Magnetic discrimination that will satisfy regulators?

Transcription

1 Magnetic discrimination that will satisfy regulators? Stephen D. Billings 1, John M. Stanley 2 and Clif Youmans 3 1 Geophysical Inversion Facility, The University of British Columbia, 2219 Main Mall, Vancouver, V6T-1Z4 2 Geophysical Technology Limited (G-tek), P.O. Box U9, Armidale NSW 2351, Australia 3 Montana Army National Guard, P.O. Box 4789, Helena, Montana, Abstract Discrimination, or the minimization of false alarms, is recognized as one of the major challenges in efficient UXO clearance. With appropriate historical knowledge as to what weapons were used at the site and geological conditions that are not too adverse, we believe that discrimination using magnetics can be structured in a way that would be very well regarded by the regulatory community. This conclusion is based on analysis of high quality total-field magnetic data collected over the Guthrie Road and Limestone Hills sites in Montana. These Montana case studies, also show that there is a significant difference between the magnetic signature of seeded UXO and that from live-site UXO which suggests that the measurement of discrimination performance on seeded sites may have disadvantaged magnetics in previous evaluations of this technology. The core of the discrimination method is an understanding of the remanent magnetic properties of live UXO and the inversion of individual anomalies for the best fitting magnetic dipole. Using the recovered dipole moment it is not possible to uniquely classify the item as ordnance/non ordnance, nor is it possible to unambiguously determine the ordnance type. Rather, the items can be placed in a list ordered by the likelihood that they are UXO. The ranking is based on how closely the recovered moment matches the predicted dipoles drawn from a library of ordnance items expected to occur in the area. Anomaly excavation is then prioritized based upon the discrimination ranking list. Data available after the validation of dig sheets by excavation in the Montana case study areas has demonstrated that all live ordnance items are recovered after less than 50% of anomalies are excavated. In the Guthrie Road case study a False Alarm Rate (FAR defined as the number of false alarms divided by the number of UXO recovered) of 3.4 was achieved and in the Limestone Hills the FAR was 4.5. In both cases these excellent FAR results were achieved with 100% detection of the UXO. At Guthrie Road, this FAR result was obtained with all of the seeded targets detected. In a clearance operation a decision needs to be made regarding how far down the list the excavation stops. An initial estimate of the number of items to be recovered can be obtained by specifying a maximum remanent magnetization. This number can be modified depending on the yield of ordnance as the excavation progresses. Excavation of additional items further down the list is undertaken as part of a QC procedure. In summary, discrimination of magnetics, based on sound physical modeling and objective decision making criteria, is potentially capable of exceeding the requirements of regulators. Introduction Recent research into UXO discrimination technology has focused upon electromagnetic induction (EMI) data rather than magnetic data. This is due to the EMI response potentially being more sensitive to object dimension and shape than is the magnetic anomaly, the ability of EMI to detect non-ferrous as well as metallic objects and its capacity to operate in areas with magnetic geology (e.g. Kaho alawe Island). Notwithstanding these perceived benefits of EMI, there are a lot of situations where magnetometry is a more viable and cost-effective alternative.

