- Email: [email protected]
Trends in total ozone column over India: 1979–2008
Atmospheric Environment 45 (2011) 1648e1654
Contents lists available at ScienceDirect
Atmospheric Environment journal homepage: www.elsevier.com/locate/atmosenv
Trends in total ozone column over India: 1979e2008 Ankit Tandon, Arun K. Attri* School of Environmental Sciences, Jawaharlal Nehru University, New Delhi 110067, India
a r t i c l e i n f o
a b s t r a c t
Article history: Received 22 May 2010 Received in revised form 25 December 2010 Accepted 5 January 2011
Time-series decomposition analysis was performed on, (1) Multi Sensor Reanalysis (MSR) Total Ozone Column (TOC) monthly mean time-series data-set [1979e2008], and (2) Total Ozone Mapping Spectrometer (TOMS) Version 8 Overpass monthly mean time-series data-set from Nimbus 7 satellite, (TOMS N7) [1979e1993] to estimate long-term linear trends in the data to assess a scale of surface UV changes over India. Long-term trend estimation, subsequent to the removal of annual cyclic variations, for MSR TOC data-set was done over Indian region covering latitude spread 0 N e 40 N, and Longitude spread 67.5 E to 97.5 E. Trend estimates for TOMS Overpass data-sets, treated on similar lines, was done for fifteen locations over India (1190 N to 34 040 N). Statistically significant declining trends ranging from () 0.8 e () 1.5 percent/decade were seen over Indian region above 25 N latitude in MSR TOC data-set (1979e2008). In case of TOMS N7 data-set (1979e1993), statistically significant declining trends were estimated over New Delhi (28 400 N) and Srinagar (34 040 N) with a value of () 2.5 and () 3.6 percent/ decade respectively. Observed TOC decline covered 40% of total geographical area of Indian region, however rest of the Indian region (peninsular) did not show statistically any significant trend. Ó 2011 Elsevier Ltd. All rights reserved.
Keywords: Total ozone column Time series analysis Long term trends Satellite data
1. Introduction The functional attribute of the Stratospheric “Ozone Layer” to modify the incoming solar radiation by filtering the high energy ultra violet (UV) portion of the spectrum is crucial. Its role in shaping the distinct temperature regime in stratosphere, different to the one prevailing in Troposphere, is important for the global circulation of material and energy (Cockell and Blaustein, 2001; Shen et al., 1995). Scientific recording of the Total Ozone Column (TOC) was initiated during the early part of the twentieth century (Fabry and Buisson, 1913; Dobson, 1930, 1957, 1968; Dutsch, 1974); more systematic recording of TOC over different latitudinal locations was initiated in 1957, the International Geophysical Year (IGY), by establishing a network of ground-based instruments (the Dobson network). These initiatives were turning point in establishing the thinning of Ozone Layer, whereby it was recognized that “Ozone Hole” was forming every spring over Antarctica (Farman et al., 1985; Jones and Shanklin, 1995). Global concern towards thinning of Ozone layer initiated number of investigations to assess the short-term and long-term trends in TOC abundance over many locations. Initial attempts in analyzing the long-term trend in the TOC contents revealed an increase in its abundance over many locations (Komhyr et al., 1971; Angell and Korshover, 1973; London and Kelley,
* Corresponding author. Tel.: þ91 11 26704309. E-mail address: [email protected] (A.K. Attri). 1352-2310/$ e see front matter Ó 2011 Elsevier Ltd. All rights reserved. doi:10.1016/j.atmosenv.2011.01.008
1974). However, investigations undertaken after the discovery of “Ozone Hole” in 1985 (Farman et al., 1985), involving statistical analysis of TOC data to estimate the long-term trends in the TOC reported a decline: data used in these analysis was post 1970s onwards for many locations (Reinsel et al., 1987, 1994; Bojkov et al., 1990, 1995; Bojkov and Fioletov, 1995; Harris et al., 2003). The analysis included representative data over locations in different climatic zones, but the stress was more on the temperate and Polar Regions where the extent of variability in TOC abundance is large. However, such investigations hold equal importance for the tropical regions housing maximal biodiversity and also receiving solar-UV irradiance much higher than that over latitudes 40 (N or S); even a small increase in UV can have significant impact on the prevalent biodiversity and balance between different species (Vashney and Attri,1995). The case in point is the tropical-subtropical climatic region of India inhabiting a human population in excess of 1 billion (Census of India, 2001). Implication of even a “small” decrease in the Ozone layer thickness over this region, expectedly, will have far reaching impact on agriculture output and food security. The findings reported, on the analysis of TOC time-series data, over several locations in Indian subcontinent to estimate trends in literature (Reinsel et al., 1987, 1994; Bojkov and Fioletov, 1995), are not recent. Investigation reported by Reinsel et al. (2002), covering the latitudinal spread from 30 to 60 , inferred that a positive change in the TOC abundance can be detected within 7e8 years from the time-point of reversal in the declining trend, which was taken as 1996e97. In this regard, Andersen et al. (2006), using the hockey stick
A. Tandon, A.K. Attri / Atmospheric Environment 45 (2011) 1648e1654
method (Mann et al., 1998), reported a positive trend in TOC for both Northern and Southern hemispheres. However, these conclusions were countered by Weatherhead and Andersen (2006). The most recent body of work, specifically estimating the TOC variability over Indian region (Chakrabarty et al., 1998; Sahoo et al., 2005; Ganguly and Iyer, 2006; Pal, 2010) have presented widely varying TOC trend estimates. Analysis of TOC time-series data-sets, in these investigations was done by using simple linear regression, and not accounting for the presence of strong autocorrelation on account of seasonal and inter-annual variations present in TOC values. Reported results, from these studies, manifest large declining TOC trends all over India, errors do not figure in their estimated trends; in short analysis lacked statistical rigor, and estimated trends stood at variance with other studies (Reinsel et al., 1994, 2005; Chipperfield et al., 2007). Estimated trends over matching locations reported by Reinsel et al. (1994) were notsignificant (significant decline was only observed over Varanasi, and Srinagar). Scientific assessment report of ozone depletion, covering different zones, reported no change in TOC trends over equatorial region (0e10 N & S). Small decline in TOC was estimated over region 10e25 (N,S) latitude, though statistically it was found to be in-significant (Chipperfield et al., 2007). Significance of calculated trend is with reference to the spread of error therein; often the trends reported in literature associate 1s error, whereas 2s error is considered more reliable to test the significance (Weatherhead et al., 1998). Additional factor of concern, while estimating TOC trend over a region, requires adequate attention to ascertain the presence of autocorrelation in the datasets used in the estimation of the trends (Weatherhead et al., 2000). In this light, the differences reported in trends between investigations focused specifically over Indian region (Chakrabarty et al., 1998; Sahoo et al., 2005; Ganguly and Iyer, 2006; Pal, 2010), and from those reported in zonal studies (Reinsel et al., 1994, 2002, 2005) stand out in stark variance. Given the differences in the estimated trends from investigations reported over India over many locations (Chakrabarty et al., 1998; Sahoo et al., 2005; Ganguly and Iyer, 2006; Pal, 2010), and that estimated from investigations of zonal TOC trends (Reinsel et al., 1994, 2002, 2005; Chipperfield et al., 2007) we in this manuscript present statistical analysis of two TOC series: (1) Multi Sensor Reanalysis (MSR) TOC monthly mean time-series data-set, Jan 1979eDec 2008, and (2) Total Ozone Mapping Spectrometer Version 8 Overpass monthly mean time-series data-set from Nimbus 7 satellite, (TOMS N7), Jan 1979eDec 1993. The first (MSR) data-set allowed us to do analysis to estimate trends as a function of the size of data-set (Number of data points in series) by punctuating the data-set of the size corresponding to TOMS N7 Overpass (180 month data). Second, the analysis on TOMS Overpass data-set, primarily was done to compare trend estimates presented in this work with the results from investigation focused over many Indian locations. Statistical de-convolution of trends was done in stepwise manner by removing the influence of annual cycles from the data methodology used earlier by Khalil and Rasmussen (1990). TOC trends over whole Indian region from MSR data-set (360 months) are plotted as contours: 0 Ne40 N 67.5 Ee97.5 E latitude/longitude (861 grid points) to provide better assessment of regions over which TOC trends are statistically significant. Significance of calculated trends was ascertained by accounting for the effect of autocorrelation present in respective de-seasonalized TOC data-sets, while estimating 2s error in trend. 2. Details of data used Multi Sensor Reanalysis (MSR) monthly mean total ozone column time-series data-set (resolution of 1 latitude 1.5 longitude) from
1649
Jan 1979 to Dec 2008 (360 months) was accessed from www.temis.nl/ protocols/O3global.html, details of appropriate correction function used and assimilation of multiple satellite data-sets are discussed by van der A et al. (2010). In the present analysis, the data-set for spatial coverage 0 Ne40 N 67.5 Ee97.5 E latitude/longitude (861 grid points) over Indian region was used. The TOMS instrument measures solar irradiance and the radiance backscattered by the Earth’s atmosphere at six wavelengths extending from 310 to 380 nm, Ozone absorption in the Huggins band is used to calculate the TOC abundance (Heath et al., 1975). TOMS Version 8, daily Overpass data-set (spatial resolution of 1 latitude 1.25 longitude) is accessible from National Aeronautics and Space Administration (NASA) at http://toms.gsfc. nasa.gov/n7toms/n7_ovplist_1.html. The monthly mean TOC time-series were generated by taking arithmetic means of daily values for a particular month. The time-series TOMS N7 (TOMS data from Nimbus 7 satellite) and TOMS TOC (l.300) by TOMS EP (from Earth Probe satellite) when the results for the 1979e1993 (n ¼ 180 month) and 1997e2005 (n ¼ 108 month) are discussed, respectively. 3. Methodology used in the analysis of TOC time-series (MSR and TOMS overpass data-sets) The underlying assumption, while subjecting the time-series to mathematical analysis, is that each observation represents the summation of many components (Khalil and Rasmussen, 1990; Gottman, 1981). Monthly mean TOC time-series data-sets (MSR and TOMS N7) are free of daily and weekly variations, but still retain seasonal, inter-annual, and other random influences. The analysis de-convolutes the TOC time-series data in steps to extract all relevant components. The general equation representing TOC time-series can be written as:
TOC ¼ trend þ annual cycle þ residuals
(1)
3.1. Removal of the annual cyclic effects The “trend” in above equation can be further expressed as the sum of the linear trend and inter-annual variations (IAV) present in the TOC time-series (Khalil and Rasmussen, 1990):
trend ¼ linear trend þ inter annual variations
(2)
where, the linear trend in functional form is equal to C0 þ b t; C0 being the intercept and b the slope, and t is the time in the units of months. Estimated value of b represents the linear long-term trend of the time-series and C0 correspond to the fitted value of TOC at the start of the time-series. Determination of b (trend) requires the removal of the seasonal influences present in the form of annual cyclic variation in the monthly mean TOC data, which was done in two steps. Respective TOC data-sets were processed by using thirteen months weight centered moving average filter represented by expression (3).
