The Kp index and solar wind speed relationship: Insights for improving space weather forecasts

Transcription

1 SPACE WEATHER, VOL. 11, , doi:1.1/swe.53, 13 The index and solar wind speed relationship: Insights for improving space weather forecasts Heather A. Elliott, 1 Jörg-Micha Jahn, 1 and David J. McComas 1 Received 3 October 1; revised April 13; accepted April 13; published June 13. [1] The geomagnetic index forecasts are currently used to predict the aurora, MeV electron fluxes at geosynchronous, spacecraft anomalies and charging events, and times when accurate geological surveys can be performed. Many forecasts rely on the upstream solar wind speed since the speed strongly correlates with the index. However, the distribution of and solar wind speed measurements is quite broad. To understand how common certain combinations of and speed are, we plot the percentage of points in two-dimensional and speed bins using a color scale. Using these color -solar wind speed distributions for compressions, rarefactions, and Interplanetary Coronal Mass Ejections separately, we find that much of the variability in the -solar wind speed distribution is attributable to the dynamic interaction between the fast and slow wind. We compare three different criteria for identifying compressions and rarefactions and find that density criteria provide greater separation between compressions and rarefactions than dynamic pressure or speed-time slope criteria. However, the speed-time slope provides enough separation to be useful given that the solar wind speed has a long autocorrelation time and can be predicted using solar observations (e.g., expansion factor models). To ensure our work can easily be incorporated into forecast models, we provide the -speed distributions files for all three methods of identifying compressions and rarefactions. We describe a method to extend forecast lead times by estimating compression strength with a speed-time profile obtained from solar wind speed predictions based on solar, coronal, and/or heliospheric imaging observations. Citation: Elliott, H. A., J.-M. Jahn, and D. J. McComas (13), The index and solar wind speed relationship: Insights for improving space weather forecasts, Space Weather, 11, , doi:1.1/swe Introduction [] Geomagnetic indices are often used to quantify how the Earth responds to a given type of solar wind structure; one commonly used index is the index, which is calculated using the horizontal magnetic field measurements from 13 subauroral ground magnetometer stations. The index is derived from the K index from each station, and the K index is based on the more disturbed component of the observed horizontal magnetic field after removing the daily quiet variations [Bartels et al., 1939; Menvielle and Berthelier, 1991]. After eliminating local time and seasonal effects, concurrent values from all 13 stations are used to create the global index [Menvielle and Additional supporting information may be found in the online version of this article. 1 Southwest Research Institute, San Antonio, Texas, USA. Corresponding author: H. A. Elliott, Space Science and Engineering, Southwest Research Institute, San Antonio, TX 7, USA. (helliott@swri.edu) 13. American Geophysical Union. All Rights Reserved /13/1.1/swe.53 Berthelier, 1991]. Thomsen [] concluded that the index is a good measure of the magnetospheric convection because the strength of the magnetic perturbations at the subauroral stations depends on the distance to the auroral zone and strong currents in the auroral zone map to the plasma sheet. The inner edge of the plasma sheet is strongly affected by the convection electric field; thus, these subauroral stations are sensitive to convection [Thomsen, ]. An early study by Snyder et al. [193] showed that the index is well correlated with the solar wind speed. This correlation seems reasonable given that magnetospheric convection electric field is strongly affected by the coupling, which occurs as the solar wind sweeps past Earth [Thomsen, ]. As Thomsen [] points out, the index is still widely used for a variety of ionospheric and magnetospheric studies because of its long baseline and connection to magnetospheric convection. [3] The index is also quite heavily used in applied space weather studies and forecasts. Both 1 and h forecasts by Wing et al. [5] are available at the National Oceanic Atmospheric Administration (NOAA) 339

2 Space Weather Prediction Center (SWPC) ( swpc.noaa.gov/wingkp/). These forecasts are produced with a neural network using the following inputs: the nowcast; upstream solar wind speed, and density; and upstream interplanetary magnetic field (IMF) magnitude, and Bz component. Bala and Reiff [1] developed a similar alert system, which predicts the witha1and3hlead times using an artificial neural network with the upstream solar wind speeds, IMF field strength, and the IMF clock angle (based on By and Bz components) as inputs. The customer requirement document at the National Oceanic Atmospheric Administration (NOAA) Space Weather Prediction Center (SWPC) website ( gov/services/) specifies several users of these index forecasts. This document states that the forecasts are used by the Federal Aviation Administration to manage high frequency and very high frequency system problems, by geological surveying crews for rescheduling to prevent erroneous survey errors, by Boeing and the International Space Station partners to shut down equipment to prevent damage, and by NASA for mission planning and orbit predictions. [] New uses for the index are still being discovered and developed. Choi et al. [11] analyzed anomalies for commercial satellites in geostationary Earth orbit and found that there is a strong relationship between the occurrence rate of