2 Clearly, we are not advocating its use in geologically magnetic areas such as Kaho alawe but we note that there are a very large number of sites where the geology is favorable to magnetics (Fort Ord in California, and the Helana Valley in Montana are notable examples). Additionally, there are very few UXOs that are completely non-ferrous and hence undetectable with magnetics. Magnetometers are generally simpler to operate than EMI sensors and can achieve production rates many times higher. For instance, at the ODDS site in Fort Ord, a man portable four-sensor magnetometer array surveyed the site 4 times faster than the EM-61 cart system (Parsons Engineering., 2002). In Montana, G- tek has consistently sustained production rates of between 8 and 10 acres a day using this same system. Typical EM-61 production rates in comparable terrain are of the order of 1 acre per day. The purpose of this paper is to show that magnetics is also capable of excellent discrimination between UXO and scrap and can produce very low false alarm rates. Furthermore, this procedure could potentially exceed the requirements of regulators. The data collection, processing and selection of anomalies from magnetometry are all well-established procedures. From historical records of site usage and/or exploratory surveying, the ordnance items expected to occur in an area can usually be determined. Based on ballistic modeling, maximum depths of occurrence for these different items can be specified. Knowledge of the proposed land-usage can also be used to set a desired maximum detection depth. Once a desired detection depth has been established it is possible to estimate (either using modeling or an historical database) the minimum amplitude anomaly that needs to be investigated. However, if this amplitude is too low then many anomalies from geologic sources, shrapnel or metallic debris will also be detected and will need to be addressed. Significant amounts of historical data, from a wide variety of sites, now exists in order to successfully set this threshold to the proper level. The normal process in the industry of basing the dig list purely on the anomaly amplitude is not well supported by scientific evidence and represents an over-simplistic approach to detection and discrimination of UXO targets. An 81 mm mortar at 3 feet depth can have the same peak amplitude as a small piece of shrapnel near the surface. However, the anomaly shapes over these two items are considerably different. Our methodology is to exploit this shape information so that we can make inferences on the probability that the item is a UXO. Our procedure is based on using a sound physical model, basing the decision of UXO likelihood on objective criteria and with sufficient flexibility to reliably capture the variability of individual UXO (different material properties, deformation on impact, etc). This process represents a structured, scientific approach, which is likely to be highly acceptable to regulators. The basis of our discrimination method is inversion for the dipole moment of an anomaly. Due to the rapid decay with distance of higher order moments we do not attempt to invert a more complicated model. The dipole model has an inherent ambiguity, in that it alone cannot be used to uniquely constrain the geometry of the object. There is also the issue of possible remanent magnetization of items, which would change the dipole moment and hence may change the interpretation. Our method to overcome these issues is to compare the recovered dipole moment to a library of ordnance items expected to occur in the area. The difference between the recovered moment and that from the closest ordnance item provides an estimate of the minimum remanent magnetization that the item must possess if it were a UXO. We take this a step further and rank all items based upon their remanence. We have observed that the lower the value of the remanence the more likely the source is to be an intact UXO. As also evidenced from these Montana case studies, the remanent properties of seeded UXO are significantly different from live-site UXO. Thus there is an indication from these results that the measurement of discrimination performance on seeded sites may be inaccurate in the case of magnetics. Moreover, because the different remanent properties make detection of seeded UXO items more difficult than live-site UXO this attribute could potentially be used to provide an effective quality assurance technique for assessing detection performance. This combination of model matching with remanence ranking provides the basis of our discrimination method. 2