xj ¼ Cj
k ¼X jþT=2 k ¼ jT=2
wk $Ck ; j
T þ1 2
(3)
In the above expression, T represents the period of annual cycle (12 months). Thus, the first value would be 7, and wk ¼ 1/24 for first and last weighting coefficients and for rest wk ¼ 1/12. In the second step, transformed TOC time-series data (TTOC) was used to calculate the magnitude of annual cycle’s contribution (Uk) present in each month’s value (January to December):
1650
Uk ¼
A. Tandon, A.K. Attri / Atmospheric Environment 45 (2011) 1648e1654
X 1 M1 x M i ¼ 1 kþTi
(4)
where, k ¼ (i mod 12), i represents the month in the time-series data-set, M stands for number of years. Uk value, one for each respective month (January U1,., December U12), is estimated. The de-seasonalized time-series (DTOCi) is obtained by subtracting from the original TOC time-series (Ci) month’s mean value, the corresponding Uk (annual cycle/seasonal affects).
3.2. Estimation of long-term linear trend in TOC abundance The de-seasonalized data-set DTOCi obtained for MSR and TOMS (N7 and EP), at each location, were subjected to linear regression to obtain C0 (intercept) and b (long-term linear trend) over a given location for the time span of the respective time-series:
Ci0 ¼ C0 þ b t
(5)
Ci0
In the Eq. (5), is the value of ith observation in the DTOC timeseries, and t is the month number for ith observation. The linear trend estimated for each location is expressed with the 2s error (Davis, 2002; Wilks, 2006). Both, TOMS and MSR data-sets, were subjected to same analysis; although MSR data-set’s spread was over whole Indian region [Latitude 0e40 N, and Longitude 67.5 to 97.5 E], whereas TOMS Overpass data-sets were for specific locations only. For comparison the equivalent relevant values from MSR analysis were punctuated for corresponding locations (latitude and longitude) of TOMS data-sets. 3.3. Calculation of 2s error in estimated trends The significance of calculated trends in TOC, for Gaussian distribution, is ascertained by associated standard errors, which is square-root of variance estimated by Mean Square Errors (MSE) as Pn ðCi00 Ci0 Þ2 i¼1 and therefore, 95% confiS2b ¼ PMSE 2 where, MSE ¼ n2 ðti tÞ qffiffiffiffiffi dence limits of b are calculated as b ð1:96 S2b Þ where, ti is the ith month’s number, t is the mean of month numbers, Ci00 is the estimated de-seasonalized TOC value for ith month, Ci0 is the observed de-seasonalized TOC value for ith month, and n is the total number of months in the de-seasonalized TOC time-series. The trends estimates are done on de-seasonalized TOC values (DTOCi), which still retains the inter-annual variations (Eq. (2)). Presence of IAVs in DTOCi data-sets adds persistence in the adjacent values, and can be measured by calculating first order autocorrelation (r1) in DTOCi. In all data-sets (MSR and TOMS N7 and EP) used for trend estimation the r1 was found to be significant (Table 1). Presence of significant r1 would reduce the effective sample size to n0 nN thus affecting the variance and standard error estimates. Variance in this case is estimated by using following corrected value of MSE (Wilks, 2006; Weatherhead et al., 1998; Pn ðCi00 Ci0 Þ2 i¼1 Weatherhead et al., 2000): MSE0 ¼ ð1þr1 1r1Þ, in effect n2 the error estimates done without accounting for IAV’s presence in TOC data-set would underestimate the error, and in turn affect the significance of estimated trend in TOC. Size of the data-set, in view of the reduced number of observations n0 from n will depend on the strength of autocorrelation present in the DTOCi values. Calculated values of autocorrelation Function (ACF) r1, for MSR and TOMS data-sets are given in Table 1 with the reduced number of observations n0 .
Table 1 Calculated values of Autocorrelation function (r1), and reduced number of observations due to the presence of Inter-annual variations associated with de-seasonalized (DTOCi) data-sets (TOMS TS and MSR). Location
ACF (r1) in TOMS TS TOC Data-set
n0 , from original n ¼ 180, for TOMS TS TOC data-set
ACF (r1) for MSR TOC data-set
n0 , from original n ¼ 360, for MSR TOC data-set
Trivandrum Kodaikanal Banglore Madras Hyderabad Poona Bombay Nagpur Dum Dum Ahmedabad Mount Abu Benares Varanasi New Delhi Srinagar
0.76 0.68 0.69 0.61 0.69 0.72 0.75 0.73 0.65 0.75 0.73 0.71 0.70 0.68 0.55
25 34 33 44 33 29 26 28 38 26 28 31 32 34 52
0.74 0.69 0.61 0.63 0.67 0.70 0.71 0.69 0.63 0.68 0.69 0.68 0.68 0.69 0.56
54 66 87 82 71 64 61 66 82 69 66 69 69 66 102
4. Results Applicability of statistical analysis methodology to analyze TOC time-series, for extracting trends and errors therein, requires prior knowledge of data’s characteristic frequency distribution (Wilks, 2006). This was ascertained by examining the parameters like mean, median, mode, range, inter quartile range and other related features of the distribution: kurtosis, skewness and variance (Storch and Zwiers, 1999; Davis, 2002; Wilks, 2006). Data-sets used in analysis manifested close to normal distribution of the monthly mean TOC values (representative Supplementary Fig. S1), which was also confirmed from ShapiroeWilk’s test to confirm the extent of normality present (Hair et al., 2006). The results are given in Supplementary Table S1. 4.1. Estimation of long-term trend in the total ozone column Calculated estimates for long-term trends in percent/decade with corresponding 2s error for MSR data-set (1979e2008) and TOMS N7 Overpass data-set (1979e1993) are given in Table 2 for 15 locations in Indian region. For comparing the trend estimates with those reported in earlier India specific-investigations, estimates were also done on TOMS EP Overpass EP data-set. Analysis involving three different data-sets of varying sizes allowed us to unambiguously determine as to how the data-size affected the statistical significance of estimated trends. Trends extracted from longer MSR data-set manifested statistically significant declining trends [() 0.8 e () 1.5 percent/decade] over Indian region above 25 N latitude; this included locations viz. Varanasi (25 270 N), New Delhi (28 400 N), and Srinagar (34 040 N). In case of TOMS N7 the declining TOC trends were significant over New Delhi (28 400 N) and Srinagar (34 040 N) having decline of () 2.5 and () 3.6 percent/decade respectively. It is apparent that the estimated trends at all locations for TOMS EP data-set were not found to be significant, as the calculated 2s errors were large. Statistical significance of the calculated TOC trends in view of the presence of 2s error as a function of the size (n ¼ 108, 180, and 360) of the MSR TOC data-set is shown in Figs. 1e3 as contour plots (0 Ne40 N 67.5 Ee97.5 E). These plots provide better grasp of the extent of region (area) of India where significant decline in TOC is estimated. Trend contours for segmented MSR data-set (1997e2005) size 108 month are plotted in Fig. 1. As observed, for equivalent TOMS EP Overpass data-set [Table 2], all over India the estimated trends
A. Tandon, A.K. Attri / Atmospheric Environment 45 (2011) 1648e1654
1651
Table 2 Long-term trend estimates (% per decade) for TOMS N7 & EP, and MSR de-seasonalized monthly mean TOC time-series data-sets; and the trend estimates reported by Reinsel et al. (1994) and Harris et al. 1995. Trends in shaded cells are statistically significant at 95% confidence limit. Areas of the region above which the estimated trends (decline) are statistically significant constitute w40% of total area of India. Location
Latitude
TOMS N7 (Jan 1979eDec 1993) n ¼ 180 month
TOMS EP (Jan 1997eDec 2005) n ¼ 108 month
MSR (Jan 1979eDec 2008) n ¼ 360 month
