DANISH METEOROLOGICAL INSTITUTE

DANISH METEOROLOGICAL INSTITUTE TECHNICAL REPORT 00-09 Evaluation of the HIRLAM Surface Analysis Scheme for 2 metre Temperature and Relative Humidity by Comparison with AMIS Gridded Observations March 2000 Sattler K. Amstrup B. Hilden A. Hansen J. ISSN 0906-897X Copenhagen 2000

Preface This report presents a comparison of gridded temperature and humidity fields from the DMI AgroMeteorological Information System (AMIS) with short-term forecasts from the DMI- HIRLAM (High Resolution Limited Area Model) system and fields generated using the HIRLAM optimum-interpolation surface analysis scheme. The investigation was carried out at DMI in 1999 as part of a joint project with Danish agricultural organisations on IT and decision support systems in agriculture, Informatik og Beslutningsstøttesystemer i Jordbruget (INF96-1), financed in part by the Danish Ministry of Food, Agriculture and Fishery. The authors wish to thank Michael Steffensen, DMI, for assistance in preparing the manuscript. DMI, March 2000 Kai Sattler Bjarne Amstrup Anna Hilden Jakob Hansen 1

List of contents 1.... Introduction 3 1.1... General 3 1.2... Methods and Data 3 1.3...Outline 1.4... Abbreviations 1.... Map of geographical names 7 2.... The Field Types 8 2.1...AMIS 8 2.2...The Operational HIRLAM Model 9 2.3...The OI-analysis Scheme 10 3....Evaluation of Cases 11 3.1...May th, 1999: Cold-air advection and Radiative Heating 11 3.2...June 17 th, 1999: A Cold Front 21 3.3... July 28 th, 1999: Weak Flow and strong Radiative Cooling 31 3.4...Conclusions 41 4....Verification 42 4.1... Data 42 4.2... Verification Methods 42 4.3...Results 42 4.3.1... Temperature 44 4.3.2... Relative humidity 47 4.4...Conclusions 47.... Conclusions and Outlook 48 6....References 0 Appendices Appendix A: Results for AMIS Appendix B: Results for OI-analysis Appendix C: Results for HIRLAM-D 2

1. Introduction 1.1 General A central component of DMI s AgroMeteorological Information System (AMIS) is the interpolation of observed meteorological data to the 10 by 10 kilometre AMIS grid. As has been documented elsewhere (Hilden and Hansen, 1998), the AMIS observational data are generally of high quality, but do contain certain systematic errors, probably stemming in part from the quite simple, isentropic interpolation scheme used to calculate the data from the raw observed values. An obvious possibility for improving the AMIS observational fields would be to exploit the methods of a state-of-the-art numerical weather prediction system. In the analysis and initialisation parts of such a system, the three-dimensional state of the model atmosphere is adjusted towards measured values in a manner which respects and makes use of the physical and dynamical laws that govern and constrain the (model) atmosphere. Likewise, the influence of local surface characteristics (represented in the model) on the atmospheric conditions near the ground are taken into account through the model parameterisations of the interactions between the atmosphere, the sea or land surface (including vegetation), and the layers below the surface. The present study addresses the question of how and to what extent the gridding of observations in AMIS might benefit from a closer connection to the Danish operational numerical weather prediction system, the DMI-HIRLAM (High Resolution Limited Area Model) system. In the current versions of DMI-HIRLAM, pressure observations from ground stations enter into the analysis and initialisation procedures implicitly, through their (modelled) influence on the lowest model layer, but other parameters measured at ground stations situated on land are not used. However, a separate HIRLAM surface analysis package may be run on top of the analysis and forecast modules to produce two-dimensional gridded fields of 2 metre temperature and relative humidity (and, if desired, 10 metre wind) from raw observations, using HIRLAM fields as a first guess in an optimum interpolation procedure. This study is a comparative evaluation of 2 metre temperature and relative humidity fields generated by AMIS, the operational HIRLAM model and the HIRLAM surface analysis scheme. 1.2 Methods and Data The investigation was performed for a period of four months for which a homogeneous set of archived HIRLAM forecast data was available: April 21, 1999, 06 UTC, through August 19, 1999, 00 UTC. Four observation hours per day were considered: 00, 06, 12 and 18 UTC. The data sets compared were: Recalculated AMIS observation fields of 2 metre temperature and relative humidity; Archived 6-hour forecasts of these parameters generated by HIRLAM-D, the operational km resolution version of the DMI-HIRLAM model; 3

Surface analyses of the same parameters performed with the HIRLAM surface analysis scheme and with the HIRLAM-D forecasts providing the first-guess fields; 16 SYNOP and automatic climate stations were selected as verification stations (Figure 1.2.1 and Table 1.1); measurements from these were left out in the surface analysis and in the AMIS recalculations. Figure 1.2.1 Stations used in verification The evaluation was done for each station and each observation hour on monthly samples using standard meteorological verification measures. In addition to the statistical verification, a qualitative validation was carried out for three days with different characteristic weather types. 4

000 Åholm 0090 Silstrup 0170 Hald 0360 Mejdrup 000 Store Jyndevad 0640 Årslev 0770 Flakkebjerg 0890 Bønsvig Strand 0910 Abed 098 Klemensker Ø 0696 Rømø 06104 Billund 06110 Skrydstrup 06124 Tåsinge 06160 Værløse 06169 Gniben Table 1.1 Stations used in verification 1.3 Outline The report is organised as follows: Chapter 2 contains brief descriptions of the three production systems: AMIS, the DMI- HIRLAM system, and the surface analysis scheme. Chapter 3 and Chapter 4 present the results of the qualitative case studies and the statistical verification, respectively. In Chapter some central conclusions are drawn, and a look is taken at the possible directions of future work on surface analysis. References are given in Chapter 6. Detailed results of the statistical verification are compiled in an Appendix. A list of abbreviations and acronyms used throughout the report is given below. Section 1. contains a map of geographical names referred to in the text. 1.4 Abbreviations AMIS AgroMeteorological Information System, see Chapter 2.1. HIRLAM-D High Resolution Limited Area Model ( km version), see Chapter 2.2. OI-analysis HIRLAM optimum-interpolation surface analysis, see Chapter 2.3. Month no. 1 April 21-May 20 1999 Month no. 2 May 21-June 20 1999

Month no. 3 June 21-July 20 1999 Month no. 4 July 21-August 19 1999 ME Mean Error, i.e. the sum of the difference between the analysed values and the observations, divided by the number of observations. MAE Mean Absolute Error, i.e. the sum of the absolute difference between the analysed values and the observations, divided by the number of observations. RMSE Root Mean Square Error, i.e. square root of the mean squared error. HR 1 Hit Rate, i.e. the relative number of analysed value that are within +/- 1 degrees Celsius of the observed temperature. HR 2 Hit Rate, i.e. the relative number of analysed value that are within +/- 2 degrees Celsius of the observed temperature. HR Hit Rate, i.e. the relative number of analysed value that are within +/- % of the observed relative humidity. HR 10 Hit Rate, i.e. the relative number of analysed value that are within +/- 10% of the observed relative humidity. All Hit Rates are given in percent hits. ME, MAE and RMSE are in degree Celsius in tables showing temperature statistics and in percent humidity in tables showing relative humidity statistics. 6

