Database of electron density profiles from Arecibo Radar Observatory for the assessment of ionospheric models

Transcription

1 SPACE WEATHER, VOL. 9,, doi: /2010sw000591, 2011 Database of electron density profiles from Arecibo Radar Observatory for the assessment of ionospheric models Vince Eccles, 1 Hien Vo, 2 Jonathan Thompson, 1 Sixto Gonzalez, 2 and Jan J. Sojka 1 Received 15 May 2010; revised 16 August 2010; accepted 22 October 2010; published 13 January [1] We describe the reduction of the Arecibo Observatory incoherent scatter radar electron density profiles from 1966 to 2006 into a standardized database of electron density profiles useful for the assessment of ionospheric models. The database of electron density profiles covers approximately 700 days of observation over nearly 4 solar cycles and all seasons. These data are averaged into climatological conditions with special attention at maintaining a normal profile shape in altitude. The reduced profile database and the climatological average profiles are provided to the Community Coordinated Modeling Center for open access by the community and their efforts to generate ionosphere model metrics and skill scores. Citation: Eccles, V., H. Vo, J. Thompson, S. Gonzalez, and J. J. Sojka (2011), Database of electron density profiles from Arecibo Radar Observatory for the assessment of ionospheric models, Space Weather, 9,, doi: /2010sw Introduction [2] With the increased importance of space weather effects on technology systems there is a need for better ionospheric specifications and forecasts. Because of the sparse observations, we must rely upon ionospheric models to fill out the three dimensional specifications of the primary ionospheric parameter, electron density. Many of the important data sets used to help prescribe the free electron density in the Earth ionosphere are actually not observations of electron density directly. Two very important data sets for current global ionospheric specification methods are GPS TEC and ionosonde frequencyheight ionograms [Mandrake et al., 2005; Schunk et al., 2004; Shim et al., 2008]. However, these two data types require significant physical interpretational insight to relate the data to the electron density profiles (EDPs) in a threedimensional specification. Ionosphere models become the interpretive tool to infuse physics into the data type interpretations. To assess the physics within physicsbased ionosphere models and to assess the empirical representations of empirical ionosphere models, it is important to use a more direct measurement of EDPs that does not depend on model profiles for interpretation. While satellites can provide measures of in situ electron density along an orbit track, these data again do not 1 Space Environment Corporation, Providence, Utah, USA. 2 Atmospheric Science Group, Arecibo Observatory, Arecibo, Puerto Rico. provide easy insights into an ionosphere model s deficiencies. The Earth s gravity is an important physical parameter for ordering of the electron densities in the ionosphere, thus, the EDP is the best benchmark observation for assessing ionosphere models [Liu et al., 2007; Luan et al., 2006]. The ionosonde vertical sounder does probe the EDP, but interpretation of the vertical height of the reflected signal requires a model of electron density in the valley region between the E and F peak. This makes model assessment by ionosonde based EDPs a slightly incestuous affair. The incoherent scatter radar (ISR) can provide EDP shapes without distortion. There are several ISRs operating around the world and some have a long database of electron density profiles over many solar and magnetic conditions. [3] For the assessment of ionosphere models the National Space Weather Program develop space weather metrics that uses the root mean square (RMS) of the differences of the measured and modeled electron densities every 20 km from 200 to 600 km altitude (The National Space Weather Program, The Implementation Plan 2nd Edition (January 1997), FCM P ). It was a simple metric for calculating an objective measure of ionospheric model accuracy. The metric and other ionospheric metrics require a ground truth database of EDPs to generate these metrics. To this end we have collected, reduced, and assess the 4 decades of electron density profiles observed by the ISR at the Arecibo Observatory (AO) in Puerto Rico. Other studies have examined the AO ISR EDPs for optimal profile modeling [Liu et al., 2007]. Makela et al. [2000] Copyright 2011 by the American Geophysical Union 1of11

