This note documents an investigation into the nature of the data that cannot be fitted by the current background model. Specifically, there are large excursions in the count rate observed during the one day background observation on 97/01/01. Light curves were extracted from the data using the standard selection criteria for faint sources (Table 1). The observed rate (sky plus internal background), model predicted rate, and observed-model rate are shown for 5 PCU, front layer, and channels 0-27 (Figure 1). Large flares are present in these data, and are clearly not accounted for in the background model. It has been verified that these large flaring events are present in the individual PCU, and are seen at a lower level in the middle layer.

The L7 rate should scale linearly with the observed background rate. However, it is clear that in these data, the linear relationship breaks down during the large flaring events (Figure 2). The data could be "cleaned" by fitting the L7/observed count rate linear relationship, and deleting data points which lie N standard deviations from the best-fit (where N is a user defined quantity). However, for variable sources, the L7 rate does not scale linearly with the observed count rate, and the data cannot be cleaned using this method.

Similar flaring events are seen in the propane layer (VpSpecPcuN; N=0...5). The propane rates for the individual detectors were co-added and the resultant rate was normalized to the background rate as measured by the Q8VxVpXeCnt rate (Figure 3). Although the propane/Q8VxVpXeCnt count rate ratio increases rapidly during the large flaring events, the ratio also increases during periods of quiescence background. Thus, the ratio does not provide a good predictor of erratic background; similar ratios were obtained using different HK rates as the denominator.

Finally, we consider the possibility that the rapid increase in the background is caused by the geomagnetically trapped particles with moderate energy. The observed-model rate is shown as a function of the angle btween the satellite pointing direction and the geomagnetic field line (Figure 4). Although there is an increase in the residual rate close to 90 degrees, in agreement with that found for the Ginga satellite, there is a large dispersion in the data.

We conclude that the outliers occur durin periods when the L7/background rate linear relationship breaks down. However, it is not clear what the nature of this breakdown is, or whether periods of similar breakdowns can be identified in source data.

Table 1: Selection Criteria

Criterion	Description
(BKGD_THETAa < -70.0 or BKGD_THETA > 150.0) or TIME_SINCE_SAAb > 30.0 or TIME_SINCE_SAA < 0.0	Outside the South Atlantic Anomaly
PCUn_ON 0.5	PCUn fully on
ELV > 10.0	Angle between the satellite pointing direction and the Earth's horizon
OFFSET < 0.01	Angle between the satellite pointing direction and the nominal field of view
VpX1L, VpX1R < 6.25	Veto rate between the propane layer and the left/right halves of the Xenon layer
(VpX1L+VpX1R)/Q6c < 0.1	Normalized propane layer veto rate