRAS) in ERA-Interim (Figure 3.3). The implementation of COSMIC data end of 2006 increased the number of GPS-RO data assimilated per day by a factor of 10, having a warming effect of the tropopause of about 0.1 to 0.2 K. (Poli et al., 2010).

3.3.2 Assimilation of GPS-RO in ERA5

As noted before, ERA5 assimilates more observations than ERA-Interim, also GPS-RO data.

Updates for the GPS-RO data have also been implemented, such as a correction of observation error for GPS-RO dataset, and a reprocessed COSMIC GPS-RO dataset, which produces results in much closer agreement with those of MetOp-A GRAS (Hersbach, Rosnay, et al., 2018). Figure 3.3 illustrates the time evolution of GPS-RO. Both reanalyses assimilate

24

(31)

Figure 3.3: Time series of GPS-RO data assimilated in ERA5 (above dashed line) and ERA-Interim (below dashed line). Satellite names are presented on the left side of the figure. Information provided by Tegtmeier et al., 2020, ECMWF ERA5 documentation (https: // confluence. ecmwf. int/ display/ CKB/ ERA5% 3A+ data+

documentation, ERA-Interim documentation (https: // confluence. ecmwf. int/

display/ CKB/ ERA-Interim% 3A+ documentation

CHAMP, METOP-A and COSMIC for the same time period. ERA5 assimilates all seven satellite missions listed in Table 3.2.

3.4 Calculating the CPT and LRT in the reanalyses

Both ERA5 and ERA-Interim are evaluated at model levels instead of pressure levels, as this provides finer vertical resolution. The cold point and lapse rate tropopause are derived for model-level data between 500 and 10 hPa at each grid point at 6-hourly temporal resolution (Tegtmeier et al., 2020), where the cold point is the coldest model level in the given span, and the lapse rate is evaluated from the WMO definition (WMO, 1957). The data sets are monthly means from January 2007 to December 2018, on 1^◦×1^◦and 0.7^◦×0.7^◦ grid resolution for ERA5 and ERA-Interim, respectively. I computed the difference between the reanalyses and GPS-RO data always in the last step, e.g. zonal means were computed for all data sets, and then the difference was calculated. The aim of this thesis is to extend and deepen

(32)

3.4. CALCULATING THE CPT AND LRT IN THE REANALYSES

the analyses of Chapter 8, the tropical tropopause layer, of the S-RIP report (see Tegtmeier et al., 2020 and Fueglistaler et al., 2009). The intention behind using data provided by Tegtmeier was to be able to compare results to Tegtmeier et al., 2020. A majority of the analysis in this thesis consist of latitude-longitude variations and I interpolated the reanalyses by bilinear interpolation to the same resolution as the GPS-RO data (5^◦×5^◦) to calculate the difference. The regridded files are used for all differences presented in the thesis, in contrast to Tegtmeier et al., 2020, who used the highest available resolution of the individual reanalyses. This is to produce consistent results through the thesis. Figure 3.4 illustrates the difference between using the regridded files and the zonal mean files with original resolution.

The zonal mean files from the reanalyses are averaged over 20^◦S to 20 ^◦N, and then the difference is computed. It is evident from Figure 3.4 that these two different methods yield different results. For instance, the ERA5 difference is mostly negative for the zonal mean method and mostly positive for the regridded method, and the mean difference between them is 0.3 K. The ERA-Interim difference is larger for the regridded method (0.3 K) but shows

Figure 3.4: Time series of ERA5 (upper panel) and ERA-Interim (lower panel) differences when compared to GPS-RO. The regridded files have a5^◦×5^◦resolution, the zonal mean files have a1^◦×1^◦resolution for ERA5,0.7^◦×0.7^◦, for ERA-Interim. Lighter color shows the rolling mean

26

(33)

the overall same pattern as the zonal mean. Both methods provide important information depending on the research questions in question, results in Chapter 4 are produced using 5^◦× 5^◦ regridded files, except for the plots illustrating the impact of ENSO and Asian summer monsoon in ERA5, where the original files and resolutions are used. It is also important to point out that the time series of the regridded files in Figure 3.4 show similar temperature differences as Figure 2.6 from Tegtmeier et al., 2020. The mean CPT difference in the time series in Tegtmeier et al., 2020 between 2007 an 2010, is ≈0.2K for ERA5, and ≈0.5K for ERA-Interim. This is due to a different GPS-RO data with finer resolution used by Tegtmeier.

The GPS-RO data was compiled by Robin Pilch Kedzierski², the ERA-Interim data by Sean Davis³and the ERA5 data by Bernard Legras⁴. Data was provided by Susann Tegtmeier in December 2019.

