Trends in tropical tropospheric column ozone from satellite data and MOZART model



[1] Trend analysis of tropical (30°S–30°N) tropospheric columnar ozone (TCO) has been done using Nimbus7 and Earth Probe satellite -Total Ozone Mapping Spectrometer (TOMS) data for the period of 1979–2005 using multifunctional regression model. Results indicate highest increasing trend (7–9% per decade) over some parts of south Asian continental region which is attributed to increasingly high emissions of ozone precursors over the region. Among different marine regions, Ozone trend is found to be highest (4–6% per decade) over the Bay of Bengal which is largely be explained by the large scale wind systems causing massive transport of continental pollutants over this region. The study also includes the comparison of the satellite results with those obtained from a 3-D chemistry transport model. In some cases, the magnitude of observed trends is consistent with the model trends. However, in some cases like East Asia and area around Philippines differences are inevitable.

1. Introduction

[2] Elevated levels of tropospheric ozone have implications to the human health, vegetation growth and most importantly for the climate. Analysis of the MOZAIC (measurement of ozone and water vapor by airbus in-service aircraft) data has revealed an increase of ∼20% in the tropical upper tropospheric ozone from 1994 to 2003 [Bortz et al., 2006]. This rate of increase is twice as large as reported by Lelieveld et al. [2004] for the surface ozone over the tropical Atlantic Ocean. In contrast to this, surface and ozonosonde data up to the mid-troposphere from Southern Hemispheric tropical station, Samoa (14°S and 170°W) show a small decrease from the late 1980s to the most recent decade in austral winter, and no change in rest of the year [Oltmans et al., 2006]. The MOZAIC data are from South America, the Atlantic, and Africa, while the ozonosonde data are from the Pacific region. Ziemke et al. [2005] report no significant trend in the tropical tropospheric column ozone over the Pacific region derived from TOMS data for the period of 1979–2003. Saraf and Beig [2004] reported on the basis of ozonosonde observations at a few Indian locations that tropospheric ozone concentrations have increased significantly over the last 30 years. Earlier, Chandra et al. [1999] used TOMS data to derive trends in TCO for a shorter period (1979–92) and could not find any significant trend. However, their analyses did not include the aerosol/sea glint corrections and made assumption of zonal inhomogeneities which might have influenced their results. Global chemical transport models (CTM) [e.g., Prather et al., 2003; Brasseur et al., 2006] suggest that surface ozone concentration could increase by as much as 25–30 ppbv in tropical India between years 2000 and 2100, assuming an economic growth described by the IPCC/SRES A2 scenario. Recently, Beig and Brasseur [2006] used CTM to simulate a decadal trend in ozone concentration of the order of around 5–10 ppbv near the surface which reaches 5–7% in the lower part of the free troposphere and 3–5% in the upper troposphere over the geographical region of tropical India in response to enhanced human activities.

[3] The differences between the various data records reflect the pronounced spatial variability of tropical tropospheric ozone trends. A recent analysis of long-term changes in tropospheric ozone emphasizes that trends vary regionally and that within a given region trends have changed over the past 25–35 years [Oltmans et al., 2006]. These problems are more serious in the tropical region where not many long-term tropospheric ozone records are available and changes are expected to be rapid and significant in magnitude. Hence, the aim of the present paper is to report, an up to date trend analysis of tropospheric column ozone for the entire tropical belt (30°S–30°N) which provides important peace of information to the tropical tropospheric ozone trend discussion. Results are derived from TOMS which span for more than two solar cycles (1979–2005) and compared with trends derived from the MOZART (Model for Ozone and Related chemical Tracers) simulations which accounts for variations in chemical emissions.

