Comparison of surface radiative flux parameterizations Part II. Shortwave radiation

Transcription

1 Ž Atmospheric Research wwwelseviercomrlocateratmos Comparison of surface radiative flu parameterizations Part II Shortwave radiation Sami Niemela ), Petri Raisanen, Hannu Savijarvi Department of Meteorology, UniÕersity of Helsinki, PO Bo 64, FIN Helsinki, Finland Received 16 January 001; accepted 5 April 001 Abstract This paper presents a comparison of several shortwave Ž SW downwelling radiative flu parameterizations with hourly averaged pointwise surface radiation observations made at Jokioinen and Sodankyla, Finland, in 1997 Both clear and cloudy conditions are considered The clear-sky comparisons included si simple SW parameterizations, which use screen level input data, and three radiation schemes from numerical weather prediction Ž NWP models: the former European Centre for Medium-Range Weather Forecast Ž ECMWF scheme, the Deutscher Wetterdienst Ž DWD scheme, and the High Resolution Limited Area Model Ž HIRLAM scheme Atmospheric-sounding profiles were used as input for the NWP schemes For the cases with clouds, three simple cloud correction methods Ž mainly dependent on the total cloud cover were tested In the SW clear-sky comparisons, the relatively simple scheme by Iqbal provided the best results, surprisingly outperforming even the NWP radiation models Simple cloud corrections performed poorly in the SW region Out of these schemes, a new cloud correction method developed using the present data provided the best results q 001 Elsevier Science BV All rights reserved Keywords: Shortwave radiation; Surface radiative flu; Empirical formulas; Cloud corrections 1 Introduction Downwelling flues of longwave Ž LW, mm and shortwave ŽSW, mm radiation are key terms of the surface energy budget and are vitally important for ) Corresponding author Fa: q addresses: saminiemela@helsinkifi Ž S Niemela, petriraisanen@helsinkifi Ž P Raisanen, hannusavijarvi@helsinkifi Ž H Savijarvi r01r$ - see front matter q 001 Elsevier Science BV All rights reserved Ž PII: S

2 14 S Niemela et alratmospheric Research 58 ( 001) climate studies and many applications such as agricultural meteorology and air sea ice interaction studies In Niemela et al Ž 001, hereafter Part I, we compared several simple parameterization formulas and radiation codes from numerical weather prediction Ž NWP models with LW flu observations in Finland This paper presents a similar SW comparison The performance of si simple SW clear-sky radiation parameterization schemes and three cloud correction methods is evaluated using data from Jokioinen and Sodankyla, Finland In addition to the simple parameterizations, three NWP radiation codes are included in the clear-sky intercomparison: the former radiation scheme from the European Centre for Medium-Range Weather Forecast Ž ECMWF model Žhereafter EC-OLD; Morcrette, 1991, the Deutscher Wetterdienst ŽDWD; Ritter and Geleyn, 199 scheme and the High Resolution Limited Area Model Ž HIRLAM; Savijarvi, 1990 scheme Physical background The downwelling shortwave flu at the surface may be written as shown in Eq Ž 1 : FSW stscosustes0cosu, Ž 1 where SseS0 is the incident solar radiation at the top of the atmosphere on a surface Ž perpendicular to the solar beam S0 s1367 Wrm is the solar constant and es accounts for the seasonal variations of the Earth Sun distance, u is the solar zenith angle, and t is a broadband atmospheric transmissivity The transmissivity can be written Ž neglecting atmospheric refraction and the curvature of Earth as shown in Eq Ž : 1 ts S epž yt rcosu qt dl Ž ` H S 0,l l l,dif 0 0 Here, l is wavelength, tl is the monochromatic optical thickness, the first term within the wavelength integral represents the contribution by the direct solar beam, and the second term represents diffuse solar radiation It is evident from Eqs Ž 1 and Ž that the solar elevation has a very strong effect on the downwelling flu at the surface The top-of-the atmosphere insolation on a horizontal surface is directly proportional to cosu, and the atmospheric transmissivity, particularly its direct beam component, decreases with decreasing cosu as the slant path lengths become larger The factors contributing to atmospheric attenuation of solar radiation include gaseous absorption Ž most importantly, by water vapour and ozone, Rayleigh scattering by air molecules, and scattering and absorption by cloud droplets, ice crystals and aerosols The total optical thickness tl needed in the computation of direct beam transmission is obtained simply as the sum of the contributions by all of these components The computation of the diffuse transmissivity tl,dif is not simple Physically, diffuse radiation is produced by all atmospheric scattering processes The diffuse radiation reaching the ground also depends on the surface albedo via atmospheric re-reflection of surface-reflected radiation