3.4.1 ENSO analysis

To investigate the impact of ENSO on the general difference between reanalyses and GPS- RO data, El Niño and La Niña events are presented as a composite of December, January, February. The ONI (see Section 2.5.1) is used to detect warm/cold events. The El Niño and La Niña composite are described in Table 3.3. It is important to note that the ENSO analysis is not deseasonalised and includes seasonal and interannual variation, such as QBO signals.

Several time periods were evaluated to see the effect of ENSO, for instance the mean of all months having±0.5 ONI values, and only the months with the maximum ONI. Overall, the ENSO pattern was similar for periods with different ONI and thus the key ENSO season DJF was selected. This also makes it easier to compare it to previous studies, as this is a common way of detecting the ENSO signal in the TTL. As described in Table 3.3, the intensity of the El Niño and La Niña periods varies which could have an impact on the results. In addition, the period investigated includes fewer El Niño years (three) than La Niña years (five).

3.4.2 Asian Summer Monsoon analysis

As described in Section 2.6, the Asian monsoon (ASM) has a substantial influence on tropopause temperature and height and is most prominent during boreal summer. Seasonal means of monthly mean of June, July, August (JJA) from 2007-2018 were computed for all

2GEOMAR Helmholtz Centre for Climate Modelling and Analyses

3Earth System Research Laboratory, National Oceanic and Atmospheric Administration

4Laboratoire de Météorologie Dynamique, CNRS/(PSL-ENS, Sorbonne Université, Ecole Polytechnique)

(34)

3.4. CALCULATING THE CPT AND LRT IN THE REANALYSES

data sets, and the difference were computed in the last step. The strength of the monsoon has not been taking into account. The observational data is limited from 30^◦S to 30^◦N, excluding the full area of interest and limits a complete understanding of the ASM on the TTL.

Time period ONI Index

La Niña composite

2007-2008 2008-2009 2010-2011 2011-2012 2017-2018

-1.59 -0.80 -1.37 -0.81 -0.87 El Niño composite

2009-2010 2014-2015 2015-2016

1.55 0.60 2.53 Table 3.3: Description of ENSO composite used

28

(35)

4 Results and Discussion

The results are structured into four sections. The first section describes the climatological tropical mean, and the difference between GPS-RO and the reanalyses. Further investigation of the seasonality is presented in Section 4.2. The impact of ENSO in the TTL is shown in Section 4.3, followed by a study of the Asian summer monsoon, Section 4.4.

4.1 Tropopause temperature and height

The climatological annual mean CPT and LRT temperature for all three data sets are presented in Figure 4.1. The CPT temperatures are lower than the LRT temperatures, and the GPS-RO data shows the lowest values for both tropopause levels, with values of 192.2 K and 192.9 K for CPT and LRT, respectively. The temperature differences are displayed in the right panel of Figure 4.1. Both reanalysis have positive temperature biases when compared to GPS-RO. The smallest difference is the LRT temperature for ERA5 and the CPT temperature for ERA-Interim.

In Table 4.1 the climatological annual mean temperature differences are listed in the two upper rows with standard deviation. The standard deviation is larger for the LRT temperature than for CPT, indicating that the LRT exhibits larger variations.

Figure 4.2 shows the climatological annual mean height, which is higher for CPT. It is clear that the GPS-RO data has the highest tropopause heights, with values of 17.2 km and 16.6

Figure 4.1: Left: climatological annual (20^◦S to20^◦ N) mean temperatures of GPS-RO, ERA5 and ERA-Interim. Right: Difference ERA5 and GPS-RO, and ERA-Interim. Ex- tended from Tegtmeier et al., 2020.

(36)

4.1. TROPOPAUSE TEMPERATURE AND HEIGHT

ERA5 ERA-Interim

CPT-T 0.2±0.3 K 0.4±0.4 K LRT-T 0.1±0.4 K 0.7±0.5 K CPT-z -0.16±0.09 km -0.20±0.11 km LRT-z -0.12±0.12 km - 0.16±0.13 km

Table 4.1: Climatological annual mean differences for ERA5 and GPS-RO and ERA- Interim and GPS-RO. The mean±standard deviation is presented (1 sigma). Temperatures are listed first and heights are listed last

Figure 4.2: Left: Climatological annual mean heights of GPS-RO, ERA5 and ERA-Interim.

Right: Difference ERA5 and GPS-RO, and ERA-Interim. After Tegtmeier et al., 2020.

km for CPT and LRT, respectively. As in Figure 4.1, the difference is presented on the right side. The best agreement is found for the LRT height for both reanalyses, -0.12 km for ERA5 and -0.16 km for ERA-Interim.

