Data_Sheet_1_Estimation Bias in Water-Quality Constituent Concentrations and Fluxes: A Synthesis for Chesapeake Bay Rivers and Streams.pdf

Flux quantification for riverine water-quality constituents has been an active area of research. Statistical approaches are often employed to make estimation for days without observations. One such approach is the Weighted Regressions on Time, Discharge, and Season (WRTDS) method. While WRTDS has been used in many investigations, there is a general lack of effort to identify factors that influence its estimation bias. This work was aimed to (1) synthesize and compare WRTDS estimation bias for constituent concentrations and fluxes for rivers and streams in the Chesapeake Bay watershed (including headwater sites) and (2) identify controlling factors from five broad categories (watershed size, sampling practice, concentration and discharge conditions, land use, and geology). Five major constituents were considered, namely, suspended sediment (SS), total phosphorus (TP), total nitrogen (TN), orthophosphate (PO4), and nitrate-plus-nitrite (NOx). For both concentration and flux, estimation bias follows the general order of SS > TP > PO4 > TN ≈ NOx. Median TN and NOx bias statistics were near zero, with an equal distribution of small positive and negative bias. TP, PO4, and SS each showed a median positive bias across sites of <18% for flux and <7% for concentration. Particulate constituents, especially SS, tend to have larger bias at sites with smaller sampling frequencies, shorter sampling record lengths, and smaller watershed sizes. Results of multivariate models showed that both flux and concentration biases are most affected by concentration and discharge variabilities and the length of concentration record. In comparison, flux bias of particulate constituents is more affected by flow variability, whereas flux bias of dissolved constituents is more affected by concentration variability. Moreover, analysis using classification and regression trees provided additional information on how the factors affected flux bias: when all site-constituent combinations are considered, large flux biases are more likely associated with sites that have large concentration and discharge variabilities, small lengths of concentration record, and small sampling frequencies. These results may be useful for identifying sites with large biases, modifying monitoring practice at existing sites to reduce those biases, and choosing new monitoring locations in the Chesapeake watershed and beyond.

河流水质组分的通量量化长期以来都是研究热点领域，统计方法常被用于估算无实测数据时段的结果，其中一类方法为基于时间、流量与季节的加权回归法（Weighted Regressions on Time, Discharge, and Season, WRTDS）。尽管WRTDS已在诸多研究中得到应用，但目前普遍缺乏对其估算偏差影响因素的识别工作。本研究旨在达成两项目标：（1）综合并对比切萨皮克湾流域（含源头水体站点）河流与溪流的水质组分浓度及通量的WRTDS估算偏差；（2）从五大类别（流域面积、采样实践、浓度与流量条件、土地利用及地质）中识别控制估算偏差的关键因素。本次研究共纳入5种主要水质组分，即悬浮泥沙（SS）、总磷（TP）、总氮（TN）、正磷酸盐（PO4）以及硝态氮与亚硝态氮（NOx）。无论是浓度还是通量估算，其偏差的普遍排序均为SS > TP > PO4 > TN ≈ NOx，TN与NOx的偏差中位数接近0，正负偏差分布相对均衡；TP、PO4及SS的通量偏差中位数均小于18%，浓度偏差中位数则小于7%，且均表现为正偏差。颗粒态组分（尤其是SS）在采样频率较低、采样记录时长较短以及流域面积较小的站点中往往呈现更大的估算偏差。多变量模型结果显示，通量与浓度偏差均受浓度与流量变异性以及浓度记录时长的影响最为显著；相较而言，颗粒态组分的通量偏差更易受流量变异性影响，而溶解态组分的通量偏差则更易受浓度变异性影响。此外，通过分类与回归树分析进一步揭示了各因素对通量偏差的影响机制：当考虑所有站点-组分组合时，较大的通量偏差更易出现在浓度与流量变异性大、浓度记录时长较短以及采样频率较低的站点。上述研究结果可用于识别高偏差站点、优化现有站点的监测实践以降低估算偏差，同时可为切萨皮克湾流域乃至其他区域的新增监测点位选择提供科学参考。