PILPS 2e Experimental Design, Response to Comments

Thank-you to everyone who took the time to review this document and submit your questions. A summary of the questions and comments we received is provided below. The revised experimental design will be posted on the PILPS 2e web site (http://www.hydro.washington.edu/~Lettenmaier/CurrentResearch/PILPS-2E/index.html) , as soon as a few outstanding issues are resolved.

1. Section 2.3, Torne/Kalix Basin Description. The location of Junosuando/Tarendo is referred to on Figure 1, but it is not identified on the map. In the following sentences, Karunki is also mentioned. Both of these should be shown clearly on the map (or another map).

Another figure has been added to show the location of precipitation, temperature and gauging stations.

2. Section 3.1, forcing data. It is preferable for our model that forcing data of cloudiness be provided, since cloudiness is used in partitioning of incoming shortwave radiation into direct and diffuse radiation.

A separate file with hourly cloud cover fraction (tor_cloud_1hr.nc) will be provided as an optional variable.

3. Section 3.1, Table 1. Each modeling group is able to change the snow/rain temperature threshold. Snow melt is said to be the key process affecting the hydrology of this basin, so won't allowing models to vary the total snow amount which falls make the inter-comparison much more difficult?

Yes. We have reconsidered this decision, this snow/rain temperature threshold will remain fixed and should not be altered by modelers.

4. Section 3.2.1, Figure 2. There is no connection between Torne and Kalix Rivers for the bifurcation at Junosuando.

Figure 2 has been revised to show this connection.

5. Section 3.2.1, topography. We use the standard deviation of height at each grid cell to represent sub-grid scale topography, and the average slope at each grid. We need to calculate them from the 5 minute-resolution elevation data (tor_ele_5min.nc) that will be provided or from the global DEM at 30 second resolution, GTOPO30. It would be convenient for us if they are included in the Land Surface Characteristics data.

A file of 30 arc second resolution elevations (GTOPO30) will be provided instead of the proposed 5 minute DEM. Since this represents a much finer resolution than the remainder of the land surface characteristics, it will be provided in a separate file: tor_ele_30sec.nc.

6. Section 3.2.1. Is it possible to get a topographic index or elevation data - 30 to 1000 m grid of each grid cell?

See above, elevation data at 30 arc second resolution will be supplied for each grid cell, in file tor_ele_30sec.nc.

7. Section 3.2.2. I would say it is not a good idea to leave each participant a chance to use a different set of soil parameters. Normally there are big problems in understanding the discrepancies between models, so it would be better to specify the soil parameters for everybody.

This experiment is designed so that soil parameters for the entire basin can be adjusted following calibration of the two 'calibration' sub-basins. Therefore, everyone will not be using identical soil parameters, even if only one parameter set is specified. We will therefore offer a choice of initial parameters sets, as originally indicated, rather then trying to justify a choice of one over another.

8. Section 3.2.2, soils. Will instructions be provided about soil thermal properties? They may affect the thawing and freezing depth, hence the runoff characteristics. Will you provide temperature data at the lower soil boundary layer?

We are not planning to specify values for thermal conductivity and volumetric heat capacity since they are so strongly dependent on soil moisture and ice content. If a constant value is required, we suggest that you could use the equations of Johnasen [as reported in Farouki, O.T., Thermal Properties of Soils, in Series on Rock and Soil Mechanics, vol 11, Trans. Tech., Clausthal-Zellerfeld, Germany, 1986] and Flerchinger and Saxton, Simultaneous heat and water model of a freezing snow-residue-soil system, II Field verification, Trans. ASAE, 32, 573-578, 1989]. We will provide a suggested thermal damping depth and lower temperature boundary condition based on the soil temperature data (5, 20, 50 and 100 cm depth) available from the Abisko Research Station in the headwaters of the Torne River.

9. Section 3.2.3, Table 4. What is the difference between closed shrubland and open shrubland? We need to understand this to determine model specific parameters.

Unfortunately, the vegetation data that will be used do not indicate a specific threshold. The best interpretation we can offer is that closed shrubland contains more shrubland than grassland, and open shrubland contains more grass than shrubs. We suggest that you assume open shrubland contains less than 50% shrubs, while closed shrubland contains greater than 50% shrubs.

10. Section 3.2.3, vegetation. Only height and albedo of landcover are referred to in the document. Will instructions be provided for other characteristics, such as root distribution and transmissivity? Will you provide roughness length, stomatal resistance and root depth data for the given vegetation classes?

Once again we are trying to avoid specifying too many parameters since many models use them differently (e.g. transmittance of solar radiation is dependent on solar zenith angle, etc.). We will provide suggested values of the following parameters for each vegetation type: root distribution, radiation transmissivity, architectural resistance, stomatal resistance, displacement height and roughness height. It is not required that these values be used, however any deviations from the suggested values should be documented. A questionaire will be provided to clarify what sort of description is needed.

11. Section 3.2.3, Table 4. Is the landcover height of woodland really 0 m?

The value for landcover height of woodland should be 10 m, which is corrected in the revised document.

12. Section 3.3, Bullet 3. Which reference height is this? The typical measurement height for station data is 1.5 - 2 m.

The reference height for wind, temperature and humidity measurements is 2 m, which is corrected in the revised document.

13. Section 3.3. It is requested that participants use the albedo thresholds such that the overall fresh snow albedo is 0.75 (as was used in PILPS-Valdai). But we feel that this value is rather low for fresh snowfall (as a simple example, the detailed snow scheme CROCUS has total fresh snow albedos often over 0.90). The snow models in ISBA have been extensively tested and validated using the method by Douville et al. (1995, Clim Dyn). We would like to request permission to use our "standard" albedo calculation method. This yields snow albedos in the range from 0.50 to 0.85. Using our standard approach would also give results more consistent with our other work. However, if this poses a problem, we are willing to use 0.75.