1. Map of geographical names Figure 1..1 Map of geographical names 7

2. The Field Types 2.1 AMIS AMIS, DMI s AgroMeteorological Information System, provides farmers and other users within the Danish agricultural community with local meteorological data on a real-time basis. All numerical data are available on a 10 by 10 kilometre grid covering Danish land area. There are 632 AMIS points, or squares, in all (Figure 2.1.1). Figure 2.1.1 The AMIS grid The AMIS observational data are computed from standard meteorological observations made at SYNOP stations in Denmark, southern Sweden and northern Germany, and at Danish automatic climate stations (potential evaporation is calculated at SYNOP stations using a dedicated model, see Christensen and Sass 1994, Christensen 1996). For each AMIS square, the value of a given parameter at a given time is obtained by interpolation of the values from stations within a predefined cutoff radius. The interpolation algorithm is simple distance weighting with weights proportional to dr, where d is distance and r is a parameter dependent (negative) power. In the rare case of very poor data coverage, all available measured values from stations near Denmark are used. Certain stations known to have a bad 8

impact on the AMIS fields for one or more parameters are left out in the interpolation for these parameters. Table 2.1 gives an overview of the observational parameters which were included in AMIS in the growing season of 1999. The approximate number of measuring stations contributing to the AMIS fields for each parameter and the interpolation power and cutoff radius are also given. Parameter Time No. of stations (approx.) Interpol. radius (km) Interpol. power Description 2MT 00,03,...,21 UTC 10 60-1.7 Temperature 2 m above ground (degrees Celsius) 2MRH 00,03,...,21 UTC 10 60-1.7 Relative humidity 2 m above ground (percent) 10MFF 00,03,...,21 UTC 10 70-1.3 Wind speed 10 m above ground (m/s) 24HAT 06-06 UTC 113 80-1.4 24 hours' accumulated precipitation (mm) 24HPEV 06-06 UTC 38 80-1.4 24 hours' accumulated potential evaporation (mm) 24HGLR 06-06 UTC 19 100-2.3 24 hours' accumulated radiation ( MJ/m 2 ) Table 2.1 AMIS observed parameters 2.2 The Operational HIRLAM Model DMI's current operational forecasting system for Denmark consists of three nested models, DMI-HIRLAM-G, DMI-HIRLAM-E and DMI-HIRLAM-D, which are all versions of the HIRLAM model (see Sass 1999 for the version which was operational in spring/summer 1999). The large scale model HIRLAM-G is driven with boundary data from the global model of the European Centre for Medium-Range Weather Forecasts (ECMWF). HIRLAM- G covers a large part of the northern hemisphere with a horizontal resolution of about 0 km. It provides the boundaries to HIRLAM-E, a nested model which covers Europe and the eastern part of the North Atlantic with a resolution of about 1 km. The third nested model HIRLAM-D makes use of the HIRLAM-E forecasts at its boundaries and covers Denmark with a resolution of approximately km. HIRLAM-D is applied four times a day with boundaries from HIRLAM-E for the times 00, 06, 12 and 18 UTC. A data assimilation is carried out when the model is started, which introduces observations from within the domain of HIRLAM-D. During the forecast run, an update of the boundaries is performed using HIRLAM-E forecasts valid every hour. The forecast length is 36 hours for the forecasts started at 00 and 12 UTC. For those started at 06 and 18 UTC, the forecast length is limited to 6 hours. Among the numerous predicted parameters are fields for temperature and relative humidity at 2 m above the ground level. 9

As the quality of the forecast slowly decreases with growing forecast length, the most reliable forecasts are obtained within the first 12 hours of the forecast. Concerning the analysis applied in the work of this project, the respective 6 hour forecasts from the operational HIRLAM-D runs were used. 2.3 The OI-analysis Scheme The analysis scheme for the 2 m temperature and relative humidity is part of the international HIRLAM reference system. It is based on the Optimum Interpolation method (OI) and takes into account the land-sea contrast when correlating the observations as well as the distance of the observations to the analysis point (Navascues 1997). It is not to be confused with the operational 3-D analysis scheme, which also uses OI. The starting point for the analysis is a first guess field, which in this case is given by the HIRLAM-D forecast. It contains a prescribed uncertainty in the temperature and humidity data, which is given in terms of the standard deviation. The second important input to the analysis are observations of temperature and humidity from the domain that is covered by HIRLAM-D. There is also an uncertainty connected to the measurements. The quality of each observation is tested in order to prevent that erroneous measurements influence the analysis. The OI method combines the first guess field and the observations by minimising the analysis error. This means that the resulting uncertainty of the analysis is less than both the uncertainty from the first guess and that from the observations (Daley 1991). In order to limit the necessary computer resources, the analysis is performed in "boxes", i.e. subdomains of the HIRLAM-D domain. The domain of each box overlaps with the neighbouring boxes to assure consistency at the box boundaries (Lorenc 1981). The analysis was applied as follows: The first guess fields for 2 m temperature and 2 m relative humidity were provided from 6 hour forecasts of the operational HIRLAM-D with a horizontal resolution of km. Their uncertainty was estimated to 2 K for the temperature and 22 % for the relative humidity. The estimated uncertainties for the observations were 0. K and 10 % respectively. The resulting analysis was determined on a grid with the same geometry as that of the first guess field. 10

3. Evaluation of Cases 3.1 May th, 1999: Cold-air advection and Radiative Heating The weather situation at the th of May 1999 was characterised by a high pressure system over Scandinavia. The anticyclone influenced also Denmark and the North Sea and lead to steady easterly winds on a large synoptical scale (Figure 3.1.1). Apart from some cirrus clouds there were only little amounts of cloudiness over Denmark this day. The temperature curve showed a significant amplitude due to the steady irradiation from the sun, but this did not result in a significant development of sea-breeze winds because of quite strong synoptic flow. The easterly winds transported relatively cold air over Denmark. This lead to a strong temperature gradient over Denmark. If we take a look at the observations in Figure 3.1.2, we can see low temperatures of about 6-7 C in the south eastern part of Zealand and on the east coast of Falster and Møn, cf the map section 1.. At the west coast of Jutland the temperature reaches 16 o C. This gradient is represented by both AMIS and the OI-analysis (Figure 3.1.3 and Figure 3.1.4). It is also found in the HIRLAM-D forecast, where it is weaker (Figure 3.1.). At this point it can be clearly seen how the OI-analysis improves the HIRLAM-D forecast by including actual observations. There is no fine detail in the field from AMIS. The OI-analysis, however, shows a more detailed structure of the gradient. The cool areas over the Baltic sea and the Kattegat have an influence on the temperatures near the east coasts due to the steady easterly wind. The observations on Bornholm and Møn as well as other coastal stations (Århus Havn, Sprogø Øst and Omø) indicate this, and the OI-analysis represents the effect very well. This is mainly due to the good forecast of Hirlam-D. The temperature gradient over Bornholm is also better represented in the OI-analysis than in AMIS. Concerning relative humidity, the general picture shows a gradient with decreasing humidity towards the west. However, it is not as clear as in the temperature field. The observations show very dry areas over Jutland for example (Figure 3.1.6). AMIS represents the low humidity quite well, whereas the OI-analysis gives higher values of humidity over Jutland (Figure 3.1.7 and Figure 3.1.8). Over Lolland, both AMIS and the OI-analysis show higher humidity than the observations. This has different reasons. In AMIS, which calculates interpolated values from the observations supplied to it, the station Abed on Lolland (% relative humidity) is not included in the calculations. Thus AMIS has not enough information about the humidity conditions over Lolland. On the other hand, the forecast of relative humidity from the HIRLAM-D forecast for this area is better than the OI-analysis (Figure 3.1.9). The performance of the OI-analysis is rather poor in this case probably due to the same lack of measurements from Lolland and also due to the influence of analysed humidity outside of Denmark. Over Southern Sweden, for example, the HIRLAM-D forecast shows very low values of relative humidity. These are raised during the analysis and this also has an influence on other areas. 11