2 Figure 1. (a d) The time distribution of days with ionosphere observations by the AO ISR. use the EDPs to compare the TEC measurements by the ISR and Global Pointing System. Pawlowski et al. [2008] use ISR EDPs from Millstone Hill to assess their global ionospheric model. We focus on producing an EDP database of the AO ISR EDPs without assumed model shapes with standardized cadence and altitude range. The reduced EDP database is called the Assessment of Ionosphere Models database or AIM database and is used by the Assessment of Ionosphere Models (AIM) software to produce metrics and skill scores of modeled profiles. The AIM database has been provided to the Community Coordinated Modeling Center (CCMC) for use by the ionospheric community. In addition, a set of tools for assessing ionosphere model results has been created for Ionospheric Model Assessment in an accompanying paper [Eccles et al., 2011]. [4] We discuss the AO data sources and the coverage of the data in section 2. We discuss the data reduction process in section 3. The assessment of the data uncertainty is described in section 4. Section 5 summarizes the details of the paper and the AIM database. 2. ISR Measurements of EDPs at Arecibo Observatory [5] The AO ISR, located at 18.3 N and W (29.0 magn), has been making ionospheric observations for over 4 solar cycles (Figure 1). The AO ISR ionosphere database has been archived both at AO, in the CEDAR database at NCAR, and the Madrigal database at Haystack Observatory. The altitude range (Figure 1d) and resolution as well as the time resolution of the measured EDPs varies over the decades. The ionospheric metric database constructed must take these differences into account and standardize the database profiles. Although the ISR operates in ionospheric mode primarily around world days, it has nonetheless amassed observations that span solar cycle, seasonal, diurnal and geomagnetic ranges. [6] Figure 2 shows the spread of conditions of Arecibo observations for solar, magnetic, and seasonal conditions. Each day of observation is represented by a symbol on the F10.7 solar flux index versus Kp daily sum plot and the F10.7 versus day of year graph. These scatterplots show that for each season there is a reasonably uniform coverage in both solar cycle (F10.7) and geomagnetic activity (Kp sum). In principle, there is sufficient observational data to test ionospheric models at this location via metrics and skill scores over solar cycle, seasonal, diurnal and geomagnetic activity. However, with over 700 days of data, the task of assessing the quality of these EDP observations is nonnegligible task. With respect to generating a standard set of EDPs from the AO ISR observations there is one difficulty that must be overcome to provide a suitable 2of11

3 Figure 2. Distribution of Arecibo data between 1966 and 2006 for (a) solar magnetic indices and (b) seasonal solar indices. The dark vertical lines indicate our division of data into low, medium, and high solar conditions. space weather metric for ionospheric models; rarely has the AO ionosphere group had a long contiguous period of access to the radar to follow space weather variations. Instead, there are many partial days of observations of local nighttime or local daytime hours as the scientific interest demanded. Most of the continuous observation periods (all local times over several days) occur during the 1990s and 2000s. Additionally, there are very few periods of multiple day observations at all universal time (UT) hours. These deficiencies have been identified and the ISR community is currently addressing the need for long observation periods and have integrated standardized observation modes at each UT hour [Sojka et al., 2009]. However, the historical database must not be neglected because it surveys several solar cycles of geophysical conditions. 3. Reduction of the AO ISR EDPs [7] We have aimed at a data reduction that provides useful metric comparisons between data EDPs and model EDPs. The second implementation plan report of the National Space Weather Program (NSWP) outlines in detail the various metrics to be used for the solar heliosphere, magnetosphere, and ionosphere thermosphere models. These metrics are prioritized based on several factors of which data availability plays a crucial role. For ionospheric model assessment, the primary metric is the F layer electron density comparison with the NSF ISR chain. These ISRs are located at equatorial, middle, and high latitudes and have databases extending over at least a solar cycle. Equation (1) is the specific format of the altitude profile metric, D: D ¼ ( 1 X 23 X 600 )1 = 2 ½n o ðh; tþ n m ðh; tþš 2 ð1þ t¼0 h¼200; 20 where n o (h, t) is the observed density at altitude h and time t, and n m (h, t) is the model value (The National Space Weather Program, The Implementation Plan 2nd Edition). Twenty four profiles at 1 h intervals are used in the metric, though each UT can be examined individually. The profile ranges from 200 to 600 km in 20 km steps. The initial study using the metric of equation (1) was presented at Space Weather Week 2002 by Tim Fuller Rowell for the NSF Metric Challenge. While the metric provided an unambiguous number to compare models, the interpretation of the numbers is ambiguous. A study of metrics and skill scores for ionospheric climatology and space weather are presented in companion papers. [8] The reduction of the 700 days of AO EDPs is aimed at metrics studies like the NSWP metric above. Each day of data in the measured EDPs does not fit identically to the altitude and time requirements of the NSWP metric. The altitude of the EDP observations rarely aligns with the 20 km grid. Additionally, most days in the first 2 decades did not cover the full altitude range of 200 to 600 km. To provide for the most versatile ground truth data set, we interpolated or fit all EDPs to a grid of universal time hours and 10 km altitude bins from 150 to 700 km. This altitude range preserved nearly all altitudes of the EDPs. The higher resolution in altitude was selected to maintain better curve fitting when the ISR EDP altitude resolution is small. The altitude and time resolution of the data varied greatly over the 40 years. Because of the variation in the ISR data resolutions we had to use two approaches for sparse and then dense resolutions to standardize the altitude and time nodes of the AIM profiles. In both cases we first reduced the ISR profiles to obtain density profiles at integer UT hours with the same altitude nodes as the original ISR profiles of that day. First, if the ISR profile time resolution was 15 min or greater, then we used cubic spline interpolation to obtain profiles at integer hours. The errors associated with the ISR density profile data were interpolated to assign the error for the AIM profiles. After obtaining 3of11

