LONG-WAVE IRRADIANCE AT SURFACE LEVEL RETRIEVAL IN THE CM-SAF ABSTRACT

Transcription

1 LONG-WAVE IRRADIANCE AT SURFACE LEVEL RETRIEVAL IN THE CM-SAF 1 C.Geijo, 2 R. Hollmann and 3 A. Gratzki 1 INM, Leonardo Prieto Castro s/n, Madrid, Spain. (cgeijo@inm.es) 2 GKSS Forschungszentrum, Max-Planck-Str,21502 Geesthacht, Germany. (rainer.hollmann@gkss.de) 3 DWD, Postfach , Offenbach am Main, Germany. (annegret.gratzki@dwd.de) ABSTRACT With the arrival of new operational meteorological satellites like MSG-I, traditional applications of the data generated by these systems in the field of physical climatology have received renewed attention. One of these applications is the determination of radiation fluxes at different levels in the atmosphere. This topic, transfer of energy in form of radiation, is a central topic for the new Satellite Application Facility on Climate Monitoring (CM-SAF). Estimates of incoming long-wave irradiance at earth s surface level (SDL) are necessary to close the radiation budget at this level. We present at this conference a poster on the method selected to retrieve these SDL estimates and on the validation work done so far to evaluate its performance. The CM-SAF SDL physical retrieval method has been tested with an accurate radiation transfer band model and validated using pyrgeometer measurements from different European locations, for some of which surface and upper-air in-situ observations were also available. ISCCP-DX data served for surrogate of CM- SAF cloud data. The validation results indicate that the algorithm performance is satisfactory (monthly mean within 10 W/m 2 ) provided that cloudiness and thermodynamic conditions at adequate scale are reasonably well specified. For typical midlatitude winter conditions over land, algorithm estimates and pyrgeometer measurements agree within stringent specifications (bias ~ 5 W/m 2, random error ~ 15 W/m 2, correlation > 0.9 ) for two out of three analysed stations. However, further investigation is necessary on how to tackle the problems of determination of cloud geometrical thickness and fog detection from satellite data, because this source of error has proved to induce monthly errors comparable to the cloud forcing effect itself (35 W/m 2 ). For the analysed summer data, the agreement in terms of mean errors between SDL estimates derived with ISCCP-DX data and those derived with in-situ data is within 5 W/m 2. The analysis of these summer data has allowed as well the detection of a notorious phase difference in the diurnal variation of near ground air temperature and LW irradiance which explains a persistent positive bias between calculations and measurements. Further investigation is necessary to establish the nature of this finding, which is present in all the stations analysed so far but in some cases is more prominent than in others. This fact remarks the need for careful analysis of radiometric validation data in order to get a complete assessment of validation results. Calculated values seem to underestimate measurements in clear-sky conditions (~10 W/m 2 ) which is in agreement with results reported by other researchers. 1. PRODUCT CHARACTERISTICS AND RETRIEVAL METHOD The spatial and time resolution specifications for CM-SAF (Werschek, 2003) products aim at exploiting the imagery data as close as feasible to pixel size and instantaneous estimates. This fact strongly advocate physical retrieval methods. Up-scaling is performed on intermediate products in order to obtain final products at nominal spatial resolution of 15 km and time scales from daily to monthly means. SDL estimates at this spatial scale should compare with hourly validation data within 10 W/m 2 over periodic validation times of one month. The extraction area will comprise initially that bounded by the co-ordinates 30 N up to 80 N and from