Trivandrum Kodaikanal Banglore Madras Hyderabad Poona Bombay Nagpur Dum Dum Ahmedabad Mount Abu Benares Varanasi New Delhi Srinagar
8 280 N 10 130 N 12 580 N 13 040 N 17 220 N 17 220 N 19 070 N 21 060 N 22 380 N 23 010 N 24 360 N 25 000 N 25 270 N 28 400 N 34 040 N
0.66 0.87 0.94 1.26 1.26 1.32 1.43 1.39 1.42 1.83 2.05 1.95 1.99 2.52 3.60
2.36 1.91 1.35 1.17 0.84 0.65 0.48 1.07 1.03 0.84 1.19 1.24 1.37 1.76 3.25
0.15 0.67 0.10 0.55 0.03 0.52 0.01 0.53 0.01 0.62 0.18 0.70 0.23 0.71 0.30 0.68 0.37 0.62 0.61 0.72 0.78 0.73 0.66 0.76 0.81 0.76 1.14 0.92 1.49 0.85
a
1.99 1.50 1.60 1.52 1.77 1.90 2.11 2.12 1.89 2.30 2.34 2.15 2.16 2.41 2.53
3.73 2.59 2.36 1.82 3.48 3.53 3.79 3.97 3.53 3.64 3.60 3.45 3.75 4.39 5.49
Reinsel et al. (1994) (Jan 1978eDec 1991) n ¼ 16 monthsa
Harris et al. (1995) (Jan 1979eFeb 1994) n ¼ 182 months
1.95 1.92
0.2 2.1
1.04 1.80
0.95 4.20
1.5 1.7
2.92 1.68 1.17 2.32 4.93 3.48
1.2 1.5 1.1 1.9
In Reinsel et al. (1994), 1s errors associated with long-term trends were reported, for comparison their errors, in above Table, are multiplied by 2 (i.e., 2s).
were statistically not-significant. Appearance of contour lines in Fig.1, compared with Fig. 2 [MSR data-set (1979e1993; n ¼ 180 month)] and Fig. 3 [MSR data-set (1979e2008; n ¼ 360 month)] seems haphazard over whole region. Fig. 2 [MSR data-set; 1979e1993], demarcates the plotted contour of TOC trends where decline is statistically significant. The same, in Fig. 3, for longer MSR TOC
data-set (360 month) show the additional Indian region where statistically decline in trends in TOC is significant. It is interesting to note that the quantum of decline in TOC trend (statistically significant) over Indian region is smaller in case of MSR TOC time-series data-set spanning from Jan 1979eDec 2008 than MSR TOC timeseries data-set spanning from Jan 1979eDec 1993 (Figs. 2 and 3).
Fig. 1. Contour plots of TOC trends (% per decade) over India for segmented MSR data-set, 1997e2005, (108 month). For this size data-set the calculated 2s error makes the trend estimates statistically not-significant. Consequence of data-set size is important in determining the significance of estimated trends.
1652
A. Tandon, A.K. Attri / Atmospheric Environment 45 (2011) 1648e1654
Fig. 2. Contour plots of TOC trends (% per decade) for segmented MSR data-set (180 months). Decline in TOC trends was statistically significant over New Delhi (28 400 N), and Srinagar (34 040 N). Geographical area of India, over which the TOC decline was significant is about 30% of the total area; horizontal thick grey line surrounded with two thick vertical arrows marks the region where trends in TOC decline is significant.
5. Discussions 5.1. Statistical significance of long-term trends in TOC over India Perhaps, while estimating the long-term trends in TOC over a location, statistical significance is the most important in determining the validity of the result. In this regard the calculation of error in estimated trend becomes crucial. This aspect has been, lucidly covered by Weatherhead et al. (1998) and Weatherhead et al., (2000). Significance of trend is judged on the basis of the magnitude of the 2s error (95% confidence). For comparison of results obtained in this study (TOMS N7; 1979e1993), we have compared the estimated trends with that of Reinsel et al. (1994) and Harris et al. (1995) for common locations (Srinagar, New Delhi, Varanasi, Ahemdabad, Poona, and Kodaikanal); their trend estimates with calculated standard error therein are tabulated in Table 2 along with the trends estimated in present work. The data-set reported in Reinsel et al. (1994) and Harris et al. (1995) was from ground-based Dobson spectrometer values; their values are comparable with that reported in the present work with slight difference. This difference most probably is due to the Dobson spectrometer based data-set used in their studies; whereas the TOMS Overpass data-set is satellite based, and MSR data-set is a re-analyzed TOC data-set obtained from multiple sensors, and accounted for factors affecting and likely to introduce variability in TOC data procured from different sources (Heath et al., 1975; van der A et al., 2010). The 2s errors associated with the estimated trends over few locations for TOMS N7 data-set in
present study are higher than the equivalent errors reported by Reinsel et al. (1994) and Harris et al. (1995). This can easily be attributed to the removal of ozone variability and extraction of the anthropogenic trend component in Reinsel et al. (1994) and Harris et al. (1995); in their analysis a multi-linear model was used, where Solar and QBO effects were accounted for. Consequently, the reported trend values would also be slightly different. 5.2. Data-set’s size and error in the estimated TOC trend The size of TOC data-set (number of months) does show its influence on, both, estimated trends and calculated error therein. Errors in the estimated trends decrease with the increase in the data-set’s size. This can be seen from trend estimates presented in this work and those reported by Reinsel et al. (1994) [Table 2]. This aspect has also been reported earlier (Weatherhead et al., 1998; Weatherhead et al., 2000). The affect of TOC data-set’s size on providing statistically more acceptable trend pattern over India can be perceived from trend contours plotted for segmented MSR data-sets (n ¼ 108, 180 months) in Figs. 1 and 2, and complete MSR data-set (n ¼ 360 month) in Fig. 3; in later case the contour trends show regular change, though the statistically significant decline is confined to over a region above 25 N latitude. These results agree well with the observations made from zonal trend estimates over tropics in WMO report, 2006 (Chipperfield et al., 2007). It is equally relevant to state that the area of the Indian region, where trends are estimated as significant, covers almost w40% area of
A. Tandon, A.K. Attri / Atmospheric Environment 45 (2011) 1648e1654
1653
Fig. 3. Contours plots of TOC trends (% per decade) for MSR data-set (360 month). Decline in TOC trends was statistically significant (2s confidence limit) over Varanasi (0.81 0.76), New Delhi (1.14 0.92 percent per decade), and Srinagar (1.49 0.85 percent per decade). For this size data-set the area over which the decline was significant is about 40% of the total geographical area of India; horizontal thick grey line surrounded with two thick vertical arrows demarcates the region where trends in TOC decline are significant.
India. Interestingly this region lies in the Northern part of the country (Himalayan and Sub-Himalayan region), encompassing Indo-Gangetic planes holding importance for agriculture produce and food security of the country. For MSR data-set, comparable to TOMS N7 (1979e1993; n ¼ 180 months) estimated declining trends are statistically significant over a region above 28 N latitude; in terms of area 30% of India’s geographical area. For dataset of size n ¼ 108 month, over no location the estimated trends were found to be significant. The size of the data-set used in estimating the trends have bearing on the calculated errors. Lastly, it can be stated that decline in TOC over large geographical region (40% of total area) of India is evident; though the decline is small, but given the much large insolation of solar-UV in this region, the expected increase in UV reaching the ground will hardly be trivial.
data-set (1979e2008). Estimated decline, respectively, for New Delhi (28 400 N) and Srinagar (34 040 N) were () 2.5 and () 3.6 percent/ decade also observed for TOMS N7 data-set (1979e1993).