Over Bornholm, the humidity gradient is better represented in AMIS than in the OI-analysis. The gradient given in the OI-analysis is stronger than observed. The gradient from the HIRLAM-D forecast is, however, even stronger. This means that there is an improvement through the analysis, although it is not perfect. Figure 3.1.1 Observed 10 m wind and clouds, May th, 1999, 12 UTC 12

Figure 3.1.2 Observed 2 m temperature, May th, 1999, 12 UTC 13

-2 2-3 3-4 4- -6 6-7 7-8 8-9 9-10 10-11 11-12 12-13 13-14 14-1 1-2 m. temperature AMI 19990012-273 -273 7N -273 6N -273 N 8E 10E 12E 14E Figure 3.1.3 AMIS temperature analysis, May th, 1999, 12 UTC 14

-2 2-3 3-4 4- -6 6-7 7-8 8-9 9-10 10-11 11-12 12-13 13-14 14-1 1-2 m. temperature ANA 19990012 7N 6N N 8E 10E 12E 14E Figure 3.1.4 OI-analysis of temperature, May th, 1999, 12 UTC 1

-2 2-3 3-4 4- -6 6-7 7-8 8-9 9-10 10-11 11-12 12-13 13-14 14-1 1-2 m. temperature D0 19990012 7N 6N 4 4 44 N 4 8E 10E 12E 14E Figure 3.1. HIRLAM-D temperature forecast, May th, 1999, UTC 16

Figure 3.1.6 Observed 2 m relative humidity, May th, 1999, UTC 17

0-1 1-30 30-3 3-40 40-4 4-0 0- -60 60-6 6-70 70-7 7-80 80-8 8-90 90-100 2 m. rel. humidity AMI 19990012 0 0 7N 0 6N 0 0 0 N 0 8E 10E 12E 14E Figure 3.1.7 AMIS relative humidity analysis, May th, 1999, UTC 18

0-1 1-30 30-3 3-40 40-4 4-0 0- -60 60-6 6-70 70-7 7-80 80-8 8-90 90-100 2 m. rel. humidity ANA 19990012 7N 6N N 8E 10E 12E 14E Figure 3.1.8 OI-analysis of relative humidity, May th, 1999, UTC 19

0-1 1-30 30-3 3-40 40-4 4-0 0- -60 60-6 6-70 70-7 7-80 80-8 8-90 90-100 2 m. rel. humidity D0 19990012 7N 6N N 8E 10E 12E 14E Figure 3.1.9 HIRLAM-D forecast of relative humidity, May th, 1999, UTC 20

3.2 June 17 th, 1999: A Cold Front The synoptic situation of this day is determined by a cold trough located over the North Sea at 12 UTC. It carried cold air eastwards, especially at the lower levels of the atmosphere. During the day-time, a cold front passed over Denmark. This changed the temperature and humidity conditions through the day. The cold front reached the north western part of Jutland at approximately 10:30 UTC and passed across the country during the following hours. At 1 UTC it was located over Funen and it reached the Sound at approximately 18 UTC. The position of the cold front at 12 UTC is shown in Figure 3.2.1. Figure 3.2.2 and Figure 3.2.3 show the observations of 2 m temperature and relative humidity respectively. The air mass behind the front is approximately C colder and relative humidity is about 20% higher than before the front. The OI-analysis locates the front along the line Esbjerg-Ålborg (Figure 3.2.4). This position is mainly prescribed by the HIRLAM-D forecast (Figure 3.2.). Compared with the actual frontal position, the tilt of the frontal line appears a little steeper than the actual tilt in the map of the HIRLAM-D forecast and thus also in the OI-analysis. In AMIS, the course of the frontal line is shifted eastwards in Northern Jutland, so that even Læsø seems to be passed by the front already at 12 UTC (Figure 3.2.6). Contrary to the OIanalysis, AMIS shows a more gentle tilt of the frontal line than the actual one. The frontal position in the OI-analysis is determined by the forecast, whereas AMIS uses the available observations, and as there are no observations available at the coast north of Djursland for 12 UTC, AMIS just interpolates between the existing surrounding observations, thus getting a wrong frontal position. Concerning the temperature gradient within the frontal region, the one given by the HIRLAM-D forecast appears too strong, which seems to be due to an overestimation of the temperature over land. This is weakened a little in the analysis. AMIS shows a weaker gradient seeming to be more realistic. The same applies to the gradient of relative humidity. Over Lolland and Funen, AMIS shows lower temperatures than observed. This is due to the fact that three stations used for verification do not go into the calculations of AMIS. The coincidence between the OI-analysis and the actual temperatures over Lolland is very good. There is one erroneous relative humidity observation: One station (Firhøje) in West Jutland shows 34%, whereas all surrounding stations measure much higher values. As this is not taken into account in AMIS, the erroneous measurement appears as a dry spot over West Jutland, which seems rather unrealistic (Figure 3.2.7). The erroneous observation was not used in the OI-analysis, because the analysis includes a check for observation errors (Figure 3.2.8). The relative humidity over Southwest Zealand is overestimated by the OI-analysis, because there is a strong SW-NE gradient already in the HIRLAM-D forecast, which is not that strong in the observations. The HIRLAM-D forecast gives a good estimation for humidity over Lolland (Figure 3.2.9). This is, like in the case from the th of May, deteriorated in the analysis because of too strong influences from other regions like for example South Sweden. AMIS does not show the dry area over Lolland, because one station here (Abed) was not used in the calculation of AMIS. 21

Figure 3.2.1 Observed wind, clouds, pressure tendency and weather, June 17 th, 1999, 12 UTC. Approximate position of the cold front is indicated. 22

Figure 3.2.2 Observed 2 m temperature, June 17 th, 1999, 12 UTC 23

Figure 3.2.3 Observed 2 m relative humidity, June 17 th,1999, 12 UTC 24

-10 10-11 11-12 12-13 13-14 14-1 1-16 16-17 17-18 18-19 19-20 20-21 21-22 22-23 23-2 m. temperature ANA 1999061712 12 7N 12 6N 12 12 N 12 8E 10E 12E 14E Figure 3.2.4 OI-analysis of temperature, 17 th, 1999, 12 UTC 2

-10 10-11 11-12 12-13 13-14 14-1 1-16 16-17 17-18 18-19 19-20 20-21 21-22 22-23 23-2 m. temperature D0 1999061712 7N 6N N 8E 10E 12E 14E Figure 3.2. HIRLAM-D forecast of temperature, June 17 th, 1999, 12 UTC 26

-10 10-11 11-12 12-13 13-14 14-1 1-16 16-17 17-18 18-19 19-20 20-21 21-22 22-23 23-2 m. temperature AMI 1999061712-273 -273 7N -273 6N -273 N 8E 10E 12E 14E Figure 3.2.6 AMIS temperature analysis, June 17 th, 1999, 12 UTC 27