The reanalysis with best agreement compared to GPS-RO data is ERA5 and holds for both tropopause definitions. The CPT and LRT climatological annual mean temperatures are 0.15 K and 0.2 K higher in ERA5 and 0.1 K and 0.2 K higher in ERA-Interim compared to the results of Tegtmeier et al., 2020. As discussed in Section 3.4, Tegtmeier et al., 2020 use the original resolution of the reanalyses data, which can explain the difference. The reanalyses differ in which of the tropopause definitions have the smallest temperature difference, which is LRT for ERA5 and CPT for ERA-Interim. This was also shown in Tegtmeier et al., 2020.

30

(37)

CHAPTER 4. RESULTS AND DISCUSSION

4.1.1 Latitude - longitude variations

The CPT and LRT properties are investigated further by analyzing latitude-longitude variations in Figure 4.3. The GPS-RO CPT variables in the upper panel reveal that the lowest temperatures (190 K) are located over the west Pacific and maritime continent. Other loca- tions of low temperatures include the equatorial Africa and South America, coinciding with deep convection centers. The height contours have small variations in the tropics, and seem to be highest north of 20^◦N. At latitudes above 20^◦S and 20^◦N, the temperature sharply increases, marking the beginning of the extratropics, where the tropopause is at lower levels and the CPT is not considered a tropopause definition anymore.

The ERA5 and GPS-RO CPT difference is displayed in the middle panel. The temperature difference has a zonal symmetry in the inner tropics with the largest values (+0.4 K) located over South America and central Africa, where the height differences are the largest (+0.2 km). Overall, the temperature and height differences are in the range of 0.1 to 0.4 K, and -0.1 to -0.2 km, which is within the bounds of the standard deviation presented in Table 4.1.

The lower panel illustrates the ERA-Interim and GPS-RO difference. The maximum temperature biases (+0.7 K) are over central Africa and the east Pacific, while the west Pacific and maritime continent show smaller differences. The largest height difference (+0.2 km) coincide with the temperature bias maximum. The temperature difference in ERA-Interim is larger than for ERA5 and shows pronounced maximum, while the ERA5 difference shows small zonal variations. An interesting point is the small differences over west Pacific, where the CPT temperature is lowest.

Figure 4.4 displays the LRT on latitude - longitude format, with the same structure as in Figure 4.3. The GPS-RO mean temperature shows a similar pattern as the CPT, with the lowest LRT temperatures over the maritime continent and the west Pacific. There is a sharp increase in temperature and decrease in height outside 20^◦S - 20^◦N, marking the abrupt change in static stability. The LRT height is more zonally structured than the CPT, with the highest altitudes in the inner tropics.

The ERA5 difference in the middle panel generates a similar structure as the GPS-RO CPT, but with a small decrease in height difference, 0.1 km. This is also the case for the ERA- Interim difference, but with an increase in temperature difference (+0.4) and an additional maximum over the maritime continent.

(38)

Figure 4.3: Latitude - longitude distribution of CPT. Temperature is shaded, height is marked with green contour lines. Dashed lines are negative values. Upper panel: Tropica mean of GPS-RO. Middle panel: Difference ERA5 - GPS-RO. Lower panel: Difference ERA-Interim - GPS-RO.

As in previous tropopause studies (Gettelman and Forster, 2002, Fueglistaler et al., 2009, Kim and Son, 2012), the CPT and LRT temperature minimums are located over the maritime continent and west Pacific. Other low temperature regions include equatorial Africa and South America. These are regions with localized deep convection. Investigating the CPT temperature difference between reanalyses and satellite observation, the largest differences are found in the inner tropics over Africa, east Pacific and Central America, not over the west Pacific. Tegtmeier et al., 2020, linked this bias distribution in the reanalyses to Kelvin wave activity, shown in Section 2.7, Figure 2.7. Kelvin waves can propagate through the TTL and have a significant impact on short-term variability, such as the temperature, height, mixing of air and transport to the stratosphere (Kim, Kiladis, et al., 2019). Previous studies have shown less wave activity over the maritime continent and more over 50^◦E (Kim and Son, 2012). During periods of high Kelvin wave activity, the tropopause temperature is lower, and was shown in Tegtmeier et al., 2020, to cause an increase in the reanalysis - GPS-RO bias, meaning that the reanalyses does not resolve low temperatures due to Kelvin waves. Thus, the large temperature bias over equatorial Africa could be linked to enhanced wave activity,

32

(39)

Figure 4.4: As for Figure 4.3 but for LRT. Note that the color bars are shifted

reshaping the vertical structure on small scales. As the ERA5 has a considerably better vertical resolution than ERA-Interim, it represents small disturbances of the temperature profile better. The LRT differences are similar to the CPT in ERA5, while ERA-Interim shows an increased temperature difference over the maritime continent.