3 3 Parameterization schemes S Niemela et alratmospheric Research 58 ( 001) 143 The simple parameterization schemes presented below in Sections 31 and 3 are based on empirical relationships derived from observed radiation flues The simplest schemes depend only on the solar zenith angle u while others also use screen level input variables such as water vapour pressure e whpa Cloud corrections Ž all-sky methods 0 use mainly total cloudiness observations The NWP SW radiation parameterization schemes are briefly described in Section The clear-sky flu Si formulas ŽEqs Ž 3 Ž 8 are considered for the calculation of the downwelling SW flu in clear-sky conditions In the first three ŽEqs Ž 3 Ž 5, only the cosine of the solar zenith angle Ž cosu is used Ž 1 The scheme by Bennett Ž 198 is the simplest of all It is based on the assumption that knowledge of the atmospheric mean transmissivity is sufficient for calculating the monthly mean flues Bennett used this scheme shown in Eq Ž 3 as: F SW,clr s07s0cosu, Ž 3 where S is the solar constant Ž1367 Wrm 0 Here, as in all the following formulas, the unit of the flu is Wrm Eq Ž 3 does not take into account the decrease of atmospheric transmissivity with increasing solar zenith angle Ž ie increasing slant path lengths, so it might not be very appropriate for the calculation of hourly values Ž The method of Paltridge and Platt Ž 1976 was derived using a long-time series of hourly averaged values of measured SW flu from Aspendale, Australia and is described in Eq Ž 4 as: ' FSW,clr s10q1411 cosuy310 cosu 4 Ž 3 The formula by Moritz Ž 1978, shown in Eq Ž 5, is based on the scheme by Lumb Ž 1964, which was intended for hourly as well as daily and monthly mean calculations However, Lumb s coefficients are very sensitive to local conditions so Moritz derived new coefficients to fit measured data from Baffin Bay, Canada: F SW,clr ss0cosu Ž 047q047cosu Ž 5 The net formulas ŽEqs Ž 6 Ž 7 add the screen level water vapour pressure e whpa 0 as an etra input parameter The short-time variability of the near-surface humidity is thus taken into account, which should make these formulas better suited for the calculation of instantaneous flues than the three previous formulas ŽEqs Ž 3 Ž 5 Ž 4 Zillman Ž 197 used radiation data from islands of the Indian Ocean for deriving his scheme shown in Eq Ž 6 : S0cos u FSW,clr s Ž 6 y3 1085cosuqe0Ž 7qcosu =10 q010 Ž 5 Shine Ž 1984 improved the scheme of Zillman Ž 197 by adjusting its coefficients to give better results in arctic winter conditions Shine noticed that Zillman s equation Ž

4 144 S Niemela et alratmospheric Research 58 ( 001) underestimates the SW flues especially in the Arctic regions Shine s version of Eq Ž 6 is shown as Eq Ž 7 : S0cos u FSW,clr s Ž 7 y3 1cosuqe0Ž 10qcosu =10 q00455 Ž 6 All the previous schemes are rough approimations; they even ignore the fact that the top-of-the-atmosphere insolation S varies with season due to the elliptical orbit of the Earth The sith parameterization Ž Iqbal, 1983 is somewhat more detailed than the previous ones Iqbal Ž 1983 presented a parameterization as shown in Eq Ž 8 : F ss qd, Ž 8 SW,clr dir where Sdir is the direct solar radiation on a horizontal surface and D is the diffuse irradiance The direct radiation is calculated as shown in Eq Ž 9 : S s 09751SŽ cosu t t t t t, Ž 9 dir R g w a o where S is the broadband solar radiation at the top of the atmosphere, tr is the transmittance by Rayleigh scattering, tg is the transmittance by uniformly mied gases, tw is the transmittance by water vapour, ta is the transmittance by aerosols and to is the transmittance by ozone More detailed documentation can be found in Iqbal Ž 1983 The version used in this paper is based on Venalainen Ž 1994 Eq Ž 10 shows his derived empirical function for the calculation of ta using SW radiation measurements from Jokioinen and Sodankyla: ta s059q001uy1336=10 y4 u Ž 10 Moreover, the estimation of the precipitable water content w wcm, which is used in the calculation of t w, was modified The precipitable water is originally estimated using screen level Ž sl temperature T wk and water vapour pressure e wpa via Eq Ž 11 : e wsl s0493, Ž 11 T 0 where e is estimated from e srh epž 63y5416rT Ž RH is relative humidity The calculated w is then adjusted by an empirical correction formula, Eq Ž 1 sl, which Venalainen Ž 1994 derived using radio soundings from Jokioinen: w s wsl y Ž 1 The diffuse irradiance D in this scheme is presented as the sum of the three terms as shown in Eq Ž 13 : DsDRqDaqD m, Ž 13 where DR is the Rayleigh-scattered and Da is the aerosol-scattered diffuse irradiance, and D is the multiple-reflected irradiance Ž m backward scattering of surface-reflected radiation Ž see Iqbal, 1983 for more details The surface albedo, which is needed in the term D m, was estimated by using the actual measured downwelling and upwelling hourly averaged SW flues Ž see Section 4 for further discussion