4 Figure 3. Data reduction of observed EDPs (red) +/ 20 min of 0900 UT to the single reduced profile at 0900 UT (black) for (a) high resolution mode (data fitting in altitude) and (b) typical resolution mode (cubic interpolation in altitude). The uncertainty bars are also represented with percent uncertainty plots. the 24 hourly profiles with ISR altitude nodes, then, if the altitude node resolution was greater than 5 km, a cubicspline fit was used to generate the 10 km resolution for the AIM profiles. Again the AIM errors were interpolated from the errors from the time node reduced profile. Second, if the ISR profile time resolution was less than 15 min resolution we used an error weighted least squares (cubic) fit to data near each UT hour (+/ 30 min). Then the profiles with the ISR altitudes were reduced to the ISR altitudes with cubic spline interpolation or error weighted least squares fit depending on the altitude resolution. Each error weighted fit produces a chi. We joined contiguous days of data when available to provide for optimal curve fitting at the UT day boundaries. At the end of the reduction of the ISR profiles to the AIM profiles each density has an associated error that is a combination of the errors associated with the ISR data reduction and the consistency of the data used in the profile reduction procedure. Figure 3 shows about 20 observation EDPs (red symbols) around 0900 UT on 13 October 1999 (+/ 20 min for each hour). The EDP on the altitude grid and time grid for the metrics is plotted in black. [9] The complete set of standardized profiles from 1964 to 2006 is the database for the AIM package. They are placed in ASCII flat files for ease of use and distribution. These reduced profiles have errors based on (1) the data error provided by the AO ISR physics analysis of the spectral shape of the return scatter and (2) the quality of the fit of multiple data points. The quality of the fit is primarily a representation error introduced by the standardized time and altitude requirements of the profiles. The typical uncertainty at the F 2 density peak during the early years of operation ( ) is around 10%. The percent uncertainty of the F 2 peak generally improves after Most years after 1990 average around 6% uncertainty. Automated programs reduced all data, but each day was visually examined to screen inappropriate data reductions. A third source of error, that is, normalization procedure to relate the ISR power profile to the electron density profile, is folded into the AIM profile database. This error is described in section Additional Error Assessment of Peak Densities [10] The error bars presented in section 3 are generated by a combination of the ISR analysis error, which is typically very small, and the profile reduction analysis for the AIM EDPs, which include some variations of the ionosphere. It is useful to examine the accuracy of the peak 4of11