2 60 W to 60 E. These specifications are considered adequate for the achievement of satisfactory climate research and monitoring activities at a regional scale. This idea, climate monitoring from space, applications and research on a regional scale, is central in the CM-SAF project. The goal is to obtain consistent and properly validated radiation and cloudiness products at high spatial resolution. The approach selected within the CM-SAF for the SDL product is a physical retrieval technique originally developed by S. Gupta at the NASA s Langley Research Centre (Gupta,1989) for TOVS data. Later (Gupta et al. 1992) an adaptation for ISCCP-C1 data was developed. The SDL Gupta algorithm consists of two terms, the first calculates the energy radiated towards the ground by cloudless air and the second estimates the long-wave cloud forcing at the surface. The scheme is driven by data on air humidity and temperature and on cloudiness. Air thermodynamic variables will be taken from NWP analysis and the radiation field in equilibrium with these analyses will be perturbed and modulated by cloudiness observations from satellite imagers. This approach is often named hybrid because it is not based exclusively on remote-sensing observations. By the beginning of next year (2004) AVHRR data will be used and soon after MSG-I, specifically SEVIRI data, will be added. This algorithm has been enhanced and adapted considering specific CM-SAF characteristics like extraction area and algorithm driving data (Geijo and Hollmann, 2002). A location dependent adaptive scheme for the calculation of the clear-sky term was included. This scheme shortened the range of mean differences between algorithm and accurate RT model calculations for several locations within the CM-SAF baseline area from (+7 W/m 2, -5 W/m 2 ) to (+3 W/m 2, -1 W/m 2 ) while the range of r.m.s was equally reduced from (10 W/m 2,4 W/m 2 ) to (6 W/m 2, 3 W/m 2 ). The low-level cloudiness adjustment presented in Gupta et al (1992) was equally modified to improve the agreement with RT calculations under typical arctic weather conditions. This new adjustment proved to be useful in suppressing bias of about -15 W/m 2. Figure 1 shows the August 1995 results from a CM-SAF SDL prototype algorithm compared with GEWEX SRB (1x1 degree) algorithm (Stackhouse, personal communication) results for the same period of time, for the CM-SAF baseline area. Over sea surface the spatial structure of SDL estimates is clearly similar in both cases. Over land surface, several orographic features are clearly discernible on the CM-SAF retrieval which were not captured by the GEWEX SRB algorithm. CM-SAF values are slightly bigger than GEWEX ones, just 4 W/m 2 for the area averaged values. 2. ALGORITHM VALIDATION TESTS Up to this date, different validation exercises have been carried out. These exercises have consisted of tests of the SDL algorithm prototype when it is driven by ECMWF analyses and ISCCP-DX cloud data (mode A) on one hand and when it is driven by in-situ surface and upper-air observations (mode B) on the other side. A comparison of the performance of the algorithm for both configurations or modes has been done for two months (August 1995 and December 1997). This comparison has been restricted to three different German observation sites (Lindenberg, Schleswig and Stuttgart) for which, together to the in-situ data, hourly LW irradiances were available. One of the sites, Lindenberg, belongs to the Baseline Surface Radiation Network (BSRN/WCRP). The results of this comparison for the summer month (August 1995) are be summarised in Table 1. The columns labelled A correspond to mode A and those labelled B correspond to mode B. First one has to note the different number of cases entering in the statistics for each case. Radiosonde ascents were launched four times a day at Lindenberg, but only twice a day from Schleswig and Stuttgart. On the other hand, the number of retrievals per day is eight (main and secondary synoptic hours) when NWP analyses and ISCCP-DX cloud data were used. NWP fields were interpolated in time from main hours to secondary hours. The most prominent features are the persistence of an important positive bias and a significant improvement in the centred statistics (i.e., r.m.s and R), improvement that becomes more notorious if one considers that the sample size for the columns labelled as B is clearly smaller. The question of course is where this positive bias comes from. It happens that the deviation between calculations and measurements has a clear signature when it is plotted versus other parameters of interest. Figure 2 just depicts these plots for Lindenberg (for the other stations they look similar). Statistical relation of the bias with near surface air temperature, time of the day and cloud cover looks apparent from this figure. The eye-catching dependency of the mean error on cloud cover led to the question if modulation of the radiation field by broken cloud fields could be responsible for this error. However, neither the sign nor the magnitude of this effect (Harshvardhan and Weinmann, 1982, Ellingson, 1981; Schmetz 1984,