5 3 The all-sky flu S Niemela et alratmospheric Research 58 ( 001) 145 The effects of clouds on the SW flu have typically been parameterized in the simpler schemes by multiplying the SW clear-sky flu by a cloud correction factor which depends on the total cloudiness c Two previously suggested corrections are considered here First, Berliand Ž 1960 presented an all-sky flu parameterization linear on c as shown in Eq Ž 14 : FSW,all s Ž 1yc qtc c F SW,clr, Ž 14 where t is the cloud transmissivity We assume t s 048 as Bennett Ž 198 c c did in sea ice eperiments in the Arctic regions Second, Laevastu Ž 1960 assumed, based on data from midlatitude oceans, that the cloud factor was rather a cubic function of total cloudiness as indicated in Eq Ž 15 : F s Ž 1y06c 3 F Ž 15 SW,all SW,clr The previous parameterizations use only the total cloudiness as an input variable but sometimes there is more information available, for eample the amount of low clouds Therefore, we tested as the third alternative, a new parameterization derived using radiation and cloudiness data from Jokioinen, Finland, 1997 This scheme needs the total cloudiness c, the amount of low clouds clow and the amount of Aother cloudsb coth Ž s c y c low as input variables This parameterization is hereafter denoted as the Alow cloud schemeb described in Eq Ž 16 : clow 47y4 cq10 y SW,all ž low oth/ SW,clr F s 1yc q031c q073c F Ž Radiation schemes for NWP models The downwelling surface clear-sky SW flu produced by the EC-OLD, DWD and HIRLAM schemes was evaluated in this study A brief description of the main features of these schemes in the SW region is given below More detailed documentation can be found in Morcrette Ž 1991 Ž EC-OLD, Ritter and Geleyn Ž 199 Ž DWD and Savijarvi Ž 1990 and Sass et al Ž 1994 Ž HIRLAM The SW radiation is calculated in the EC-OLD scheme and the DWD scheme using a d-two-stream approach In the EC-OLD scheme, the SW part of the spectrum is divided into two intervals, whereas the DWD scheme has three intervals Both schemes treat separately gas absorption by H O and O3 in the SW part of the spectrum The uniformly mied gases Ž CO, O, CH, N O and CO are treated as a single Ahybrid gasb in both 4 schemes Aerosols are divided into five AstandardB aerosol types both in the EC-OLD and DWD schemes; the continental type is assumed in the present study The HIRLAM scheme differs considerably from EC-OLD and DWD There is only one SW interval, and only H O is treated eplicitly whereas other gases and aerosols are added using empirical coefficients The HIRLAM scheme is much faster than the other NWP schemes, so it could be less accurate