6. Conclusions
Appendix. Supplementary data
In MSR TOC data-set (1979e2008), statistically significant declining trends ranging from () 0.8 e () 1.5 percent/decade were observed over Indian region above 25 N latitude. The analysis also show that the quantum of decline between MSR data-set (1979e2008) was smaller when compared with that estimated from MSR data-set (1979e1993); at the same time the statistically significant decline in TOC covered much larger Indian region for MSR
Supplementary data associated with this article can be found in the online version, at doi:10.1016/j.atmosenv.2011.01.008.
Acknowledgements Authors acknowledge Goddard Space and Flight Centre, NASA for providing TOMS TOC data-sets and Tropospheric Emission Monitoring Internet Service (TEMIS) for MSR TOC data-sets used in present study. The financial assistance provided by CSIR (India) in the form of a research project to AKA is acknowledged. One of the authors, AT, acknowledges UGC (India) and DST (India) for their financial support in the form of fellowships and award of Research Project for Young Scientists. Authors extend their gratitude to two anonymous referees of this manuscript for their in-depth review; their suggestions and queries have resulted in much improved manuscript.
References Andersen, S.B., et al., 2006. Comparison of recent modeled and observed trends in total column ozone. J. Geophys. Res. 111, D02303. doi:10.1029/2005JD006091.
1654
A. Tandon, A.K. Attri / Atmospheric Environment 45 (2011) 1648e1654
Angell, J.K., Korshover, J., 1973. Quasi-biennial and long-term fluctuations in total ozone. Monthly Weather Rev. 101, 426e443. Bojkov, R.D., Fioletov, V.E., 1995. Estimating the global ozone characteristics during the last 30 years. J. Geophys. Res. 100 (D-8), 16,537e16,551. Bojkov, R.D., Bishop, L., Fioletov, V.E., 1995. Total ozone trends from quality-controlled ground-based data (1964e1994). J. Geophys. Res. 100 (D-12), 25,867e25,876. Bojkov, R.D., Bishop, L., Hill, W.J., Reinsel, G.C., Tiao, G.C., 1990. A statistical trend analysis of revised Dobson total ozone data over the northern hemisphere. J. Geophys. Res. 95 (D-7), 9785e9807. Census India, 2001. Available at: http://www.censusindia.gov.in/. Chakrabarty, D., Peshin, S., Pandya, K., Shah, N., 1998. Long-term trend of ozone column over the Indian region. J. Geophys. Res. 103 (D-15), 19245e19251. Chipperfield, M.P., Fioletov, V.E., et al., 2007. Global Ozone: Past and Present, Chapter 3 in Scientific Assessment of Ozone Depletion: 2006. Global Ozone Research and Monitoring ProjectdReport No. 50. World Meteorological Organization, Geneva, Switzerland. Cockell, C.S., Blaustein, A.R., 2001. Ecosystems, Evolution and Ultraviolet Radiation. SpringerVerlag, New York, USA. Davis, J.C., 2002. Statistics and Data Analysis in Geology. Jhon Wiley & Sons, New York, USA. Dobson, G.M.B., 1930. Observations of the amount of ozone in the Earth’s atmosphere and its relation to other geophysical conditions. Proc. R. Soc. London, Ser. A. 129, 411. Dobson, G.M.B., 1957. Observers handbook for the ozone spectrophotometer. In: Annals of the International Geophysical Year V, Part 1. Pergamon Press, New York, USA, pp. 46e89. Dobson, G.M.B., 1968. Forty years’ research on atmospheric ozone at Oxford: a history. Appl. Opt. 7, 387e405. Dutsch, H.U., 1974. The ozone distribution in the atmosphere. Can. J. Chem. 52, 1491e1504. Fabry, C., Buisson, M., 1913. L’absorption de l’ultraviolet par l’ozone et la limite du spectre solaire. J. Phys. 3, 196e206. Farman, J.C., Gardiner, B.G., Shanklin, J.D., 1985. Large losses of total ozone in Antarctica reveal seasonal ClOx/NOx interaction. Nature 315, 207e210. Ganguly, N.D., Iyer, K.N., 2006. Long-term trend in ozone and erythemal UV at Indian latitudes. J. Atmos. Chem. 55, 227e239. Gottman, J.M., 1981. Time Series Analysis. Cambridge University Press, Cambridge, U.K. Hair, J.F., Black, B., Babin, B., Anderson, R.E., Tantham, R.L., 2006. Multivariate Data Analysis. Pearson Prentice Hall, London, UK. Harris, J.M., Oltmans, S.J., Bodeker, G.E., Stolarski, R., Evans, R.D., Quincy, D.M., 2003. Long-term variations in total ozone derived from Dobson and satellite data. Atmos. Environ. 37, 3167e3175. Harris, N.R.P., et al., 1995. Observed changes in ozone and source gases: total and vertical-column ozone, chapter 1. Global Ozone Research and Monitoring ProjectdReport No. 37. In: Scientific Assessment of Ozone Depletion: 1994. World Meteorological Organization, Geneva, Switzerland. Heath, D.F., Krueger, A.J., Roeder, H.A., Henderson, B.H., 1975. The solar Backscatter ultraviolet and total ozone mapping spectrometer (SBUV/TOMS) for Nimbus G. Opt. Eng. 14, 323e331.
Jones, A.E., Shanklin, J.D., 1995. Continued decline of total ozone over Halley, Antarctica, since 1985. Nature 376, 409e411. Khalil, M.A.K., Rasmussen, R.A., 1990. Atmospheric methane: recent global trends. Environ. Sci. Technol. 24 (4), 549e553. Komhyr, W.D., Barrett, E.W., Slocum, G., Weickmann, H.K., 1971. Atmospheric total ozone increase during the 1960s. Nature 232, 390e391. London, J., Kelley, J., 1974. Global trends in total atmospheric ozone. Science 184 (4140), 987e989. Mann, M.E., Bradley, R.S., Hughes, N.K., 1998. Global scale temperature climate forcing over the past six centuries. Nature 392, 777e787. Pal, C., 2010. Variability of total ozone over India and its adjoining regions during 1997e2008. Atmos. Environ. 44, 1927e1936. Reinsel, G.C., Tiao, G.C., Miller, A.J., Wuebbles, D.J., Connell, P.S., Mateer, C.L., Deluisi, J.J., 1987. Statistical analysis of total ozone and stratospheric Umkehr data for trends and solar cycle relationship. J. Geophys. Res. 92 (D-2), 2201e2209. Reinsel, G.C., Tiao, G.C., Wuebbles, D.J., Kerr, J.B., Miller, A.J., Nagatani, R.M., Bishop, L., Ying, L.H., 1994. Seasonal trend analysis of published ground-based and TOMS total ozone data through 1991. J. Geophys. Res. 99 (D-3), 5449e5464. Reinsel, G.C., Weatherhead, E.C., Tiao, G.C., Miller, A.J., Nagatani, R.M., Wuebbles, D.J., Flynn, L.E., 2002. On detection of turnaround and recovery in trend for ozone. J. Geophys. Res. 107 (D-10), 4078. Reinsel, G.C., Miller, A.J., Weatherhead, E.C., Flynn, L.E., Nagatani, R.M., Tiao, G.C., Wuebbles, D.J., 2005. Trend analysis of total ozone for turn around and dynamical contributions. J. Geophys. Res. 110. doi:10.1029/204JD004662. Sahoo, A., Sarkar, S., Singh, R.P., Kafatos, M., Summers, M.E., 2005. Declining trend of total ozone column over the northern parts of India. Int J Remote Sens. 26 (16), 3433e3440. Shen, T.L., Wooldridge, P.J., Molina, M.J., 1995. Stratospheric pollution and ozone depletion. In: Singh, H.B. (Ed.), Composition, Chemistry, and Climate of the Atmosphere. Van Nostrand Reinhold, New York, USA. Storch, H.V., Zwiers, F.W., 1999. Statistical Analysis in Climate Research. Cambridge University Press, Cambridge, UK. van der A, R.J., Allaart, M.A.F., Eskes, H.J., 2010. Multi sensor reanalysis of total ozone. Atmos. Chem. Phys. Discuss. 10, 11401e11448. Vashney, C.K., Attri, A.K., 1995. In: Majumdar, S.K., Brenner, F.J., Miller, E.W., Rosenfield, L.M. (Eds.), Impact of Stratospheric Ozone Depletion on Terrestrial Ecosystem, in Environmental Contaminants and Health. Pennsylvania Academy of Sciences, USA. Weatherhead, E.C., et al., 1998. Factors affecting the detection of trends: statistical considerations and applications to environmental data. J. Geophys. Res. 103, 17,149e17,161. Weatherhead, E.C., et al., 2000. Detecting the recovery of total column ozone. J. Geophys. Res. 105, 22,201e22,210. Weatherhead, E.C., Andersen, S.B., 2006. The search for signs of recovery of the ozone layer. Nature 441, 39e45. Wilks, D.S., 2006. Statistical Methods in the Atmospheric Sciences. Academic PressElsevier, San Diego, USA.