reported anomalies and the index. O Brien [9] found the surface charging hazard quotient for geosynchronous satellites peaks when the index is between and 7-. Ling et al. [1] developed a neural network model to predict the > MeV electron flux at geosynchronous orbit using the previous 1 days of electron flux measurements, and 7 days of the daily total of the index. Recently, new empirical auroral models have been created which are parameterized by the index [Zhang and Paxton, ; Carbary, 5]. Newell et al. [1] used the hemispheric power derived from Polar Ultraviolet Imager (UVI) images to test the hemispheric power estimates from four auroral precipitation models. They found that the Hardy model [Hardy et al., 199], which is parameterized by the index was the second best when 1 min UVI images were used and was the third best when hourly images were used, and at both, these cadences it outperformed the Brautigam IMF-based model. Sigernes et al. [11] coupled the real-time Wing forecast available at SWPC to the Starkov [199a, 199b] and Zhang and Paxton auroral models to obtain real-time auroral oval forecasts ( [5] Forecasts of are derived from prior values, upstream solar wind, and IMF conditions, or some combination of those two. Ground magnetometer measurements are available in real-time. However, many forecasts models use solar wind and IMF measurements taken upstream at L1 since these provide a 3 to min warning time. Several key relationships are at the heart of the forecast models. Early on Snyder et al. [193] found a correlation coefficient of.73 between the daily average solar wind speed and the total daily for a 15 day interval, and a correlation coefficient of.5 at the h resolution. More recently, Gholipour et al. [] found that the autocorrelation coefficient for the index (.79) is much higher than cross-correlation coefficients of with solar wind and IMF parameters, and developed a model using only prior values to forecast. However,theGholipour et al. [] study also found the solar wind speed to have the highest cross-correlation coefficient (.), while IMF Bz, VBs, nv Bs, and n all had cross-correlation coefficients below.. In contrast to the Gholipour et al. [] findings, Newell et al. [] found a complex expression that included the solar wind speed, density, clock angle (determined from IMF By and IMF Bz), and Bt (magnitude of IMF By and IMF Bz) that had a cross-correlation coefficient with of.. The Newell expression is shown below: = v 3 B 3 t sin 3 (c /) +. 1 n 1 v. (1) The Newell function does have an IMF Bz dependence but is a strong function of the solar wind speed since it has both v 3 and v terms. [] Forecast warning times could potentially be extended to a day or more if models incorporated estimates of the solar wind conditions at Earth derived from solar, coronal, and heliospheric imaging observations [Richardson and Cane, 11]. A fast Interplanetary Coronal Mass Ejections (ICME) moving at 1 km/s takes about 1.7 days to reach Earth, and a parcel moving at the average wind speed (5 km/s) takes about 3. days. Most solar wind and IMF parameters have short autocorrelation times and are difficult to predict. Borovsky et al. [199] found the solar wind speed to have a long autocorrelation time and IMF Bz to have a short autocorrelation time. Even though IMF Bz southward intervals are geoeffective in several different ways, forecasting IMF Bz remains a challenge because IMF Bz has a short autocorrelation time and cannot be estimated en route using solar, coronal, and heliospheric imaging. In contrast to the difficulty predicting IMF Bz, there has been some success predicting the solar wind speed, which is also a geoeffective parameter. Solar wind speed predictions using the photospheric magnetic field expansion factor are updated in real-time at the National Oceanic and Atmospheric Administration Space Weather Prediction Center ( noaa.gov/) [Wang and Sheeley, 3; Arge and Pizzo, ]. Recently, the removal of background sources in the Solar Terrestrial Relations Observatory (STEREO) Heliospheric Imager (HI) observations has been improved such that fainter and smaller scaled features can be tracked from the Sun to 1 AU [DeForest et al., 11]. This tracking information allows the speed-time profile of structures to be obtained. The brightness in the imaging observations provides information about the integrated density along the line of sight. Therefore, predicting the index using the solar wind speed and predictions/estimates of the solar wind speed based on solar photospheric, coronal, and heliospheric imaging observations is appealing. 3

3 Autocorrelation hour values 1% Coverage Autocorrelation Time: 7 [hrs] Lag [hrs] Figure 1. Autocorrelation for the 3 h index derived from data from 199 through the end of 9. [7] In this paper, we first examine the autocorrelation times for the index and all the solar wind and IMF parameters since the autocorrelation provides basic timescale information relevant for predicting the solar wind and using prior measurements. Then, we explore the -solar wind speed relationship in more detail to further understand some of the sources of the broadness in the -speed distribution. We find one source to be the amount of compression in the solar wind structures, and we can identify compressions using the solar wind speed time profile. We discuss how our findings can be used to improve index forecasts.. Autocorrelations [] In order to forecast a time series quantity, it is useful to know the autocorrelation time of that quantity. In Figure 1, we show the autocorrelation function verses lag time for the 3 h magnetic index. We use a standard exponential folding time (e-folding time) to define the autocorrelation time. This corresponds to the time at which the correlation