SWE-maps

SWE-maps were produced with observation from Special Sensor Microwave Imager (SSM/I) and Advanced Microwave Scanning Radiometer for EOS (AMSR-E) sensors, using Helsinki University of Technology (HUT) snow emission model (Pulliainen et al. 1999), and corrected with FMI’s synoptic weather station snow depth data, using assimilation method (Pulliainen 2006). Maps on SWE produced this way were compared with in situ observations from SYKE’s snowlines and with areal snow water equivalents calculated thereof.

Maps of daily values and seasonal averages were compared visually and root mean squared error (RMSE) rates were calculated. To take account effects of different scales between data sources; SWE on snow lines was recalculated using land cover class distributions data from Corine 2006 for radiometer pixels containing snow lines.

3. RESULTS AND DISCUSSION

A strong seasonally changing bias was discovered. Microwave radiometer-based technique tends to overestimate SWE during early- and mid-winter accumulation period and severely underestimate it once spring melt has started. We also identified several sites, where difference between radiometer and in-situ observations remains high and consistently biased throughout the seasons and between years. First type of bias can be mostly tracked to spatially and temporally changing attributes of snow cover. Latter cases point to presence of some systematic error source such as sparseness of gauging network, quick changes in elevation or wrongly interpreted terrain type. Figure 3 shows evolution of differences between satellite and in-situ based products during the spring 2011.

Figure 3. Evolution of differences between satellite and in-situ based products during the spring 2011.

68

Snow density and depth, and how they evolve over time, can vary greatly between different terrain types. Weather stations are mostly located on open and relatively flat areas and interpolating snow depth values over terrain with varying elevation and vegetation cover can lead to errors in background maps of snow depth. These errors can be reduced by using in situ (e.g. snow line) information on how snow depth varies between different terrain types, especially between open and densely vegetated areas.

Currently there are no daily observations available on how snow density evolves over time.

Interpreting SWE from satellite images has to rely on using constant values or use functions fitted to time series in order to describe how snow’s density changes. Both methods disregard quick changes in snow density caused by rapid melting or heavy snow falls. In addition, when interpreting SWE over larger areas such as whole country problem raises that start and end dates of winter and thus snow’s density can and usually will vary significantly over different parts of the area observed. This means that using single constant or function to whole area will cause errors.

Correcting SWE values with local values of snow density and depth was investigated. Results were promising with RMSE rates considerably reduced. More research on how to best include this data to the assimilation process itself is needed. One possibility is that data from snow lines could be used to improve snow depth maps used in assimilation of radiometer observations.

Significant part of the seasonal bias is caused by using same density function for whole country. Winter usually starts earlier and last longer in Northern Finland than in the south.

Also in the south melting and refreezing occurs often during a winter. Therefore density of the snow varies greatly between different parts of the country. Country was divided in two at the latitude 63.5° and new regional density functions were calculated. RMSE rate for northern subset dropped from 28 mm to 20 mm. Southern subset was more heterogeneous and the new density function did not improve results. RMSE values were reduced only at 40% of sites with overall RMSE rising from 31 mm to 33 mm (see figure 4 for effects of using regional density functions).

New versions of the algorithm used to interpret the microwave signal take into account better such things as layered structure of snow and effects of frozen water bodies in instruments field of view. These improvements increase the accuracy of the method and reduce the need for auxiliary data but do not remove it completely.

69

Figure 4. Effects of using regional density functions for two different snow lines Ullava and Tuusula 2011.

4. CONCLUSIONS

Satellite based microwave radiometer observations provide daily information on SWE over large areas. Comparison of in-situ observations and satellite SWE products showed seasonal bias, which was highest during the snow melt period. To improve method’s accuracy to requirements of operational observations, challenges caused by temporally and spatially changing snow attributes need to be first solved. Radiometer measurements need ground truth information on snow density and snow depth to give more accurate results. For purposes of operational observations work, best way to use satellite products is to combine them with the snow course observations data.

70

5. REFERENCES

Chang A.T.C., Foster J.L., Hall D.K., Rango A. & Hartline B.K. 1982 Snow water equivalent estimation by microwave radiometry. Cold Reg. Sci. Technol. 5(3), 259–267.

Moisander, M. 2014 Supporting operational observations of snow water equivalent with remote sensing data. Master’s Thesis, Aalto University. 75 p. (in Finnish)

Pulliainen J. 2006 Mapping of snow water equivalent and snow depth in boreal and sub-arctic zones by assimilating space-borne microwave radiometer data and ground-based observations. Remote Sensing of Environment issue 101:257-269.

Pulliainen J., Grandell J. & Hallikainen M. 1999 HUT Snow Emission Model and its Applicability to Snow Water Equivalent Retrieval. IEEE Transactions on Geoscience and Remote sensing Vol. 37. No 5.

Pulliainen, J. & Hallikainen, M. 2001 Retrieval of Regional Snow Water Equivalent from Space-Borne Passive Microwave Observations. Remote Sens. Environ. 75:76–85.

71

Experiences and recommendations on automated