Contents lists available at ScienceDirect
Atmospheric Environment journal homepage: www.elsevier.com/locate/atmosenv
Trends in total ozone column over India: 1979e2008 Ankit Tandon, Arun K. Attri* School of Environmental Sciences, Jawaharlal Nehru University, New Delhi 110067, India
a r t i c l e i n f o
a b s t r a c t
Article history: Received 22 May 2010 Received in revised form 25 December 2010 Accepted 5 January 2011
Time-series decomposition analysis was performed on, (1) Multi Sensor Reanalysis (MSR) Total Ozone Column (TOC) monthly mean time-series data-set [1979e2008], and (2) Total Ozone Mapping Spectrometer (TOMS) Version 8 Overpass monthly mean time-series data-set from Nimbus 7 satellite, (TOMS N7) [1979e1993] to estimate long-term linear trends in the data to assess a scale of surface UV changes over India. Long-term trend estimation, subsequent to the removal of annual cyclic variations, for MSR TOC data-set was done over Indian region covering latitude spread 0 N e 40 N, and Longitude spread 67.5 E to 97.5 E. Trend estimates for TOMS Overpass data-sets, treated on similar lines, was done for fifteen locations over India (1190 N to 34 040 N). Statistically significant declining trends ranging from () 0.8 e () 1.5 percent/decade were seen over Indian region above 25 N latitude in MSR TOC data-set (1979e2008). In case of TOMS N7 data-set (1979e1993), statistically significant declining trends were estimated over New Delhi (28 400 N) and Srinagar (34 040 N) with a value of () 2.5 and () 3.6 percent/ decade respectively. Observed TOC decline covered 40% of total geographical area of Indian region, however rest of the Indian region (peninsular) did not show statistically any significant trend. Ó 2011 Elsevier Ltd. All rights reserved.
Keywords: Total ozone column Time series analysis Long term trends Satellite data
1. Introduction The functional attribute of the Stratospheric “Ozone Layer” to modify the incoming solar radiation by filtering the high energy ultra violet (UV) portion of the spectrum is crucial. Its role in shaping the distinct temperature regime in stratosphere, different to the one prevailing in Troposphere, is important for the global circulation of material and energy (Cockell and Blaustein, 2001; Shen et al., 1995). Scientific recording of the Total Ozone Column (TOC) was initiated during the early part of the twentieth century (Fabry and Buisson, 1913; Dobson, 1930, 1957, 1968; Dutsch, 1974); more systematic recording of TOC over different latitudinal locations was initiated in 1957, the International Geophysical Year (IGY), by establishing a network of ground-based instruments (the Dobson network). These initiatives were turning point in establishing the thinning of Ozone Layer, whereby it was recognized that “Ozone Hole” was forming every spring over Antarctica (Farman et al., 1985; Jones and Shanklin, 1995). Global concern towards thinning of Ozone layer initiated number of investigations to assess the short-term and long-term trends in TOC abundance over many locations. Initial attempts in analyzing the long-term trend in the TOC contents revealed an increase in its abundance over many locations (Komhyr et al., 1971; Angell and Korshover, 1973; London and Kelley,
* Corresponding author. Tel.: þ91 11 26704309. E-mail address: [email protected] (A.K. Attri). 1352-2310/$ e see front matter Ó 2011 Elsevier Ltd. All rights reserved. doi:10.1016/j.atmosenv.2011.01.008
1974). However, investigations undertaken after the discovery of “Ozone Hole” in 1985 (Farman et al., 1985), involving statistical analysis of TOC data to estimate the long-term trends in the TOC reported a decline: data used in these analysis was post 1970s onwards for many locations (Reinsel et al., 1987, 1994; Bojkov et al., 1990, 1995; Bojkov and Fioletov, 1995; Harris et al., 2003). The analysis included representative data over locations in different climatic zones, but the stress was more on the temperate and Polar Regions where the extent of variability in TOC abundance is large. However, such investigations hold equal importance for the tropical regions housing maximal biodiversity and also receiving solar-UV irradiance much higher than that over latitudes 40 (N or S); even a small increase in UV can have significant impact on the prevalent biodiversity and balance between different species (Vashney and Attri,1995). The case in point is the tropical-subtropical climatic region of India inhabiting a human population in excess of 1 billion (Census of India, 2001). Implication of even a “small” decrease in the Ozone layer thickness over this region, expectedly, will have far reaching impact on agriculture output and food security. The findings reported, on the analysis of TOC time-series data, over several locations in Indian subcontinent to estimate trends in literature (Reinsel et al., 1987, 1994; Bojkov and Fioletov, 1995), are not recent. Investigation reported by Reinsel et al. (2002), covering the latitudinal spread from 30 to 60 , inferred that a positive change in the TOC abundance can be detected within 7e8 years from the time-point of reversal in the declining trend, which was taken as 1996e97. In this regard, Andersen et al. (2006), using the hockey stick
A. Tandon, A.K. Attri / Atmospheric Environment 45 (2011) 1648e1654
method (Mann et al., 1998), reported a positive trend in TOC for both Northern and Southern hemispheres. However, these conclusions were countered by Weatherhead and Andersen (2006). The most recent body of work, specifically estimating the TOC variability over Indian region (Chakrabarty et al., 1998; Sahoo et al., 2005; Ganguly and Iyer, 2006; Pal, 2010) have presented widely varying TOC trend estimates. Analysis of TOC time-series data-sets, in these investigations was done by using simple linear regression, and not accounting for the presence of strong autocorrelation on account of seasonal and inter-annual variations present in TOC values. Reported results, from these studies, manifest large declining TOC trends all over India, errors do not figure in their estimated trends; in short analysis lacked statistical rigor, and estimated trends stood at variance with other studies (Reinsel et al., 1994, 2005; Chipperfield et al., 2007). Estimated trends over matching locations reported by Reinsel et al. (1994) were notsignificant (significant decline was only observed over Varanasi, and Srinagar). Scientific assessment report of ozone depletion, covering different zones, reported no change in TOC trends over equatorial region (0e10 N & S). Small decline in TOC was estimated over region 10e25 (N,S) latitude, though statistically it was found to be in-significant (Chipperfield et al., 2007). Significance of calculated trend is with reference to the spread of error therein; often the trends reported in literature associate 1s error, whereas 2s error is considered more reliable to test the significance (Weatherhead et al., 1998). Additional factor of concern, while estimating TOC trend over a region, requires adequate attention to ascertain the presence of autocorrelation in the datasets used in the estimation of the trends (Weatherhead et al., 2000). In this light, the differences reported in trends between investigations focused specifically over Indian region (Chakrabarty et al., 1998; Sahoo et al., 2005; Ganguly and Iyer, 2006; Pal, 2010), and from those reported in zonal studies (Reinsel et al., 1994, 2002, 2005) stand out in stark variance. Given the differences in the estimated trends from investigations reported over India over many locations (Chakrabarty et al., 1998; Sahoo et al., 2005; Ganguly and Iyer, 2006; Pal, 2010), and that estimated from investigations of zonal TOC trends (Reinsel et al., 1994, 2002, 2005; Chipperfield et al., 2007) we in this manuscript present statistical analysis of two TOC series: (1) Multi Sensor Reanalysis (MSR) TOC monthly mean time-series data-set, Jan 1979eDec 2008, and (2) Total Ozone Mapping Spectrometer Version 8 Overpass monthly mean time-series data-set from Nimbus 7 satellite, (TOMS N7), Jan 1979eDec 1993. The first (MSR) data-set allowed us to do analysis to estimate trends as a function of the size of data-set (Number of data points in series) by punctuating the data-set of the size corresponding to TOMS N7 Overpass (180 month data). Second, the analysis on TOMS Overpass data-set, primarily was done to compare trend estimates presented in this work with the results from investigation focused over many Indian locations. Statistical de-convolution of trends was done in stepwise manner by removing the influence of annual cycles from the data methodology used earlier by Khalil and Rasmussen (1990). TOC trends over whole Indian region from MSR data-set (360 months) are plotted as contours: 0 Ne40 N 67.5 Ee97.5 E latitude/longitude (861 grid points) to provide better assessment of regions over which TOC trends are statistically significant. Significance of calculated trends was ascertained by accounting for the effect of autocorrelation present in respective de-seasonalized TOC data-sets, while estimating 2s error in trend. 2. Details of data used Multi Sensor Reanalysis (MSR) monthly mean total ozone column time-series data-set (resolution of 1 latitude 1.5 longitude) from
1649
Jan 1979 to Dec 2008 (360 months) was accessed from www.temis.nl/ protocols/O3global.html, details of appropriate correction function used and assimilation of multiple satellite data-sets are discussed by van der A et al. (2010). In the present analysis, the data-set for spatial coverage 0 Ne40 N 67.5 Ee97.5 E latitude/longitude (861 grid points) over Indian region was used. The TOMS instrument measures solar irradiance and the radiance backscattered by the Earth’s atmosphere at six wavelengths extending from 310 to 380 nm, Ozone absorption in the Huggins band is used to calculate the TOC abundance (Heath et al., 1975). TOMS Version 8, daily Overpass data-set (spatial resolution of 1 latitude 1.25 longitude) is accessible from National Aeronautics and Space Administration (NASA) at http://toms.gsfc. nasa.gov/n7toms/n7_ovplist_1.html. The monthly mean TOC time-series were generated by taking arithmetic means of daily values for a particular month. The time-series TOMS N7 (TOMS data from Nimbus 7 satellite) and TOMS TOC (l.300) by TOMS EP (from Earth Probe satellite) when the results for the 1979e1993 (n ¼ 180 month) and 1997e2005 (n ¼ 108 month) are discussed, respectively. 3. Methodology used in the analysis of TOC time-series (MSR and TOMS overpass data-sets) The underlying assumption, while subjecting the time-series to mathematical analysis, is that each observation represents the summation of many components (Khalil and Rasmussen, 1990; Gottman, 1981). Monthly mean TOC time-series data-sets (MSR and TOMS N7) are free of daily and weekly variations, but still retain seasonal, inter-annual, and other random influences. The analysis de-convolutes the TOC time-series data in steps to extract all relevant components. The general equation representing TOC time-series can be written as:
TOC ¼ trend þ annual cycle þ residuals
(1)
3.1. Removal of the annual cyclic effects The “trend” in above equation can be further expressed as the sum of the linear trend and inter-annual variations (IAV) present in the TOC time-series (Khalil and Rasmussen, 1990):
trend ¼ linear trend þ inter annual variations
(2)
where, the linear trend in functional form is equal to C0 þ b t; C0 being the intercept and b the slope, and t is the time in the units of months. Estimated value of b represents the linear long-term trend of the time-series and C0 correspond to the fitted value of TOC at the start of the time-series. Determination of b (trend) requires the removal of the seasonal influences present in the form of annual cyclic variation in the monthly mean TOC data, which was done in two steps. Respective TOC data-sets were processed by using thirteen months weight centered moving average filter represented by expression (3).
