Data underlying the publication: Sensitivity analysis of the Substance Emission Model v2.1.2 component of the Greenhouse Emission Model

In the analysis ensembles of 7-year SEM simulations were performed for 100 assessments for different scenarios and substances. For each assessment, an ensemble of 365 simulations was performed with varying dates of substance application, covering every day of the year. For each simulation the following postprocessing was performed on the daily substance emission (g.m-2.d-1) from the greenhouse and its 10-day moving average:* Determine the of the annual maximum for each of the 7 simulation years.* Calculate the 50th and 90th percentiles over the 7 annual maxima (referred to as PEC50 and the PEC90, respectively).<br>This results in four PEC values (PEC50--daily, PEC90--daily, PEC50--10-day-average, PEC90--10-day-average) for each of the 100x365 simulations. Next, for each of the 100 assessments, the results of the 365 simulations were processed as follows:* Calculate the 90th percentile over 365 values for the four PEC values--this is referred to as the "true" 90th percentile.* Remove 5 simulations for application dates 7-Feb, 21-Apr, 3-Jul, 14-Sep and 26-Nov, resulting in a set of 360 simulations. This is done because 360 has more divisors than 365.<br>Subsequently, processing was performed on subsamples of different sizes N, taken from the 360 simulations. The following subsample sizes were considered: 12, 15, 18, 20, 24, 30, 36, 40, 45, and 60. For each subsample size N, M_N = 360/N sets of subsamples were taken with application date evenly spread over the year. For example, for N=12, M_12=30 sets of application dates were selected, with each set one day offset to the next. This results in 10 sets of subsamples of varying size. For each set N, the following processing was performed:* For each M_N values for the four PECs, calculate the relative difference compared to the true 90th percentile (based on the full 365 set of simulations; see above) as follows: RD = (PEC_est-PEC_365)/PEC_365.* Calculate the 10th percentile over the M_N relative differences for each of the four PECs; this is referred to as the 90th percentile underestimation* For each M_N values for the four PECs, calculate the multiplication factor relative to the true 90th percentile as follows: MF = PEC_est/PEC_365.* Calculate the 90th percentile over the M_N multiplication factors for each of the four PECs.<br>This results in 4000 values for the relative difference and multiplication factor for each combination of assessment (100), subsample size N (10), and PEC quantity (4). The relative underestimations form the data underlying Figure 13.3 in Braakhekke et al. (2024). The multiplication factors for N=12 form the data underlying table 13.1 in Braakhekke et al. (2024).<br>

本分析针对不同情景与物质开展了100次评估，每次评估均运行7年的SEM集合模拟（ensemble）。
每次评估中，均开展包含365次模拟的集合模拟，施药日期逐日变化，覆盖全年每日。
针对每次模拟，均对温室每日物质排放量（单位：克每平方米·天，g·m⁻²·d⁻¹）及其10日滑动平均值开展如下后处理：
* 确定7个模拟年各自的年度最大值。
* 计算7个年度最大值的50百分位数与90百分位数，分别记为PEC50与PEC90（PEC，Predicted Environmental Concentration）。

由此，100×365次模拟各自均可得到4个PEC数值：日值PEC50、日值PEC90、10日滑动平均值PEC50、10日滑动平均值PEC90。

随后，针对100次评估中的每一次，均对365次模拟的结果开展如下处理：
* 针对4个PEC数值的365个结果计算90百分位数，该值记为"真实"90百分位数。
* 移除施药日期为2月7日、4月21日、7月3日、9月14日与11月26日的5次模拟，最终得到360次模拟的数据集。此举因360的约数数量多于365。

随后，从360次模拟中抽取不同规模N的子样本开展处理，考虑的子样本规模包括：12、15、18、20、24、30、36、40、45与60。针对每个子样本规模N，共抽取M_N=360/N个子样本集，各子样本的施药日期在全年均匀分布。例如当N=12时，M_12=30个子样本集被选取，相邻子样本集的施药日期相差1天。由此共得到10种不同规模的子样本集。

针对每种规模N的子样本集，均开展如下处理：
* 针对4个PEC的M_N个估算值，计算其与真实90百分位数（基于完整365次模拟数据集，详见前文）的相对差异，公式如下：RD=(PEC_est - PEC_365)/PEC_365。
* 针对4个PEC各自的M_N个相对差异值计算10百分位数，该值记为90百分位数低估量。
* 针对4个PEC各自的M_N个估算值，计算其相对于真实90百分位数的乘数因子，公式如下：MF=PEC_est/PEC_365。
* 针对4个PEC各自的M_N个乘数因子计算90百分位数。

由此，针对评估次数（100次）、子样本规模（10种）与PEC类型（4种）的每种组合，均可得到4000个相对差异与乘数因子数值。上述相对低估量即为Braakhekke等人2024年发表的图13.3的基础数据；而N=12时的乘数因子则为其表13.1的基础数据。