3 Figure 1. Results from a CM-SAF SDL prototype algorithm compared with GEWEX SRB results. Killen and Ellingson, 1994) could explain satisfactorily the validation results. This arc-shaped curve results from an implicit connection between cloud cover and time of the day, partly cloudy skies being reported more frequently during midday hours than at other times of the day. Other interesting feature is the negative bias ( about -10 W/m 2 ) found in clear sky cases, a fact in agreement with reports by other researches and attributed to poor formulation of the water vapour continuum in many RT models (Ohmura et al., 1998). Lind (A) Lind (B) Schl (A) Schl (B) Stut (A) Stut (B) Monthly Mean In situ (W/m²) Monthly Mean Calculated (W/m²) Number of cases Bias (W/m²) r.m.s (W/m²) R Table 1 Table summarising 08/95 validation exercise

4 Figure 2 Error analysis of SDL prototype algorithm at Lindenberg (08/1995) The layout is as follows: error vs. SDL measured (top left), error vs. air temperature at two meters height (top right), error vs. cloud cover fraction (bottom left) and error vs. time of the day (bottom right). A careful inspection of the data used in the validation indicates that this bias is caused by an interesting phase difference between near surface air temperature and LW irradiance records (Figure 3). On this figure one can see column integrated water vapour amount, two meters air temperature and measured LW irradiance for the three German stations during a few days in August Note the phase lag between the blue (temperature) and black (LW irradiance) lines. For one of the stations (Stuttgart) this lag is nearly half cycle, so that maximum on one curve and minimum on the other almost happen at the same time. At the moment it is not totally clear the reason for this. It might well be a measurement inaccuracy. It is known that the measurement of terrestrial radiation is still more difficult and less understood than the measurement of solar irradiance (WMO, 1996), the radiation exchange between the thermopile and the filtering dome is more complex in the pyrgeometer than in the pyranometer. Studies with instruments of similar manufacture in a variety of measurement configurations have indicated that, following careful calibration, fluxes measured at night agree within 2%, but in high solar radiation this difference can reach 13%. By following very cautious procedures, a state-of-the-art accuracy of 5% is possible (WMO, 1996). It is worth to mention at this point that one can find parameterisations like that developed by Martin and Berdahl (1984) which explicitly include in the clear-sky long-wave emissivity a time-of-the-day dependent correction term. In an attempt to clarify some more this interesting point, RT model calculations (Fu and Liou, 1992) were compared with in-situ measurements. The results are shown on Figure 4. They clearly resemble those obtained with the prototype algorithm shown on Figure 2. The overall monthly difference is W/m 2 and the standard deviation is 17.8 W/m 2. The mean difference at midday is +25 W/m 2 (about 7% - 8% of the midday mean value). For clear sky cases (they all were reported at night time) the bias is 5.5 W/m 2. One could note too, that the mean difference between calculations based on ISCCP-DX cloud data and those with in-situ cloud data is about only 5 W/m 2, as long as we disregard the results from Stuttgart.

5 Figure 3 Timeseries of PWC (dash line),t2m and LW SDL measured for first days of August 95 for three DWD stations (from top to bottom, Stuttgart, Lindenberg and Schleswig). Figure 4 Error analysis of RT model calculations for Lindenberg (08/1995).