3 Magnetic modeling Spheroids have been proposed as an approximate parameterization of ordnance by several authors (McFee, 1989; Altshuler, 1996; Bulter et al., 1998). While the spheroid does not capture the top-bottom asymmetry of many ordnance items, close agreement between observed anomalies over test stands and spheroids have been demonstrated (McFee, 1989; Bulter et al., 1998). Furthermore, the magnetic anomaly from a solid spheroid has been shown to be very similar to a hollow spheroid (Altshuler, 1996). Therefore, we will use solid spheroids as the basis of our magnetic modeling of UXOs. To model the response of a spheroid we need to know its spatial position, (x,y,z), orientation (φ,θ), diameter (a), length (L=ae, where e is the aspect ratio) and magnetic permeability (µ), along with the extent of any remanent magnetization. We adopt the convention that φ is the angle clockwise from North of the projection of the semi-major axis of the spheroid onto a horizontal plane, while θ is the dip angle (positive upwards) of the axis relative to that plane. We also need to know the direction and strength of the Earth s magnetic field and use the convention that x is positive to the East, y positive to the North and z positive upwards. The magnetic anomaly of a buried ferrous item arises from both remanent M rem and induced magnetization M in, the total magnetization M, is then M = M rem + M in (1) Remanent magnetism is present even in the absence of an inducing field and is due to magnetic moments being locked into alignment with an external field at some stage in the history of the steel casing. Induced magnetism arises because magnetic domains in a ferrous material tend to align with the direction of the Earth s field. The ease with which the moments align, and hence the strength of the magnetism, depends on the magnetic permeability of the steel. For a compact body such as a spheroid, demagnetization effects become important. This phenomenon refers to the extent that the induced field is reduced due to the shape of the spheroid. It arises due to the boundary conditions that the field must satisfy across a discontinuity in magnetic permeability. A solution of the boundary value problem (Stratton, 1941) will return demagnetization factors, F, that together with the Earth s field B o, determine the strength of the induced magnetization, M in = FB o / µ o, (3) where we have written F as a 3x3 diagonal matrix with (F 2,F 2,F 3 ) along the diagonal (axial symmetry implies that F 1 =F 2 ). An important consequence of self-demagnetization is that the induced magnetism can change significantly with orientation. For instance, with an aspect ratio of 4 we find F 2 =2.1 and F 3 =12.7 so that the magnetism when the spheroid semi-major axis is aligned with the field is around 6 times greater than when it is perpendicular. With this uniform magnetism the magnetic field of the spheroid can be calculated exactly using prolatespheroidal harmonics (Stratton, 1941) but we chose to use a multipole expansion; the details can be found in McFee (1989). The first non-zero moment is the dipole m, which is given by the expression, m V π 3 = M in = M in 6 ea, (4) where V is the volume of the spheroid i.e. the dipole is the product of the magnetization with volume. The magnetic field above a dipole will be dependent on the magnitude and orientation of the dipole moment, as well as the distance and orientation of the observation point. Typically, the anomaly over a dipole consists of a negative peak in the direction of the dipole and a positive peak in the opposite direction. Given an observed magnetic anomaly we can invert the data to recover an estimate of the dipole magnitude, orientation and position. 3

4 For a spheroid, the octupole is the next non-zero moment after the dipole but its field dies off rapidly with distance. This means that once the sensor distance exceeds a few body lengths, the field is essentially dipolar. Discrimination methodology Any magnetic discrimination methodology must account for each of the following aspects 1. UXO and scrap will exhibit remanent magnetization. While there is some evidence to suggest that on impact UXO s lose most of this remanence due to shock demagnetization (Altshuler, 1996; Nelson et al., 1998) it has been our experience that the remanence of such UXO s becomes realigned with the Earth s field. Unlike UXO, the remanence of fragmentation and scrap will be randomly oriented relative to the Earth s magnetic field; 2. Anomalies are dominated by the dipole moment; 3. The dipole moment alone cannot constrain the geometry of a buried item; 4. The dipole moment for a given spheroid varies in both magnitude and direction depending on the spheroid s orientation relative to the Earth s field We want to invert the data about each anomaly in such a way that we can recover information on the likelihood the item is a UXO. Due to the dipole nature of anomalies and the ambiguity of the dipole moment, attempting to directly invert for the spheroid s dimensions is unlikely to meet with success. Rather, the approach we take is to invert for the dipole moment, determine its remanence and then use these properties to make inferences about UXO likelihood. We use standard geophysical inversion techniques to estimate a dipole moment for each anomaly. From historical records (if available) or the results of existing excavations, we can infer what ordnance items are likely to be present in an area. We will refer to such a list as an ordnance library. For each item, i, in our library, we calculate the orientation that causes the minimum difference, m, between the moment of the ordnance and the one recovered by inversion, m. For each item in the library we will then have an estimate of the minimum percentage of remanent magnetization required to make it match the observed dipole, m rem i = 100, (5) m For example, assume that we are analyzing two separate anomalies. For the first item we calculate (Table 1) the amount of remanent magnetization required to match each of the five items in the ordnance library (assumed to consist of 60 and 81 mm mortars, 76, 105 and 155 mm projectiles). Of most