4.1.2 Time evolution

A time series of the temperature and height difference between reanalyses and satellite data displays the annual and interannual changes, and is plotted in Figure 4.5. The GPS-RO tropical mean temperatures in the upper panel show a seasonal variability, with low temperatures during boreal winter and high temperatures during boreal summer. The LRT temperature is always warmer than the CPT temperature, and the difference between them is largest in mid- summer and mid-winter. The red and blue vertical lines mark El Niño and La Niña years, respectively. There is little evidence of a clear ENSO signal when averaging over the tropics.

For instance, the CPT temperature is close to 190K in 2008, 2009 and 2012, which is the average temperature in the tropics during boreal winter, while the temperature is almost 2 K warmer in 2011, even though they are all La Niña years. The lowest temperature for the period evaluated is January 2013 (189 K) and the highest is August 2008 (196 K).

The ERA5 - GPS-RO difference is presented in the middle panel. The CPT difference (blue

(40)

line) is larger than the LRT difference (green line) for the whole period, except in the beginning of 2013, occurring at the same time as the lowest temperature in GPS-RO. Both temperatures show similar patterns but is more distinct in the LRT mean. In the beginning of most years, a sharp decrease in the temperature difference is present, followed by a pronounced increase shortly after. Thus, ERA5 captures some of the low temperatures in GPS-RO. The ERA-Interim difference in the lower panel is different. The LRT temperature difference is always greater than the CPT, and they are similar in behaviour. In contrast to ERA5, most years start with a large temperature bias which decreases shortly after, and indicates that ERA-Interim does not capture the annual cycle of low temperatures in GPS-RO.

The ENSO signal is mixed, with no clear pattern. The smallest difference in ERA5 is found in late 2010 and the beginning of 2011 during a strong La Niña, when GPS-RO temperatures are higher than normal, which could be a bias in the GPS-RO. The temperature bias can also be sensitive to assimilation of new satellite data and the termination of satellite missions.

For instance, Metop-A was assimilated by both reanalyses in the mid of 2008 (see Figure 3.3), and there is an increase in temperature difference during this period. As discussed in Chapter 3, there was an increase in ERA5 and ERA-Interim temperatures after the inclusion of COSMIC data in 2006, as a consequence of a reduced temperature bias at model levels (Tegtmeier et al., 2020).

Other changes in the temperature difference may be explained by interannual variability, as the QBO. The ENSO signal in the reanalyses is studied in Section 4.3.

34

(41)

Figure 4.5: Time series of tropical mean (20^◦S -20^◦N) monthly mean temperature. Blue vertical lines are El Niño years, red lines are La Niña years. Upper panel: time series of GPS-RO. Middle panel: ERA5 - GPS-RO. Lower panel: ERA-Interim - GPS-RO.

(42)

4.2. SEASONALITY OF TROPOPAUSE TEMPERATURE AND HEIGHT

4.2 Seasonality of tropopause temperature and height

As the largest differences in tropopause temperatures are due to seasonal variation, illustrated in Figure 4.5, studying the monthly climatology is of interest. Figure 4.6 presents the climatology of monthly means for the CPT temperature, and the differences between the data sets in the lower panel. There is a clear seasonality for all data sets, with the lowest temperature during boreal winter and highest temperature during boreal summer, a difference of about 4.6 K. The temperature is higher in both reanalyses throughout the year when compared to GPS-RO, evident in the lower panel of Figure 4.6. ERA5 has the largest difference in March and April (+0.3 K), which decreases in May and is stable throughout the year. ERA-Interim shows larger temperature differences compared to ERA5, with maximum during boreal winter (+0.5 K). The monthly climatology of the LRT temperature (Figure A.1) follows a similar pattern as CPT, but is larger for all months of the year. ERA5 and ERA-Interim have a similar increased temperature difference during boreal spring and winter, respectively.

The CPT height is displayed in Figure 4.7, showing the reversed seasonality of the temperature, with the highest tropopause during boreal winter and the lowest during boreal summer.

The GPS-RO CPT height is always above the two reanalyses. Studying the height differences in the lower panel, ERA5 agrees well with GPS-RO data, and the bias does not change

Figure 4.6: Climatology of CPT temperature, 2007-2018, 2007-2018,20^◦S -20^◦N. Upper panel: Climatology for GPS-RO, ERA5, ERA-Interim. Lower panel: Temperature difference ERA5 - GPS-RO and ERA-Interim - GPS-RO.