5 Figure 4. Example comparison of F2 peak density measured by the AO ISR (black squares), AO Digisonde (red circles), Ramey Digisonde (green triangles), and the AO CADI (blue crosses). density of the ISR EDPs. The radar measures the scatter of the transmitted signal, which is proportional to the electron density. Historically the altitude profile of the ISR return power must be normalized at the peak return to the density inferred by the observed f o F 2 from a colocated ionosonde. The f o F 2 value is a very accurate measure of peak density and easily has less than 5% error if the ionograms are hand scaled. The ionosonde data originally used to normalize the ISR EDP observations from the 1966 through 1990 are not available. Therefore, we examined the error of the normalization of the reduced EDP density peaks through a comparison with available ionosonde data from two ionosondes: the Canadian Digital Ionosonde (CADI) located at AO, the Digisonde located at AO, and the Ramey Digisonde located 50 km west of AO. The CADI data were provided by the AO staff. The AO Digisonde data were obtained from the National Astronomy and Ionosphere Center at Arecibo Observatory Web site. The Ramey Digisonde data were obtained from the National Geophysical Data Center. Figure 4 shows example comparisons of F 2 peak densities. Most frequently the ionosonde and ISR peak densities compare favorably as in Figure 4 (top). However, there are some days that are off substantially. Figure 4 (bottom) shows a substantial difference between two Digisonde measurements. There were only several days of the 700+ days that had ISR peak densities that did not follow the ionosonde data. These were removed from the database. Much of the oldest ISR data (pre 1998) we were not able to obtain suitable ionosonde data to assess the ISR peak densities. Without the ionosonde data for each day of ISR data we cannot remove all suspect days from the AIM ISR database. [11] We compared available ionosonde measures of the F 2 peak density with the AO ISR EDP peak densities. The average of the absolute value of the relative difference was or 9.5% average relative difference from 2000 through 2006 using the CADI, the Ramey Digisonde, and the AO Digisonde (Figure 5). The RMS of the relative difference of peak densities was 0.15 for the same years. For ionosonde data available in the 1990s the average relative difference was approximately 17% and the RMS was With no better estimate of uncertainty, we applied a minimum uncertainty to the EDPs, if there is no supporting ionosonde data confirming the peak normalization. Comparisons of the CADI ionosonde and Ramey Digisonde observations of F 2 peak densities over the same time period show they differ by an average of about 10%. There is an inherent uncertainty in the determination of 5of11

6 Figure 5. Comparison of the AO Digisonde and the Ramey Digisonde peak density measurements with the AO ISR peak density measurements. the fof2 values of ionograms associated with representation error associated with the space time differences of the observations of the ionosphere around AO. Newer methods of self normalization of the AO ISR profile observations will help reduce the EDP uncertainties but for this study 10% uncertainty was the best that can be used. [12] The errors associated with the EDPs determined by our automated data reduction methods are based on the data characteristics and fitting characteristics. This relative uncertainty was often less than 5% near the density peak (Figure 3). The analysis of the peak density for the most recent decade (2000s) indicated that the uncertainty was closer to 10%. The 1990s had even higher errors closer to 18%, but the available radar data with simultaneous ionosonde data very limited. Similar studies of the 1960s, 1970s, and 1990s could not be performed because the analyzed ionosonde data were not available in the current World Data Center databases. Archived ionograms stored on film at the National Geophysical Data Center could be reanalyzed for better normalization error estimates for these early decades. Our experiences with the EDPs from all decades of data indicate that the most recent data are improved in cadence, resolution, and quality. We desire to weight the more recent decades in any climatological averages therefore we have chosen to impose require higher average uncertainty with increasing decadal age of the data. Thus, the peak density uncertainty uncertainties were adjusted to average 10% for the 2000s as the error study indicated. The average peak density error of each decade was adjusted to be 20% for 1990s, 30% for the 1980s, 40% for the 1970s, and 50% for the 1960s. The relative errors of the electron densities related to ISR analysis and the AIM profile reduction are maintained within each decade. The data errors of the AIM database are used in the determination of error weighted averages to obtain climatological trends, metric assessments, and skill score determinations. The later data will have more influence on the ionospheric representations, but the earlier data will ensure that the database spans all geophysical conditions and universal times are represented well. [13] This study is a reminder that careful review of instrument procedures with multi instrument comparisons needs to be performed to determine how claims of accuracy and precision must be limited. The ISR measurement is excellent for profile shape and may have excellent EDP accuracy with careful power profile normalization, but there is still an element of uncertainty associated with representation errors associated with the volume and time of the observations. This will generally create a larger uncertainty than the instrument uncertainty if the ionosphere is completely quiescent and well behaved. 5. Overview of the AIM Database [14] The AIM database is composed of ASCII flat files of electron density profiles and uncertainty estimates on the densities for each UT day of observations. The files have 24 line headers describing the file. The AIM Package for ionosphere model assessment has FORTRAN programs to read the profiles and generate key profile parameters for model profile comparisons. From each EDP profile in the AIM EDP database we generated the F 2 peak density (n M F 2 ), the F 2 peak height (h M F 2 ), the scale height at the half peak density point above the peak (H S ), and the total electron content from 200 to 600 km altitude (TECp) and added these to the AIM database. These parameters and their associated uncertainties are generated and written in day tagged text files with a 24 line header containing metadata. The parameters are generated from a cubicspline fit to the profiles. A normalized topside profile 6of11