6 146 S Niemela et alratmospheric Research 58 ( 001) The EC-OLD and DWD schemes have formal wavelength limits of and mm However, the top-of-the-atmosphere solar flu actually equals the Ž season-corrected solar constant; this is also true for the HIRLAM scheme 4 Data and measurements SYNOP observations, radio soundings and radiation measurements around 1 UTC X X X X collected from the Jokioinen Ž60849 N, 3830 E and Sodankyla Ž678 N, 6839 E observatories, Finland, were used in the comparison The Sodankyla observation site was described in Part I The climate conditions in Jokioinen are a little warmer; summers are temperate Ž July mean temperature 1588C and winters are fairly cold Ž January mean temperature y758c The annual mean precipitation amount in Jokioinen Ž 58 mm is slightly larger than in Sodankyla The area around the Jokioinen observatory consists mainly of field and forest terrain The elevation is 103 m above the sea level and the area is basically flat The comparison periods were from 1 January to 31 December 1997 and from 6 January to 13 November 1997 at Jokioinen and Sodankyla, respectively The data for Jokioinen included 6 clear-sky cases and 358 all-sky cases The data for Sodankylä included 7 clear-sky cases; cloudy conditions were not eamined Ž The downwelling and upwelling SW flues F and F SW SW and diffuse SW flu Ž D SW were measured using a Moll Gorczynski pyranometer with an estimated accuracy of "5% The calculated SW flues were compared directly with the measured hourly averaged SW flues The SYNOP observations Ž 1 UTC were made between 15 and 30 min, and the radio soundings were launched between 0 and 15 min of the same hour over which the radiation measurements were averaged The SW parameterizations considered here use similar input variables as the LW schemes presented in Part I In addition, the visually estimated amount of low clouds clow was collected for the SW all-sky calculations The conditions were considered cloudless when the visually observed total cloudiness was zero or one octas However, some of these cases were eliminated as possibly cloud-contaminated, based on a suspectibly large ratio of diffuse to total downwelling Ž SW radiation D rf SW SW compared to clear cases with similar solar elevation A total of 10 Jokioinen and 3 Sodankyla cases were ecluded Temperature, pressure and humidity profiles from radio soundings were the input data for the EC-OLD, DWD and HIRLAM schemes The ozone profiles and the vertical grid were the same as used in Part I For the SW calculations, the vertically integrated optical thickness of aerosols at ls055 mm was set to a constant value 01 in the EC-OLD and DWD schemes Ž continental aerosol assumed The surface albedo for all schemes was given by the actual measured local albedo FSW rfsw It should be noted that this is another source of uncertainty: the effect of surface albedo on the downwelling flu Ždue to the backward scattering of surface-reflected radiation is not determined by the local albedo but by the AeffectiveB albedo of a larger area The latter may differ substantially from the local albedo in horizontally heterogeneous conditions, especially in winter, when there are large albedo differences

7 S Niemela et alratmospheric Research 58 ( 001) 147 Fig 1 Aerosol optical thickness in clear-sky 1 UTC cases in 1997 The solid line gives the constant value taer s01 used in the standard set of calculations The crosses connected by dashed line give taer estimated from visibility observations via Eq Ž 18 Ž a Jokioinen Ž b Sodankyla between open snowy areas and forest A sensitivity test was made with the Iqbal Ž 1983 scheme in which the winter albedo was given a constant value 05 Žwinter albedo reduced on average by 05 This reduced FSW on average by Wrm only Two sensitivity eperiments were also made with the EC-OLD scheme In the first test, a rather large uncertainty Ž "30% of the total ozone was assumed in all input profiles The effect on FSW was only about "3 Wrm, so the typical errors of the total ozone do not affect greatly the SW irradiance In the second eperiment, the aerosol optical thickness was estimated from local visibility observations assuming that the aerosol concentration decreases eponentially upward with a scale height H s 1 km When Junge Ž 1963 standard aerosol assumption was made, the aerosol volume etincw y1 tion coefficient km, shown in Eq Ž 17, at height z wkm is approimately z baer s0ž /ž / epž y /, Ž 17 V l H where V is visibility wkm and l is wavelength wmm The optical thickness of aerosols is then obtained by integrating b over the air column as shown in Eq Ž 18 : aer /ž / ` taer sh baerd zs0ž H Ž 18 V l 0 Values of t computed from Eq Ž 18 for l s 055 mm Ž aer the reference wavelength used in the EC-OLD and DWD schemes for aerosols are plotted in Fig 1 The mean value is 010 for Jokioinen and 008 for Sodankyla, indicating slightly cleaner air in the north 5 Results Similar to the LW comparisons in Part I, the quality of the SW flu parameterizations is estimated by considering the mean difference to observations Ž bias, the standard