interest for the purpose of discrimination is the minimum remanent magnetization, which is 17% (for the 81 mm mortar). For the second recovered moment, we also calculate the minimum remanence which turns out to be ~38% (Table 1). The basis of our discrimination method is that item 1 with a remanent magnetization of 17% is more likely to be a UXO than item 2 with a remanence of 38%. At no point do we ever make a hard prediction that a given item is a UXO or scrap. Rather, using the recovered dipole moments for each anomaly, we rank them according to the remanent magnetization required to match an item in our ordnance library. The practical application of this ranking may be treated in two ways. Because items with small remanence are assumed more likely to be UXO these may be dug first. One continues excavating according to the prioritized list until no further UXO are recovered during the last, say, 50 holes. Alternately, a priority dig list may be compiled containing those items for which the remanence is less than a defined value (for example 50%). A further QC dig list providing thise items with a marginally higher value of remanence (for example, between 50% and 60%) may then also be dug in order to confirm the validity of the chosen magnetization cut-off. 4

5 Ordnance Remanence: % of moment Item 1 Item 2 60 mm mortar mm projectile mm mortar mm projectile mm projectile Table 1: Remanent magnetization required for each item in the ordnance library to match two different recovered dipole moments. On this basis Item 1 was classified as an 81 mm mortar and Item 2 as a 105 mm projectile, with item 1 deemed to be more likely a UXO. In summary our discrimination method is implemented as follows, 1. Fit a dipole to each anomaly; 2. Calculate the minimum remanent magnetization required to match an item in the ordnance library; 3. Prioritize the dig list based on the remanent magnetization; 4. Continue digging until no UXO s were found in the last 50 holes or until a predefined maximum remanence has been reached; and 5. If digging is continued until all seeded items have been accounted for, then confidence that all live-site UXO have been recovered is further enhanced. A discrimination strategy to satisfy regulators? Aspects of the magnetometer processing strategy advocated in this paper were implemented during ordnance clean-up operations at Guthrie Road and Limestone Hills areas in Montana. At these sites a combination of hand-held (quad-sensor array) and vehicle-towed (8-sensor array) magnetometer systems were used depending upon terrain conditions. At Guthrie Road a total of 840 anomalies were identified of which 804 have now been validated. For Limestone Hills 360 validated anomalies are available for analysis. This data now provides an objective test of our discrimination method and the outcome achieved provides convincing evidence of its effectiveness. Initial processing The raw data are subjected to a pre-processing stage before anomaly selection takes place. We apply a 1- D (along-line) high pass median filter with a window width of typically 5 to 10 meters depending upon the ballistic penetration or required search depth. This effectively removes variations in the magnetic field that occur from items deeper than half the window width and from diurnal changes in the field strength (obviating the requirement to use a base-station magnetometer), while leaving the anomalies from shallower items unaffected. Detection / selection of anomalies The next step in the sequence is to identify all possible UXO anomalies in the processed data. This involves picking anomalies with a peak-to-peak amplitude above a defined threshold. As a general rule, lowering the detection threshold, (the signal amplitude above which any response is interpreted as potentially having a metallic source) will increase the number of items detected. This is because weaker signals arising from smaller and/or deeper items become detectable. As the detection threshold is reduced to a level approaching the noise floor in the data (the background variation due to minerals in the soil or system noise), the number of detected signal peaks can be expected to increase considerably. In most situations, adopting a high threshold will result in relatively few false alarms (arising from fragmentation or geological sources) but the detection depth for the target UXO will also be decreased. By lowering the 5

6 threshold to the noise floor the detection depth for UXO will be maximized but at the expense of incurring a very high anomaly detection rate as every item of fragmentation and variation in ground mineralization will be detected. If too many of these false targets are included in the collected data, then it will be extremely difficult to provide fully effective discrimination. Therefore, it is proposed that a maximum detection depth for the type of UXO present be introduced - based either on the maximum expected penetration depth or the future land use. G-tek has collected sufficient historic data to enable an objective determination of this threshold to be made. This then determines the depth to which UXO detection has been achieved for each type of UXO. In practice, an appropriate amplitude cut-off that meets the requirements is first agreed with the regulators. This cut-off may be most stringently determined using ballistic modeling to determine the maximum penetration