xj ¼ Cj
k ¼X jþT=2 k ¼ jT=2
wk $Ck ; j
T þ1 2
(3)
In the above expression, T represents the period of annual cycle (12 months). Thus, the first value would be
1650
Uk ¼
A. Tandon, A.K. Attri / Atmospheric Environment 45 (2011) 1648e1654
X 1 M1 x M i ¼ 1 kþTi
(4)
where, k ¼ (i mod 12), i represents the month in the time-series data-set, M stands for number of years. Uk value, one for each respective month (January U1,., December U12), is estimated. The de-seasonalized time-series (DTOCi) is obtained by subtracting from the original TOC time-series (Ci) month’s mean value, the corresponding Uk (annual cycle/seasonal affects).
3.2. Estimation of long-term linear trend in TOC abundance The de-seasonalized data-set DTOCi obtained for MSR and TOMS (N7 and EP), at each location, were subjected to linear regression to obtain C0 (intercept) and b (long-term linear trend) over a given location for the time span of the respective time-series:
Ci0 ¼ C0 þ b t
(5)
Ci0
In the Eq. (5), is the value of ith observation in the DTOC timeseries, and t is the month number for ith observation. The linear trend estimated for each location is expressed with the 2s error (Davis, 2002; Wilks, 2006). Both, TOMS and MSR data-sets, were subjected to same analysis; although MSR data-set’s spread was over whole Indian region [Latitude 0e40 N, and Longitude 67.5 to 97.5 E], whereas TOMS Overpass data-sets were for specific locations only. For comparison the equivalent relevant values from MSR analysis were punctuated for corresponding locations (latitude and longitude) of TOMS data-sets. 3.3. Calculation of 2s error in estimated trends The significance of calculated trends in TOC, for Gaussian distribution, is ascertained by associated standard errors, which is square-root of variance estimated by Mean Square Errors (MSE) as Pn ðCi00 Ci0 Þ2 i¼1 and therefore, 95% confiS2b ¼ PMSE 2 where, MSE ¼ n2 ðti tÞ qffiffiffiffiffi dence limits of b are calculated as b ð1:96 S2b Þ where, ti is the ith month’s number, t is the mean of month numbers, Ci00 is the estimated de-seasonalized TOC value for ith month, Ci0 is the observed de-seasonalized TOC value for ith month, and n is the total number of months in the de-seasonalized TOC time-series. The trends estimates are done on de-seasonalized TOC values (DTOCi), which still retains the inter-annual variations (Eq. (2)). Presence of IAVs in DTOCi data-sets adds persistence in the adjacent values, and can be measured by calculating first order autocorrelation (r1) in DTOCi. In all data-sets (MSR and TOMS N7 and EP) used for trend estimation the r1 was found to be significant (Table 1). Presence of significant r1 would reduce the effective sample size to n0 nN thus affecting the variance and standard error estimates. Variance in this case is estimated by using following corrected value of MSE (Wilks, 2006; Weatherhead et al., 1998; Pn ðCi00 Ci0 Þ2 i¼1 Weatherhead et al., 2000): MSE0 ¼ ð1þr1 1r1Þ, in effect n2 the error estimates done without accounting for IAV’s presence in TOC data-set would underestimate the error, and in turn affect the significance of estimated trend in TOC. Size of the data-set, in view of the reduced number of observations n0 from n will depend on the strength of autocorrelation present in the DTOCi values. Calculated values of autocorrelation Function (ACF) r1, for MSR and TOMS data-sets are given in Table 1 with the reduced number of observations n0 .
Table 1 Calculated values of Autocorrelation function (r1), and reduced number of observations due to the presence of Inter-annual variations associated with de-seasonalized (DTOCi) data-sets (TOMS TS and MSR). Location
ACF (r1) in TOMS TS TOC Data-set
n0 , from original n ¼ 180, for TOMS TS TOC data-set
ACF (r1) for MSR TOC data-set
n0 , from original n ¼ 360, for MSR TOC data-set
Trivandrum Kodaikanal Banglore Madras Hyderabad Poona Bombay Nagpur Dum Dum Ahmedabad Mount Abu Benares Varanasi New Delhi Srinagar
0.76 0.68 0.69 0.61 0.69 0.72 0.75 0.73 0.65 0.75 0.73 0.71 0.70 0.68 0.55
25 34 33 44 33 29 26 28 38 26 28 31 32 34 52
0.74 0.69 0.61 0.63 0.67 0.70 0.71 0.69 0.63 0.68 0.69 0.68 0.68 0.69 0.56
54 66 87 82 71 64 61 66 82 69 66 69 69 66 102
4. Results Applicability of statistical analysis methodology to analyze TOC time-series, for extracting trends and errors therein, requires prior knowledge of data’s characteristic frequency distribution (Wilks, 2006). This was ascertained by examining the parameters like mean, median, mode, range, inter quartile range and other related features of the distribution: kurtosis, skewness and variance (Storch and Zwiers, 1999; Davis, 2002; Wilks, 2006). Data-sets used in analysis manifested close to normal distribution of the monthly mean TOC values (representative Supplementary Fig. S1), which was also confirmed from ShapiroeWilk’s test to confirm the extent of normality present (Hair et al., 2006). The results are given in Supplementary Table S1. 4.1. Estimation of long-term trend in the total ozone column Calculated estimates for long-term trends in percent/decade with corresponding 2s error for MSR data-set (1979e2008) and TOMS N7 Overpass data-set (1979e1993) are given in Table 2 for 15 locations in Indian region. For comparing the trend estimates with those reported in earlier India specific-investigations, estimates were also done on TOMS EP Overpass EP data-set. Analysis involving three different data-sets of varying sizes allowed us to unambiguously determine as to how the data-size affected the statistical significance of estimated trends. Trends extracted from longer MSR data-set manifested statistically significant declining trends [() 0.8 e () 1.5 percent/decade] over Indian region above 25 N latitude; this included locations viz. Varanasi (25 270 N), New Delhi (28 400 N), and Srinagar (34 040 N). In case of TOMS N7 the declining TOC trends were significant over New Delhi (28 400 N) and Srinagar (34 040 N) having decline of () 2.5 and () 3.6 percent/decade respectively. It is apparent that the estimated trends at all locations for TOMS EP data-set were not found to be significant, as the calculated 2s errors were large. Statistical significance of the calculated TOC trends in view of the presence of 2s error as a function of the size (n ¼ 108, 180, and 360) of the MSR TOC data-set is shown in Figs. 1e3 as contour plots (0 Ne40 N 67.5 Ee97.5 E). These plots provide better grasp of the extent of region (area) of India where significant decline in TOC is estimated. Trend contours for segmented MSR data-set (1997e2005) size 108 month are plotted in Fig. 1. As observed, for equivalent TOMS EP Overpass data-set [Table 2], all over India the estimated trends
A. Tandon, A.K. Attri / Atmospheric Environment 45 (2011) 1648e1654
1651
Table 2 Long-term trend estimates (% per decade) for TOMS N7 & EP, and MSR de-seasonalized monthly mean TOC time-series data-sets; and the trend estimates reported by Reinsel et al. (1994) and Harris et al. 1995. Trends in shaded cells are statistically significant at 95% confidence limit. Areas of the region above which the estimated trends (decline) are statistically significant constitute w40% of total area of India. Location
Latitude
TOMS N7 (Jan 1979eDec 1993) n ¼ 180 month
TOMS EP (Jan 1997eDec 2005) n ¼ 108 month
MSR (Jan 1979eDec 2008) n ¼ 360 month
Trivandrum Kodaikanal Banglore Madras Hyderabad Poona Bombay Nagpur Dum Dum Ahmedabad Mount Abu Benares Varanasi New Delhi Srinagar