8 148 S Niemela et alratmospheric Research 58 ( 001) deviation around the mean difference, and the RMS difference As pointed out in Part I, these differences are not only caused by model inaccuracy, but also by inaccuracy in the input data Ž screen-level observations and radio soundings used, and by errors in the measured radiative flues used for validation 51 Clear-sky results The differences of the parameterized downwelling SW flues F SW,clr in clear-sky situations are shown in Fig The bias, standard deviation and RMS difference of the parameterized flues for Jokioinen and Sodankyla are given in Table 1 Most of the schemes overestimate the SW flu in Jokioinen, whereas all schemes underestimate the SW flu in Sodankyla This difference is discussed further below The most accurate parameterization was Iqbal s Ž 1983 scheme; its bias, RMS difference and standard deviation were the lowest of all schemes in Jokioinen The HIRLAM scheme had the lowest bias in Sodankyla, while the DWD scheme and Iqbal s Ž 1983 scheme had the smallest standard deviations Surprisingly, the parameterization used by Iqbal Ž 1983 was slightly better than the EC-OLD, DWD and HIRLAM schemes The DWD and HIRLAM schemes had almost the same values of the standard deviation, whereas for the EC-OLD scheme the standard deviation was about 5 6 Wrm higher The HIRLAM scheme was more transparent than the DWD and EC-OLD schemes The results of the simplest schemes contained much more scatter Bennett s Ž 198 scheme performed surprisingly well in Jokioinen, while Shine s Ž 1984 scheme was good in Sodankyla The good performance of Iqbal s Ž 1983 scheme could be due to the modifications by Venalainen Ž 1994, which were based on data from the same stations The net step was to remove these adjustments from Iqbal s scheme and recalculate the flues using cloudless data from Jokioinen Thus, the empirical Eq Ž 10 for transmittance by aerosols was modified so that the vertical optical thickness of aerosols was set to constant 01 Table 1 Results of the SW comparison in cloudless situations The average measured SW flu values were 378 Wrm for Jokioinen and 356 Wrm for Sodankylä Schemes Jokioinen 1997 Sodankyla 1997 Bias SD RMS Bias SD RMS Paltridge and Platt Ž 1976 y y Moritz Ž 1978 y y Bennett Ž 198 y y Zillman Ž y Shine Ž y Iqbal Ž 1983 y y EC-OLD y DWD y HIRLAM y Bias ŽWrm sparameterizedymeasured SD ŽWrm sstandard deviation RMS ŽWrm sroot-mean-square difference

9 S Niemela et alratmospheric Research 58 ( 001) 149 Fig Differences Ž Diff sparameterizationymeasurement in downwelling SW flu in cloud-free conditions Ž a Paltridge and Platt Ž 1976 Ž b Moritz Ž 1978 Ž c Bennett Ž 198 Ž d Zillman Ž 197 Ž e Shine Ž 1984 Ž f Iqbal Ž 1983 Ž g EC-OLD scheme Ž h DWD scheme Ž i HIRLAM scheme Jokioinen Ž q and Sodankylä Ž =

10 150 S Niemela et alratmospheric Research 58 ( 001) Žt s epž y01rcosu a as for the other schemes The correction of the precipitable water ŽEq Ž 1 was also removed The surface albedo was given constant values; in summer Ž May Oct 01 and in winter Ž Nov Apr 071 This AstrippedB Iqbal s scheme still outperformed the other schemes as regards with standard deviation The bias was y3 Wrm and the standard deviation became 17 Wrm Ž cf Table 1 When aerosol optical thickness from Eq Ž 18 was used in the EC-OLD scheme, the bias and standard deviation for Jokioinen were 85 and 190 Wrm, whereas those for Sodankyla were y16 and 149 Wrm Thus, the biases were slightly Žabout 3 Wrm more positive than in the AstandardB set of calculations, in which a constant value taer s 01 was assumed The differences in standard deviation were small Thus, even if the use of Eq Ž 18 brings visibility-dependent variation into the aerosol optical thickness, this did not greatly improve the calculated SW flues The above test also suggests that differences in atmospheric transparency cannot eplain the main part of the systematic difference in biases between the Jokioinen and Sodankyla results Evidently, systematic measurement errors at one or both stations contribute to this difference When using constant t aer, a value of 014 was needed to eliminate the bias of the EC-OLD scheme in Jokioinen, whereas the corresponding value was only 003 for Sodankyla This difference is substantially greater than that suggested in Fig 1 5 All-sky results The differences of the calculated SW all-sky flues are shown in Fig 3 Žfor Jokioinen, 1997 The bias, standard deviation and RMS difference of the parameterized Fig 3 Differences Ž Diff sparameterizationymeasurement in downwelling SW flu in all cases in Jokioinen, 1997 F is calculated by Iqbal s Ž 1983 scheme Ž a No cloud correction Ž b Berliand Ž 1960 Ž SW,clr c Laevastu Ž 1960 Ž d ALow cloud schemeb