0-1 1-30 30-3 3-40 40-4 4-0 0- -60 60-6 6-70 70-7 7-80 80-8 8-90 90-100 2 m. rel. humidity AMI 1999061712 0 0 7N 0 6N 0 0 0 N 0 8E 10E 12E 14E Figure 3.2.7 AMIS analysis of relative humidity, June 17 th, 1999, 12 UTC 28

0-1 1-30 30-3 3-40 40-4 4-0 0- -60 60-6 6-70 70-7 7-80 80-8 8-90 90-100 2 m. rel. humidity ANA 1999061712 7N 6N N 8E 10E 12E 14E Figure 3.2.8 OI-analysis of relative humidity, June 17 th, 1999, 12 UTC 29

0-1 1-30 30-3 3-40 40-4 4-0 0- -60 60-6 6-70 70-7 7-80 80-8 8-90 90-100 2 m. rel. humidity D0 1999061712 7N 6N N 8E 10E 12E 14E Figure 3.2.9 HIRLAM-D forecast of relative humidity, June 17 th, 1999, 12 UTC 30

3.3 July 28 th, 1999: Weak Flow and strong Radiative Cooling The weather situation was characterised by a large high pressure system that was located over the British Isles. The pressure gradient over Denmark was weak and there was hardly any large scale flow. There was clear sky during the night between 27th and 28th of July (Figure 3.3.1). So, radiative emission was strong, leading to a large amplitude in the diurnal course of temperature. The distinct radiative conditions thermally induced land-sea-breeze circulations at the coastlines of Denmark. They dominated the wind field over Denmark at this time. This had an influence on the temperature and humidity fields near the coast. As the night during this season is rather short and sunrise already occurs at about 03 UTC, the fields at 00 UTC were taken in order to compare the conditions at night time. As can be seen in the temperature observations, there was a distinct difference between temperature over sea and over land. Relatively high temperatures of 12 to 1 C occur at the coastal stations, whereas the land stations show temperatures below 10 C (Figure 3.3.2). This temperature gradient is well represented in the OI-analysis, where it originates in the HIRLAM-D forecast (Figure 3.3.3 and Figure 3.3.4). However, temperatures over land are represented higher in the OI-analysis than observed. It seems that there is too little structure in the temperature field of the OI-analysis over land this time (Figure 3.3.3). AMIS represents temperature over Jutland and Zealand quite well, but the temperature over Lolland does not seem realistic (Figure 3.3.). This deviation is due to the lack of observations from inner Lolland. The same effect occurs over Falster, Læsø and to some extend over Funen. The field over Bornholm is represented very well in the OI-analysis, whereas AMIS cannot cope with the strong temperature gradient at the coast. The difference in temperature between the OI-analysis and the observations shall be discussed a little further. It seems that in this case the difference between the forecast field and the observations is so strong that the analysis has difficulties in calculating a better solution. The error in the forecast fields changes in space and time. In the analysis, however, it is treated as a constant, because it can hardly be determined. Thus, there may occur cases like this one where the error is underestimated, leading to poor analysis quality over some regions. The relative humidity during the night between the 27th and 28th of July is determined by high values of over 90% (Figure 3.3.6). As there is no significant spatial structure, the representation in AMIS is very good (Figure 3.3.7). But one must be careful on islands where there are no observations, like Læsø. AMIS is not reliable there. Contrary to the observations, the HIRLAM-D forecast and thus also the OI-analysis show a structure in the humidity field (Figure 3.3.8 and Figure 3.3.9). It is difficult to say something about the quality of these structures, but there seems to be a slight negative bias in the HIRLAM-D forecast. This bias is larger in the OI-analysis. The analysis of relative humidity does not really seem to work well in this case. 31

Figure 3.3.1 Observed wind, weather and cloudiness, July 28 th, 1999, 00 UTC 32

Figure 3.3.2 Observed 2 m temperature, July 28 th,1999, 00 UTC 33

-8 8-9 9-10 10-11 11-12 12-13 13-14 14-1 1-16 16-17 17-18 18-19 19-20 20-21 21-2 m. temperature ANA 1999072800 11 11 7N 11 6N 11 N 8E 10E 12E 14E Figure 3.3.3 OI-analysis of temperature, July 28 th, 1999. 00 UTC 34

-8 8-9 9-10 10-11 11-12 12-13 13-14 14-1 1-16 16-17 17-18 18-19 19-20 20-21 21-2 m. temperature D0 1999072800 7N 6N N 8E 10E 12E 14E Figure 3.3.4 HIRLAM-D forecast of temperature, July 28 th, 1999, 00 UTC 3

-8 8-9 9-10 10-11 11-12 12-13 13-14 14-1 1-16 16-17 17-18 18-19 19-20 20-21 21-2 m. temperature AMI 1999072800-273 -273 7N -273 6N -273 N 8E 10E 12E 14E Figure 3.3. AMIS temperature analysis, July 28 th, 1999, 00 UTC 36

Figure 3.3.6 Observed relative humidity, July 28 th, 1999, 00 UTC 37

0-1 1-30 30-3 3-40 40-4 4-0 0- -60 60-6 6-70 70-7 7-80 80-8 8-90 90-100 2 m. rel. humidity AMI 1999072800 0 0 7N 0 6N 0 0 0 N 0 8E 10E 12E 14E Figure 3.3.7 AMIS analysis of relative humidity, July 28 th, 1999, 00 UTC 38

0-30 30-3 3-40 40-4 4-0 0- -60 60-6 6-70 70-7 7-80 80-8 8-90 90-9 9-100 2 m. rel. humidity D0 1999072800 7N 6N N 8E 10E 12E 14E Figure 3.3.8 HIRLAM-D forecast of relative humidity, July 28 th, 1999, 00 UTC 39

0-30 30-3 3-40 40-4 4-0 0- -60 60-6 6-70 70-7 7-80 80-8 8-90 90-9 9-100 2 m. rel. humidity ANA 1999072800 7N 6N N 8E 10E 12E 14E Figure 3.3.9 OI-analysis of relative humidity, July 28 th, 1999, 00 UTC 40

3.4 Conclusions The case studies show some characteristics of both AMIS and the analysis. AMIS completely relies on the observations and does not perform any error check. Thus, the quality of the fields depends mainly on the number of observations and the quality of each of them. Structures which are not caught by the observations are not caught in AMIS either. But the method is straightforward and fast. The analysis of 2 m temperature and relative humidity makes use of HIRLAM forecasts and observations and combines these in an optimised way. But the cases show that especially the analysis of relative humidity does not always work satisfactorily. There is especially a tendency towards deterioration of the bias when the forecast already shows high values of relative humidity. On the other hand the forecast is able to catch strong gradients and other structures in more detail than observations can (see the cases from th of May and 28 th of July), and these are also reflected in the analysis. The analysis is better when temperature and humidity are determined by processes on the synoptic scale. This is presumably due to the fact that the first guess error correlations are determined for a large scale similar to the synoptic one. Thus, the analysis may lead to deterioration during weather conditions where local influences are dominant, like the case from 28 th of July. The local conditions are actually represented in a more detailed manner in a model with higher resolution. This means that the occurrence of local effects has to be taken much more into account during the analysis. However, there are no investigations yet about the changes in the scale of the forecast error correlations when going to higher resolution, and the inhomogeneity of the forecast error also still plays a role. But further improvements in the analysis procedure with special concentration on these points seem to be feasible in the future. 41

