Improving Hydrological Models With the Assimilation of Crowdsourced Data
Improving Hydrological Models With the Assimilation of Crowdsourced Data
复制标题
DOI：
10.1029/2019wr026325
复制
发表时间：
2020-05
影响因子：
5.4
通讯作者：
P. Avellaneda;D. Ficklin;C. Lowry;J. Knouft;D. M. Hall
中科院分区：
文献类型：
--
作者：
P. Avellaneda;D. Ficklin;C. Lowry;J. Knouft;D. M. Hall
小溪流往往缺乏可靠的水文数据。环境机构在提供此类数据方面发挥着关键作用;然而，这些机构往往面临日益增长的监测需求和资金短缺的挑战。鉴于观测数据与小流域/水源之间的空间不匹配，当地志愿者可以作为潜在的宝贵研究伙伴。我们研究如何CrowdHydrology，公民科学计划，收集流阶段和流温度观测，改善了美国密歇根州博因河的水文模型。志愿者在四个校准点提供观测，观测的间隔时间不同。我们测试了流阶段和流温度观测（由志愿者测量）是否提高了博因河的土壤和水评估工具（SWAT）模型的性能。使用集合卡尔曼滤波器将观测结果纳入模型。这个框架使我们能够整合观测误差，跟踪模型参数的变化，并模拟每天的径流量和流温度的分水岭。每日模型性能的衡量标准包括Nash-Sutcliffe效率、修改后的Nash-Sutcliffe效率（Ef-mod）、精确一致性指数（dr）和相对偏差（Bias）。对于所有校准地点，与基于初始/默认SWAT参数的模拟相比，数据同化后的流量估计有所改善。模型性能的不同措施出现的基础上的间隔时间的观察。结果表明，当地志愿者收集的观测，具有一定的时间分辨率，可以改善SWAT水文模型和捕捉集中趋势。
Small streams often lack reliable hydrological data. Environmental agencies play a key role in providing such data; however, these agencies are often challenged by the growing monitoring needs and lack of funding. Given the spatial mismatch between observed data and small watersheds/headwaters, local volunteers can act as potentially valuable research partners. We examine how CrowdHydrology, a citizen science program that collects stream stage and stream temperature observations, improves a hydrologic model of the Boyne River, Michigan, USA. Volunteers provided observations at four calibration sites with different interarrival times of the observations. We tested whether stream stage and stream temperature observations (measured by volunteers) improved the performance of a Soil and Water Assessment Tool (SWAT) model of the Boyne River. Observations were integrated into the model using the ensemble Kalman filter. This framework allowed us to integrate observation error, track the variability of model parameters, and simulate daily streamflow and stream temperature across the watershed. Measures of daily model performance included the Nash‐Sutcliffe efficiency, modified Nash‐Sutcliffe efficiency (Ef‐mod), refined index of agreement (dr), and relative bias (Bias). For all calibration sites, estimates of streamflow improved after data assimilation compared to simulations based on initial/default SWAT parameters. Different measures of model performance emerged based on the interarrival times of the observations. Results demonstrate that observations collected by local volunteers, with a certain temporal resolution, can improve SWAT hydrological models and capture central tendency.