This dataset is no longer supported and will be removed from the Climate Data Store (CDS) at a later date. Documentation below is provided as is.
Contributors: K. Lemmens (Offshore Navigation Ltd), L. Eriksson (Chalmers University), L. Jonasson (Chalmers University), W. Mao (Chalmers University)
Issued by: OSM / Sverre Dokken
Issued Date: 01/09/2018
Ref: C3S_D422Lot1.OSM.2.2_201809_Scientific_Indicators_(Ice)_Technical_Note_v1
Official reference number service contract: 2018/C3S_D422_Lot1_OSM/SC2
Introduction
This document describes scientific indicators in the C32-442 service related to sea ice. Three indicators are documented: Route availability index, Sea ice concentration along Arctic ship routes and Sea ice thickness along Arctic ship routes. Additionally, data requirements and initial validation for two of the indicators are presented.
Two of the scientific indicators presented in this report is meant to be used as input to the operational indicators related to cost and safety of Arctic shipping. The third product (Route availability index) can be seen as an operational indicator in the service.
Data description
Data from multiple sources are needed to produce the sea ice scientific indicators. Reanalysis data for sea ice concentration are provided by ERA-interim to validate and construct the Route Availability index and Sea ice concentration along Arctic ship routes. Historical data for sea ice thickness is provided by the CDS product from the Envisat and CryoSat-2 missions.
Information about future Arctic sea ice concentration and thickness are collected from the CMIP5 climate projections. The most skillful models with respect to reconstruction of historical sea ice area in the Arctic are selected. Mean and spread of this model ensemble subset is then calculated and used as input data for the indicators. The Route Availability index requires daily CMIP5 data while the Sea ice concentration along Arctic ship routes and Sea ice thickness along Arctic ship routes uses monthly mean data.
Arctic ship routes
Two different ship routes have been defined and digitized into the standard route format of the project. Both routes utilize the Northeast Passage at two different latitudes. At present, none of the indicators are defined along the Northwest Passage and no standard routes have been created. The reason for this being that model resolution in both ERA-interim and the CMIP5 ensemble are to coarse to resolve the narrow straits east of Greenland.
The two Northeast Passage routes are shown in the figure below. The routes have been defined in [Mulherin et. al. 1999] and are created by using historical voyages together with a route optimization model based on ice and weather information. The original routes have been altered in a few locations where the route points were positioned on ERA-interim land points. In these situations, the route points have been moved away from the coast as much as required to be covered by a wet grid cell in ERA-interim.
Figure 1: Northerly and southerly Northeast Passage
Route Availability Index
The Route Availability Index calculates the transit window in days for a Northeast Passage route. The algorithm follows the steps presented in Khon et al., 2016 and utilized the same subset of CMIP5 models. The selected CMIP5 models have been chosen based on their skill in predicting historical transit windows.
An example of the indicator product is shown in the figure below. In this case the calculations are done on the southern Northeast Passage route. The criteria for an open transit is determined using sea ice concentration threshold for ice-free conditions and the percentage of the daily length of ice free conditions with respect to the full route.
Figure 2: Opening and closing dates of the Northeast Passage. Gray and green shaded areas indicate dates calculated from ERA-interim and CMIP5 RCP4.5 scenarios, respectively. The dashed line shows the one standard deviation spread of the CMIP5 models.
Step-by-step description of the algorithm
- Define a route for the North East Passage
- Interpolate sea ice concentration to the route for each ensemble member
- Take ensemble average for each route grid point
- For each time step:
- calculate the sea ice concentration in grid points that covers parts of the route.
- Apply sea ice concentration threshold and determine if grid point is ice covered
- Get the percentage of ice free grid points
- Apply transit window threshold and determine if route is open or closed for this time step
- Summarize open dates for each year
Validation
The algorithm is validated in Khon et al., 2016 and only verification of the implementation are required.