coefficient decreases to.3. Using this definition, we find to have an autocorrelation time of 7 h, which spans 9 values. The long correlation time combined with the autocorrelation being greater than the cross correlation with other parameters is why Gholipour et al. [] created a neural network to predict based on the prior 9 values. In Figure, we show the autocorrelation function for the plasma (Figure a) and the magnetic field (Figure b) parameters derived from the 1 h OMNI data set from the National Space Science Data Center (NSSDC). The OMNI data set is a collection of solar wind observations from many spacecraft in the solar wind near Earth [King and Papitashvili, 5]. More detailed documentation is available online ( omniweb.gsfc.nasa.gov), which describes how measurements taken upstream are time shifted, and how the data sets are normalized to compensate for differences in calibration. Since the autocorrelation function algorithm requires no gaps in the time series, we use a subset of the data from 199 through the end of 9 when there are few data gaps. We did not want to interpolate over long gaps since that would impact the resulting autocorrelation time. Elsewhere in the paper, we use data from 193 through the end of 9 to improve statistics because our subsequent analysis is not as sensitive to larger data gaps.tosummarizetheseresults,intable1,welistthe Autocorrelation OMNI a) Plasma b) Magnetic Field Autocorrelation Time [hr] Autocorrelation Time [hr] Vp 59 B. Np 1. Bz GSM Tp 19 By GSM 1 Pdyn 1 Bx GSM Lag [hrs]. 1 Lag [hrs] Figure. (a) Autocorrelations for the solar wind proton speed (black), density (blue), temperature (purple), and dynamic pressure (green). (b) Autocorrelations for the interplanetary magnetic field magnitude (black), IMF Bx (green), By (purple), and Bz (blue) in GSM coordinates. These autocorrelations were determined using the hourly OMNI data from 199 through the end of 9. The percent coverage for the solar wind parameters ranged from 9.1% to 99.%, and the coverage for the IMF parameters was 99.7%. 31

4 Table 1. Autocorrelations for, Solar Wind, and IMF Parameters a Quantity Autocorrelation Time (h) Vp 59 Bx GSM 9 7 B Tp 19 Np 1 Pdyn (n p m p v ) 1 By GSM 1 Bz GSM a The autocorrelation time was derived from 3 h values, and the solar wind and IMF parameter times were determined from the hourly OMNI measurements. e-folding autocorrelation times for the index, and the solar wind and IMF parameters in descending order. The speed has the longest autocorrelation time (59 h) out of the solar wind parameters and the dynamic pressure (1 h) the shortest. The Bz GSM component of the IMF has a much shorter time ( h) than IMF By (1 h) or IMF Bx (9 h). A consequence of these numbers is that it is difficult to forecast IMF Bz using prior values, but much easier to forecast the solar wind speed using prior values. 3. -Solar Wind Speed Relationship [9] Next, we examine the -solar wind proton speed (Vp) relationship in more detail. First, we create a simple scatter plot of all the valid and Vp measurements in the hourly OMNI. In the OMNI data set, a given 3 h value is repeated for each hour in that 3 h time period such that each hourly speed value has a corresponding value. Since the community often uses the OMNI data directly, we have chosen to keep this 1 h cadence. We calculated 3 h solar wind parameters and compared those to the 3 h index. The effects we show in this paper are long duration (which implies large-scale) effects and still present in 3 h averages. Even though the index correlates fairly well with the solar wind speed, the -speed distribution is quite broad. From a simple scatter plot, it is not possible to determine how common various combinations of and Vp are (Figure 3a). To illustrate this point, we also show a color plot of the same data binned in both -index and solar wind speed (Figure 3b). In this representation, we have divided the number of points in a given bin by the total number of valid and Vp points and multiplied by 1 to obtain the percentage. Presented this way, it is clear that most of the points have < 5and have speeds between 75 and 5 km s 1 and that the distribution is quite broad at all speeds. [1] In order to investigate why the -speed distribution is so broad, we perform additional sorting. Since it is possible that some of the outlying points could be associated with fast Interplanetary Coronal Mass Ejections (ICMEs), we examine separately the ICME and non-icme -speed distributions (Figure ). We define the non- ICME wind to be all the data remaining after having removed the ICMEs and regions adjacent to the ICMEs (Figure a). To remove the ICMEs, we remove any data points that satisfy any one of the following criteria: (1) proton beta <.1, () alpha-to-proton density ratio >., (3) occurring within the ICME start and stop times listed on the Richardson and Cane ICME list [Cane and Richardson, 3; Richardson and Cane, 1], or occurring within 15 h ahead and h after satisfying criteria 1 3 [Elliott et al., 1]. These adjacent regions are removed to ensure that we do not include any points affected by ICMEs interacting with the surrounding wind. In Figure b, we show only data points that are most likely to be associated with an ICME; therefore, we show only the points that satisfy criteria 1 3 mentioned above. The main part of the ICME -speed distribution is steeper than in the non- ICME distribution particularly at low speeds. Most of the high speed and high points shown in Figure 3 were associated with fast ICMEs. Since the non-icme -speed distribution is quite broad (Figure a), we perform additional sorting of the non-icme wind to look for additional sources of variability.. Compressions and Rarefactions [11] In this section, we investigate if dynamic