4. Verification 4.1 Data For the objective verification data from the whole verification period were used, i.e. April 21 to August 19, 1999. The verification period was divided into four sub-periods, each of about one months length, cf. 1.4. For both temperature and relative humidity, and for each of the three field types (AMIS, OIanalysis, HIRLAM-D), the field data were stratified according to month and observation hour, and verification was performed by comparing measured values from each of the 16 verification stations (see Table 1.1) to grid point values; for AMIS, the value at the grid square to which the station belonged was used, for the other field types a value was interpolated from the four nearest grid points. 4.2 Verification Methods The following verification measures were computed for each sample (cf. 1.4): ME, MAE, RMSE, and, in addition, HR1 and HR2 for temperature, HR and HR10 for relative humidity. Summaries were produced in the form of means for all stations and/or months and/or observation hours. In addition, time series for temperature at selected stations were inspected. 4.3 Results Table 4.1 shows verification results for temperature and humidity for the whole period and each of the four observation hours, averaged over all 16 stations. Results for all data samples can be found in the Appendix. Below we comment on the results for each parameter separately. 42

T e m pe ra tu re R e la tiv e H u m id ity A ll M on th s A ll M on th s 00 U T C M E M AE R M SE HR 1 HR 2 00 U T C M E M AE R M SE HR HR 10 A M IS 0.3 1.03 1.89 9.9 86.21 A M IS -2.07 3.98.30 70.76 93.33 O I-ana ly s is 0.38 1.12 1.43.93 83.38 O I-ana ly s is -4.39 6.26 8.07.77 79.0 H IR L A M -D -0.0 1.27 1.8 47.77 78.68 H IR L A M -D -0.10 4.79 6.47 67.2 87.6 T e m pe ra tu re R e la tiv e H u m id ity A ll M on th s A ll M on th s 06 U T C M E M AE R M SE HR 1 HR 2 06 U T C M E M AE R M SE HR HR 10 A M IS 0.10 0.69 0.92 79.46 9.26 A M IS -2.0 4.28.46 68.4 90.66 O I-ana ly s is -0.10 0.84 1.11 69.91 92.74 O I-ana ly s is -1.43.2 6.8 60.72 86.70 H IR L A M -D -0.73 1.14 1.44 4.8 84.18 H IR L A M -D -3.40.64 7.23 7.04 82.03 T e m pe ra tu re R e la tiv e H u m id ity A ll M on th s A ll M on th s 12 U T C M E M AE R M SE HR 1 HR 2 12 U T C M E M AE R M SE HR HR 10 A M IS -0.17 0.94 1.26 67.2 89.39 A M IS 0.97 6.32 8.29 3.90 81.42 O I-ana ly s is -0.01 1.21 1.8 3.89 81.94 O I-ana ly s is 0.21 7.88 10.07 42.11 69.80 H IR L A M -D 0.01 1.40 1.80 44.0 76.06 H IR L A M -D -6.34 9.62 11.81 32.14 9.62 T e m pe ra tu re R e la tiv e H u m id ity A ll M on th s A ll M on th s 18 U T C M E M AE R M SE HR 1 HR 2 18 U T C M E M AE R M SE HR HR 10 A M IS 0.11 0.68 0.93 79.66 9.19 A M IS -0.18.46 7.3 61.41 84.91 O I-ana ly s is 0.20 0.93 1.24 6.67 90.37 O I-ana ly s is -0.38 7.69 10.1 44.0 72.39 H IR L A M -D 0.06 1.11 1.44 6.1 84.70 H IR L A M -D -4. 8.76 11.24 38.98 6.27 T a b le 4.1 Su mm a r y o f v er ific a tio n s ta tis tic s 43

4.3.1 T e m p er a tu re L ook in g firs t a t th e s u mm a ry fig u re s o f T a b le 4.1, w e s ee th a t th e ov e ra ll qu a lity o f th e a n a ly s e d te m p e ra tu re fie ld s is qu ite h ig h fo r a ll th ree a n a ly s is m e thod s, w ith M A E v a lu e s a round 1 C a nd R M S E v a lu e s a lw a y s b e lo w 2 C. F o r m o s t c o m b in a tion s o f a n a ly s is m e thod a nd ob s e rv a tion tim e th e b ia s is b e lo w 0.2 C b y a b s o lu te v a lu e, a lthou g h th e re a re e x ce p tion s to th is. B e tw ee n 76 a nd 9 p e rce n t o f th e v e rifie d fie ld v a lu e s fe ll w ith in 2 C fr o m t h e v e rify in g ob s e rv a tion. A s jud g e d b y M A E, R M S E a nd H R v a lu e s, th e A M IS m e thod is fo r a ll fou r ob s e rv a tion hou rs b e tte r th a n th e O I-a n a ly s is, w h ic h is in tu rn b e tte r th a n H IR L A M -D. T hu s, it s ee m s th a t - a t lea s t in th is s e tup - th e O I m e thod g iv e s too m u c h w e ig h t to th e H IR L A M firs t- g u e ss fie ld a nd too littl e to th e ob s e rv a tion s. A ll a n a ly s is m e thod s p e rf o rm b e tte r a t 06 a nd 18 U T C th a n a t 00 a nd 12 U T C. T h is is p re s u m a b ly b eca u s e th e la tte r v a lu e s te nd to b e m o re e x tre m e. T h e re s u lts s ho w n in th e A pp e nd ice s (e.g., p a g e s A 10, B 10, C 10 ) s ho w th a t th e re is qu ite a la rg e v a ria tion b e tw ee n s ta tion s in th e qu a lity o f th e a n a ly s e s. T h is ho ld s fo r a ll th ree m e thod s a nd ob s e rv a tion hou rs, thou g h m o s t p ro mi n e n tly fo r A M IS a t 00 a nd 12 U T C. T h e fo llo w in g e x a m p le illu s tra te s th e fac to rs th a t m a y in flu e n ce th e qu a lity o f th e a n a ly s e s a t eac h s ta tion : F ig u re 4.3.1 a nd F ig u re 4.3.2 s ho w tim e s e rie s o f m ea s u re d a nd a n a ly s e d te m p e ra tu re fo r tw o s ta tion s, 06104 B illund a nd 0890 B øn s v ig S tra nd. T h e fo rm e r is a n in la nd s ta tion, th e la tte r is s itu a te d b y th e s ea (F ig u re 1.2.1 ). It is e v id e n t th a t a ll th ree a n a ly s is s y s te m s p e rf o rm th e b e tte r a t th e in la nd s ta tion w h e re th e te m po ra l v a ria b ility is le ss th a n a t th e c o a s t. F o r bo th s ta tion s th e A M IS c u rv e fo llo w s th e ob s e rv a tion s c lo s e r th a n th e c u rv e s o f th e o th e r a n a ly s is s y s te m s ; in p a rtic u la r, th e p ea k s a nd trou g h s in th e ob s e rv a tion s e rie s a re b e tte r ca p tu re d. T h e A M IS s e rie s fo r B illund fits th e ob s e rv a tion s v e ry w e ll ind ee d. N o c lea r d iff e re n ce s a re found b A 12, B 12, C 12 ). e tw ee n th e re s u lts fo r th e fou r m on th s (A pp e nd ice s, p a g e s 44