2. Data and Regression Model

[4] Most of the satellite measurements that commenced in 1979 promised to provide data with systematic geographical coverage and regarded to be more reliable, however, focus of these observations was to measure either total or stratospheric ozone. After the invention of some algorithms especially the convective cloud differential (CCD) technique, which has been tested and established to be reliable [Ziemke et al., 2005, 2001], it became possible to derive the TCO from the TOMS data of total integrated ozone. A unique property of the CCD method for deriving TCO is that it is not affected by inter-instrumental calibration error and at the same time the data is also corrected for several other factors and hence considered to be most reliable [Ziemke et al., 2001]. In this work, the TCO time series derived from Nimbus7 and Earth Probe –TOMS footprint measurements for the period of 1979–2005 has been used to derive the linear trend. The CCD data used in this work are from TOMS version 8 level 2. The CCD method assumes that one can make an accurate estimate of stratospheric column ozone (SCO) using high reflecting convective clouds that reach at or near the tropopause level in the tropics. The CCD method also assumes that SCO is zonally invariant within the tropical latitudes. With these assumptions, TCO in tropical latitudes can be calculated by differencing low-reflectivity (<0.2) gridded total column ozone and high-reflectivity (>0.9) SCO. The TCO time series covers the latitude band of 30°S to 30°N and longitudinally full coverage from East to West with 5° resolution. To analyze the TCO time series obtained from TOMS, a multi-functional regression model has been used in this work, details of which are given elsewhere [Saraf and Beig, 2004; Ziemke et al., 1997] in detail and hence described here briefly. It consists of seasonal cycle, linear trend, solar cycle, Quasi-Biennial Oscillation (QBO) and El Niño Southern Oscillations (ENSO). The general expression for the regression model equation used for the analysis can be written as follows:

equation image

[5] Where t is the month index (1–324 for 1979–2005), O3(t) is TCO time series ozone, α(t) is the seasonal cycle coefficient, A(t) is the seasonal trend coefficient, B(t) is the seasonal ENSO coefficient, C(t) is the seasonal solar cycle coefficient, D(t) is the seasonal QBO coefficient, and Res(t) is the residual error time series for the regression model which includes all other parameters not considered including noise and also the sporadic events like Kalimantan big fire of 1997. Trend models often use a particular harmonic expansion to represent the seasonality of the interaction between O3(t) and a particular surrogate (proxy). In the present statistical model, we have included the ENSO proxy time series based on the actual measured values of Southern Oscillation Index ( In this model, time series of each proxy are used from literature [Ziemke et al., 1997; Saraf and Beig, 2004]. Proxy time series have been de-trended and de-seasonalized to avoid any tampering of the derived seasonal fits and to remove any fictional trends caused by sudden changes in the proxies on the long term variations of ozone.

3. MOZART Model Integrations

[6] The MOZART-2 chemical-transport model [Horowitz et al., 2003], which is used to study the distribution of ozone, provides global distribution of 63 chemical compounds at a horizontal resolution of ∼1.8 degrees in longitude and latitude, and on 31 vertical levels extending from the surface to approximately 10 hPa pressure level. It takes into account the surface emissions of several chemical compounds. The emissions due to fossil fuel combustion, agricultural burning, bio-fuel, etc are adopted from the recent estimates by the project named ‘Precursors of Ozone and their Effect on the Troposphere’ (POET) [Olivier et al., 2003] (except for India, see below). Over the Indian subcontinent, recent high-resolution emission estimates from the Indian national inventory [Dalvi et al., 2006] account for the rapid temporal variability and small-scale geographical variations (e.g. hotspots). In the adopted Indian emission inventory, the CO emission for 2001 from bio-fuel sources (wood and cow dung burning) (34,282 Gg/yr) represents almost 50% of total CO emissions in India. The total CO emissions from all sources over India for 2001 represent 69,400 Gg CO/yr [Dalvi et al., 2006]. The total surface NOx emissions in the same region for 2001 are estimated to be 4,260 Gg NO2/yr. The maximum contribution is provided by coal burning followed by fossil fuel combustion. Model is driven by assimilated meteorological data from the European Center for Medium Range Weather Forecasts. For further details, the reader is referred to Beig and Brasseur [2006]. The model integrations are performed for the period 1991 and 2001 and the decadal trend in TCO is derived. We used model simulations for 1991 to 2001 because we trust that this is the period in which emission inventory of ozone precursors are reliably and accurately available [Dalvi et al., 2006; Beig and Brasseur, 2006] for tropics and we have low confidence in the available emission inventory of pollutants in 80s, specially for the tropics which may likely to introduce more error than signal. For further details about model integration, the reader is referred to Beig and Brasseur [2006]. We have deduced the tropospheric column ozone by integrating the data below 100 hPa pressure level. In the context to the present paper, model results should be viewed with two constraints. Firstly, the model simulated results reflect the trend in TCO caused due to variation in chemical emissions only and secondly, the model provide the trend for 1990s whereas satellite data covers the period from 1980s to preset time. We could not simulate the trend for the same period as observational data because of lack of reliable emission inventory during 1980s for the tropical region. This is a severe limitation.