11 S Niemela et alratmospheric Research 58 ( 001) 151 Table Results of the SW comparison in all-sky situations The average measured SW flu value was 88 Wrm F is calculated using Iqbal s Ž 1983 scheme SW,clr Schemes Jokioinen 1997 Ž Ž Ž Bias Wrm SD Wrm RMS Wrm No correction Berliand Ž 1960 y Laevastu Ž ALow cloud schemeb y Biassparameterizedymeasured all-sky flues are given in Table The all-sky flues were produced by multiplying the clear-sky flu by cloud-correction factors ŽEqs Ž 14 Ž 16 The clear-sky flues were calculated using Iqbal s Ž 1983 scheme The bias of Iqbal s scheme was very low Žy11 Wrm in Jokioinen so most of the biases in the all-sky calculations are caused by the cloud correction factors If a greatly biased clear-sky scheme was used, the conclusions regarding the cloud corrections would also be erroneous The all-sky SW flu is of course heavily overestimated if no cloud correction is used Ž Fig 3a The corrections improve the results but the scatter still remains very large The low cloud scheme gave the best overall results The standard deviation was smallest Ž Fig 4 Shortwave cloud correction coefficient F rf SW,all SW,clr as a function of total cloudiness The triangles represent the measured F divided by the calculated clear-sky flu F Ž Iqbal, 1983 SW,all SW,clr The lines are for the referred all-sky schemes The lines Alow cloudsb and Aother cloudsb are both related to the Alow cloud schemeb: the former gives the correction in cases in which the whole visible cloud cover consists of low clouds while the latter represents cases with middle andror high clouds only

12 15 S Niemela et alratmospheric Research 58 ( 001) using Laevastu s Ž 1960 scheme but its bias is quite large Ž190 Wrm Although the bias was smallest for Berliand s Ž 1960 scheme, the performance of this method is poor because of the large standard deviation Ž967 Wrm In Fig 4, the different cloud corrections are compared with observations as a function of total cloudiness AMeasuredB correction over unity implies reflections from cloud sides; this strong three-dimensional radiative transfer effect of the inhomogeneous sky is seen to occur occasionally at almost all cloud amounts Berliand s Ž 1960 linear scheme underestimates the all-sky flu for almost all values of the total cloudiness The curves of the low cloud scheme and Laevastu s Ž 1960 scheme follow the mean of the AmeasuredB values much better A major problem with the cloud corrections is the uncertainty in cloud optical properties, the optical thickness being the most important of these The low cloud scheme accounts implicitly for the fact that low clouds tend to be optically thicker than higher clouds This brings some advantage over the other schemes although the improvement is limited: physically, it is indeed the cloud optical thickness rather than the cloud height that is important for the downwelling solar flu 6 Conclusions The goals of this study were to evaluate the performance of several simple and some more comple radiation schemes in computing the downwelling SW radiative flues, and to find the optimum simple SW parameterization in both clear and cloudy conditions The calculated flues were compared to hourly averaged radiation observations made at the Jokioinen and Sodankyla observatories in southern and northern Finland in 1997 at 1 UTC All the simpler parameterization schemes were empirical methods, whose input variables were screen level weather observations Žeg water vapour pressure along with the solar zenith angle Comparisons were also made with three different radiation schemes used for NWP Input to these was provided by radio soundings As these schemes also need cloud water profiles, which are not observed, their comparisons were restricted to clear-sky cases only The SW clear-sky flues were estimated most accurately using the modified Ž Venalainen, 1994 parameterization of Iqbal Ž 1983 It was anticipated that this scheme would outperform all the simpler schemes Surprisingly, it was also better than the NWP schemes, which used atmospheric sounding profiles as input The DWD and HIRLAM schemes gave the net best results The clear-sky results of the EC-OLD scheme did not improve significantly when we used varying aerosol optical thickness values estimated from local visibility observations instead of the constant value 01 On the average, the atmospheric SW transmissivity was a bit larger when using optical thickness deduced from visibility The all-sky flues were calculated using simple cloud correction factors, which depend mainly on total cloudiness These cloud factors were used together with Iqbal s Ž 1983 scheme The SW all-sky flues were produced most accurately using the new Alow cloudb scheme introduced in this article ŽEq Ž 16 Laevastu s Ž 1960 scheme was nearly as good However, the new scheme may be site specific due to fact that it was