8 280 N 10 130 N 12 580 N 13 040 N 17 220 N 17 220 N 19 070 N 21 060 N 22 380 N 23 010 N 24 360 N 25 000 N 25 270 N 28 400 N 34 040 N
0.66 0.87 0.94 1.26 1.26 1.32 1.43 1.39 1.42 1.83 2.05 1.95 1.99 2.52 3.60
2.36 1.91 1.35 1.17 0.84 0.65 0.48 1.07 1.03 0.84 1.19 1.24 1.37 1.76 3.25
0.15 0.67 0.10 0.55 0.03 0.52 0.01 0.53 0.01 0.62 0.18 0.70 0.23 0.71 0.30 0.68 0.37 0.62 0.61 0.72 0.78 0.73 0.66 0.76 0.81 0.76 1.14 0.92 1.49 0.85
a
1.99 1.50 1.60 1.52 1.77 1.90 2.11 2.12 1.89 2.30 2.34 2.15 2.16 2.41 2.53
3.73 2.59 2.36 1.82 3.48 3.53 3.79 3.97 3.53 3.64 3.60 3.45 3.75 4.39 5.49
Reinsel et al. (1994) (Jan 1978eDec 1991) n ¼ 16 monthsa
Harris et al. (1995) (Jan 1979eFeb 1994) n ¼ 182 months
1.95 1.92
0.2 2.1
1.04 1.80
0.95 4.20
1.5 1.7
2.92 1.68 1.17 2.32 4.93 3.48
1.2 1.5 1.1 1.9
In Reinsel et al. (1994), 1s errors associated with long-term trends were reported, for comparison their errors, in above Table, are multiplied by 2 (i.e., 2s).
were statistically not-significant. Appearance of contour lines in Fig.1, compared with Fig. 2 [MSR data-set (1979e1993; n ¼ 180 month)] and Fig. 3 [MSR data-set (1979e2008; n ¼ 360 month)] seems haphazard over whole region. Fig. 2 [MSR data-set; 1979e1993], demarcates the plotted contour of TOC trends where decline is statistically significant. The same, in Fig. 3, for longer MSR TOC
data-set (360 month) show the additional Indian region where statistically decline in trends in TOC is significant. It is interesting to note that the quantum of decline in TOC trend (statistically significant) over Indian region is smaller in case of MSR TOC time-series data-set spanning from Jan 1979eDec 2008 than MSR TOC timeseries data-set spanning from Jan 1979eDec 1993 (Figs. 2 and 3).
Fig. 1. Contour plots of TOC trends (% per decade) over India for segmented MSR data-set, 1997e2005, (108 month). For this size data-set the calculated 2s error makes the trend estimates statistically not-significant. Consequence of data-set size is important in determining the significance of estimated trends.
1652
A. Tandon, A.K. Attri / Atmospheric Environment 45 (2011) 1648e1654
Fig. 2. Contour plots of TOC trends (% per decade) for segmented MSR data-set (180 months). Decline in TOC trends was statistically significant over New Delhi (28 400 N), and Srinagar (34 040 N). Geographical area of India, over which the TOC decline was significant is about 30% of the total area; horizontal thick grey line surrounded with two thick vertical arrows marks the region where trends in TOC decline is significant.
5. Discussions 5.1. Statistical significance of long-term trends in TOC over India Perhaps, while estimating the long-term trends in TOC over a location, statistical significance is the most important in determining the validity of the result. In this regard the calculation of error in estimated trend becomes crucial. This aspect has been, lucidly covered by Weatherhead et al. (1998) and Weatherhead et al., (2000). Significance of trend is judged on the basis of the magnitude of the 2s error (95% confidence). For comparison of results obtained in this study (TOMS N7; 1979e1993), we have compared the estimated trends with that of Reinsel et al. (1994) and Harris et al. (1995) for common locations (Srinagar, New Delhi, Varanasi, Ahemdabad, Poona, and Kodaikanal); their trend estimates with calculated standard error therein are tabulated in Table 2 along with the trends estimated in present work. The data-set reported in Reinsel et al. (1994) and Harris et al. (1995) was from ground-based Dobson spectrometer values; their values are comparable with that reported in the present work with slight difference. This difference most probably is due to the Dobson spectrometer based data-set used in their studies; whereas the TOMS Overpass data-set is satellite based, and MSR data-set is a re-analyzed TOC data-set obtained from multiple sensors, and accounted for factors affecting and likely to introduce variability in TOC data procured from different sources (Heath et al., 1975; van der A et al., 2010). The 2s errors associated with the estimated trends over few locations for TOMS N7 data-set in
present study are higher than the equivalent errors reported by Reinsel et al. (1994) and Harris et al. (1995). This can easily be attributed to the removal of ozone variability and extraction of the anthropogenic trend component in Reinsel et al. (1994) and Harris et al. (1995); in their analysis a multi-linear model was used, where Solar and QBO effects were accounted for. Consequently, the reported trend values would also be slightly different. 5.2. Data-set’s size and error in the estimated TOC trend The size of TOC data-set (number of months) does show its influence on, both, estimated trends and calculated error therein. Errors in the estimated trends decrease with the increase in the data-set’s size. This can be seen from trend estimates presented in this work and those reported by Reinsel et al. (1994) [Table 2]. This aspect has also been reported earlier (Weatherhead et al., 1998; Weatherhead et al., 2000). The affect of TOC data-set’s size on providing statistically more acceptable trend pattern over India can be perceived from trend contours plotted for segmented MSR data-sets (n ¼ 108, 180 months) in Figs. 1 and 2, and complete MSR data-set (n ¼ 360 month) in Fig. 3; in later case the contour trends show regular change, though the statistically significant decline is confined to over a region above 25 N latitude. These results agree well with the observations made from zonal trend estimates over tropics in WMO report, 2006 (Chipperfield et al., 2007). It is equally relevant to state that the area of the Indian region, where trends are estimated as significant, covers almost w40% area of
A. Tandon, A.K. Attri / Atmospheric Environment 45 (2011) 1648e1654
1653
Fig. 3. Contours plots of TOC trends (% per decade) for MSR data-set (360 month). Decline in TOC trends was statistically significant (2s confidence limit) over Varanasi (0.81 0.76), New Delhi (1.14 0.92 percent per decade), and Srinagar (1.49 0.85 percent per decade). For this size data-set the area over which the decline was significant is about 40% of the total geographical area of India; horizontal thick grey line surrounded with two thick vertical arrows demarcates the region where trends in TOC decline are significant.
India. Interestingly this region lies in the Northern part of the country (Himalayan and Sub-Himalayan region), encompassing Indo-Gangetic planes holding importance for agriculture produce and food security of the country. For MSR data-set, comparable to TOMS N7 (1979e1993; n ¼ 180 months) estimated declining trends are statistically significant over a region above 28 N latitude; in terms of area 30% of India’s geographical area. For dataset of size n ¼ 108 month, over no location the estimated trends were found to be significant. The size of the data-set used in estimating the trends have bearing on the calculated errors. Lastly, it can be stated that decline in TOC over large geographical region (40% of total area) of India is evident; though the decline is small, but given the much large insolation of solar-UV in this region, the expected increase in UV reaching the ground will hardly be trivial.
data-set (1979e2008). Estimated decline, respectively, for New Delhi (28 400 N) and Srinagar (34 040 N) were () 2.5 and () 3.6 percent/ decade also observed for TOMS N7 data-set (1979e1993).