4. Results and Discussion

[7] Figures 1 and 2present annually averaged linear trend coefficient (% per decade) in TCO over the tropical region (30°S–30°N) of the globe as calculated from TOMS data (for the period 1979–2005) and MOZART model (for the period 1991–2001), respectively. The spatial correlation coefficient of the trend between the model simulated data and that of TOMS data is computed to be 0.51 which is reasonable within the constraints as mentioned above. To calculate the spatial correlation of the different sampled data, TOMS trend is interpolated from the centre of the grid (5 × 5) to the MOZART grid (1.875 × 1.875) with the square of the inverse distance weights and within the area of 12 points. It should be noted that the trend is shown in Figure 1 (top) and the quantity, following ± sign, shown in the bracket along with trend is the 2 sigma error in the TOMS calculated trend coefficients and is shown in Figure 1 (bottom). The area of significant trend, values having magnitude more then the 2 sigma error, is highlighted by the contour lines. Figure 1 shows high magnitude of significant trend over south Asian region (China, Myanmar, India, Indonesia, Thailand, etc.) with the highest trend coefficient of 7–9(±3.4–4.6)% per decade which is equivalent to about 3–4 DU per decade. Magnitude of the trend coefficient is also high in some parts of China and Taiwan. Among the oceanic region, the Bay of Bengal is found to exhibit the highest trend coefficient of 4–6(±2.9–4.3) % per decade due to horizontal advection. In both the model and the observed trend plots (Figures 1 and 2), major parts of the Bay of Bengal, the Arabian Sea and some parts of the Pacific Ocean along their coastlines show positive trends higher than those over remote oceanic regions. A systematic pattern of outflow in the TCO from continent to oceanic region is noticed. The TCO trend coefficient over the Bay of Bengal and over some part of the Pacific Ocean close to China is high in magnitude as compared to those over Atlantic Ocean and over the remote Indian Ocean. The effect of advective transport is noticed quite deep into the Indian Ocean (80–100°E, 5°S) where trend is found to be around 3.5(±2.9–3.8)% per decade but it diminishes over regions further deep into the Ocean and becomes insignificant. The region of the Pacific Ocean close to China as shown in Figure 1 indicates an increasing trend 3–5(±3.4–4.3)% per decade in TCO. In satellite data, the highest trend is noticed in the grid covering the latitude belt of 15°N–30°N and the longitude 20°W–180°E which includes the Arabian Sea region. Some parts of South America (Brazil) show significant positive trends in ozone at the rate of 2–5 (±3.4–3.8)% per decade. Some parts of Atlantic Ocean indicate significant positive trend with the highest trend, 4–8(±3.9–6.4)% per decade, seen close to north Africa, the likely cause is the substantial increase of anthropogenic emissions of nitrogen oxides (NOx) associated with energy use in Africa, which has added to NOx from biomass burning and natural sources [Lelieveld et al., 2004]. On the other hand, the observations show significant trend over southern Atlantic Ocean between 10°S and the equator, which is not simulated by the model. No significant trend is observed over other parts of the Oceanic region. Satellite data indicate a trend of around 3–6(±3.4–4.3)% per decade in Saudi Arabia which is in agreement with the model results. Positive trend seen over Sudan, Mali and Mauritania are significant with highest trend observed over Mali, 4–5 (±3.4–3.8)% per decade, at the same time, calculated trend over Egypt, Mali, and Niger are not significant. Ziemke et al. [2005] had zonally averaged the data for the Pacific region in their analysis and they could not find any significant trend over the tropics using the TOMS data for 1979–2003. In the present analysis also, much smaller or insignificant trend is found over most part of the Pacific Oceanic regions which is in agreement with their results. Some differences are even understandable which could be due to the assumption of zonal homogeneities made by Ziemke et al. [2005] whereas no such assumptions are made in the present work. It should be noted here that the TOMS data resolution is 5° × 5°. Due to this we find interpolated trends between 2 grid points. In addition, it may also introduce some error and artificial trends near the edges at the beginning and ending points along the zonal axis.

