• No se han encontrado resultados

PRINCIPIOS TECNICOS PARA EL DISEÑO DE PRODUCTOS MULTIMEDIA

6. MARCO REFERENCIAL

6.2 PRINCIPIOS TECNICOS PARA EL DISEÑO DE PRODUCTOS MULTIMEDIA

When the ground is covered by snow, spatial differences in surface tempera- ture are most certainly reduced, so that it becomes feasible to compare a point measurement of the surface temperature with the satellite record of MODIS LST. The procedure yet remains problematic due to of the large scaling gap between the point measurement and the satellite footprint of more than 1 km2.

As spatially distributed measurements of the surface temperature are not avail- able, the long time series of measurements of Lout is employed to improve the

statistical significance and to account for the wide range of synoptic conditions encountered during the winter season (see Sect. 4.2.3). The comparison is per- formed for the period from 1 November until 15 April for seven winter seasons from 2002/2003 to 2008/2009. During the chosen period, the ground is usually covered by snow and melt processes do not occur, except for sporadic rain-on- snow events (see Sect. 4.2.3). For each season, on the order of 2500 MODIS granules are available, which fulfill the overlap criterion of 60% with the target area depicted in Fig. 3.7. However, the frequent cloudiness results in a rather limited success rate of about 25% to 30%, which is nevertheless better than the success rates of less than 20% during the snow-free season.

The average surface temperatures for each of the years derived from MODIS L2 LST and the “true” surface temperature inferred from measurements of Lout

(Eq. 2.130) at the Bayelva station (columns in bold face, TM

1 and T2B) is pre-

sented in Table 4.4. It is clear that the average surface temperature derived from the satellite records is systematically biased to colder temperatures by on average more than 3 K. Using the incremental averaging procedure (first hourly, then six-hourly, daily and weekly averages, see Sect. 3.3.1) for MODIS LST does not significantly improve the results (TM

2 , Table 4.4). Three reasons

1-Nov 1-Dec 31-Dec 30-Jan 29-Feb 30-Mar -40 -30 -20 -10 0

T

surf

/ °C

2003/2004

1-Nov 1-Dec 31-Dec 30-Jan 1-Mar 31-Mar

-40 -30 -20 -10 0 Bayelva station MODIS L2 LST

T

surf

/ °C

2005/2006

Figure 4.28: Comparison of surface temperature measurements from MODIS L2 and surface temperature measured at the Bayelva climate station for the periods between 1 November and 15 April of the winter seasons 2003/2004 and 2005/2006.

Table 4.4: Average surface temperatures for the winter seasons (defined as 1 November to 15 April) of seven years derived from hourly averages of the MODIS L2 LST product and the hourly record of the Bayelva station. TM

1 : average of

MODIS L2 LST; TM

2 : average of MODIS L2 LST, incremental averaging proce-

dure; TB

1 average of surface temperatures at the Bayelva station for times, when

MODIS LST measurements are available; TB

2 average of surface temperatures

at the Bayelva station.

winter MODIS Bayelva station bias season TM 1 /◦C T2M/◦C T1B/◦C T2B/◦C (T1M-T2B)/◦C 2002/2003 -17.0 -17.9 -14.4 -15.1 -2.6 2003/2004 -22.9 -20.1 -20.2 -16.1 -6.8 2004/2005 -17.1 -16.5 -15.8 -14.0 -3.1 2005/2006 -14.0 -14.0 -10.8 -10.3 -3.7 2006/2007 -14.9 -15.8 -13.3 -12.7 -2.2 2007/2008 -15.8 -16.2 -15.6 -14.0 -1.8 2008/2009 -17.2 -17.8 -15.2 -14.2 -3.0 average -17.0 -16.9 -15.0 -13.8 -3.2

1. During polar night, clear-sky conditions feature systematically colder tem- peratures than overcast conditions (see Sect. 4.2.3). Therefore, the single satellite measurements can be accurate, but as satellite detection of LST is limited to clear-sky conditions, cold conditions are systematically over- represented in averages.

2. Strongly erroneous measurements, as they have been observed for the sum- mer season, exist in the MODIS data set during winter. The systematic contamination with colder cloud top temperatures could be responsible for the cold-bias in the MODIS-derived LST averages.

3. The point measurements at the Bayelva station are not representative for the satellite footprint, so that a strong spatial heterogeneity of the surface temperatures must exist within the satellite footprint.

The first point can be investigated by calculating the average of the surface temperature recorded at the Bayelva station at times, when a satellite LST measurement is available. The result is displayed in Table 4.4 (T1B): if the

single MODIS LST measurements would conform to the surface temperature at the Bayelva station, TB

1 would have to be equal to T1M, the average derived

from MODIS L2 LST. However, the temperature is considerably warmer than the average of MODIS LST, though still colder than the terrestrial average for most of the years. Fig. 4.28 displays the surface temperature at the Bayelva station and MODIS LST for the two example years. The winter 2003/2004 is the coldest in the record, and TB

1 is close to the T1M, the average of MODIS

LST, so that the strong deviation between satellite and terrestrial average can be largely explained by reason 1. Indeed, the MODIS LST measurements are strongly concentrated in periods with cold surface temperatures, while measure- ments are almost completely missing during the regular events when the surface temperature is in the range of 0◦C. The agreement between single MODIS LST

measurements and the terrestrial record is generally good. In the second exam- ple year 2005/2006, which features the warmest average surface temperatures in the record, the value of TB

1 is almost equal to the terrestrial average, so

that systematic overrepresentation of cold periods can only play a minor role in explaining the cold bias of -3.7◦C of MODIS LST. Instead, a large number of

strongly cold-biased MODIS LST measurements are observed during prolonged periods of warm surface temperatures in the range from -5◦C and 0C. The

cold-bias of single measurements regularly exceeds 10 K, which clearly classifies them as measurements errors with likely admixing of cloud top temperatures, as it has been reported in the previous section. The same is actually observed in the much colder winter 2003/2004, e.g. in beginning of November, beginning of January and beginning of March. However, as the warm periods are much shorter than in 2005/2006, these erroneous measurements play a much smaller role.

The systematic cold-bias of MODIS LST in case of warm surface temperatures, that predominantly occur in case of overcast conditions, is corroborated by Fig. 4.29. It displays the scatter plot between MODIS LST and the record of the Bayelva station (i.e. the hourly value closest in to the acquisition time of MODIS). Despite of a considerable spread, the agreement between terrestrial and satellite measurements is excellent for surface temperatures colder than −15◦C, but the performance rapidly degrades towards 0C, where MODIS LST

features a cold bias of on average more than 10 K.

Finally, the possibility, that spatially heterogeneous surface temperatures do ex- ist during winter, must be critically elucidated. In this case, the systematic cold- bias of MODIS LST could only be explained by areas with significantly colder surface temperatures than those at the Bayelva station. Apart from the Kongs- fjorden, which would be much warmer than the land areas, high-lying mountain slopes and glaciers characterize the immediate surrounding of the MODIS tar- get area (Fig. 3.7). The surface temperature of these features is most likely on the order of the lowland area around the Bayelva station, or even higher due to the prevailing stable atmospheric stratifications (see Sect. 4.2.3). Therefore, the magnitude of the cold-bias of average MODIS LST measurements cannot be explained by spatial heterogeneity of the footprint area. However, it is well pos- sible that the spread between single terrestrial and MODIS LST measurements (Fig. 4.29) is at least partly caused by areas in the satellite footprint featuring slightly different surface temperatures than those at the Bayelva station.

Documento similar