Implementation updates
Due to the lack of daily CMIP5 results in the CDS database the background data had to be changed to monthly mean data. The algorithm described above for the Route availability index determines on a daily basis if the North East Passage is open or close. Using this technique on monthly data is possible but will obviously reduce the temporal resolution. To present the indicator like in Figure 2 and in Kohn et. al., 2016 the monthly data has been up-sampled to daily data through linear interpolation. On one hand this produces a smoother visualization but on the other hand it also a somewhat misleading representation of the underlying data. However, it was decided to up-sample the data as long as the general trend and average dates did not change significantly.
The CMIP-5 model selection was also modified during the implementation phase since many of the realizations used during the development phase is not available in the CDS. The new subset of models is described in Section 4.4.
Sea ice concentration along an Arctic route
This indicator describes the projected sea ice concentration along a Northeast Passage route. The route is divided into 8 sections and in each route section the average sea ice concentration is calculated. The figure below shows the average April to June sea ice concentration in section 4. The thin lines represent each ensemble member, the thick orange line the ensemble average, the gray region the spread of one standard deviation and the thick blue line shows the historical sea ice concentration from ERA-interim.
The final product will likely only show decadal or 5-year average of a season instead of annual values. The exact temporal precision will depend on, among other things, the requirements from the operational indicators.
Figure 3: Average sea ice concentration in route section 4 during April to June. The blue line shows ERA-interim data while the thick orange line shows the mean CMIP5 projection. Thin lines show individual CMIP5 model runs and the shaded grey region indicates one standard deviation spread in CMIP5.
Description of the algorithm
Direct interpolation of model data to geographical points along a route has the drawback of using very limited information from the climate models. A Northeast passage route of 1-degree resolution would only use about 100 data points from the climate model. A remedy for this is to use climate information from a larger geographic region to represent the ice conditions along a section of the route, provided that there exists a clear correlation between the two.
The approach taken here to produce an average sea ice concentration along a section of a Arctic route is to look at regional Arctic sea ice area as a proxy. The Northeast passage is divided into 8 sections and each section is paired with a geographic region (Fig. 4). A correlation between sea ice area and average sea ice concentration along the route section is developed using reanalysis data. The correlation between regional sea ice area and average route section sea ice concentration are shown in the left panel of Fig 4. The results show a strong correlation in all route section with increased ice concentration along the route with growing sea ice area in the corresponding Arctic subregion.
Once this correlation has been established it is enough to calculate the regional sea ice area in the climate models and apply to correlation to get the average sea ice concentration along the specific route section. Here, 9 different CMIP5 models have been selected to produce a sea ice area ensemble (Fig 4, labels in left panel). The selection is based on the model's skill in estimating historical sea ice area in the eastern Arctic (Fig. 5). The average and spread of the sea ice area is calculated in each subregion of this ensemble and the correlation described above is applied to the statistics to the final product.
Figure 4: Left: Northeast passage division and corresponding regions used to calculate sea ice area in the regression. Right: Correlation between the monthly mean average sea ice area and average sectional sea ice concentration along the route.
Using this technique has at least two advantages over direct interpolation: the amount of data points used from the climate models increased approximately an order of magnitude and the results are less sensitive to the model resolution.
Step-by-step description of the algorithm
Step I: Develop correlation between sea ice area and sea ice concentration along a subroute
- Define a route for the North East Passage
- Define Arctic subregions that encompass different parts of the route
- Develop an indicator (e.g. correlation of historical data) between sea ice area in each subregion and sea ice concentration along the corresponding subroute
Step II: Apply the indicator to a subset of CMIP5 projections
- Select a subregion/subroute
- For each ensemble member:
- Get sea ice area in subregion
- Average data for specific month for each year
- Apply the indicator and get sea ice concentration along the subroute
- Store sea ice concentration along the subroute
- Store sea ice area in subregion
- Calculate model spread of sea ice concentration along the route
- Apply indicator to the model average sea ice area and get average sea ice concentration along the subroute
Validation
Model selection
As mentioned above, the selection of CMIP5 models is carried out by comparing sea ice area of the eastern Arctic from the historical simulation to ERA-interim estimations. The results of this exercise for 25 models are shown in Fig. 5 below. The 9 most skillful models have been used for the production of the indicator.