interactions in the solar wind can contribute to the broadness in the -speed distribution. The development of compressions and rarefactions owing to the dynamic interaction of different speed parcels as they propagate away from the Sun causes systematic relationships between the solar wind and IMF parameters to develop. Therefore, if measurements taken in Earth s magnetosphere or ionosphere correlate well with a specific solar wind or IMF parameter, then they will likely correlate well with several other solar wind and IMF parameters as well. Detailed knowledge of these systematic relationships can be useful when sorting magnetospheric measurements. Borovsky and Denton [1] performed a superposed epoch analysis of 7 Corotating Interaction Regions (CIRs). They found as others have before that when the fast wind emitted from coronal holes runs into slower wind emitted at an earlier time, a compression region develops in the solar wind [e.g., Gosling et al., 197]. In these compression regions, the density, temperature, pressure, and field strength become enhanced (Figure 5). Similar signatures of compression have even been found in high speed streams associated with polar coronal holes where differences in parcel speeds are small [Neugebauer et al., 1995]. Compressions can also develop when fast Interplanetary Coronal Mass Ejections (ICMEs) interact with slower surrounding wind emitted at an earlier time and are referred to as ICME sheaths which is why we removed times adjacent to ICMEs. It important to note that a CIR is a result of the interaction between the fast and slow wind, and CIRs are followed by a High Speed Stream (HSS) of fast solar wind. This fast wind originates from coronal holes and contains Alfvén waves [Belcher and Davis, 1971]. The high dynamic pressure in CIRs will affect the magnetosphere differently 3

5 a) All Data b) All Data Percentage 1 1 Figure 3. (a) Scatter plot of all of the hourly OMNI and solar wind speed measurements from 193 through the end of 9. The hourly OMNI data repeats the 3 h value for each hour within a given 3 h period. (b) The hourly measurements shown on the left are binned in both solar wind speed and, and the color for each bin is the number of points in each bin is divided by the total number of valid and solar wind speed measurements and multiply by 1. than the high wind speed, low pressure, and Alfvén waves found in HSSs. [1] As mentioned above compressions and rarefactions can have a variety of sources; however, the basic cause is interacting wind parcels of differing speed. Since the density, temperature, pressure, and field strength peak during a speed rise, it is possible to use the speedtime profile to identify compressions. We determine the compressions and rarefactions using similar criteria as Elliott et al. [5] and show an example in Figure. We chose a higher threshold than the. 1 km s in Elliott et al. [5] because we wanted to examine stronger compressions and rarefactions in order to determine if there is a clear difference in the compression and rarefaction -speed distributions. When the day running speed-time slope is greater than km s,itis labeled compression (orange). Similarly, when it is less than km s, it is labeled rarefaction (blue). [13] Next, we sort the non-icme wind (Figure a) into compressions and rarefactions using density, dynamic a) Non-ICME b) ICMEs Percentage Figure. (a) Same format as Figure 3b, with any data points having proton beta <.1, alpha to proton density ratio >., or occurring within the ICME start and stop times listed on the Richardson and Cane ICME list removed. Additionally, we remove any data within 15 h ahead and h after any of the above mentioned criteria to ensure that we do not include any points affected by ICMEs interacting with the surrounding wind. (b) Data points most likely to be associated with an ICME, that is points having proton beta <.1, alpha to proton density ratio >., or occurring within the ICME start and stop times listed on the Richardson and Cane ICME list. To allow for comparison with Figure 3b, in both Figures a and b, we have divided each bin by the total number of valid and speed measurements and multiplied by 1. 33

6 [km/sec] Solar Wind Speed Plasma Parameters /p T [ ev] i -3 n [cm ] -1 1 Days from Stream Interface IBI [nt] Figure 5. Superposed epoch analysis of 7 CIRs. Top panel shows the average speed profile, and the bottom panel shows the average profiles for density (red), temperature (green), field strength (blue), and alpha to proton density ratio (purple). Adapted from Figure of Borovsky and Denton [1]. pressure, and speed-time slope criteria to look for systematic differences in the -speed distributions. We separately apply the density, dynamic pressure, and speed-time slope criteria to identify compressions and rarefactions. Specifically, for compressions, we use the following separate and independent sorting criteria: densities 1 cm 3 (Figure 7a), dynamic pressure 3nPa (Figure 7b), and the day running speed-time slope > km s (Figure 7c). Likewise, for the rarefactions, 3 we apply the following separate and independent sorting criteria: densities cm 3 (Figure 7d), dynamic pressure 1.3 npa (Figure 7e), and the day running speed-time slope < km s (Figure7f).Itcanbeseenin Figure 5 that the speed-time slope on either side of the compression is quite low in both the fast and slow wind. Although the density is quite high in the compression, the density is quite low in the fast wind but higher in the slow wind than in the fast wind. We set the density threshold higher than is typical for the slow wind. Note that to determine the percentages in Figure 7, we divided by the number of non-icme points with valid and proton speed values for the slope criteria, and for the density