Progress

Presentations

• ACTRIS meeting in Rennes 13-16 May 2024
  Presented first results in the session on Long-term trends and variability of the atmospheric composition :
  ACTRIS_Rennes_2024.pptx 
• DORA midterm evaluation 10 June 2024
  Presented results, progress, and prospects of the project in front of the committee of DORA
  DORA_midterm.pptx
   
• NDACC/IRWG-TCCON-COCCON annual meeting 8-12 July 2024
  Presented results and progress of the DORA project in the NDACC/IRWG session
  NDACC_IRWG_2024.pptx
   
• Quadrennial Ozone Symposium in Boulder, Colorado 14-19 July 2024
  Accepted talk in the session on Stratospheric Ozone Science
  QOS_Boulder.pptx  
  
  
• European Geosciences Union General Assembly in Vienna, Austria 28 April-2 May 2025 
  Accepted PICO presentation in the session on Atmospheric composition variability and trends 
  EGU2025_PICO
  
  
• BE-Polar Conference in Brussels, 11th September 2025  
  Accepted Talk in the Atmospheric Science session
  BE_Polar_Talk
  
  
• NDACC 2025, Virginia Beach, VA , USA, 27-30 October 2025
  Online Poster
  P_E03_Jonas.pdf
  

Preprint with results

A preprint article containing all the results of this project is available at EGU Preprint DORA. Below we report on the main results obtained.

Project results

Sodankyla ozone time series

In the Arctic, we have 6 FTIR stations measuring with the NDACC specifications, however we have 1 additional FTIR available from TCCON at Sodankyla. Measurements are made in a different spectral region (3040 cm-1) so the retrieval strategy needs to be adapted. The used retrieval strategy is based on the strategy by Zhou et al. (2020) in Xianghe, China and García et al. (2014) in Izaña, Spain.

Using this, we are able to provide total column data and 2 partial columns (DOFS is around 2.5) from 2012 onwards of a completely new ozone time series product.

Transmission signal in the spectral windows around 3040 cm-1 used for the retrieval of ozone from the TCCON FTIR instrument in Sodankylä, together with the obtained altitude profile.

Comparison with satellites: evaluation of ground-based datasets and of satellite's drifts.

 We used two satellites datasets, IASI-CDR (2007-2024) obtained from the AERIS portal and MEGRIDOP (2001-2024) from the Ozone_cci project, see also Keppens et al, 2025 and Sofieva et al, 2021. We considered monthly mean gridded values for all stratospheric columns for MEGRIDOP and for all columns for IASI-CDR.

In all those columns, we computed the drift of those two satellites datasets with respect to each ground-based dataset. The drift is the linear trend of the difference between the two datasets. Here, we base the difference on the anomalies of each dataset. The anomalies are obtained by subtracting and dividing the ozone value by the mean ozone value of the corresponding month.

We obtain the following results:

• The ozonesonde located in Scoresbysund presents a strong negative drift with respect to both satellite datasets in all its columns. This drift is most probably associated with ample instrumental changes around 2016 that cannot be fully corrected for. We therefore excluded this time series in our trend analysis.
• We find hints of a potential jump in the Resolute ozonesonde dataset before 2005, possibly due to instrumentation changes. We restricted this time series to 2005−2024 for the trend analysis.
• By comparing satellites datasets to the mean of all Arctic ground-based datasets (after excluding the above problematic datasets), we find that:
  • IASI-CDR and MEGRIDOP's drifts in all stratospheric columns are always smaller than |3%/decade|.
  • IASI-CDR presents a significant negative drift in the troposphere of (−3.47±2.35)%/decade. However, we have not accounted for the low sensitivity of IASI in the high-latitude troposphere.
  • MEGRIDOP shows significant positive drifts for the zonal band in the lower, upper and total stratosphere of (2.76±1.48), (1.97±0.97) and (0.87±0.82)%/decade respectively.

IASI-CDR's drift with respect to all ground-based datasets in the mid stratosphere column. Scoresbysund's sonde drift clearly stands out.MEGRIDOP's drift with respect to all ground-based datasets in the mid stratosphere column. Drifts with respect to FTIR are in blue and those wrt ozonesondes are in red.

Drift of IASI wrt the mean of all ground-based datasets in the Arctic for all columns.Drift of MEGRIDOP wrt the mean of all ground-based datasets in the Arctic for all stratospheric columns.

Representativeness study

Within DORA we analyzed the representativeness of the ozone measurements for the Arctic. An obvious limitation of ground-based observations of ozone trends is the representativeness of these sparse data:

• Are the detected trends at a specific site representative for ozone behavior at a regional scale?
• How does the network of ground-based stations improve the representativeness of ozone trends in the Arctic, as compared to a single site?

