Arctic Transportation Network

Permafrost Laboratory task: Ground Temperature Forecast Modeling 

The simulation model from the Geophysical Institute Permafrost Laboratory (GIPL) used in this study was first described in Tipenko and Romanovsky (2001) and used for “permafrost temperature reanalysis” by Romanovsky et al. (2002). GIPL is a sophisticated numerical model (Tipenko and Romanovsky, 2001; Sergueev et al., 2003), which takes into account the temperature-dependent latent heat effects. The model is capable of modeling spatially distributed permafrost temperatures as described in Sazonova and Romanovsky (2003) and Marchenko et al. (2008). The structure of the model is illustrated in Figure 1.

The model needs to be calibrated using available data for air temperature, snow depth and density, vegetation, soil properties, and soil temperatures. The first step in that calibration process we use a few years of soil surface temperatures to calculate and subsequently calibrate the ground temperature calculations including permafrost temperatures. When a match is found between measure and calculated soil temperatures we start the procedure on using the air temperature as an input variable. The air temperature is more variable and the ground is insulated with snow. The snow thermal conductivity is variable based on snow density and snow cover extend. The first step is assuming a constant snow density over the winter, this can be modified for a more detailed simulation during projection runs. 

For ground temperature projections we can optimize parameters to match observed and predicted ground temperatures as freezing progresses. There are likely small differences in soil moisture from year to year that have an effect on the thermal properties. The optimization will need to be done manually.

An example for Climate Station DFM4 is show in Figure 2. 

Figure 2.  Soil-temperature comparisons between measured values at 20 cm depth for Climate Station DFM4 and simulated values after model calibration.

The next phase of the project included running the model tool using air temperature and snow data. The model performance is reduced due to the complex nature of snow insulation, but the overall match between observed and simulated soil temperatures is still good, see Figure 3.

Figure 3.  Soil-temperature comparisons between measured values at 5 cm depth for Climate Station DFM4 and simulated values after model calibration using measured air temperature and snow depth, with the later illustrated in the graph.

The project team has tested different forward simulation techniques to produce 14-day forecast models of soil temperatures at 20-cm depths. Figure 4 illustrates a forecast simulation using simulations starting from a current day and using a fixed forward air temperature and an adjusted snow boundary condition.

Figure 4.  Soil-temperature predictive simulations showing one example of using a running 5-day average air temperature as input showing the 20 cm depths simulation, compared with the measured temperature at 20 cm depth for Climate Station DFM4.

The individual lines show 14 day predictions on that day. The initial conditions are always updated as starting temperatures in the model. They are then calculated forward using the fixed average daily air temperature of the past 5 days.

Similarly we have calibrated the model for a second site called DFM3. Figure 5 shows the calibration step using soil surface temperature as the input for the model.

Figure 5. Soil-temperature comparisons between measured values at 20 cm depth for Climate Station DFM3 and simulated values after model calibration.

The model calibration run with air temperature and snow depth is illustrated in Figure 6.

Figure 6. Soil-temperature comparisons between measured values at 5 cm depth for Climate Station DFM3 and simulated values after model calibration using measured air temperature and snow depth, with the later illustrated in the graph.

The predictions for 2007 for the DFM3 site are presented in Figure 7. 

Figure 7 – Soil-temperature predictive simulations showing one example of using a running 5-day average air temperature as input showing the 20 cm depths simulation, compared with the measured temperature at 20 cm depth for Climate Station DFM3.

The goal of our part in the ATN project is to predict soil temperatures automatically. Therefore we have to implement a routine that will run the soil temperature forecast tools for index sites.  Currently we are in the process of implementing this procedure on a Supper Computer at ARSC. Data communication has been setup to download field data on a daily basis. Data will feed into the model through a set of input files. These files are currently generated automatically.

References

Marchenko, S.S., Romanovsky, V.E. and Tipenko, G.S.: Numerical Modeling of Spatial Permafrost Dynamics in Alaska. Editor(s): D.L. Kane and K.M. Hinkel Collection: Proceedings of the Ninth International Conference on Permafrost, June 29-July 3, Fairbanks, Alaska, 2008 Bibliography: Vol.2: pp. 1125-1130, (2008).

Romanovsky, V.E., Burgess, M., Smith, S.L., Yoshikawa, K., and Brown, J.: Permafrost temperature records: indicators of climate change. EOS, AGU Transactions, 83: 589-594, (2002).

Sazonova, T.S. and Romanovsky, V.E.: A model for regional-scale estimation of temporal and spatial variability of active layer thickness and mean annual ground temperatures. Permafrost and Periglacial Processes, 14: 125-139, (2003).

Sergueev, D., Tipenko, G., Romanovsky, V., and Romanovskii, N.: Mountain permafrost thickness evolution under influence of long-term climate fluctuations (results of numerical simulation). In: Proceedings of the VII International Permafrost Conference, Switzerland, July 21-25, pp. 1017-1021, (2003).

Tipenko, G.S. and Romanovsky, V.E.: Simulation of Soil Freezing and Thawing: Direct and Inverse Problems, EOS, Trans. AGU, 82 (47), Fall Meet. Suppl., Abstract, F551, (2001).