and dynamic pressure criteria, we divided by the number non-icme points with valid, protonspeed, and proton density values. From Figure 7, it is clear that a distinct difference between the compression and rarefaction -speed distributions is achieved when using the density criteria. The density compression criterion seems to identify a more localized (narrow) portion of the overall non-icme -speed distribution than either the dynamic pressure or speed-time slope criteria. The separation between the compression and rarefaction -speed distributions is not as distinct for the dynamic pressure and speed-time slope criteria, but the dynamic pressure (Figure 7b) and speed-time slope criteria (Figure 7c) seem to identify the same compression populations, and the density (Figure 7d) and slope criteria (Figure 7f) identify the same rarefaction population. [1] In order to examine more closely how different the compression and rarefaction -speed distributions are, we created 1-D binned plots (Figure ) using the same sorting as in Figure 7. In Figure, we binned only in speed and calculated the average (solid) index for a given bin. To provide an indication of the variability, we also show the average index one standard deviation (dashed) for given speed bins. To improve statistics, the speed bins are 3 km s 1 wide above km s 1, but at lower speeds ( km s 1 ) where there are more measurements, the bins are 15 km s 1 wide. The compression and rarefaction curves for the density criteria (Figure a) on average overlap less than when using 1 Compression Rarefaction Year Figure. An example of compression (orange) and rarefaction (blue) sorting using the day running average of the speed-time slope with a threshold of km s.inblackare points not satisfying these slope criteria. 3

7 Density b) Np 1 [cm-3] P 3. [npa] 1 Np [cm ] -3 f) P 1.3 [npa] Slope < -5.7 x 1- km s- 1 1 e) d) Speed-Time Slope Slope > 5.7 x 1- km s- Rarefaction c) Pressure Figure 7. Additional sorting applied to the non-icme data shown in Figure a to examine the (top row) compressions and (bottom row) rarefactions separately. We use three different sorting techniques. In the first column, we identify the compressions as points with densities 1 cm 3, and the rarefactions as points with densities cm 3. Similarly, in column, the compressions are points with dynamic pressures 3 npa, and rarefactions are points with dynamic pressures 1.5 npa. In the third column, we use speed-time slope criteria to identify compressions (increasing speed-time profile) and rarefactions (decreasing speedtime profile). Np [cm ] Np 1 [cm ] ICME Pdyn 1.3 [npa] Pdyn 3. [npa] ICME 1 Rarefaction (slope) Compression (slope) ICME Percentage a) 1 1 Figure. The same sorting as in Figure 7, but calculating the average value for given speed bins. Above km s 1, the speed bins are 3 km s 1, and for speeds km s 1, the bins are 15 km s 1. The dashed lines is the average one standard deviation for a given bin. Any bins with less than five points were removed. 35 Percentage Compression ELLIOTT ET AL.: KP AND SOLAR WIND SPEED RELATIONSHIP

8 Difference Between The Average Compressions & Rarefaction Speed-Time Slope Density Pressure Figure 9. The difference between the average compression and rarefaction values for given speed bins shown in Figure. The average difference in for compressions and rarefactions is greater when sorting compressions and rarefactions using density criteria particularly at high speeds. Although the difference in for compressions and rarefactions is less when speed-time slope criteria are used, a distinct separation is still found, and the solar wind speed profile is easier to estimate from solar, coronal, and heliospheric measurements. the dynamic pressure (Figure b) or speed-time slope criteria (Figure c). To more clearly illustrate the differences between the compression and rarefaction sorting for all three methods, in Figure 9, we show the difference between the average compression and rarefaction value for given speed bins. Even though the separation is less for the speed-time slope criteria, this method still could be quite useful if applied in forecast models since the solar wind speed has a long autocorrelation time and is easier to predict using solar observations (e.g., expansion factor speed predictions). Also, the speed profile of given structures can be tracked using coronal and heliospheric imaging. 5. Discussion and Conclusions [15] Despite its age and limitations, the index continues to be a useful index for assessing global geomagnetic activity and is used for long timescale studies, space weather predictions, and as an input for other operational space weather data products. Predicting is of both scientific and operational interest. Past research has shown that the index is strongly correlated with the solar wind speed; however, the -solar wind speed distribution is quite broad. This variability makes predicting more difficult. We sorted the and speed measurements using several different techniques to determine if we could identify the source(s) of some of the width in the -speed distribution. We show that solar wind compression and rarefaction regions on average occupy different portions of the -speed distribution, with compressions generally resulting in higher values than rarefactions for the same speed values. In contrast, ICMEs and compressions occupy a similar portion of the -solar wind speed distribution. [1] Out of the three sorting methods used, we find that the compression and rarefaction -speed distribution separation is most distinct if data are sorted using density criteria and is least distinct when solar wind speed-time slope criteria are used. The speed-time slope method can distinctly separate compression from the slow wind since