depth for the relevant UXO types or it may be determined less stringently by the intended land use. In some cases the geological conditions will determine the lowest cut-off that can be applied and if this is higher than that desired from the previous two considerations then all stakeholders should be made aware of the detection depths that actually are achievable. For the Guthrie Road and Limestone Hills data the interpretation threshold was determined from ballistic considerations to be 7.5 nt representing a depth threshold of approximately 28 inches for a 76mm projectile and 65 inches for a 155mm projectile. Discrimination The first step in the discrimination process is to fit dipoles to each of the anomalies using our dipoleinversion routine. Before we utilize the recovered dipole moment we first calculate the correlation coefficient between the predicted and observed data. If this number is below 0.7 then we consider that we were not able to fit the data adequately and hence can t reliably make inferences about the UXO likelihood. The anomalies with failed fits go into a list that needs to be treated separately. Of the 804 anomalies inverted at Guthrie Road, 65 (or 8%) have failed fits. Of these none were from items confirmed to be UXO. At Limestone Hills, 360 anomalies were inverted, with 17 failed fits (5% failure), again with no items validated as UXO. Before committing to digging all the items in the list we first go back to the processed data and determine why the items were picked in the first place and whether there is a data issue that can be easily remedied. Where a sensible fit cannot be obtained after reanalysis or the anomaly cannot be rejected due to an objective reason, the items go into a separate list that needs to be excavated. Almost all the anomalies with failed fits were very small in amplitude (typically 10 nt or less peak to peak) and/or consisted of two overlapping anomalies. Guthrie Road results: Because of the military history at this site we need only two items in our ordnance library (76 mm projectile and 81 mm mortar) in order to determine the minimum remanent magnetization of the anomalies with good fits. Once we have calculated the minimum remanent magnetization we place the anomalies into a prioritized list. Figure 1a shows the percentage of UXO recovered as the excavation progress and also shows the situation at 50% remanent magnetization. There is a very high yield of ordnance with holes dug until about 85% of the ordnance has been recovered. The yield then levels off slightly but all ordnance are recovered by the time 200 holes have been dug. Note that it is more difficult to recover the additional 25 items that were buried as part of the QC process because the remanence of these items is less likely to be aligned with the Earth s field. We note that all UXO s (except two items in the QA list) are recovered using a cutoff of 50% remanent magnetization. If we adopt the system that we stop digging after 50 holes have been excavated with no UXO being recovered then we need to dig only 243 holes (because the last UXO was found in hole 193). Assuming as a worse case, that we also need to dig all anomalies with failed fits (65) then we have to dig a total of 303 holes. This is a significant reduction from the 776 holes (804 minus the 25 QA items) required if we don t use discrimination. In summary, only 39% of holes need to be excavated and yet we were able to recover all UXOs. The FAR has been reduced to only 3.4 by this process. The Receiver Operating Characteristic for the Guthrie Road data is shown in Figure 2. The ROC curve is shown as a close up and also at the scale used by Parsons Engineering (2002) in the compilation of the 6

7 Fort Ord Ordnance Detection and Discrimination Study (ODDS). Plotting at this scale emphasizes the excellent discrimination results that can be achieved with this new method. (a) Guthrie Road (b) Limestone Hills Percentage of UXO recovered UXO including QA items UXO only 20 Less than 10 50% Greater than 50% remanent magnetization remanent Number of holes to dig 20 Less than 10 50% Greater than 50% remanent magnetization remanent Number of holes to dig Figure 1: Yield of ordnance with the number of holes dug for (a) Guthrie Road and (b) Limestone Hills. For Guthrie Road/Limestone Hills, 100% recovery equals 83/68 items for UXO including QA items and 55/26 items for UXO only. The red vertical line represents the situation at 50% remanent magnetization for the UXO only case. Percentage of UXO recovered Unmodeled projectiles UXO including QA items UXO only (a) Guthrie Road Limestone Hills (b) Guthrie Road Limestone Hills Percentage of UXO recovered Percentage of UXO recovered False Alarm Rate FAR Figure 2: Receiver Operating Characteristics (ROC) curves for Guthrie Road and Limestone Hills showing the % of live UXO recovered as a function of the false alarm rate. (a) Shown on the same scale as the ROC curves at Fort Ord (Parsons Engineering, 2002); and (b) Close up of the ROC curve. Limestone Hills results: Previous surface sweeps for the Limestone Hills site revealed five different caliber projectiles had been used (76 mm AP/T, White Phosphorous (WP) and HE, 90 mm AP/T, WP and HE, 105 mm illumination, WP and HE, 4.2 illumination and HE, and 155 mm illumination, HE and WP). The size variability of the different types for a given caliber are quite small except for the 4.2 illumination and HE rounds. Therefore, a single set of dimensions were used for each ordnance caliber except for the 4.2 rounds. Prioritizing on remanent magnetization using the six-item ordnance library results in the discrimination results shown in Figure 1b. Again, there is a high yield of UXO until about 85% are recovered at which point there is a slight reduction. There is only one item with greater than 50% remanent magnetization (66%) and it turns out to be a 50-caliber projectile that was not included in our 7