S ta tion no. 0890 30 2 O I a n a ly s is H IR L A M -D A M IS O b se r va tion 20 1 Temperature 10 0 1/8 1999 22/7 1999 20/6 1999 19/ 1999 21/4 1999 D a y F ig u re 4.3.1 T im e s er ie s o f m e a s u re d a nd a n a ly s e d te m p er a tu re fo r s ta tio n 0890 B ø n s v ig S tr a nd 4

1/8 1999 S ta tion no. 6104 3 30 2 O I-a n a ly s is H IR L A M -D A M IS O b se r va tion 20 1 Temperature 10 0 22/7 1999 20/6 1999 19/ 1999 21/4 1999 D a y F ig u re 4.3.2 T im e s er ie s o f m e a s u re d a nd a n a ly s e d te m p er a tu re fo r s ta tio n 6104 B illund 46

4.3.2 R e la tiv e hu m id ity F o r r e la tiv e hu mi d ity, th e g e n e ra l qu a lity a s d e s c rib e d b y th e fig u re s T a b le 4.1 is g ood, bu t no t e x ce lle n t. M A E v a lu e s o f th e th ree m e thod s a nd th e fou r ob s e rv a tion hou rs ra n g e fr o m 4 to 10 %, a nd fr o m 60 to 93 p e rce n t o f th e a n a ly s e d v a lu e s a re c o rr ec t w ith in 10 %. A ll m e thod s p e rf o rm l ea s t w e ll a t 12 U T C, w h e re - p re s u m a b ly - th e ho riz on ta l v a ria b ility o f th e re la tiv e hu mi d ity te nd s to b e la rg e. E x ce p t a t 00 U T C w h e re H IR L A M -D ou tp e rf o rm s th e O I-a n a ly s is, th e A M IS fie ld s h a v e a h ig h e r qu a lity th a n th e O I fie ld s w h ic h a re b e tte r th a n th e H IR L A M fie ld s. T hu s, on ce a g a in, th e O I-a n a ly s is do e s in g e n e ra l im p rov e on th e H IR L A M firs t g u e ss b y a d ju s tin g to w a rd s m ea s u re d v a lu e s, bu t A M IS is th e b e s t. It s hou ld b e no te d h e re th a t s o m e 10 s ta tion s w h ic h g o in to th e A M IS g ridd in g, fo r tec hn ica l rea s on s a re no t u s e d in th e O I-a n a ly s is ; th is mi g h t e x p la in p a rt o f th e d iff e re n ce s in qu a lity. A s fo r te m p e ra tu re, th e re a re qu ite la rg e d iff e re n ce s fr o m s ta tion to s ta tion in ho w w e ll th e a n a ly s e d d a ta fit th e ob s e rv e d v a lu e s (A pp e nd ice s, p a g e s A 11, B 11, C 11 ). A M IS w a s b e tte r in m on th s 2 a nd 3 th a n in m on th s 1 a nd 4 ; H IR L A M -D a nd th e O I-a n a ly s is w e re b e tte r in m on th s 1 a nd 2 th a n in th e re m a in in g p a rt o f th e v e rifica tion p e riod (A pp e nd ice s, p a g e s A 13, B 13, C 13 ). 4.4 C o n c lu s io n s T o s u m up, th e v e rifica tion s ho w s th a t a ll th ree m e thod s p e rf o rm s a tis fac to rily fo r bo th te m p e ra tu re a nd re la tiv e hu mi d ity, a lthou g h th e re s u lts fo r te m p e ra tu re a re b e tte r th a n tho s e fo r hu mi d ity. T h e O I s u rf ace a n a ly s is s c h e m e im p rov e s on th e H IR L A M firs t-g u e ss, bu t s till th e s im p le A M IS in te rpo la tion w ith g round ob s e rv a tion s a s its s o le d a ta s ou rce a nd no bu iltin ph y s ic s o r in fo rm a tion on s u rf ace c h a rac te ris tic s do e s a b e tte r job in th e m ea n. 47

. C o n c lu s io n s a nd O u tloo k In th e p re s e n t s tud y th e qu a lity o f A M IS g ridd e d fie ld s o f 2 m e tre te m p e ra tu re a nd re la tiv e hu mi d ity h a s b ee n c o m p a re d to th a t o f th e c o rr e s pond in g fie ld s p rodu ce d b y th e op e ra tion a l D M I-H IR L A M -D m od e l a nd th e H IR L A M O p tim u m In te rpo la tion s u rf ace a n a ly s is s c h e m e u s in g th e H IR L A M -D fie ld s a s firs t g u e ss. T h e inv e s tig a tion h a s b ee n ca rr ie d ou t th rou g h ca s e e v a lu a tion a nd s ta tis tica l v e rifica tion. T h e imm e d ia te c on c lu s ion s fr o m bo th th e ca s e s tud ie s a nd th e v e rifica tion a re th a t a ll th ree a n a ly s is m e thod s w o rk rea s on a b ly w e ll, a lthou g h b e tte r f o r te m p e ra tu re th a n fo r r e la tiv e hu mi d ity. A s m ea s u re d b y th e s ta tis tic s, A M IS p e rf o rm s b e s t fo r bo th p a ra m e te rs, fo llo w e d b y th e O I-a n a ly s is. T h e ca s e e v a lu a tion s uppo rts th is c on c lu s ion a nd a ls o po in ts to s o m e im po rta n t s tre n g th s a nd w ea kn e ss e s o f th e th ree m e thod s : T h rou g h th e ca s e s it h a s b ee n illu s tra te d ho w th e u s e o f a h ig h -r e s o lu tion nu m e rica l w ea th e r p re d ic tion s y s te m i n th e a n a ly s is ca n c on fe r f ea tu re s on s ca le s w h ic h a re no t re s o lv e d b y th e ob s e rv a tion s to th e fie ld s, fea tu re s w h ic h in th e ca s e s inv e s tig a te d h e re g e n e ra lly s ee m ph y s ica lly rea lis tic. T h e hu mi d ity fie ld s fr o m H IR L A M a nd th e O I-a n a ly s is m a y a t tim e s b e qu ite poo r. T h e e x a m p le s h a v e a ls o s ho w n th a t th e qu a lity o f th e A M IS fie ld s is v e ry m u c h d e p e nd e n t on th e qu a lity a nd a v a ila b ility o f s in g le ob s e rv a tion s ; in p a rtic u la r, s in ce on ly a c o a rs e c lim a to lo g ica l c h ec k is p e rf o rm e d on th e inpu t v a lu e s, A M IS is v e ry p ron e to ob s e rv a tion e rr o rs w h ic h m a y ru in s in g le fie ld s lo ca lly. H o w ca n it b e th a t in c rea s in g th e g rid re s o lu tion a nd a dd in g s oph is tica te d ph y s ic s a nd d y n a mi c s, s u rf ace ph y s io g ra ph y a nd a n a ly s is tec hn iqu e s to th e g ridd in g p ro ce du re ac tu a lly im p a irs th e qu a lity o f th e re s u ltin g fie ld s? P a rt o f th e e x p la n a tion lie s p rob a b ly w ith th e fac t th a t th e O I-a n a ly s is s c h e m e is tun e d to p e rf o rm w e ll in a c lim a te (th e S p a n is h ) w h ic h is m o re c on tin e n ta l a nd le ss ho riz on ta lly v a ria b le th a n th e D a n is h on e, a nd w ith a fo reca s t m od e l h a v in g a lo w e r r e s o lu tion th a n th a t o f D M I-H IR L A M -D. M u c h c ou ld undoub te d ly b e g a in e d fr o m t un in g a nd op timi s in g th e O I p ro ce du re fo r D a n is h a pp lica tion s. T h e qu e s tion po s e d b ea rs upon a ce n tra l iss u e in nu m e rica l w ea th e r p re d ic tion, n a m e ly th a t o f th e re la tion b e tw ee n, on on e h a nd, in c rea s e d s p a tia l a nd te m po ra l re s o lu tion in th e m od e ls a nd th e re s u ltin g s h a rp e r a nd m o re v a ria b le m od e l fie ld s - a nd, on th e o th e r, th e qu a lity o f th e s e fie ld s a s m ea s u re d a t s in g le po in ts. In g e n e ra l, la rg e r v a ria b ility w ill te nd to g iv e poo re r po in t s ta tis tic s, e v e n w h e n th e v a ria b ility is ph y s ica lly rea lis tic. T h e w a y ou t o f th is d ile mm a is p rob a b ly no t to g iv e up w o rk in g on in c rea s e d m od e l s h a rpn e ss a nd re s o lu tion, bu t, ra th e r, to w o rk in p a ra lle l on s c ie n tifica lly w e ll-f ound e d w a y s o f qu a n tify in g th e e x p ec te d m od e l e rr o rs, e.g., b y th e u s e o f e n s e m b le p re d ic tion tec hn iqu e s. L ook in g a b it a h ea d in tim e, it is c lea r th a t s ub s ta n tia l im p rov e m e n ts in A M IS, e s p ec ia lly a s re g a rd s ho riz on ta l v a ria tion s on s ca le s no t re s o lv e d b y th e ra w ob s e rv a tion s, w ill h a v e to b e m a d e th rou g h m o re in te n s iv e u s e o f a nu m e rica l w ea th e r p re d ic tion s y s te m. O f in te re s t h e re 48

