6
Uncertainty in DS86

It had been part of the planning for the DS86 to produce a complete uncertainty assessment. However, in spite of the aspirations and plans of the working group and the Committee on Dosimetry for the RERF, a complete assessment of uncertainties did not materialize; and at the time of publication of DS86, a complete uncertainty assessment had not been carried out. A temporary assessment of uncertainty in the DS86 kerma estimates was performed. Chapter 9 of the DS86 report (Roesch 1987) describes the DS86 method of computing the uncertainty in the dose of an individual survivor. Uncertainties associated with various components of the system were also described.

The temporary uncertainty assessment yielded estimated fractional standard deviations (FSDs) for various key parameters and preliminary crude estimates of the correlation among system components. It then combined uncertainties by using standard nonparametric methods that were valid as long as the component FSDs were relatively small (<40%). It made no assumptions regarding probability distributions for these parameters, and it relied heavily on estimates of uncertainty and correlation coefficients given by the various authors of the model components. Additional uncertainty estimates were based mainly on the judgments of the DS86 authors. The correlation estimates, in particular, were based on very little analytic support. Some of the preliminary estimates of FSDs were quite tentative, particularly for the neutron component, in which case the apparent disagreement between measured and calculated thermal-neutron activation suggested a possible unresolved problem in the neutron-transport models.

The review of DS86 report by the National Research Council in 1987 (NRC 1987) recommended that a rigorous uncertainty analysis be undertaken with improved uncertainty input values for each aspect of the dosimetry system. The review



The National Academies | 500 Fifth St. N.W. | Washington, D.C. 20001
Copyright © National Academy of Sciences. All rights reserved.
Terms of Use and Privacy Statement



Below are the first 10 and last 10 pages of uncorrected machine-read text (when available) of this chapter, followed by the top 30 algorithmically extracted key phrases from the chapter as a whole.
Intended to provide our own search engines and external engines with highly rich, chapter-representative searchable text on the opening pages of each chapter. Because it is UNCORRECTED material, please consider the following text as a useful but insufficient proxy for the authoritative book pages.

Do not use for reproduction, copying, pasting, or reading; exclusively for search engines.

OCR for page 79
Status of the Dosimetry for the Radiation Effects Research Foundation (DS86) 6 Uncertainty in DS86 It had been part of the planning for the DS86 to produce a complete uncertainty assessment. However, in spite of the aspirations and plans of the working group and the Committee on Dosimetry for the RERF, a complete assessment of uncertainties did not materialize; and at the time of publication of DS86, a complete uncertainty assessment had not been carried out. A temporary assessment of uncertainty in the DS86 kerma estimates was performed. Chapter 9 of the DS86 report (Roesch 1987) describes the DS86 method of computing the uncertainty in the dose of an individual survivor. Uncertainties associated with various components of the system were also described. The temporary uncertainty assessment yielded estimated fractional standard deviations (FSDs) for various key parameters and preliminary crude estimates of the correlation among system components. It then combined uncertainties by using standard nonparametric methods that were valid as long as the component FSDs were relatively small (<40%). It made no assumptions regarding probability distributions for these parameters, and it relied heavily on estimates of uncertainty and correlation coefficients given by the various authors of the model components. Additional uncertainty estimates were based mainly on the judgments of the DS86 authors. The correlation estimates, in particular, were based on very little analytic support. Some of the preliminary estimates of FSDs were quite tentative, particularly for the neutron component, in which case the apparent disagreement between measured and calculated thermal-neutron activation suggested a possible unresolved problem in the neutron-transport models. The review of DS86 report by the National Research Council in 1987 (NRC 1987) recommended that a rigorous uncertainty analysis be undertaken with improved uncertainty input values for each aspect of the dosimetry system. The review