Figure 5: Relative average error in annual SIA of eastern Arctic for historical cmip5 model runs
Historical sea ice concentration along Northeast Passage
A sample form the preliminary validation of the final product can be seen in Fig. 6 below. Sea ice concentration along route section 8 is shown for the direct interpolation of the reanalysis data and for the regression-based data from the 9 CMIP5 historical model runs. The overlapping years for the two datasets spans 1980-2005 with ERA-interim limiting the start year and CMIP5 the end year. The statistical mean of the CMIP5 ensemble agrees well with the ERA-interim data for this section, both in terms of absolute value and trend.
It should be noted that other sections show larger discrepancies and specific seasons show larger bias than the annual average. A more thorough validation will be presented in an updated report.
Figure 6: Same as figure 3 but using historical CMIP5 runs
Implementation updates
The methodology used to produce this indicator was changed in the implementation phase due to two reasons: 1) technical difficulties in producing the indicator in CDS and 2) to produce a consistent dataset between SIC and SIT. As described below, a different approach had to be used for the SIT indicator because of the lack of observation to produce the regression. This approach was also selected for the SIC during the implementation in the CDS. It follows the method describe in section 5. That is, direct interpolation of native gridded CMIP-5 data onto the route waypoints through nearest neighbor and subsequent averaging of the route waypoints into 8 sub-routes.
The CMIP-5 model selection was also modified in the implementation phase. During the development of the indicator the CDS did not contain any CMIP-5 data. For this reason, the climate projections ware downloaded offline and the selection was based on all available models. However, at the time of implementation in CDS only a few of the models used during the development and validation phase where available in the CDS database. Thus, the model selection has changed to incorporate only those models available in the CDS database. These models where the top five models in Figure 5 and HADGEM2-CC, however HADGEM2-CC missed the RCP8.5 and was removed for that reason. The new model selection for this indicator is thus:
- GFDL-ESM2M
- NorESM1-M
- IPSL-CM5A-LR
- GFDL-CM3 (sea ice thickness is missing)
- CNRM-CM5
Sea ice thickness along an Arctic route
In theory, the sea ice thickness along an Arctic route can be produce with the approach that was used to calculate the sea ice concentration. However, lack of observational data of Arctic sea ice thickness makes it difficult to develop a statistical regression model. The CDS database contains observations of sea ice thickness from the Envisat and CryoSat-2 missions. It is available for from 2002 to 2015 and only for the winter months (Oct-Apr). Because of this, the sea ice thickness data from remote sensing and CMIP5 are directly interpolated onto the Northeast passage route. In order to keep consistency with the sea ice concentration product the route is divided into the same route sections where the average and spread of the sea ice thickness is calculated. The climate projection of sea ice thickness along the route uses the same subset of CMIP5 models as in the sea ice concentration indicator.
Figure 7: Same as figure 3 but for sea ice thickness
Implementation updates
The model selection has been changed according to the models used in SIC indictor describe in section 4.4. However, GFDL-CM3 was also removed due to missing SIT parameter in RCP8.5. Thus, in total 4 CMIP-5 models was used to derived the indicator.