6 The results of this comparison for the winter month (December 1997) can be summarised in the following table (Table 2). The columns labelled A correspond to mode A and those labelled B to mode B. Lind (A) Lind (B) Schl (A) Schl (B) Stut (A) Stut (B) Monthly Mean In situ (W/m²) Monthly Mean Calculated (W/m²) Number of cases Bias (W/m²) Rms (W/m²) R Table 2. Table summarising 12/97 validation exercise We see that in this case the use of surface and upper-air observations suppresses the moderate (in Lindenberg) to important (in Schleswig) negative biases obtained with the satellite data. In fact, the agreement between algorithm and measurements when the atmospheric conditions are specified reasonably well is remarkable, particularly if we again leave out Stuttgart. Figure 5 displays the time series of measurements and calculations by, both, algorithm and RT model for Lindenberg during December Analysis of this validation case reveals that the cloudiness observed from ground and that derived from ISCCP-DX data do not match up well. The cloud-base height estimates from satellite data turned out to be inaccurate, in general, the cloud-base level was overestimated from the ISCCP-DX cloud top information by a significant amount (see Figure 6). The calculated values were therefore clearly below the values measured on the ground. In addition, some fog and haze situations reported in the synop messages with a very important impact on the LW radiation field (~ 50 W/m 2 ) passed undetected. One outstanding problem with this SDL retrieval methodology is the lack of information on cloud base. The original version of the algorithm takes a constant cloud thickness of 50 hpa. Frouin et al (1988) used satellite data (TOVS profiles and GOES images) to retrieve SDL fluxes over ocean surface and validated their results with MILDEX measurements (off the central Californian coast; 25 October- 13 November 1983). They considered a number of different alternatives of increasing degree of sophistication to estimate cloud-base height. The most complex method made use of CLWP (cloud liquid water path), equally derived from satellite data, and a statistical distribution of liquid water concentration within the clouds to compute a cloud-base height. The simplest method considered a constant cloud thickness of roughly 50 hpa, a value extracted from an old climatology elaborated by Telegadas and London (1954). Although the statistics for the most sophisticated method turned out to be slightly better than for the other methods, the answer to which method performed best remained uncertain. As pointed out by Frouin at al. (1988) themselves, this climatological value is not representative of nimbostratus and cumulonimbus type clouds. For these clouds, the algorithm will underestimate the SDL flux. This fact encourages the search for possible improvements in this critical aspect of the retrieval. Figure 6 displays the error in which one incurs with a fixed cloud thickness of 50 hpa if visual observations from ground are taken as truth. The observation site is Lindenberg (52.22N, 14.12E, 125 m above m.s.l) and only situations where single cloud layer was reported are considered. The difference between estimate and observed cloud-base height is plotted versus (from top to bottom) cloud top temperature, cloud type and cloud cover. On the left the results for August 1995 and on the right the results for December 1997 are shown. We see that negative differences (i.e., thickness underestimated) are frequently confirmed by observations. However, at this stage it is important to notice that this error does depend on satellite cloud top temperature, cloud type and cloud cover. Other aspects relevant to this discussion are: Cloud overlap assumptions can be another important source of error in situations with several cloud layers. Algorithm adaptations for partly cloudy skies might be necessary too, although this effect seems to be second order for SDL fluxes (Ellingson, 1982; Killen and Ellingson,1994). Harshvardan and Weiman (1982) and more recently Takara and Ellingson (1996) relax the black-body cloud assumption in a series of numerical experiments in order to assess the significance of scattering effects in the calculation of LW fluxes in broken cloud fields. However, their studies focus on particular

7 spectral ranges (atmospheric window) instead on broad band simulations and disregard gas absorption (the so-called non-intervening atmosphere assumption) which clearly is a too idealised concept in the context of SDL retrieval. Figure 5. Timeseries of RTM calculated, SDL algorithm calculated and measured LW irradiance values for Lindenberg December Figure 6. Cloud-base height error analysis at Lindenberg. Left column for 08/95 and right column 12/97