OCR for page 79
Status of the Dosimetry for the Radiation Effects Research Foundation (DS86) stated “that the full usefulness of the DS86 system could not be realized until the uncertainty in the organ dose estimates have been properly codified and incorporated into the DS86 system. Quantitative information on uncertainty as a function of distance is an important parameter in the analysis of radiation effects.” The present committee was presented with a first draft of such an analysis (Kaul and Egbert 1989). Although it is more complete and rigorous than the preliminary analysis in the DS86 report itself, it has not undergone formal peer review. Furthermore, some of the assumptions and values assigned to various parameters are debatable, as are some of the estimates of correlation between various parameters. On the basis of temporary uncertainty assessment in DS86 and the draft report mentioned above, errors in kerma to a specific organ of a single survivor have been estimated to be represented by an FSD of about 25–40% (NCRP 1997). The two analyses suggested that the largest contribution to uncertainty is house-shielding, which depends primarily on the location of the survivor in the house and his or her shielding conditions. The committee feels that the uncertainty estimates are too low. Technical improvements in the transport models alone—cross sections, energy bin structure, and delayed neutron transport (see Chapter 4, Kaul 2000) — have suggested an increase of about 30% in neutron kerma at Hiroshima at a 1300-m slant range. As discussed elsewhere in this report, the apparent discrepancy between predicted and measured neutron fluences suggests that the uncertainty in neutron kerma for any given person is larger than in the present uncertainty assessments. It is also possible that some unknown sources of error have not been considered in the uncertainty assessments and that the extent of others has been underestimated because of a lack of information. As pointed out in NCRP report 126 (NCRP 1997), uncertainty analysis of the atomic-bomb survivor data that accounts fully for all sources of error in dosimetry would be very difficult even if all the sources could be fully characterized. Recent reevaluations of potential errors in shielding assignments suggest that gamma doses are more uncertain than indicated in the current assessments (Kaul and Egbert 1989). Evidence from biodosimetry data and the Nagasaki factory-worker effects history indicates further that the preliminary DS86 estimates of uncertainty in individual gamma doses are lower than they should be, even though the overall agreement between DS86 calculations and TLD measurements for gamma rays is very good. Uncertainty in DS86 can be divided into three types: systematic uncertainty that would affect estimates of doses to all people or groups at about the same distance in the same manner, random errors resulting from the method, and random errors resulting from the input data. The random errors would affect estimated doses to individuals independently. SYSTEMATIC UNCERTAINTY System components that can contribute to systematic error are device yield, radiation output, height of burst, location of hypocenter, air density, air and soil

OCR for page 79
Status of the Dosimetry for the Radiation Effects Research Foundation (DS86) moisture, transport methodology, fission-product radiation, shielding methodology, organ-dose calculation methodology, and transport cross sections. The evaluation of the yield of the Hiroshima device and its uncertainty should be improved. The uncertainty in the Hiroshima yield is quite high—±2 kT (CV=10%), and recent reevaluations indicate that a higher yield than the 15 kT used in DS86 could be more appropriate (Kaul and Egbert 1998). However, this reevaluation relies heavily on the comparison of measurements and calculations of the 32S activation near the epicenter and might be modified on the basis of the results of forthcoming 63Ni measurements. (The improvement in air cross sections and energy bin structure lower the calculated 32S activation compared with that using the original DS86 method). The number of neutrons escaping the bomb casing is also uncertain and should be reevaluated. The estimated confidence in the Nagasaki yield is much higher than that in the Hiroshima yield (CV=5%) (Roesch 1987). The uncertainty in the height of burst at Hiroshima was estimated to be ±15 m (99% CI) (Roesch 1987). However, a larger error than reflected by this uncertainty could account for some of the discrepancy in measured and calculated thermal-neutron activities close to the epicenter, so this should be reevaluated for any new dose system. The uncertainty in the height of burst at Nagasaki is estimated to be only ±10 m (Roesch 1987). Errors in height of burst translate into uncertainties that vary with distance, with the greatest impact close to the epicenter. Because an error in height of burst at Hiroshima could explain some of the apparent bias observed in comparisons of thermal activation near the epicenter, height of burst at Hiroshima should be reexamined. The location of the hypocenter is believed to be a relatively minor contributor to overall uncertainty. The fission product and thus gamma and neutron delayed sources were calculated on the basis of thermal neutron (reactor) fission yields, and this might have resulted in an error in the radiation-source terms of about 10% for neutrons and 5% (CV) for gamma rays. Shortly before publication of DS86, the energy spectrum of delayed neutrons used in DS86 was found to be harder (that is to contain more-energetic neutrons) than estimated, and the delayed-neutron spectrum given in the DS86 publication is thus not the actual spectrum used in the current official DS86 system (Egbert 1999). Later improvements in the delayed-neutron source and transport indicate that the contribution from delayed neutrons was significantly underestimated in DS86 (Egbert 1999). The revised delayed-neutron transport, which has not yet been implemented in DS86 (see Chapter 4), would tend to reduce the uncertainty associated with the delayed-neutron contribution to kerma, according to the observed improvement in the agreement between activation calculations and measurements at Nagasaki and for NTS tests of devices similar to the Nagasaki device. The delayed neutrons were a relatively small contributor to the neutron kerma at both Hiroshima and Nagasaki (<5%) even after the above improvements, but