Sea ice extent
The sea ice extent indicator calculates the monthly ice edge limit averaged over a decade and visualizes it on a map over the Arctic. Historical maps are calculated using sea ice information from ERA-interim while the subset of the CMIP-5 ensemble describe in Section 4.4 is used in the future projections. The calculations carried out to derive the map are as follows:
- Load data
- If needed: regularize the grid to 1x1 degree mesh using the interpolation tool in CDO
- If model ensemble (CMIP-5): Calculate average and standard deviation across models
- If ensample (CMIP-5): do step 4-6 for average data and for mean (+/-) 1 std
- Extract specific month and calculate decadal monthly means
- Apply a SIC threshold of 15% to determine ice-free grid point
- Plot the ice-covered grid points on a map
Ice class limits
Ship navigating in regions with ice need to have a certain ice class. The ice class determines in what ice conditions a ship is allowed to sail with or without the support of ice-breaker. There exist different version of ice class certification e.g. Finnish-Swedish certification and Russian certification and although similar the definitions differ. The ice class of a ship is determined by its hull strength and rudder and propeller protection among other things. Since the ice class definition is more or less defined from sea ice concentration and sea ice thickness model data can be used to show where a ship of a specific ice class is allowed to sail. For this indicator the Russian Maritime Register of Shipping definition is used which is defined in the document for the operational indicators. Since the ice class definition differentiate between first year and multiyear ice as well as floating and fast ice while model data only contain SIC and SIT, some assumption had to be made. First year ice is assumed to me less than 1.7 m thick and free-floating ice is assumed when SIC is less than 95%.
The algorithm follows the one described in Section 6 but differs in point 6 (thresholding). Instead of taking the 15% SIC as was done for the ice extend, the following thresholds is used for the specific ice class:
ARC-4: free-floating 1-year ice with less than 0.8m thickness (SIC<95% and SIT< 0.8m)
ARC-5: free-floating 1-year ice with less than 1.0m thickness (SIC<95% and SIT< 1.0m)
ARC-6: free-floating 1-year ice with less than 1.3m thickness (SIC<95% and SIT< 1.3m)
ARC-7: 1-year ice (SIT< 1.7m)
ARC-8: multi-year ice with thickness less than 3m (SIT< 3m)
ARC-8: multi-year ice with thickness less than 4m (SIT< 4m)
Summary
This report describes the technical details of three sea ice indicators included in the C32-442 service. The algorithms are described in detail together with an initial validation of methods not already published in literature. Two Northeast passage routes have been defined, one southerly and one more northerly. To describe the future availability of ice free navigation along these routes we apply a published and validated approach which produces an easily understandable product.
The methods for presenting projected future sea ice concentration and thickness along the routes are also described. The routes are divided into sections where average sea ice conditions are calculated. Future ice information is collected from a subset of the CMIP5 models. The models are selected based on their skill in reproducing historical sea ice area in the eastern Arctic. We propose to use a regression model to calculate the sea ice concentration along Arctic routes instead of direct interpolation. Historical relationship between
regional sea ice area and local ice conditions is developed on the ERA-interim reanalysis. This relation can be applied to calculated local ice conditions from the CMIP5 climate projections. Because of the limited amount of observations for the Arctic sea ice thickness this method cannot be applied for this parameter. Instead, the data from CMIP5 models are directly interpolated onto the routes and statistics is calculated on the same route sections as used in the sea ice concentration product.
The Route Availability Index indicator should be of direct interest to end-users and can as such be viewed as on operational indicator. The other two indicators might be more suitable as input to the Arctic operational indicators, such as the Cost of Arctic routes, which are described in a separate report.
Updates summary
The reports have been updated (2019-04-13) with Section 3.3, 4.4, 5.1, 6 and 7. Section 6 and 7 describe 2 new indicators in detail while the updates in Section 3, 4 and 5 describe modifications that has been made during the implementation phase. The main reason for the modifications has been that data that was used during the development phase is not available in the CDS. Thus, the indicators have been modified to handle different datasets or only a subset of the data that was assumed during development.
References
[Khon et al., 2016] Khon V. C., Mokhow, I. I., Semenov, V. A., Transit navigation through Northern Sea Route from satellite data and CMIP5 simulations, 2017, Environmental Research Letters
[Mulherin et. al. 1999] Mulherin D. N., Eppler T. D., Sodhi S. D., NSRSIM2A A Time and Cost Prediction Model for Northern Sea Route Shipping, 1999, INSROP working