For the sites that are found representative of the same area in the Arctic, we will construct merged time series of ozone anomalies and study the impact on the detection of the trends (e.g., reduced uncertainties due to higher sampling), by applying our MLR model on the anomalies.

To tackle these questions of representativeness we use model reanalysis data from CAMS (Copernicus Atmosphere Monitoring Service) which supply ozone data every 3 hours from 2003 to 2024. The CAMS ozone data combines model data and observations to give gridded monthly means time series (0.75°x0.75°) for 25 different pressure levels and is taken from https://doi.org/10.24381/d58bbf47.

For the spatial representativeness, we follow the methodology of Weatherhead et al. (2017), who were using satellite data instead of reanalysis. Here we calculate the correlation of each station with the other spatial points within the Arctic. This is done based on the anomalies calculated in each partial and total column.

1. The first step is to convert the provided mass-mixing ratios of CAMS to column densities (in DU) using N_A/M_air/g*r_m*Δp/DU
   
   (where N_A is Avogadro’s constant, M_air is molecular mass of air, g is the gravitational acceleration, r_m is mass mixing ratio of ozone, Δp is pressure difference between layer boundaries, and DU is a conversion number 2.6867e20 molec/m²).
   
   
2. In a second step the vertical profiles of CAMS are integrated into partial columns which approximate the columns as calculated for the ground-based measurements by summing over the column densities of the levels within these partial column layers. These are defined in pressure levels from 1000-300hPa for the tropospheric column, 300-100hPa, 100-20hPa, and 20-1hPa for the stratospheric columns.
   
   
3. The third step is to calculate the anomaly from the ozone timeseries by subtracting from each column value the mean value of its corresponding month in the year:   

                                      anomaly^{month i, year j} = ozone^{month i, year j} – 1/22·∑_{k=[2003,2024]} ozone^{month i, year k} 

1. Lastly, we iterate over each point of the CAMS grid to compare the anomalies of that point to the anomalies at the location of one of the measurement sites. This gives us a map of the Pearson correlation coefficient at each point of the CAMS grid which is different for each station. We also report correlations between each station in tables, helping the visual identification of 

                                                                                          Pearson coefficient of the anomalies correlations between the Eureka station and all the rest of the Arctic for the total column of ozone. The total column is highly correlated (r>0.95) between Eureka and the 3 other Canadian Arctic stations (Resolute, Alert, Thule).

                                                                                Correlation table with Pearson coefficient between each stations for the total column of ozone. Correlations above (resp. below) 0.8 are in black (resp. white). Groups of stations are determined for each column based on these tables.

Merging datasets

The CAMS correlation analysis is used to decide which ground-based measurements to merge into one merged dataset. We decide to merge stations if the Pearson correlation coefficient exceeds a value of 0.8.

This allows us to define several regions within the Arctic which are highly correlated and thus eligible for merging. These regions differ by columns and are summarized in the table below:

The merging is performed by calculating the weighted mean of the merged datasets, anom_w-m, with a weight corresponding to the inverse uncertainties squared (we only use random uncertainties because we consider anomalies for which systematics errors are irrelevant):

with ω_α = 1/(Δanom_α)² . 

Anomalies time series (blue) and their weighted mean (red) for the Canada group in the total column .

Trend analysis

The trends were calculated using a Multiple-Linear Regression method including proxies to capture the natural variability of ozone. We used a stepwise regression procedure which decides for each trend the most relevant set of proxies to use to avoid overfitting, while maximizing how well the model fits to the data (i.e., the coefficient of determination, R²). The stepwise procedure also enables to include proxies with high correlation without worry. If two highly correlated proxies are used simultaneously, we have verified that their combined use improved the model fit (higher R² together with lower trend uncertainties). The processes that we considered to account for the natural variability and their corresponding proxies used are listed here:

1. Solar cycle: Bremen composite Mg II index
2. Quasi-Biennal Oscillation (QBO): Principal components zonal mean wind 6◦S−6◦N (ERA5)
3. El-Niño Southern Oscillation (ENSO): Multivariate ENSO Index (MEI)
4. AO: Arctic Oscillation Index
5. Brewer-Dobson Circulation (BDC): Eliassen-Palm (EP) Flux at 100hPa
6. VPSC: Volume of Polar Stratospheric Clouds (not used in the upper stratosphere)
7. TP: Tropopause Pressure at each station location (ERA5)
8. EL: For each of the three stratospheric columns: Local Equivalent Latitude averaged over the corresponding column (ERA5)
9. T: For each of the three stratospheric columns: Local Temperature averaged over the corresponding column (ERA5) 