OCR for page 79
Status of the Dosimetry for the Radiation Effects Research Foundation (DS86) they contributed about one-third of the neutron activation at 1500 m (ground range) at Nagasaki and 8% at Hiroshima according to the original DS86 calculations (Roesch 1987). The revised delayed neutron contribution to activation should be considered in comparing calculated and measured thermal activation. The prompt-neutron output from the Hiroshima device is also estimated to have an uncertainty (CV) of about 10%. However, both the number and energy distribution of the neutrons from the Hiroshima source might be considerably more uncertain, as discussed below. The new calculation of this source being carried out at LANL (see Chapter 4) should provide an improved estimate of uncertainty for the total radiation output and for the energy and angular source spectra. Errors in air and soil density and moisture content can affect the transport of low-energy neutrons in particular but would probably have only a small impact on the kerma estimates—CV about 5%, according to Kaul and Egbert (1989). However, these errors might have a substantial impact on the calculation of thermal-neutron activation for some locations. The DS86-calculated thermal and epithermal neutron fluences vary by as much as about 25–50% as the altitude increases from 1 m to 25 m (see Chapter 3). A sensitivity analysis of the effect of the uncertainty in these values on the calculated low-energy component of the fluence at various distances and heights should be included in the uncertainty assessment. The variations in low-energy fluence and their possible impact on the comparison between measured and calculated activation are discussed in more detail in Chapter 3. Errors and limitations in the shielding and organ-dose methodology (forward-adjoint fluence coupling) could have had a relatively small impact on the estimated kerma (around 5–10% CV), as discussed by Roesch (1987). The uncertainty in kerma and activation due to uncertainty in air cross-section values increases with distance, particularly for the neutron component (Lillie and others 1988). The effect on the neutron-kerma uncertainty was estimated to be about 15% (CV) at 1500 m; the prompt and secondary gamma CVs were estimated to be about 3% and 6%, respectively. The uncertainty in nitrogen and oxygen cross sections and improvements in the transport code energy bin structure have been extensively investigated since 1986 and appear to have a substantial impact on the calculated kerma in air. Kaul and Egbert (1989) estimate a CV of about 15% for both Hiroshima and Nagasaki for the uncertainty in the neutron kerma due to uncertainty in the new cross sections (see Chapter 4). RANDOM ERRORS RESULTING FROM METHOD Components that can result in random error are uncertainties in the assumed survivor-shielding and organ-dose model. The shielding assignment was estimated to contribute substantially to total uncertainty (Roesch 1987). Estimates of terrain shielding at Nagasaki and the model