Figure 1.

(Top) Geographical distribution of trend coefficient (% per decade) in TCO over the tropics (30°S–30°N) obtained in the present work from TOMS satellite data for the period 1979–2005. Blue contour lines show the area of significant trend. (Bottom) The 2-sigma error for the same.

Figure 2.

Geographical distribution of trend coefficient (% per decade) in TCO over the tropics (30°S–30°N) obtained in the present work from MOZART model simulations for the period 1991–2001.

[8] Model simulated trend coefficient shown in Figure 2 over major south Asian region is 7–8% per decade (i.e. 2–3 DU per decade) which agrees reasonably well with the results obtained from TOMS data. In the south East Asia, area around Philippines, model simulates positive trend of 3–4% per decade but there is no such signature in the TOMS calculated trend. TOMS trend in some parts of South America (Brazil) are comparable to the model results. Model does not indicate any positive trend over most of the parts of Africa but it simulates significant positive trends of 4–6% per decade over Peru coast and Colombia. Apart from this, positive trend of 2–3% per decade simulated by the model over Australia is not consistent with the insignificant trend shown by the TOMS. The broad patterns of trend in TCO over the South Asian region appear to be consistent with both satellite derived and model simulated results. In some cases like few parts of Bay of Bengal, India, Arabian sea, parts of North Africa and also parts of northern Atlantic ocean and Pacific ocean near China, model simulated trends are underestimated which is due to the fact that model integrations are performed to assess the changes in the tropospheric distribution of ozone that might have occurred as a result of changes in chemical emissions by keeping the identical wind conditions. In addition to the trends discussed above, present model analysis also provides information on seasonal, QBO, solar components and residues in the time series which is listed in Table 1 for a few specific locations. It is clear from Table 1 that the solar component is negative on all the locations whereas the magnitude of seasonal coefficient is substantial.

Table 1. Calculated Value of Different Coefficient in Some Locations for a Particular Grid Point as Obtained From TOMS-TCO Data Analysis
Grid, long., lat.Location of GridTOMS-TCO: CoefficientsResidue, DU
Seasonal Cycle, DUQBO, DU/10ms−1Solar, DU/100sfu
97.5°, 2.5°Indonesia24.25−0.59−0.430.03
72.5°, 17.5°India27.580.40−1.23−0.04
102.5°, 22.5°China (lower south)23.75−0.22−1.30−0.14
22.5°, −2.5°Congo34.42−1.74−0.150.01
−67.5°, −7.5°Brazil28.830.07−0.490.13
−2.5°, 17.5°Algeria28.08−0.43−0.53−0.06