8 modeling list as it is not considered to be UXO. This emphasizes the importance of having an accurate ordnance library, but we note that this item would have been recovered as part of the QC process. Recovering the QA items turns out to be much harder at this site because there are many emplaced UXOs with large remanent magnetization. The most difficult is a 2.75 rocket warhead that was not included in our ordnance library but was part of the QA list (it is the last item recovered at about hole 240). The last UXO (excluding the QA set and the unmodeled item) was found in hole 77, so assuming 50 more holes plus the 17 failed fits implies that 144 holes need to be excavated. Given that minus the QA set there were a maximum of 318 holes, this means that only 45% require excavation to recover all the UXO. The FAR at the point where 100% detection is achieved is 2.5 (Figure 2). As there are a low number of live UXO s at this site (26), by the time the extra QC holes have been dug the FAR is increased to 4.5 Discussion The results from Limestone Hills demonstrate the importance of including all ordnance types within the library. If an ordnance type is not included but its shape and size are similar to other items in the library (e.g. different models of a given caliber) the method has enough flexibility to accommodate the difference in magnetic characteristics. It is important to ensure that there is no omission of an item with significantly different dimension. Thorough archival search, surface clearance and preliminary site characterization should eliminate this possibility. Both the Limestone Hills and Guthrie Road case studies illustrate that emplaced QA items have different magnetic properties to live-sites UXO. This fact reduces the ability to identify these as UXO and it demonstrates that the performance of UXO discrimination using magnetics cannot be reliably measured on seeded sites. The results of these case studies indicate that continued excavation until all seeded UXO have been recovered may provide an excellent Quality Control process which may assure that all live-site UXO will also have been recovered. This is because seeded UXO will tend to have been ranked lower than live-site UXO. Conclusions We have demonstrated the feasibility of a magnetic discrimination method based on ranking items according to the minimum remanent magnetization required to match an item in an ordnance library. At Guthrie Road, 76 mm projectiles and 81 mm mortars were the only ordnance present. By using this discrimination method less than 40% of identified anomalies needed to be excavated to achieve 100% recovery of UXO. More ordnance types were present at Limestone Hills but 100% recovery could still be achieved with less than 45% anomaly excavation. The discrimination method is based on: Sound physical modeling; Objective decision making criteria (that only requires identification of the ordnance likely to be present); and Has enough flexibility to accommodate the variation in shape and material properties within a given ordnance class. Such a scientific and systematic process should be able to provide the necessary results for regulators. While all ferrous UXO will display some magnetic remanence, in the cases of live-sites UXO this remanence will be predominantly oriented in the same direction as the Earth s magnetic field. In the Guthrie Road case study a False Alarm Rate of 3.4 was achieved and in the Limestone Hills the FAR was 4.5. In both cases these excellent FAR results were achieved with 100% detection of the UXO. 8

9 Acknowledgement The authors would like to thank the University of British Columbia, Professor D.W. Oldenburg and L.R. Pasion for their cooperation and collaboration with these case studies and with their help in the preparation of the information used to create this paper. References Altshuler, T. W., 1996, Shape and orientation effects on magnetic signature prediction for unexploded ordnance: Proc. UXO Forum 1996, Bulter D. K., Cespedes E. R., Cox C. B., and Wolfe P. J. Multisensor methods for buried unexploded ordnance detection, discrimination and identification: Technical Report 98-10, SERDP, September McFee, J. E., Electromagnetic remote sensing; low frequency electromagnetics: Technical Report 124, Defence Research Establishment Suffield, January Nelson, H. H., Altshuler, T. W., Rosen, E. M., McDonald, J. R., Barrow, B., and Khadr, N., 1998, Magnetic modeling of UXO and UXO-like targets and comparison with signatures measured by MTADS: Proc. UXO Forum 1998, Parsons Engineering, 2002, Ordnance Detection and Discrimination Study (ODDS), Final Report, US Army Corp of Engineers, Sacramento District. Stratton, J., 1941, Electromagnetic theory: McGraw Hill. Youmans, C., and Daehn, L., 1999, Quality assurance and quality control in UXO remediation: A case study from Montana. In Proc. UXO Forum