the slope is significantly lower in the both the slow and fast wind. The separation for the density compression and rarefaction found here is partially more distinct because we raised our threshold to distinguish the compressions from the slow wind which has a higher density than the fast wind, but not as high as in the compressions. Nevertheless, the slope method provides enough separation between compressions and rarefactions to be useful for improving forecasts. The history of the solar wind speed (running speed-time slope) can provide an estimate of how much compression is in the solar wind and consequently provide information about the corresponding enhancements in the density, field strength, and temperature of the wind. Additional knowledge about the solar wind conditions driving the magnetosphere can be deduced by examining a longer stretch of solar wind speed history on a multiday timescale (e.g., day running speed-time slope) as opposed to an hour-by-hour basis. [17] Our results could have implications for solar wind and predictions. The solar wind speed has a long autocorrelation time, and it is a relatively easy parameter to predict for extended time periods from solar observations alone (e.g., expansion factor speed predictions) [Wang and Sheeley, 3; Arge and Pizzo, ]. By extension, such solar wind speed predictions can be used to improve predictions. Knowledge of the solar wind context and history (i.e., detecting compressions and rarefactions sorting) may improve the accuracy of predictions, whereas reliance on solar observations as opposed to in situ measurements at L1 alone may also improve the forecast lead-time. This approach could easily be incorporated into existing or future prediction models. There are no practical limitations preventing this from becoming an improvement for operational forecast methods. In Figure 1, we show a flow diagram of how our work can be applied to the current forecast system. In yellow is the current system that uses L1 solar wind and IMF values, and prior index values from ground magnetometers. In blue, we show how our work could be applied using L1 observations, in orange, we show how our work could be applied using photospheric expansion factor solar wind speed forecasts, and in green, we show how our work could be applied using speed profiles and densities determined from coronal and heliospheric imaging. [1] In our color -D binned plots, we specifically chose to plot the percentage of points meeting our criteria to enable additional studies using our results. As a 3

9 Figure 1. Flow diagram of index forecasts. The yellow boxes indicate how current forecasts are made. The blue boxes indicate how forecasts could be adapted using L1 observations and our results. The narrow blue box refers to the running speed-time which we use to identify compressions and rarefactions. The wide blue box refers to the -solar wind speed distribution arrays from Figure 7 which are included as an electronic supplement. The arrays from Figure 7 are -solar wind speed distributions for compressions and rarefactions sorted using density, dynamic pressure, and speed-time slope criteria. The orange boxes indicate how forecasts could be adapted by using our results and photospheric expansion factor predictions. Similarly, the green boxes indicate how forecasts could be adapted by using our results and coronal and heliospheric imaging measurements. supplement to this paper, we provide our -D binned arrays for each color panel in Figures 3,, and 7 in a simple ASCII formatted files. It is important to note that the color in Figures 3 and indicate the percentage based on the total number of valid and speed points. In Figure 7, the percentage is based on the number of valid non-icme points. We chose different color scales to emphasize this difference. The wide blue box in Figure 1 refers to the files for the arrays from Figure 7. [19] There are two reasons that using the running average of the speed-time slope can improve current forecasts, which use solar wind and IMF measurements at L1. The first reason is that no alpha measurements are reported in the real-time stream from the Advanced Composition Explorer (ACE) located at L1 [Zwickl et al., 199]; therefore, the full dynamic pressure and ion density are not known in real-time. The alpha to proton density ratio is highly variable [McComas et al., ]. The alpha to proton density ratio can be very enhanced in ICMEs, and an alpha to proton density ratio (n(he ++ )/n(h + )) ratio of. or greater is often used as one of several ICME indicator [e.g., Cane and Richardson, 3]. In ICMEs, the n(he ++ )/n(h + ) ratio can exceed. [Hirshberg et al., 197]. Even though the alpha to proton density ratio is low in the fast wind (only %), the alphas contribute about 15% to the total dynamic pressure in the fast wind because the alpha mass is much higher than the proton mass, and the alpha speed is about the same as the proton speed [McComas et al., 3]. The running average of the speed-time slope provides additional information about the density and pressure since the density and pressure are both elevated when the speed rises with time. Similarly, the density and pressure are low when the speed is decreased with time. [] The second reason for using the running average of the speed-time slope with L1 measurements to 37