is th e c h a n g e, p la nn e d fo r 2000, in th e op e ra tion a l D M I-H IR L A M d a ta a ss imil a tion s y s te m to a 3 D v a ria tion a l s c h e m e. T h is w ill a llo w n e w d a ta ty p e s to b e in c lud e d in th e a ss imil a tion, m o s t im po rta n tly s e v e ra l n e w ty p e s o f s a te llite d a ta, a nd w ill a ls o g iv e a b e tte r u s e o f th e a lrea d y in g o in g d a ta ty p e s. S u rf ace ob s e rv a tion s fr o m S YNO P, a u to m a tic c lim a te, a nd ro a d w ea th e r s ta tion s o f p a ra m e te rs lik e te m p e ra tu re, re la tiv e hu mi d ity a nd, a t a la te r s ta g e, po ss ib ly e v e n p rec ip ita tion a m oun t a nd in te n s ity, w ill e n te r d irec tly in to th e a ss imil a tion. O f s p ec ia l in te re s t to th e a g ric u ltu ra l c o mm un ity is th e po ss ib ility o f im p rov in g th e re p re s e n ta tion o f s u rf ace c h a rac te ris tic s a nd th e p a ra m e te ris a tion o f s u rf ace p ro ce ss e s in D M I-H IR L A M ; no ta b ly, th e in c lu s ion o f la nd -u s e in fo rm a tion, v a ry in g th rou g h th e c ou rs e o f th e s ea s on, a nd c o rr e s pond in g a dv a n ce m e n ts in th e p a ra m e te ris a tion o f v e g e ta tion, p re s e n t in trig u in g po ss ib iliti e s fo r f u rth e r im p rov e m e n ts in th e a n a ly s is a nd s ho rt-te rm fo reca s tin g o f s u rf ace fie ld s. S o, e v e n if th e imm e d ia te c on c lu s ion s fr o m t h is s tud y do no t po in t in th a t d irec tion, it is e x p ec te d th a t A M IS u s e rs in th e c o mi n g y ea rs w ill b e n e fit fr o m c lo s e r c onn ec tion s b e tw ee n A M IS a nd th e D M I-H IR L A M s y s te m. 49

6. R e fe r e n ce s C h ris te n s e n, O.B. a nd B.H. S a ss, 1994 : A d e s c r ip tion o f th e D MI ev apo r a tion fo r ec a s t m od e l. D M I T ec hn ica l R e po rt 94-3, D M I 1994. C h ris te n s e n, O.B.,1996 : C hang e s to th e D MI ev apo r a tion -p r e d ic tion s y s te m in 1996. D M I T ec hn ica l R e po rt 96-7, D M I 1996. D a le y, R., 1991 : A tm o s ph e r ic D a ta A na ly s is. C a m b rig e U n iv e rs ity P re ss. H ild e n, A. a nd J. H a n s e n, 1998 : Q ua lity and a v a ilab ility o f A MI S da ta fo r th e g r o w ing s e a s on s o f 1997 and 1998. D M I T ec hn ica l R e po rt 98-19, D M I 1998 L o re n c, A., 1981 : A G loba l T h r ee -D im e n s iona l M u ltiv a r ia te S ta tis tic a l In te r po la tion S c h e m e. M on th ly W ea th e r R e v ie w 109, 701-721. N a v a s c u e s, B., 1997 : A na ly s is o f 2 m e te r T e m p e r a tu r e and R e la tive H u m id ity. H IR L A M T ec hn ica l R e po rt N o. 28, N o rr köp in g, J a nu a ry 1997. S a ss, B.H., 1999 : O p e r a tiona l H IR L A M a t D MI. H IR L A M N e w s le tte r 33, 18-23. 0

A pp e nd ix A T h is a pp e nd ix c on ta in s ta b le s w ith d e ta ile d s ta tis tic o f th e v e rifica tion o f th e A M IS fie ld fo r eac h v e rify in g s ta tion. T h e fo llo w in g p a ra m e te rs a re u s e d : M on th no. 1 A p ril 21 -M a y 20 1999 M on th no. 2 M a y 21 -J un e 20 1999 M on th no. 3 J un e 21 -J u ly 20 1999 M on th no. 4 J u ly 21 -A u g u s t 19 1999 M E M ea n E rr o r, i.e. th e s u m o f th e d iff e re n ce b e tw ee n th e a n a ly s e d v a lu e s a nd th e ob s e rv a tion s, d iv id e d b y th e nu m b e r o f ob s e rv a tion s. M A E M ea n A b s o lu te E rr o r, i.e. th e s u m o f th e a b s o lu te d iff e re n ce b e tw ee n th e a n a ly s e d v a lu e s a nd th e ob s e rv a tion s, d iv id e d b y th e nu m b e r o f ob s e rv a tion s. R M S E R oo t M ea n S qu a re E rr o r, i.e. s qu a re roo t o f th e m ea n s qu a re d e rr o r. H R 1 H it R a te, i.e. th e re la tiv e nu m b e r o f a n a ly s e d v a lu e th a t a re w ith in + /- 1 d e g ree s C e ls iu s o f th e ob s e rv e d te m p e ra tu re. H R 2 H it R a te, i.e. th e re la tiv e nu m b e r o f a n a ly s e d v a lu e th a t a re w ith in + /- 2 H R d e g ree s C e ls iu s o f th e ob s e rv e d te m p e ra tu re. H it R a te, i.e. th e re la tiv e nu m b e r o f a n a ly s e d v a lu e th a t a re w ith in + /- % o f th e ob s e rv e d re la tiv e hu mi d ity. H R 10 H it R a te, i.e. th e re la tiv e nu m b e r o f a n a ly s e d v a lu e th a t a re w ith in + /- 10 % o f th e ob s e rv e d re la tiv e hu mi d ity. A ll H it R a te s a re g iv e n in p e rce n t h its. M E, M A E a nd R M S E a re in d e g ree C e ls iu s in ta b le s s ho w in g te m p e ra tu re s ta tis tic s a nd in p e rce n t hu mi d ity in ta b le s s ho w in g re la tiv e hu mi d ity s ta tis tic s. A pp e nd ix A -1