13 S Niemela et alratmospheric Research 58 ( 001) 153 derived using data from Jokioinen only The accuracy of the SW all-sky schemes was generally poor because of the large scatter in the results A major limitation of the present study is that it was made using data from two stations only Thus, the results cannot be generalized to climatic conditions significantly different from Jokioinen and Sodankyla For obtaining better geographic coverage, the comparisons should be made for a larger set of stations It could also be interesting to compare the results with ECMWF Reanalysis Ž ERA, Gibson et al, 1997 or National Center for Environmental Prediction Ž NCEP, Kalnay et al, 1996 reanalysis data sets, although it should be noted that reanalysed radiative flues are naturally model-dependent Acknowledgements We thank the Finnish Meteorological Institute for providing all the data used in this study Comprehensive comments made by an anonymous reviewer who helped to improve the original manuscript significantly Sami Niemela and Petri Raisanen have been financed by the Academy of Finland Ž Project References Bennett, TJ, 198 A coupled atmosphere sea ice model study of the role of sea ice in climatic predictability J Atmos Sci 39, Berliand, TC, 1960 Method of climatological estimation of global radiation Meteorol Gidrol 6, 9 1 Gibson, JK, Kallberg, P, Uppala, S, Hernandez, A, Nomura, A, Serrano, E, 1997 ERA description ECMWF Re-Analysis Project Rep Ser 1, European Centre for Medium-Range Weather Forecasts, Reading, United Kingdom, 66 pp Iqbal, M, 1983 An Introduction to Solar Radiation Academic Press, 390 pp Junge, CE, 1963 Air Chemistry and Radioactivity Academic Press, 38 pp Kalnay, E, Kanamitsu, M, Kistler, R et al, 1996 The NCEPrNCAR 40-year reanalysis project Bull Am Meteorol Soc 77, Laevastu, T, 1960 Factors affecting the temperature of the surface layer of the sea Commentat Phys-Math, 5 Lumb, FE, 1964 The influence of cloud on hourly amounts of total solar radiation at the sea surface Q J R Meteorol Soc 90, Morcrette, JJ, 1991 Radiation and cloud radiative properties in the European Centre for Medium-Range Weather Forecasts forecasting system J Geophys Res 96 Ž D5, Moritz, RE, 1978 A model for estimating global solar radiation Energy budget studies in relation to fast-ice breakup processes in Davis Strait Occas Pap- Univ Colorado, Inst Arct Alps Res 6, Niemela, S, Raisanen, P, Savijarvi, H, 001 Comparison of surface radiative flu parameterizations: Part I Longwave radiation Atmos Res 58, 1 18 Paltridge, GW, Platt, CMR, 1976 Radiative Processes in Meteorology and Climatology Elsevier, Amsterdam, 318 pp Ritter, B, Geleyn, J-F, 199 A comprehensive radiation scheme for numerical weather prediction models with potential applications in climate simulations Mon Weather Rev 10, Sass, BH, Rontu, L, Raisanen, P, 1994 HIRLAM- radiation scheme: Documentation and tests HIRLAM technical report 16, SMHI, Norrkoping, Sweden, 43 pp Savijarvi, H, 1990 Fast radiation parameterization schemes for mesoscale and short-range forecast models J Appl Meteorol 9,

14 154 S Niemela et alratmospheric Research 58 ( 001) Shine, KP, 1984 Parametrization of the shortwave flu over high albedo surfaces as a function of cloud thickness and surface albedo Q J R Meteorol Soc 110, Venalainen, A, 1994 The spatial variation of mean monthly global radiation in Finland Ph lic thesis, Department of Meteorology, University of Helsinki, 47 pp Zillman, JW, 197 A study of some aspects of the radiation and heat budgets of the southern hemisphere oceans Meteorological Study, vol 6, Bur of Meteorol, Dep of the Inter, Canberra, Australia