7 Figure 6. (left) Normalized profile and (right) the number of profiles used in the average profile at each altitude. The red line in the normalized profile graph is the uncorrected normalized profiles, and the black line is the corrected profile shape. shape (T S (z)), and normalized bottomside profile shape (B S (z)) are also generated for use in profile metrics measures. The AIM package uses the same subroutines to produce identical profile parameters and profile shapes for ionosphere model results to generate data to model metrics as discussed in the two companion papers on climatological [Eccles et al., 2011]. [15] The data are not enough to generate a definitive climatology for the ionosphere over Arecibo. It would be best to combine the ISR profiles with the long term database of peak density obtained from ionosondes. The AO ISR EDPs should be used to generate the definitive profile shapes and definitive peak height in combination with the ionosonde peak density determinations. This will change in the future with new normalization procedures and long term monitoring effort at the ISRs. We determine the profile parameter averages for an ISR climatology using a Bayesian weighted average hxi ¼ X P i x i with P i as a probability of a single observation instance based on the uncertainty of the data. Frequently, the Bayesian probability is taken to be the inverse of the variance (P i / 1/å 2 ), but this works best for combining many observations of a stationary value. Each observation can be assigned a probability using the uncertainty variance. Each measurement can have a different uncertainty ð2þ associated with the technique and particular observation characteristics (e.g., noise levels, signal conditions, etc.). However, the ionosphere profiles have weather variance as part of any current ordering of the data, that is, the observations are not of stationary values. The uncertainty variance is generally much smaller than the weather variance. To properly respect the weather variance, we assign the probability using weighting based on the standard error uncertainty: 1. " i P i ¼ P1. ð3þ " i First, we obtain the peak parameter climatology of density and height using weighted averages (equations (2) and (3)) and, second, we average normalized profile shapes. The average normalized profile shapes are then used to generate average profiles by combining the average peak parameters and average normalized profiles. The data profiles do not all cover the same altitude ranges so the average profile has a different number of profiles averaged into different altitudes. Figure 6 (left) plots the normalized profile, and Figure 6 (right) plots the number of profiles in the average. The changing number of profiles with altitude permits unrealistic jumps in the profile shape. The peak always had more profiles in the average, i 7of11

8 Figure 7. Average parameters for season and solar conditions. Day and night are indicated by blue and yellow lines on each graph frame. thus, we assume the densities around the peak defines the density reference for the profile and the shape of the profile away from the peak is redefined using scale heights away from the profile. The red line in Figure 6 (left) is the uncorrected normalized profile obtained through averaging and the black line is the corrected normalized profile. [16] Figure 7 shows the average peak and profile parameters for low, medium, and high solar conditions of the data for the F10.7 ranges show. Figure 7 (left) is of peak density, Figure 7 (middle) is of peak height, and Figure 7 (right) is of the topside scale height at half peak density. The summer high solar climatology bin had few EDPs to average and the average characteristics probably not valid. The daytimes and nighttimes are indicted on the plots with blue and yellow axes lines. It is noteworthy how the half density scale height, H 1/2, decreases after sunrise in spring, autumn and winter. Figure 8 plots the 1 sigma variation within the bin reflecting the natural weather variation about the average solar condition response of these same profile parameters. The nighttime variation is especially large during solar minimum and just before 8of11

9 Figure 8. The percent deviation of the weather variation about the average parameter. Day and night indicators are same as Figure 7. sunrise during solar medium and maximum. These averages as well as the AIM database can be downloaded from [17] The average profiles for solar conditions and seasons for noontime are shown in Figure 9. The true altitude/ density profiles (Figure 9 (left)) are reconstructed from the average normalized profiles and the average peak parameters (n M F 2 and h M F 2 ). There is considerable roughness in the scale height values (Figure 9 (right)) above the 200 km distance above the peak. This is due in part to the various numbers of profiles being averaged into the profile shape (see Figure 6), but also reflects modulations in each profile s topside in the database. However, the smoothed scale height curve could be used to create normalized profiles in empirical models with better fidelity than using averages of real profiles. Notices that the topside scale heights above a certain altitude difference above the peak seem to have the same values regardless of solar conditions. The bottom right panel in Figure 9 for the best example showing identical smoothed 9of11

