Assessing the Efficacy of High-Resolution Satellite-Based PERSIANN-CDR Precipitation Product in Simulating Streamflow Journal of Hydrometeorology

This study aims to investigate the performance of Precipitation Estimation from Remotely Sensed Information Using Artificial Neural Networks–Climate Data Record (PERSIANN-CDR) in a rainfall–runoff modeling application over the past three decades. PERSIANN-CDR provides precipitation data at daily and 0.25° temporal and spatial resolutions from 1983 to present for the 60°S–60°N latitude band and 0°–360° longitude. The study is conducted in two phases over three test basins from the Distributed Hydrologic Model Intercomparison Project, phase 2 (DMIP2). In phase 1, a more recent period of time (2003–10) when other high-resolution satellite-based precipitation products are available is chosen. Precipitation evaluation analysis, conducted against stage IV gauge-adjusted radar data, shows that PERSIANN-CDR and TRMM Multisatellite Precipitation Analysis (TMPA) have close performances with a higher correlation coefficient for TMPA (~0.8 vs 0.75 for PERSIANN-CDR) and almost the same root-mean-square deviation (~6) for both products. TMPA and PERSIANN-CDR outperform PERSIANN, mainly because, unlike PERSIANN, TMPA and PERSIANN-CDR are gauge-adjusted precipitation products. The National Weather Service Office of Hydrologic Development Hydrology Laboratory Research Distributed Hydrologic Model (HL-RDHM) is then forced with PERSIANN, PERSIANN-CDR, TMPA, and stage IV data. Quantitative analysis using five different statistical and model efficiency measures against USGS streamflow observation show that in general in all three DMIP2 basins, the simulated hydrographs forced with PERSIANN-CDR and TMPA have close agreement. Given the promising results in the first phase, the simulation process is extended back to 1983 where only PERSIANN-CDR rainfall estimates are available. The results show that PERSIANN-CDR-derived streamflow simulations are comparable to USGS observations with correlation coefficients of ~0.67–0.73, relatively low biases (~5%–12%), and high index of agreement criterion (~0.68–0.83) between PERSIANN-CDR-simulated daily streamflow and USGS daily observations. The results prove the capability of PERSIANN-CDR in hydrological rainfall–runoff modeling application, especially for long-term streamflow simulations over the past three decades. Grant no. NA14NES4320003

本研究旨在探究基于人工神经网络的遥感信息降水估算气候数据记录（Precipitation Estimation from Remotely Sensed Information Using Artificial Neural Networks–Climate Data Record, PERSIANN-CDR）在近三十年降雨-径流模拟应用中的性能表现。PERSIANN-CDR可提供1983年至今、南北纬60°纬度带及0°-360°经度范围内的日尺度降水数据，时空分辨率为0.25°。本研究在分布式水文模型比对计划第二阶段（Distributed Hydrologic Model Intercomparison Project, Phase 2, DMIP2）的三个试验流域内分两阶段实施。第一阶段选取2003-2010年这一较近时段，此时已有多款高分辨率星载降水产品可供使用。以四级雨量校正雷达数据（Stage IV gauge-adjusted radar data）为参照开展降水评估分析，结果显示：PERSIANN-CDR与TRMM多卫星降水分析产品（TRMM Multisatellite Precipitation Analysis, TMPA）性能相近——TMPA的相关系数更高（约0.8，PERSIANN-CDR约为0.75），二者的均方根偏差基本一致（均约6）。TMPA与PERSIANN-CDR的表现优于原始PERSIANN产品，原因在于二者均经过雨量校正，而原始PERSIANN未采用此类校正手段。随后，本研究采用PERSIANN、PERSIANN-CDR、TMPA及四级雨量校正雷达数据驱动美国国家气象局水文发展办公室水文实验室研究型分布式水文模型（National Weather Service Office of Hydrologic Development Hydrology Laboratory Research Distributed Hydrologic Model, HL-RDHM）。以美国地质调查局（United States Geological Survey, USGS）的河道流量观测数据为基准，采用五种不同统计及模型效率指标开展定量分析。结果表明：在全部三个DMIP2试验流域中，由PERSIANN-CDR与TMPA驱动得到的模拟水文过程线吻合度相近。鉴于第一阶段已取得良好结果，本研究将模拟时段回溯至1983年，此时仅可获取PERSIANN-CDR的降水估算数据。结果显示，基于PERSIANN-CDR得到的河道流量模拟结果与USGS观测数据具有较好可比性：PERSIANN-CDR模拟的日流量与USGS日观测数据的相关系数约为0.67-0.73，相对偏差约为5%-12%，一致性指数约为0.68-0.83。本研究结果证实了PERSIANN-CDR在水文降雨-径流模拟应用中的适用性，尤其适用于近三十年的长期河道流量模拟。本研究受资助编号NA14NES4320003资助。