10 improve forecasts is that some ICMEs with associated intense Solar Energetic Particle events can produce a high penetrating radiation background in solar wind instruments that can affect the ability to produce reliable solar wind parameters. The basic solar wind parameters most affected by the penetrating radiation background are the density and temperature. However, the solar wind speed can usually be determined despite the presence of a background. For example, during the 3 Halloween event, the ACE Solar Wind Electron Proton Alpha Monitor (SWEPAM) onboard tracking algorithm did not work because of a high penetrating radiation background, and the energy range for the tracking sweeps was below (5 17 ev q 1 ) the solar wind energy at that time. The search mode data consists of an energy sweep from 5 ev q 1 to 35. kev q 1, and the same energy steps are used all the time [McComas et al., 199]. Theenergyrangeforthetracking mode data varies since a track mode sweep is centered on the energy step with the peak count rate from either a prior tracking or a prior search mode sweep McComas et al., [199]. The tracking mode sweeps have a s cadence and are used to calculate the solar wind parameters. The search mode data have a 33 min cadence [Skoug et al., ]. Since the tracking algorithm did not work during the Halloween ICME, Skoug et al. [] determined the solar wind parameters by using the 33 min search mode data and removing the penetrating radiation background. This kind of specialized processing is not done for the realtime feed. Information about the solar wind density and pressure can still be obtained when there is penetrating radiation if the running average of the speed-time slope is used. [1] The expansion factor speed prediction can be used to estimate the day running speed-time slope used in our analysis. Our -speed percentage results for compressions (Figure 7c) and rarefactions (Figure 7f) sorted based on the speed-time slope criteria could be incorporated directly into a forecast model using the electronic supplement. Then, the predicted solar wind speed and predicted day running speed-time slope, both derived from photospheric expansion factors, could be used as input to the forecast model and extend the lead-time of the model (Figure 1). [] Recent research has also shown that our instantaneous knowledge of the solar wind state could be extended beyond a combination of in situ L1 and direct solar observations through the use of coronal and heliospheric imaging. DeForest et al. [11] improved the removal of background sources in STEREO Heliospheric Imager (HI) observations to the point that they can track much fainter and smaller scale features associated with a CME and even with the quiet wind from the Sun to 1 AU. Similarly, CIRs have been tracked using STEREO-HI measurements [Rouillard et al., ; Williams et al., 11]. The newstereoprocessingby DeForest et al. [11] could also be used to track the compressions associated with CIRs from the Sun to 1 AU. This opens up the possibility of a much improved tracking between the Sun and Earth of many solar wind structures, not only ICMEs. Tracking solar wind structures in imager observations provide a speed profile, which allows a more accurate prediction of even before such structures reach the L1 point. The brightness in the heliospheric imaging observations could be used to provide a direct measure of the number density of structures if the size (volume) of the structure, and the absolute brightness of the background could be accurately estimated [DeForest et al., 1], then that density information could be used to further leverage the density criteria for compression and rarefaction sorting (Figures 7a and 7d) presented here for additional improvements of forecasts. [3] Improving magnetospheric activity predictions via tracking of solar wind structures with heliospheric imagers from side-viewing space platforms is presently a cutting edge research topic. Space assets are currently not in place to transition such an approach into an operational product anytime soon. However, we have shown that the activity-solar wind speed connection, as measured by the index, is more nuanced than previously thought. This realization can be used to improve our ability to understand and predict magnetospheric activity levels. Our prediction abilities could be further enhanced by leveraging solar photospheric observations and employing expansion factor speed predictions to extend the lead-time of forecasts. While this specific approach requires some further research, the tools are available now to turn this into an operational method. [] Acknowledgments. We would like to thank Ian Richardson and Hilary Cane for making their Interplanetary Coronal Mass Ejection list readily available. We thank Joseph King and Natalia Papitashvili for creating the combined OMNI data set. We also thank the reviewers for their careful review of this paper. References Arge, C. N., and V. J. Pizzo (), Improvement in the prediction of solar wind conditions using near-real time solar magnetic field updates, J. Geophys. Res., 15, 1,5 1,, doi:1.19/1999ja9. Bala, R., and P. Reiff (1), Improvements in short-term forecasting of geomagnetic activity, Space Weather, 1, S1, doi:1.19/1sw779. Bartels, J., N. H. Heck, and H. F. Johnston (1939), The three-hourrange index measuring geomagnetic activity, J. Geophys. Res.,, 11, doi:1.19/teip11. Belcher, J. W., and L. Davis Jr. (1971), Large-amplitude Alfvén waves in the interplanetary medium,, J. Geophys. Res., 7, 353, doi:1.19/ja7i1p353. Borovsky, J. E., and M. H. Denton (1), Solar wind turbulence and shear: A superposed-epoch analysis of corotating interaction regions at 1 AU, J. Geophys. Res. Space Physics, 115, A111, doi:1.19/9ja19. Borovsky, J. E., M. F. Thomsen, and R. C. Elphic (199), The driving of the plasma sheet by the solar wind, J. Geophys. Res. Space Physics, 13, 17,17 17,, doi:1.19/97ja9. Cane, H. V., and I. G. Richardson (3), Interplanetary coronal mass ejections in the near-earth solar wind during 199, J. Geophys. Res. Space Physics, 1, 115, doi:1.19/ja917. Carbary, J. F. (5), A -based model of auroral boundaries, Space Weather, 3, S11, doi:1.19/5sw1. 3