10 Figure 9. The noontime (left) average profiles, (middle) average normalized profiles, and (right) scale height of average profiles for seasons and solar conditions. 10 of 11

11 scale height values above 100 km. Only the altitudes near the peak show differences in the scale heights between the solar conditions for noon. 6. Conclusion [18] The Arecibo Incoherent Scatter Radar data from 1966 to 2006 has been collected, assessed, and reduced to provide a standard data set of electron density profiles and ionospheric parameters for the assessment of ionospheric models. The AIM database is available for download from the Community Coordinated Modeling Center (CCMC). These reduced data have hourly electron density profiles, peak density, peak height, and profile shape as the key benchmarks for assessing ionospheric model accuracy and precision. The reduced data types are averaged into a climatology for the Arecibo ionosphere with season and solar conditions. Additionally, the CCMC has access to the metrics software to compare ionospheric model results with the AIM database. Plots similar to Figures 7 9 for all hours are at vince/aim.html. [19] Acknowledgments. This work was preformed under the NWP/NSF ATM grant Data of the Arecibo Radar and ionosondes were made available by the Arecibo Observatory ionosphere group, the NSF CEDAR Database, and NGDC Spyder Ionosonde database. References Eccles,V.,J.Thompson,J.J.Sojka,H.Vo,andS.Gonzalez(2011), Assessment of Ionospheric Models package for the Community Coordinated Modeling Center: Climatology, Space Weather, 9, S01004, doi: /2010sw Liu, L., H. Le, W. Wan, M. P. Sulzer, J. Lei, and M. L. Zhang (2007), An analysis of the scale heights in the lower topside ionosphere based on the Arecibo incoherent scatter radar measurements, J. Geophys. Res., 112, A06307, doi: /2007ja Luan, X., L. Liu, W. Wan, J. Lei, S. R. Zhang, J. M. Holt, and M. P. Sulzer (2006), A study of the shape of topside electron density profile derived from incoherent scatter radar measurements over Arecibo and Millstone Hill, Radio Sci., 41, RS4006, doi: /2005rs Makela, J. J., S. A. González, B. MacPherson, X. Pi, M. C. Kelley, and P. J. Sultan (2000), Intercomparisons of total electron content measurements using Arecibo Incoherent Scatter Radar and GPS, Geophys. Res. Lett., 27(18), , doi: /2000gl Mandrake, L., B. Wilson, C. Wang, G. Hajj, A. Mannucci, and X. Pi (2005), A performance evaluation of the operational Jet Propulsion Laboratory/University of Southern California Global Assimilation Ionospheric Model (JPL/USC GAIM), J. Geophys. Res., 110, A12306, doi: /2005ja Pawlowski, D. J., A. J. Ridley, I. Kim, and D. S. Bernstein (2008), Global model comparison with Millstone Hill during September 2005, J. Geophys. Res., 113, A01312, doi: /2007ja Schunk, R. W., et al. (2004), Global Assimilation of Ionospheric Measurements (GAIM), Radio Sci., 39, RS1S02, doi: / 2002RS Shim, J. S., L. Scherliess, R. W. Schunk, and D. C. Thompson (2008), Spatial correlations of day to day ionospheric total electron content variability obtained from ground based GPS, J. Geophys. Res., 113, A09309, doi: /2007ja Sojka,J.J.,R.L.McPherron,A.P.vanEyken,M.J.Nicolls,C.J. Heinselman, and J. D. Kelly (2009), Observations of ionospheric heating during the passage of solar coronal hole fast streams, Geophys. Res. Lett., 36, L19105, doi: /2009gl V. Eccles, J. J. Sojka, and J. Thompson, Space Environment Corporation, Ste. A, 221 N. Spring Creek Pkwy., Providence, UT , USA. (vince.eccles@spacenv.com) S. Gonzalez and H. Vo, Atmospheric Science Group, Arecibo Observatory, Arecibo, Puerto Rico, of 11