[9] An attempt to understand the different increasing trend in TCO can be made. In general, the tropospheric ozone budget is believed to be controlled mainly through photochemistry involving anthropogenically produced nitrogen oxides (NOx), hydrocarbons and carbon monoxides over the ground surface. In addition to this, NOx produced over the tropics through lightning activities and enhanced aircraft emissions of NOx aloft have also been found to greatly influence the tropospheric ozone budget during the past few decades. It may also be noted that occurrence of thunderstorm and lightning are maximum over north Indian region (62% of the total number of occurrences over Indian sub-continent) based on the 30 years of data analysis [Kandalgaonkar et al., 2005]. These are a few factors which are responsible for the elevated level of TCO over the India subcontinent and surrounding South Asian region as evident from Figures 1 and 2. The fleets of aircrafts have been continuously increasing since 1970s. Beig and Brasseur [2006] have recently reported that the surface emissions of NOx in the Indian region have increased by 22% during the 1990's. They have also reported that CO emission over the tropical Indian subcontinent has increased by about 16% during 1990's, whereas the world's CO concentration has declined during the same period [Novelli et al., 1998]. This appears to be the main reason of an increase in TCO over the Indian region. Variations in the emissions of some of the organic compounds (e.g., C2H6, CH3COCH3, CH2O) during the past two decades have been substantial whereas, the estimates under the project POET (which are used) reveals that the changes in the emissions of other VOCs are marginal. For example, C2H6 emissions have increased by 28% from 1990 to the present time, primarily due to its increasing contribution from transport sector. These data indicate that there has been substantial increase in the emissions of secondary pollutants from various sources in the tropical regions. It is understood that the high growth rates in the TCO observed over South Asian countries like China, Indonesia, Thailand, etc., are due to enhanced emissions of NOx from biomass burning and fossil fuel combustion. The ozone produced over these continental regions is then transported towards the western Pacific Ocean by large-scale dynamics controlled by the prevailing wind systems and thus causing significantly positive trend in the TCO over the coastal marine region of western Pacific Ocean (Figure 1). During monsoon months, because of the weak northerly to northeasterly winds prevailing over the Gulf countries, Pakistan and Afghanistan, ozone converges over the Arabian Sea and is further transported towards the central and eastern part of India where high trends are noticed in Figure 1. Due to the prevailing winds, this chemical compound is transported towards the north-eastern part of India and towards Nepal. Among the oceanic region, the Bay of Bengal is found to exhibit the highest trend coefficient due to horizontal advection of ozone and ozone precursor rich air from north Indian continental region in case of northwesterly winds and from china in case of northeasterly winds.

[10] The outflow of secondary pollutants from South and South-East Asia is likely to be the most intensive where high level of bio-fuel emissions and bio-mass burning activities have been reported [Beig and Brasseur, 2006]. As a result of this, the TCO trend coefficient over the Bay of Bengal and over some part of the Pacific Ocean close to China is high in magnitude. The significance trend noticed (Figure 1) quite deep into the Indian Ocean (75–100°E, 5°S) is attributed due to advective transport. Some parts of North Africa and South America (Brazil) show significant positive trends in ozone which may be attributed mainly to savanna bio-mass burning.

5. Conclusions

[11] The TCO trend coefficient is found to be highest over the South Asian region led by part of China, Taiwan and Thailand and followed by India, Japan, Indonesia and Malaysia where significant trend is of the order of 7–9(±3.4–4.6)% per decade. Consistently high trend patterns in the tropical TCO over continental and neighboring oceanic regions clearly demonstrate significant role of continental pollutants and advective processes in influencing the distribution of TCO over Oceans. The TCO trend over the Bay of Bengal is found to be maximum, 4–6(±2.9–4.3) % per decade, of the TCO trends over all the oceanic regions as it is surrounded by continental boundaries of South Asian countries where emissions of ozone precursors from bio-fuel and biomass-burning are large. The trend in the TCO over Atlantic Ocean, which is mainly influenced by easterly winds originating from South African region, North America and Europe, is found to be relatively less in magnitude. The TCO trend is moderate over Indian Ocean but the continental flow from Indian subcontinent and part of South Asia substantially enhances the TCO trends. In most cases, the magnitude of strong and significant trends obtained by satellite data are found to be broadly consistent with the model results which accounts only for increasing level of chemical emissions. Hence, it is concluded that trend in the TCO is controlled, in the present case, by chemical emissions which have shown incessant increase in recent times, in particular, over the South Asian region.

[12] According to the third assessment report of International Panel of Climate Change [Intergovernmental Panel on Climate Change (IPCC), 2001], there has been an annual global mean increase in the tropospheric ozone by 6–13 DU since the pre-industrial times. The magnitude of the maximum increase in the TCO obtained in the present work over the South Asian region is found to be around 8–12 DU during the past two and half decades which is of the same order as above. If the scaling factor of radiative forcing is adopted from IPCC [2001], then the above mentioned increase in the tropospheric ozone will result in a radiative forcing of about 0.3–0.5 Wm−2 over the tropics which may have caused significant climatic implications.


[13] We sincerely thank with gratitude the support provided by Pawan Bhartia and Jerry R. Ziemke, GSFC, NASA for satellite data. We acknowledge the Department of Science and Technology, New Delhi, for financial assistance.