36

(43)

Figure 4.7: Climatology of CPT height, 2007-2018,20^◦S -20^◦N . Upper panel: CPT height climatology for GPS-RO, ERA5, ERA-Interim. Lower panel: Height difference ERA5 - GPS-RO and ERA-Interim - GPS-RO.

much during the year. CPT height in ERA-Interim exhibits large seasonal variation relative to GPS-RO. When comparing Figure 4.6 and Figure 4.7, the largest height differences do not coincide with the largest temperature differences for each reanalyses. Thus, the explanation for the temperature could be different from the reason behind the height difference. The LRT height climatology (Figure A.2) shows a similar seasonal cycle and is always below the CPT height. The height differences are smaller to Figure 4.7.

Investigating the seasonal differences further, monthly zonal mean climatology of CPT temperature is displayed in Figure 4.8. The upper panel provides information about the meridional climatology structure of GPS-RO CPT temperature. There is a clear latitudinal symmetry, with the lowest temperature at the equator during boreal winter. The ERA5 - GPS-RO difference is displayed in the middle panel, which reveals that the greatest differences are found between 10^◦S - 10^◦N, with a maximum between equator and 10^◦N during March and April and is consistent with Figure 4.6. The latitude - longitude distribution for the maximum period is shown in Figure 4.9, together with ERA5. Central Africa, the maritime continent and the east Pacific are areas with pronounced maximal differences, and could correspond to the ITCZ pattern.

The lower panel of Figure 4.8 shows the ERA-Interim - GPS-RO difference, and also here

(44)

Figure 4.8: Zonal mean climatology for CPT temperature, 2007-2018. Upper panel: GPS- RO zonal mean. Middle panel: ERA5 - GPS-RO zonal mean. Lower panel: ERA-Interim - GPS-RO zonal mean.

Figure 4.9: Upper panel: GPS-RO CPT temperature for March and April. Lower panel:

ERA5 and GPS-RO difference for CPT temperature for March and April

are the largest differences found in the inner tropics. The maximum difference in January and February is investigated further in Figure 4.10, revealing a doubling of the bias (+0.8 K) compared to the tropical annual mean over central Africa and the east Pacific.

The LRT zonal mean climatology (Figure A.3) shows very symmetric GPS-RO temperatures, and a larger temperature bias for ERA-Interim.

The GPS-RO zonal mean height (Figure 4.11) shows that the CPT is higher during boreal 38

(45)

Figure 4.10: Upper panel: GPS-RO CPT temperature for January and February. Lower panel: ERA5 and GPS-RO difference for CPT temperature for January and February

winter, at 10^◦- 20^◦N. It does not show the same meridional gradient as the CPT temperature in Figure 4.8. The differences in the middle and lower panel also have a different structure than that of the temperature. For ERA5, the maximum difference is located 10^◦ S from February to April and 10^◦ N in October and November. There is a second maximum during boreal spring, also at 10^◦ N, which resembles the temperature differences found in Figure 4.8. The smaller differences in boreal summer are consistent with Figure 4.7. For ERA- Interim, the largest differences are in the 10^◦ S - 10^◦ N , with a pronounced maximum in April. The LRT height is shown in Figure A.4 and is more zonally symmetric than the CPT

Figure 4.11: As Figure 4.8, but for CPT height

(46)

height, as are the differences.

It is evident that all three data sets show a clear seasonal cycle in temperature and height for both CPT and LRT, as mentioned in other studies (Kim and Son, 2012, Seidel et al., 2001).

The seasonal cycle of tropopause properties are largely controlled by large-scale processes, as the BDC, and less by small scale processes, as eddies. ERA5 resembles the GPS-RO seasonal cycle more closely than ERA-Interim, showing small differences in temperature and height climatology when compared to GPS-RO. The temperature bias is small (+ 0.3 K) in the inner tropics for the months with the lowest temperatures. The largest differences are found in March and April located north of equator. There is no clear evident reason for this maximum. The ERA-Interim CPT temperature bias shows a zonal symmetry with the largest values at the equator during the months with the lowest temperature. This could have a significant impact on the representation of water vapor entrainment in the lower stratosphere when using ERA-Interim CPT temperatures. As this feature only occurs in ERA-Interim, it could be due to a coarser vertical resolution, not representing the vertical temperature profiles accurately. The varying height difference during the year indicates that some of the annual mean difference between GPS-RO and the reanalysis might be due to the representation of the seasonality in ERA-Interim. The similarity between CPT and LRT temperature differences indicates that the seasonal differences are not due to CPT or LRT characteristics.

40

Comparison of the tropical tropopause in ECMWF reanalyses and GPS-RO satellite observation