This is acceptable. The albedo of fresh snow can also be considered a 'suggested' variable, with any variations to be documented.

14. Section 4.1, Calibration/validation runs. I have some questions about the selection of the 4 test basins. Both of the selected validation basins are located in more extreme mountain locations where modeling runoff will likely prove difficult for most groups. Kaalasjärvi includes the highest point in Sweden. On the other hand, both of the calibration basins lie at lower elevation with quite different characteristics. Kilpisjärvi is higher and more exposed than Övre Lansjärvi, but not to the same extent as the other two basins. Therefore, I would suggest that you change your pairing so that one of the higher mountain basins is calibrated and the other is used for validation. Likewise, Kilpisjärvi and Övre Lansjärvi could be paired as calibration/validation basins. However, it might be even better to choose yet a different basin (than Kilpisjärvi) to be matched with the lower elevations and vegetation types at Övre Lansjärv.

The calibration and validation pairs have been switched as suggested, and the revised document so indicates.

15. Section 4.1, Can the saturated hydraulic conductivity be changed for the calibration basins and within what ranges?

Yes, saturated hydraulic conductivity can be changed for the calibration basins. However, since this value can reasonably vary over orders of magnitude, we are not restricting the range to within 20% of the values in Table 3. Values for saturated hydraulic conductivity should be within two orders of magnitude of the values in Table 3.

16. Section 4.1, Calibration/validation runs. One thing that has not been discussed is the extreme variation of precipitation in the mountains. We don't have many precipitation stations and the ones that do exist are at low elevations. SMHI has typically added a correction of some 10-20% per 100 meters when we model in these areas to try to compensate. This will be a particular problem for you in the two higher test basins - Övre Abiskojokk and Kaalasjärvi.

We have attempted to calculate a representative precipitation lapse rate to apply to the mountainous areas, through the evaluation of annual precipitation at three station pairs. Two of these pairs are in the mountainous region to the northwest (Katterjakk (elev. 560 m)/Abisko (elev. 386 m) and Katterjakk/Kiruna (elev. 450 m) and one is in the north (Keinovuopio (elev. 450 m)/Karesuando (elev. 329 m). There was some concern that the Abisko station lies in a rain shadow, and in fact the lapse rate calculated for this station was much higher than the others and was discarded. Following interpolation, the daily precipitation values were lapsed to the mean grid cell elevation. We are still discussing the validity of this technique with SMHI, but the final data set will include some sort of orographic enhancement of precipitation.

In addition, our original proposal to disaggregate daily precipitation to hourly is not possible. Therefore, the hourly precipitation will be calculated as 1/24^th of the daily value. Since the annual hydrograph in this area is dominated by snow melt, the error associated with this assumption should not be large. Andy Pitman, (Macquarie University) has proposed an experiment (possibly run on some sub-set of the models) to test the sensitivity of simulated results to the method of interpolation of meteorological data. We will explore the possibility of testing model sensitivity to the disaggregated precipitation at the same time.

17. Section 4.2, Table 8, average surface temperature and surface radiative temperature. I understand that the temperatures used in calculating the energy balance and turbulent fluxes should be used to calculate the average surface temperature. In our model, the energy balance is solved at four surfaces at each grid, namely on canopy and ground surface in snow-covered and snow-free portion, considering the snow-covered ratio and LAI. Is the average surface temperature a simple average of the four surface temperature, or weighted average for snow-covered ratio and LAI?

The average surface temperature for the grid cell should be a weighted average for each surface category over which surface temperature is calculated.

18. Section 4.2, Table 8, average surface temperature and surface radiative temperature. In calculating long-wave radiation, the same temperatures as those used to calculate energy balances are used. Then, the difference between the average surface temperature and surface radiative temperature is the difference between the average of temperature and fourth root of the average of fourth powered temperature?

Exactly. Other differences between both temperatures may arise from the numerical implementation of the energy balance. We cannot assume that the same temperature is used to compute the turbulent flux and the long-wave radiation. Thus the distinction between both temperatures is needed but very simple models may not show any difference between the two values.

19. Section 4.2, Table 8, Furthermore, could you explain a little what the averaging process means, that takes into account the fact that the outgoing long-wave flux has to be conserved? I also would like to ask if the surface radiative temperature is different from that calculated from the grid-mean upward longwave radiation by using Stefan-Boltzman equation (LW=sigma*T^4)?

Averaging the T⁴ values (weighted for surface area) for all radiative surfaces and taking the fourth root results in the effective radiative temperature of all surfaces in the model grid cell. This temperature to the fourth power results in the average outgoing longwave flux for the entire grid cell, whereas a weighted average of surface temperature will not necessarily conserve this longwave flux. So yes, the surface radiative temperature should be the same as that calculated from the weighted grid mean upward longwave radiation.

20. Section 4.2, Table 8, Total soil wetness. The soil moisture will be integrated to the whole depth to calculate the total soil wetness. The total depth of soil is however different among the models, which may lead to a different characteristic.

Yes, but that is what we are looking for. The other soil moisture parameters in Table 4 should be adequate for evaluation of absolute quantities.

21. General. As far as the time schedule goes, I think you are still a bit unrealistic. (And many other similar comments…)

The schedule will be relaxed by popular request!

A final version of the experimental design and forcing data, which incorporates all of these comments, will be released in one to two weeks, once we finalize decisions regarding the choice of calibration basins and complete our quality checks on the revised precipitation data set (see Comments 14 and 16).

We will aim for simulated results to be returned to the University of Washington by October 15, 2000.