OCR for page 79
Status of the Dosimetry for the Radiation Effects Research Foundation (DS86) used to estimate the shielding of factory workers also have large uncertainty. Kaul and Egbert (1998) estimated that errors in the shielding calculations for the Nagasaki factory workers could be responsible for an uncertainty of a factor of 2 in the organ-dose estimates. The global and nine-parameter models used to estimate shielding contribute to overall uncertainty. Uncertainty in shielding was estimated to be the largest contributor to overall uncertainty in the total-kerma estimate—CV about 20–40% (Kaul 1999). The biodosimetry data are consistent with a large random uncertainty of about 45% (Sposto and others 1991; Kaul and Egbert 1998) in DS86 dose assignments (see Chapter 5). The uncertainties in the organ-dose model are believed to be relatively minor contributors to overall uncertainty. RANDOM ERRORS RESULTING FROM INPUT DATA Random error due to uncertainty in input data arises in connection with survivor location, shielding, and survivor orientation. Survivor location and shielding description were estimated to have the greatest contribution to total random uncertainty, primarily because of uncertainty in survivor recall. Kaul estimated that the overall uncertainty in gamma-dose estimates due to uncertainty in survivor recall was around 15% (CV) (Kaul 1999). DS86 kerma estimates might be even more uncertain because of additional systematic bias in the methodology (Kaul and Egbert 1989) that would affect doses to some or all subjects nonrandomly. Comparisons of activation measurements with calculations of activation based on DS86-calculated fluences indicate additional bias in kerma estimates due to a systematic bias in free-field neutron transport. This systematic bias, unlike most of the systematic errors discussed above, appears to depend on the distance from the epicenter. The apparent discrepancies between measured and calculated neutron activation close to the epicenter and at great distances in Hiroshima, and perhaps also in Nagasaki, imply that such bias exists at least for the thermal and epithermal components of the neutron radiation field; at great distances, this bias is large enough to imply that neutron kerma is also affected. Fluctuations of only about a factor of 2 can occur in the low-energy fluence that produces the activation, because the thermal and epithermal neutrons that contribute to the activation of surface and near-surface samples have a small range in air. (About 30–40% of the total calculated fluence is in the thermal bin; even if all the neutrons were thermalized in the sample, the activation would be increased by at most a factor of 2–3.) A large excess (a factor of 2 or more) of thermal and epithermal neutrons at any distance can arise only if there is down-scattering of higher-energy neutrons in the immediate vicinity of the sample. The apparent discrepancy of about a factor of 3–5 that remains at Hiroshima (see Chapter 3) suggests a bias in the relative number of higher-energy neutrons emitted from either the device or the fireball or in the number of higher-energy neutrons able to survive long-range transport in

OCR for page 79
Status of the Dosimetry for the Radiation Effects Research Foundation (DS86) air (the effective relaxation length). Comparisons of calculated and measured fluence from the Aberdeen fission-reactor experiment suggest that uncertainty in transport in air based on DS86 was probably relatively small (less than 50%) and thus cannot account for much of the observed discrepancy. However, many of the comparisons were made with improved transport codes and cross sections rather than with the actual DS86 models and cross-section values (see Chapter 4). The comparison of the TLD data and the DS86 calculations suggests the possibility of a smaller discrepancy (20% or less; see Chapter 2) in Hiroshima in the gamma fluence and thus gamma kerma. However, the possible discrepancy in the gamma fluence might be directly or at least partially related to the apparent discrepancy in the neutron fluence. The preliminary uncertainty assessment in DS86 recognized the apparent discrepancy between the Hiroshima neutron calculations and measurements, and was one of the reasons for the decision to defer a final uncertainty assessment until its cause was resolved. If the improvements in neutron-transport calculations since DS86 described in Chapter 4 are applied, it appears that the neutron-activation measurements and revised calculations in Nagasaki agree fairly well. However, as discussed in Chapter 4, the activation-measurement results for Nagasaki at low activity (at neutron fluences corresponding to those in Hiroshima at distances of 1000–1500 m) are few, and one cannot exclude the possibility that a similar, although perhaps smaller, discrepancy also exists in Nagasaki. It is important to recognize that the disagreement between neutron-activation measurements and calculations in Hiroshima might result from a combination of systematic errors (including errors in yield, height of burst, and energy spectrum of source neutrons), errors in delayed-neutron source and transport, measurement errors (including background subtraction errors), and activation-calculation errors. An additional possible location-dependent discrepancy might also be due to the shielding models used in DS86. The failure to account for shielding by other than immediately adjacent structures could have resulted in underestimating the shielding at great distances. A benchmark study carried out by SAIC for subjects whose doses were calculated with the globe model (Kaul and Egbert 1989) indicates that DS86 has a tendency toward too low a dose, particularly for the gamma-ray component. A more rigorous modeling (which is funded) of the survivor shielding in both Hiroshima and Nagasaki that includes taking account of adjacent structures and terrain features should reduce the uncertainty in these components of the dose system considerably. Similarly, the shielding estimates for the factory workers in Nagasaki are uncertain and the DS86 doses could be biased too high (by as much as 50%), to judge from chromosomal aberration data and ESR data (see Chapter 5). A new study is under way to model the shielding of these workers much more rigorously than was done in DS86. An additional segment of the population for which a potentially large bias in the doses might have resulted is the survivors in Hiroshima who resided behind the hill known as Hijiyama. The two-dimensional air-ground model used in DS86 might reflect the scattering at low activity inaccurately because of terrain variations. Kaul (1999) estimated that the