T e m p e ra tu re T e m p e ra tu re M on th n o. 1 M on th n o. 1 0 0 U T C M E M AE R M SE HR 1 HR 2 0 6 U T C M E M AE R M SE HR 1 HR 2 0 0 0 0.33 0.48 0.71 8 6.6 7 9 6.6 7 0 0 0 0.44 0.6 0.76 8 3.3 3 9 6.6 7 0 0 9 0-0.3 9 0.99 1.2 6.6 7 9 0.0 0 0 0 9 0 0.24 0.34 0.42 1 0 0.0 0 1 0 0.0 0 0 1 7 0-0.1 9 0.90 1.19 6 0.0 0 9 3.3 3 0 1 7 0 0.32 0.6 0.8 8 0.0 0 9 6.6 7 0 3 6 0 0.21 0. 0.73 9 0.0 0 9 6.6 7 0 3 6 0 0.37 0.68 0.82 7 6.6 7 1 0 0.0 0 0 0 0 0.29 0.63 0.87 8 3.3 3 9 6.6 7 0 0 0-0.7 4 0.76 0.86 7 3.3 3 1 0 0.0 0 0 6 4 0 0.17 0.69 0.83 8 0.0 0 9 6.6 7 0 6 4 0 0.18 0.8 0.77 8 6.6 7 9 6.6 7 0 7 7 0 0.8 0.72 1.10 7 6.6 7 9 6.6 7 0 7 7 0 0.26 0.29 0.36 1 0 0.0 0 1 0 0.0 0 0 8 9 0-0.1 3 0.83 1.08 7 0.0 0 9 3.3 3 0 8 9 0 0.1 0.2 0.64 9 3.3 3 1 0 0.0 0 0 9 1 0 0.71 1.11 1.4 6 3.3 3 7 6.6 7 0 9 1 0-0.2 0 0.63 0.76 8 3.3 3 1 0 0.0 0 0 9 8 0.2 0.83 1.01 6 0.0 0 9 6.6 7 0 9 8 0.2 0.67 0.76 8 3.3 3 1 0 0.0 0 0 6 0 9 6-1.1 4 1.98 6.12 3 6.6 7 3.3 3 0 6 0 9 6-0.8 4 1.7 2.34 0.0 0 7 3.3 3 0 6 1 0 4 0.03 0.68 2.17 7 3.3 3 9 3.3 3 0 6 1 0 4 0.21 0.47 0.80 8 6.6 7 9 3.3 3 0 6 1 1 0 0.48 0.61 2.00 7 6.6 7 9 3.3 3 0 6 1 1 0 0.4 0.6 0.90 7 6.6 7 9 3.3 3 0 6 1 2 4 0.60 0.81 1.02 7 0.3 7 9 2. 9 0 6 1 2 4-0.6 0.8 1.06 6 8.0 0 9 2.0 0 0 6 1 6 0 0.80 0.86 1.16 6 3.3 3 9 0.0 0 0 6 1 6 0 0.71 0.84 1.46 7 6.6 7 8 3.3 3 0 6 1 6 9-0.2 0 0.62 0.74 8 3.3 3 1 0 0.0 0 0 6 1 6 9 0.01 0.8 0.83 8 0.0 0 9 3.3 3 M EA N : 0.17 0.83 1.47 7 0.6 9 1.0 0 M EA N : 0.08 0.67 0.90 8 1.1 3 9 4.9 2 T e m p e ra tu re T e m p e ra tu re M on th n o. 1 M on th n o. 1 1 2 U T C M E M AE R M SE HR 1 HR 2 1 8 U T C M E M AE R M SE HR 1 HR 2 0 0 0 0.12 0.42 0.1 8 9.6 6 1 0 0.0 0 0 0 0 0.17 0.24 0.32 9 6.6 7 1 0 0.0 0 0 0 9 0 0.4 0.67 0.92 8 0.0 0 9 6.6 7 0 0 9 0 0.37 0. 0.68 8 6.6 7 1 0 0.0 0 0 1 7 0 0.26 0.73 0.97 7 6.6 7 9 6.6 7 0 1 7 0 0.41 0.7 0.7 8 0.0 0 9 6.6 7 0 3 6 0 0.22 0. 0.74 8 6.6 7 9 6.6 7 0 3 6 0 0.31 0.0 0.8 9 0.0 0 1 0 0.0 0 0 0 0-0.6 6 0.78 1.00 6. 2 9 6. 0 0 0-0.0 9 0.36 0.46 9 3.3 3 1 0 0.0 0 0 6 4 0-1.3 8 1.41 1.67 3 3.3 3 8 0.0 0 0 6 4 0-0.9 0 1.01 1.40 6.6 7 8 6.6 7 0 7 7 0-0.1 0.41 0.48 9 6.6 7 1 0 0.0 0 0 7 7 0-0.4 9 0.6 0.82 8 6.6 7 9 6.6 7 0 8 9 0 1.16 1.87 2.19 3 6.6 7 3.3 3 0 8 9 0 0.14 1.06 1.27 0.0 0 9 0.0 0 0 9 1 0-2.3 4 2.34 2.60 1 0.3 4 3 7.9 3 0 9 1 0-1.2 2 1.24 1.49 4 6.6 7 8 0.0 0 0 9 8-0.6 3 1.48 1.90 4 3.3 3 8 0.0 0 0 9 8-0.1 0.80 1.09 6 6.6 7 9 3.3 3 0 6 0 9 6 0.22 2.36 3.9.1 7 6. 2 0 6 0 9 6 0.23 2.02 2.89 0.0 0 6 3.3 3 0 6 1 0 4-0.1 9 0.36 0.47 9 3.3 3 1 0 0.0 0 0 6 1 0 4-0.2 0.47 0.71 9 0.0 0 9 3.3 3 0 6 1 1 0-0.0 7 0.69 0.9 7 6.6 7 9 6.6 7 0 6 1 1 0 0.22 0.48 0.61 9 0.0 0 1 0 0.0 0 0 6 1 2 4-1.4 2 1.8 2.11 3 7.9 3 6 8.9 7 0 6 1 2 4-0.4 0 0.60 0.78 8 9.6 6 9 6. 0 6 1 6 0-0.1 6 0.0 0.68 8 6.2 1 1 0 0.0 0 0 6 1 6 0 0.0 0.3 0.49 9 3.3 3 1 0 0.0 0 0 6 1 6 9 0.6 0.99 1.1 4 6.6 7 9 3.3 3 0 6 1 6 9 0.47 0.9 0.69 9 0.0 0 1 0 0.0 0 M EA N : -0.2 1.07 1.37 6 3.4 3 8.1 4 M EA N : -0.0 7 0.71 0.94 7 8. 2 9 3. 3 T e m p e ra tu re s ta tis tic s fo r th e 1 s t m on th a t tim e s 00, 06, 12 a nd 18 U T C. A pp e nd ix A -2