The proxies contributing to each trend are sensible and consistent throughout regions, comforting our trust in the regression. However,some of those proxies themselve exhibit a trend. For instance, the volume of polar stratospheric clouds (VPSC) presents a positive trend over the 2000-2020 period (von der Gathen 2021, Pazmiño 2023). This trend is due to climate change but leads to an increase of ozone depletion. When we include the VPSC proxy, this means that the increased ozone depletion related to the VPSC increase is not reported in our trend results. This project aims to assess the ozone recovery due to the decrease of ozone depleting substances, and therefore our final results include the VPSC proxy with its trend. We have however analysed the impact of detrending the VPSC proxy and found a mean difference of -0.8%/decade with respect to the non-detrended proxy. This leads to non-significant negative trends in the lower and mid stratosphere (8-26 km). Other proxies such as the temperature, equivalent latitude or arctic oscillation also possess trends but the detrending impact is even smaller on average.

Now we present our trend results: Annual trends account for the full ozone anomalies time-series, while seasonal trends account respectively for the months of December-January-February (Winter), March-April-May (Spring), June-July-August (Summer) and September-October-November (Autumn). If there are lessthan 80 datapoints in the anomalies annual time series or less than 25 datapoints on the seasonal timeseries, we do not calculate the trend to avoid non-representative trend results.This is why is Winter, regions within the polar night where only FTIR (dependent on sunlight) give measurements may not have a trend, as in the North Pole region in the upper stratosphere or in the Ny-Ålesund region in the total column.

1. Total column of ozone:

Total column ozone trends are mostly positive, with many positive significant trends especially in Spring: this is a strong signal of ozone recovery in the Arctic. These results are in qualitative and quantitative agreement with other recent studies on total column ozone in the Arctic as Bernet 2023 and Anjali 2025. The clear added value of this project concerns the representativeness study which enables a finer description than latitudinal bands but also more general than individual stations.

2. Stratospheric columns:

 The lower stratosphere (8-17 km/300-100hPa) presents strong seasonal and regional variations of the trends, but most trends are non-significant and very small annually. Only Lapland trends are significant, positive in Autumn and negative in Winter. Lower stratospheric trends are considerably smaller or more negative when detrending the VPSC, meaning that the positive VPSC trend is delaying the expected recovery of ozone there. 

The mid stratosphere exhibits positive (but non-significant) Spring trends everywhere in Arctic. In Canada, trends are positive significant annually and in Winter. The VPSC detrending also lower mid stratospheric trends, clearly delaying ozone recovery. We further observe a clear seasonal cycle with highest positive trends in Spring and lowest (sometimes negative) trends in Summer.

In the upper stratosphere, the annual trend above North Pole (Canada +Ny-Ålesund) is significant and positive,supporting the ozone recovery detection seen in the total column. Seasonal trends are allnon-significant, and therefore we have also considered  a higher altitude layer, from 32-48 km, where we find larger positive and significant trends inSpring and annually for the North Pole region.

The mid and upper stratosphere both exhibit a strong zonal asymmetry which is known in the literature and attributed to decadal changes in the dynamics of the polar vortex above the Arctic (see Arosio 2024).

3. Tropospheric column:

The final trend results concern the tropospheric column of ozone between 0-8km (surf.-300hPa). In the North Pole, all seasonal and the annual tropospheric trends are negative, and even significant (and larger) in Spring and annually. In Scandinavia, trends are non-significant and follow a seasonal cycle, positive in Autumn-Winter and negative in Spring-Summer. Our trends are smaller than those found recently in Van Malderen 2025 for the 2000-2022 period, especially for the Scandinavian region. This is explained by the addition of three years in part, but mostly by our exclusion of the Scoresbysund ozonesonde datasets, whose false negative drift was pulling the trends toward more negative values.

Finally, we compare the total column trends to the sums of partial column trends to evaluate how well the different instruments used for the different columns agree with one another. The impact of tropospheric trends in DU per decade is negligible compared to the total trend budget in DU per decade (the green bands corresponding to 0-8km are very small). Within 2σ errors, there is overall agreement between the total column trends and the sums of partial columns trends, with a better agreement in Spring and an extremely good agreement for Ny-Ålesund throughout the seaons, despite the change of instruments between the different columns.

Explicit trend numbers in percent per decade and DU per decade, as well as more details on the regression model used and a complete analysis of the proxies is available in the paper preprint here.

Questions and comments are highly welcomed and can be addressed to This email address is being protected from spambots. You need JavaScript enabled to view it. .