OCR for page 79
Status of the Dosimetry for the Radiation Effects Research Foundation (DS86) overall uncertainty in gamma dose estimates due to uncertainty in survivor recall are on the order of 15% (CV), which is less than the estimate Jablon once made based on T65D (Jablon 1971). It is clear that any modification of DS86 should be accompanied by a comprehensive uncertainty analysis that treats all the possible sources of error discussed above and combines them properly, accounting for known correlation, to provide a reasonable estimate of uncertainty in the neutron and gamma components as a function of distance, location, and details of exposure (shielding). The analysis should carefully distinguish between random and systematic error because random error can result in a statistical bias in risk estimates that are based on dose estimates (NCRP 1997). Such an analysis is much more feasible now than in 1986 and 1989 because additional information is available on the possible sources of uncertainty, as are faster computers that will allow benchmark and sensitivity studies of the various model components. The analysis should include the sensitivity of the calculated kerma to the uncertainty in the various cross sections used in the model, to small changes in the energy and angular distribution of the radiation emitted from the device and the fireball, and to the use of a two-dimensional air-ground model, as opposed to a model that reflects the varied topography of the city (this might be particularly important for Nagasaki, where the terrain is very irregular). The uncertainty analysis should clearly indicate which kinds of uncertainty are possibly underestimated and what sources of potential error have not been considered because of insufficient information. A comprehensive biodosimetry analysis can also provide additional information and identify bias in the dose-system results for some populations. It would be instructive to estimate probability distributions for the most important contributors to uncertainty. These could be used in a detailed stochastic analysis of the distributions of possible doses for selected representative exposure scenarios, including a wide range of distances from hypocenter and shielding configurations. Such an analysis would be much more informative than the simple estimation of total coefficients of variation based on combining fractional standard deviations, and it would provide more-reasonable estimates of the confidence limits on the dose estimates for various representative exposed subjects. Finally, a comprehensive uncertainty analysis should include a rigorous estimate of the uncertainty in the calculations of neutron activation and TLD dose used to confirm the transport models; additional uncertainties that are not included in the uncertainty model for kerma come into play in these calculations. These include uncertainties in activation cross sections, attenuation in samples, sample orientation, backscattering effects, and the shape of the calculated neutron spectrum at low energy; the latter uncertainty is due to the limited energy bin structure of the transport models. As seen in Chapter 3 in Table 3–1, DS86 contains only a single-thermal energy bin that includes all neutrons below 0.4 eV. An average thermal cross section must be applied to the uncalculated spectral distribution of neutrons below 0.4 eV, and group-averaged cross sections must be used

OCR for page 79
Status of the Dosimetry for the Radiation Effects Research Foundation (DS86) for the limited number of epithermal-energy groups. Note that the latter does not effect the kerma calculations, which are dominated by high-energy radiation. Although uncertainty in the calculation of activation is unlikely to account for a sizable fraction of the observed neutron discrepancy, it might have an impact on the comparison of thermal-activation measurements versus calculation. The discrepancy discussed above in the shielding model with respect to intervening buildings may also be relevant to the thermal neutron activation calculations at great distances in creating additional scattering and thus a possibly greater fraction of lower-energy neutrons; this is discussed in more detail in Chapter 3. The apparent discrepancy between measured and calculated activation and TLD measurements might not be completely resolved by a revised dosimetry system that incorporates improved source and transport calculations. Nevertheless, a thorough uncertainty assessment can provide credible estimates of the confidence limits on the major component of the dose, the gamma rays, and for the lesser neutron component for representative exposed subjects. It should increase the confidence of both the scientific community and the Japanese population in the validity of the new dose system.