6. Conclusions
Appendix. Supplementary data
In MSR TOC data-set (1979e2008), statistically significant declining trends ranging from () 0.8 e () 1.5 percent/decade were observed over Indian region above 25 N latitude. The analysis also show that the quantum of decline between MSR data-set (1979e2008) was smaller when compared with that estimated from MSR data-set (1979e1993); at the same time the statistically significant decline in TOC covered much larger Indian region for MSR
Supplementary data associated with this article can be found in the online version, at doi:10.1016/j.atmosenv.2011.01.008.
Acknowledgements Authors acknowledge Goddard Space and Flight Centre, NASA for providing TOMS TOC data-sets and Tropospheric Emission Monitoring Internet Service (TEMIS) for MSR TOC data-sets used in present study. The financial assistance provided by CSIR (India) in the form of a research project to AKA is acknowledged. One of the authors, AT, acknowledges UGC (India) and DST (India) for their financial support in the form of fellowships and award of Research Project for Young Scientists. Authors extend their gratitude to two anonymous referees of this manuscript for their in-depth review; their suggestions and queries have resulted in much improved manuscript.
References Andersen, S.B., et al., 2006. Comparison of recent modeled and observed trends in total column ozone. J. Geophys. Res. 111, D02303. doi:10.1029/2005JD006091.
1654
A. Tandon, A.K. Attri / Atmospheric Environment 45 (2011) 1648e1654
Angell, J.K., Korshover, J., 1973. Quasi-biennial and long-term fluctuations in total ozone. Monthly Weather Rev. 101, 426e443. Bojkov, R.D., Fioletov, V.E., 1995. Estimating the global ozone characteristics during the last 30 years. J. Geophys. Res. 100 (D-8), 16,537e16,551. Bojkov, R.D., Bishop, L., Fioletov, V.E., 1995. Total ozone trends from quality-controlled ground-based data (1964e1994). J. Geophys. Res. 100 (D-12), 25,867e25,876. Bojkov, R.D., Bishop, L., Hill, W.J., Reinsel, G.C., Tiao, G.C., 1990. A statistical trend analysis of revised Dobson total ozone data over the northern hemisphere. J. Geophys. Res. 95 (D-7), 9785e9807. Census India, 2001. Available at: http://www.censusindia.gov.in/. Chakrabarty, D., Peshin, S., Pandya, K., Shah, N., 1998. Long-term trend of ozone column over the Indian region. J. Geophys. Res. 103 (D-15), 19245e19251. Chipperfield, M.P., Fioletov, V.E., et al., 2007. Global Ozone: Past and Present, Chapter 3 in Scientific Assessment of Ozone Depletion: 2006. Global Ozone Research and Monitoring ProjectdReport No. 50. World Meteorological Organization, Geneva, Switzerland. Cockell, C.S., Blaustein, A.R., 2001. Ecosystems, Evolution and Ultraviolet Radiation. SpringerVerlag, New York, USA. Davis, J.C., 2002. Statistics and Data Analysis in Geology. Jhon Wiley & Sons, New York, USA. Dobson, G.M.B., 1930. Observations of the amount of ozone in the Earth’s atmosphere and its relation to other geophysical conditions. Proc. R. Soc. London, Ser. A. 129, 411. Dobson, G.M.B., 1957. Observers handbook for the ozone spectrophotometer. In: Annals of the International Geophysical Year V, Part 1. Pergamon Press, New York, USA, pp. 46e89. Dobson, G.M.B., 1968. Forty years’ research on atmospheric ozone at Oxford: a history. Appl. Opt. 7, 387e405. Dutsch, H.U., 1974. The ozone distribution in the atmosphere. Can. J. Chem. 52, 1491e1504. Fabry, C., Buisson, M., 1913. L’absorption de l’ultraviolet par l’ozone et la limite du spectre solaire. J. Phys. 3, 196e206. Farman, J.C., Gardiner, B.G., Shanklin, J.D., 1985. Large losses of total ozone in Antarctica reveal seasonal ClOx/NOx interaction. Nature 315, 207e210. Ganguly, N.D., Iyer, K.N., 2006. Long-term trend in ozone and erythemal UV at Indian latitudes. J. Atmos. Chem. 55, 227e239. Gottman, J.M., 1981. Time Series Analysis. Cambridge University Press, Cambridge, U.K. Hair, J.F., Black, B., Babin, B., Anderson, R.E., Tantham, R.L., 2006. Multivariate Data Analysis. Pearson Prentice Hall, London, UK. Harris, J.M., Oltmans, S.J., Bodeker, G.E., Stolarski, R., Evans, R.D., Quincy, D.M., 2003. Long-term variations in total ozone derived from Dobson and satellite data. Atmos. Environ. 37, 3167e3175. Harris, N.R.P., et al., 1995. Observed changes in ozone and source gases: total and vertical-column ozone, chapter 1. Global Ozone Research and Monitoring ProjectdReport No. 37. In: Scientific Assessment of Ozone Depletion: 1994. World Meteorological Organization, Geneva, Switzerland. Heath, D.F., Krueger, A.J., Roeder, H.A., Henderson, B.H., 1975. The solar Backscatter ultraviolet and total ozone mapping spectrometer (SBUV/TOMS) for Nimbus G. Opt. Eng. 14, 323e331.
Jones, A.E., Shanklin, J.D., 1995. Continued decline of total ozone over Halley, Antarctica, since 1985. Nature 376, 409e411. Khalil, M.A.K., Rasmussen, R.A., 1990. Atmospheric methane: recent global trends. Environ. Sci. Technol. 24 (4), 549e553. Komhyr, W.D., Barrett, E.W., Slocum, G., Weickmann, H.K., 1971. Atmospheric total ozone increase during the 1960s. Nature 232, 390e391. London, J., Kelley, J., 1974. Global trends in total atmospheric ozone. Science 184 (4140), 987e989. Mann, M.E., Bradley, R.S., Hughes, N.K., 1998. Global scale temperature climate forcing over the past six centuries. Nature 392, 777e787. Pal, C., 2010. Variability of total ozone over India and its adjoining regions during 1997e2008. Atmos. Environ. 44, 1927e1936. Reinsel, G.C., Tiao, G.C., Miller, A.J., Wuebbles, D.J., Connell, P.S., Mateer, C.L., Deluisi, J.J., 1987. Statistical analysis of total ozone and stratospheric Umkehr data for trends and solar cycle relationship. J. Geophys. Res. 92 (D-2), 2201e2209. Reinsel, G.C., Tiao, G.C., Wuebbles, D.J., Kerr, J.B., Miller, A.J., Nagatani, R.M., Bishop, L., Ying, L.H., 1994. Seasonal trend analysis of published ground-based and TOMS total ozone data through 1991. J. Geophys. Res. 99 (D-3), 5449e5464. Reinsel, G.C., Weatherhead, E.C., Tiao, G.C., Miller, A.J., Nagatani, R.M., Wuebbles, D.J., Flynn, L.E., 2002. On detection of turnaround and recovery in trend for ozone. J. Geophys. Res. 107 (D-10), 4078. Reinsel, G.C., Miller, A.J., Weatherhead, E.C., Flynn, L.E., Nagatani, R.M., Tiao, G.C., Wuebbles, D.J., 2005. Trend analysis of total ozone for turn around and dynamical contributions. J. Geophys. Res. 110. doi:10.1029/204JD004662. Sahoo, A., Sarkar, S., Singh, R.P., Kafatos, M., Summers, M.E., 2005. Declining trend of total ozone column over the northern parts of India. Int J Remote Sens. 26 (16), 3433e3440. Shen, T.L., Wooldridge, P.J., Molina, M.J., 1995. Stratospheric pollution and ozone depletion. In: Singh, H.B. (Ed.), Composition, Chemistry, and Climate of the Atmosphere. Van Nostrand Reinhold, New York, USA. Storch, H.V., Zwiers, F.W., 1999. Statistical Analysis in Climate Research. Cambridge University Press, Cambridge, UK. van der A, R.J., Allaart, M.A.F., Eskes, H.J., 2010. Multi sensor reanalysis of total ozone. Atmos. Chem. Phys. Discuss. 10, 11401e11448. Vashney, C.K., Attri, A.K., 1995. In: Majumdar, S.K., Brenner, F.J., Miller, E.W., Rosenfield, L.M. (Eds.), Impact of Stratospheric Ozone Depletion on Terrestrial Ecosystem, in Environmental Contaminants and Health. Pennsylvania Academy of Sciences, USA. Weatherhead, E.C., et al., 1998. Factors affecting the detection of trends: statistical considerations and applications to environmental data. J. Geophys. Res. 103, 17,149e17,161. Weatherhead, E.C., et al., 2000. Detecting the recovery of total column ozone. J. Geophys. Res. 105, 22,201e22,210. Weatherhead, E.C., Andersen, S.B., 2006. The search for signs of recovery of the ozone layer. Nature 441, 39e45. Wilks, D.S., 2006. Statistical Methods in the Atmospheric Sciences. Academic PressElsevier, San Diego, USA.