8 3. CONCLUSIONS The conclusions learnt through this validation can then be summarised as follows. Gupta s parameterisation performance is satisfactory (monthly mean within 10 W/m 2 ) when atmospheric conditions at adequate scale are reasonably well specified. For typical midlatitude winter conditions over land, algorithm estimates and pyrgeometer measurements agree within stringent specifications (bias ~ 5 W/m 2, random error ~ 15 W/m 2, correlation > 0.9) for two out of three analysed stations. However, further investigation should be spent on ways to tackle the problem of determination of cloud geometrical thickness from satellite data, because this source of error has proved to induce errors of about the same order as the cloud forcing effect (Schleswig 12/97; -35 W/m 2 ). This problem stems from the inappropriate assumed climatological thickness of 50 hpa for the cloud types prevailing at this time of the year over the examined area. For the summer data analysed so far, the agreement in terms of mean errors between SDL estimates derived with ISCCP-DX data and those derived with in-situ data is within 5 W/m². This agreement is promising because the expected enhancement in observation sampling rates with the new space based European meteorological missions should have a clear impact in reducing random errors. The analysis of these summer data has allowed as well the detection of a notorious phase difference in the diurnal variation of near ground air temperature and LW irradiance which explains a persistent positive bias between calculations and measurements. Further investigation is necessary to establish the nature of this finding, which is present in all the stations analysed so far but in some cases is more prominent than in others. Calculated values seem to underestimate measurements in clear-sky conditions. This is in agreement with results reported by other researchers. Comprehensive testing of the algorithm performance with satellite cloud data at space and time resolution specified for the CM-SAF SDL product is still pending due to delay in the MSG program. 4. REFERENCES Ellingson R.G (1982) On the Effects of Cumulus Dimensions on Long-wave Irradiance and Heating Rate Calculations. J. Atmos. Sci., 39, pp Geijo C. and Hollmann R. (2003) Analysis and Improvement of Surface Downwards Long-wave (SDL) Flux CM-SAF Algorithm, CM-SAF VS Report, March 2003 pp 31. Available from CM-SAF scientific publications library (Deutscher Wetterdienst). Gupta S. (1989) A Parameterization for Long-wave Surface Radiation from Sun-Synchronous Satellite Data. J. Climate, 2, pp Gupta S. et al (1992) Long-wave Surface Radiation over the Globe from Satellite Data: recent Improvements. J. Appl. Meteor, 31, pp Fu Q and Liou KN (1992) On the Correlated k-distribution Method for Radiative Transfer in Non-homogeneous Atmospheres. J. Atmos. Sci., 49, pp Frouin R. et al (1988) Downward Longwave Irradiance at the Ocean Surface from Satellite Data: methodology and in situ Validation. J. Geophys. Res., 93, NoC1, pp Killen R.M and Ellingson R.G (1994) The Effect of Shape and Spatial Distribution of Cumulus Clouds on Long-wave Irradiance. 51, 14, pp Martin M and Berdhal P (1984) Characteristics of Infrared Sky Radiation in the United States, Solar Energy, 33, pp Harshvardhan and Weinman J.A (1982) Infrared Radiative Transfer through a regular Array of cuboidal Clouds. J. Atmos. Sci., 39, pp Ohmura A. et al (1998) Baseline Surface Radiation Network (BSRN/WCRP):New Precision Radiometry for Climate Research». Bull. Amer. Meteor. Soc., 79, pp Takara E.E and Ellingson R.G (1996) Scattering Effects on Long-wave Fluxes in Broken Cloud Fields. 53, 10, pp Telegadas and London (1954) A Physical Model of the Northern Hemisphere Troposphere for Winter and Summer. Scientific Rep 1: Dept of Meteorology and Oceanography, New York University, Contract No AF 19 (122)-165, pp 55. Schmetz J (1984) On the Parameterization of the Radiative Properties of Broken Clouds. Tellus, 36A, pp Werschek M., (2003) The Satellite Application Facility on Climate Monitoring. In these proceedings. WMO (1996) Guide to Meteorological Instruments and Methods of Observation (6 Edition) WMO-No8, World Meteorological Organisation, Geneve. ACKNOWLEDGEMENTS. This study was sponsored by the Visiting Scientist Programme of the CM-SAF.