11 Choi, H.-S., et al. (11), Analysis of GEO spacecraft anomalies: Space weather relationships, Space Weather, 9, 1, doi:1.19/1sw597. DeForest, C. E., T. A. Howard, and S. J. Tappin (11), Observations of detailed structure in the solar wind at 1 AU with STEREO/HI-, The Astrophys. J., 73, 13, doi:1.1/-37x/73/1/13. DeForest, C. E., T. A. Howard, and D. J. McComas (1), Disconnecting open solar magnetic flux, The Astrophys. J., 75, 3, doi:1.1/-37x/75/1/3. Elliott, H. A., D. J. McComas, N. A. Schwadron, J. T. Gosling, R. M. Skoug, G. Gloeckler, and T. H. Zurbuchen (5), An improved expected temperature formula for identifying interplanetary coronal mass ejections, J. Geophys. Res., 11, A,13, doi:1.19/ja179. Elliott, H. A., C. J. Henney, D. J. McComas, C. W. Smith, and B. J. Vasquez (1), Temporal and radial variation of the solar wind temperature-speed relationship, J. Geophys. Res. Space Physics, 117, A91, doi:1.19/11ja1715. Gholipour, A., C. Lucas, and B. N. Araabi (), Black box modeling of magnetospheric dynamics to forecast geomagnetic activity, Space Weather,, S71, doi:1.19/3sw39. Gosling, J. T., J. R. Asbridge, S. J. Bame, and W. C. Feldman (197), Solar wind stream interfaces, J. Geophys. Res., 3, 11 11, doi:1.19/ja3iap11. Hardy, D. A., M. S. Gussenhoven, and D. Brautigam (199), A statistical model of auroral ion precipitation, J. Geophys. Res. Space Physics, 9, 37 39, doi:1.19/ja9ia1p37. Hirshberg, J., S. J. Bame, and D. E. Robbins (197), Solar flares and solar wind helium enrichments: July 195 July 197, Sol. Phys., 3, 7, doi:1.17/bf119. King, J. H., and N. E. Papitashvili (5), Solar wind spatial scales in and comparisons of hourly wind and ACE plasma and magnetic field data, J. Geophys. Res., 11, A,1, doi:1.19/ja19. Ling, A. G., G. P. Ginet, R. V. Hilmer, and K. L. Perry (1), A neural network-based geosynchronous relativistic electron flux forecasting model, Space Weather,, S93, doi:1.19/1sw57. McComas, D. J., S. J. Bame, P. Barker, W. C. Feldman, J. L. Phillips, P. Riley, and J. W. Griffee (199), Solar wind electron proton alpha monitor (SWEPAM) for the advanced composition explorer, Space Sci. Rev.,, 53 1, doi:1.13/a: McComas, D. J., H. A. Elliott, J. T. Gosling, D. B. Reisenfeld, R. M. Skoug, B. E. Goldstein, M. Neugebauer, and A. Balogh (), Ulysses second fast-latitude scan: Complexity near solar maximum and the reformation of polar coronal holes, Geophys. Res. Lett., 9, 19, doi:1.19/1gl11. McComas, D. J., H. A. Elliott, N. A. Schwadron, J. T. Gosling, R. M. Skoug, and B. E. Goldstein (3), The three-dimensional solar wind around solar maximum, Geophys. Res. Lett., 3, Menvielle, M., and A. Berthelier (1991), The K-derived planetary indices - Description and availability, Rev. Geophys., 9, 15 3, doi:1.19/91rg99. Neugebauer, M., B. E. Goldstein, D. J. McComas, S. T. Suess, and A. Balogh (1995), Ulysses observations of microstreams in the solar wind from coronal holes, J. Geophys. Res., 1, 3,39 3,39, doi:1.19/95ja73. Newell, P. T., T. Sotirelis, K. Liou, and F. J. Rich (), Pairs of solar wind-magnetosphere coupling functions: Combining a merging term with a viscous term works best, J. Geophys. Res. Space Physics, 113, A1, doi:1.19/7ja15. Newell, P. T., T. Sotirelis, K. Liou, A. R. Lee, S. Wing, J. Green, and R. Redmon (1), Predictive ability of four auroral precipitation models as evaluated using Polar UVI global images, Space Weather,, S1, doi:1.19/1sw. O Brien, T. P. (9), SEAES-GEO: A spacecraft environmental anomalies expert system for geosynchronous orbit, Space Weather, 7, S93, doi:1.19/9sw73. Richardson, I. G., and H. V. Cane (1), Near-Earth interplanetary coronal mass ejections during solar cycle 3 (199-9): Catalog and summary of properties, Sol. Phys.,, pp. 9 +, doi:1.17/s Richardson, I. G., and H. V. Cane (11), Geoeffectiveness (Dst and ) of interplanetary coronal mass ejections during and implications for storm forecasting, Space Weather, 9, S75, doi:1.19/11sw7. Rouillard, A. P., et al. (), First imaging of corotating interaction regions using the STEREO spacecraft, Geophys. Res. Lett., 35, L111, doi:1.19/gl3377. Sigernes, F., M. Dyrland, P. Brekke, S. Chernouss, D. A. Lorentzen, K. Oksavik, and C. Sterling Deehr (11), Two methods to forecast auroral displays, J. Space Weather Space Climate, 1, A3, doi:1.151/swsc/113. Skoug, R. M., J. T. Gosling, J. T. Steinberg, D. J. McComas, C. W. Smith, N. F. Ness, Q. Hu, and L. F. Burlaga (), Extremely high speed solar wind: 9 3 October 3, J. Geophys. Res. Space Physics, 19, A91, doi:1.19/ja19. Snyder, C. W., M. Neugebauer, and U. R. Rao (193), The solar wind velocity and its correlation with cosmic-ray variations and with solar and geomagnetic activity, J. Geophys. Res.,, 31. Starkov, G. V. (199a), Mathematical model of the auroral boundaries, Geomagn. Aeron., 3(3), Starkov, G. V. (199b), Statistical dependences between the magnetic activity indices, Geomagn. Aeron., 3(1), Thomsen, M. F. (), Why is such a good measure of magnetospheric convection, Space Weather,, S11, doi:1.19/sw9. Wang, Y.-M., and N. R. Sheeley Jr. (3), The solar wind and its magnetic sources at sunspot maximum, The Astrophys. J., 57, 1, doi:1.1/33. Williams, A. O., N. J. T. Edberg, S. E. Milan, M. Lester, M. Fränz, and J. A. Davies (11), Tracking corotating interaction regions from the Sun through to the orbit of Mars using ACE, MEX, VEX, and STEREO, J. Geophys. Res. Space Physics, 11, A13, doi:1.19/1ja Wing, S., et al. (5), forecast models, J. Geophys. Res. Space Physics, 11, A3, doi:1.19/ja15. Zhang, Y., and L. J. Paxton (), An empirical -dependent global auroral model based on TIMED/GUVI FUV data, J. Atmos. Sol.-Terr. Phys., 7, 131 1, doi:1.11/j.jastp..3.. Zwickl, R. D., et al. (199), The NOAA real-time solar-wind (RTSW) system using ACE data, Space Sci. Rev.,, 33, doi:1.13/a: