1、基于基于ArcGIS的的水水利利大大数数据据及及应应用用团队简介水利大数据及其面临的挑战基于水利大数据的多灾害信息集成与风险预警案例主主要内要内容容123二、水利大数据及其面临的挑二、水利大数据及其面临的挑战战 水利工作关系到国计民生，尤其是我国水资源 分布存在严重的时空分布不均特性，旱灾洪涝 易发多发。水利行业在经济、生态、社会等方 面都扮演着重要角色，对水利大数据的研究具 有重要的现实意义和应用价值。水利大数据是在大数据的理论指导及技术支 撑下的水利科学和工程的重要实践。水水利工作利工作及及水利大水利大数数据据的的重重要要性性水水利大数利大数据据 水利大数据是指产生于各种水文监测网络、水利
2、设施、用 水单位和水利相关经济活动，并通过现代化信 息技术高效传输、分布存储于各地存储系统、但又可以快速读取集中于云端、实现深度数据 挖掘并可视化的海量多源数据总和。ValueVelocityVolume海量快速价值Variety多样Veracity真实交叉性，由于水利和其它领域具有交叉性，因此水利大数 据和遥感大数据、气象大数据、海洋大数据等交叉；时空分布性，需要依赖先进大数据技术进行处理分析，包 括分布式大数据存储框架、机器学习等数据挖掘方法；多元循环性，由水的多元循环决定的水利大数据在经济、社会、生态等领域的价值循环。水水利大数利大数据据的的外延外延挑战一：水挑战一：水利利大数据大数据的
3、收集与的收集与集集成成 水利大数据来源广泛，不同的监测平台得到的 数据具有不同的数据结构、存储系统，非结构 化数据、半结构化数据、结构化数据并存；由于观测条件的差异，数据可信度层次不齐，对数据清洗和质量的确保提出了很高的要求；大数据的存储与管理需要新型数据库的支持，水利大数据的信息化还未与新型数据库接轨。水水利大数利大数据据面临面临的的挑战挑战挑战二：水挑战二：水利利大数据大数据的时空多的时空多维维度分度分析析 水利大数据具有明显的时空分布特性，时间、空间双维度下的数据分析具有难度；水利大数据在其应用领域讲究实时性，比如洪 水预报等，这对大数据的处理分析速度提出了 高要求；水利大数据的深度挖掘
4、有赖于引入先进的人工 智能算法，两者的有效结合至关重要。水水利大数利大数据据面临面临的的挑战挑战挑战三：水利大数据的共享与挑战三：水利大数据的共享与安安全全 众多水利数据掌握在政府机关部门，为非公 开数据，形成数据孤岛现象；水利数据是国家安全的重要组成部分，水利 数据的共享与安全是一个值得探讨的问题。水水利大数利大数据据面临面临的的挑战挑战三、基于水利大数据三、基于水利大数据的的多灾害信息多灾害信息集集 成与风成与风险险预警案例介绍预警案例介绍基于水利大数 据的多灾害信 息集成与风险 预警案例介绍1、天、地、空、海，多基多源降水数据采集2、移动众包信息收集可视化云平台mPing3、基于水利大数
5、据的全球洪水泥石流灾害预测预报4、基于概率洪水风险预报EF55、城市洪水模型Urban CREST介绍6、全球风暴数据库及CI-FLOW7、中国区域多尺度洪水模拟及预警系统的建立8、基于ArcGIS的FFG介绍9、基于ArcGIS平台开发的ArcCREST介绍基基于水利于水利大大数据的数据的多多灾害信灾害信息息集成与集成与风风险预警险预警案案例例介介绍绍3小时临近预报（250米2.5分钟）36小时模型预报（1公里小时）1.天天、地地、空、海空、海多多基多源基多源降降水数据水数据采采集集 双双偏振雷偏振雷达达卫星卫星站点站点模型模型PERSIANN 全球卫星产品(4km,hourly)Hong
6、et al.,2004,JAM;5颗地球静止卫星（可见光颗地球静止卫星（可见光红红外）外）以以及及4颗颗极轨卫极轨卫星星（雷（雷达达和被动和被动微微波）波）通过通过人工神经网人工神经网络络ANN机器学机器学习习训练训练反演反演 High Quality 卫星降水产品卫星降水产品Merge Satellites,ground(Radar&Gauge),and Model(NWP)TRMMAquaDMSPNOAAMETEOSAT(Europe)GOESGMS/MTSAT(Japan)TMPA uses 4 Polar-orbital microwave satellites(NOAA,DoD,NA
7、SA)and 5 Geo-IR satellites(GOES8-10,GMS,MYSAT,MeteoSAT);all calibrated by TRMM Preci Radar17+years(98-16)of data;Most requested TRMM product from NASAWith Huffman et al.2007:（1700+引用）引用）2005 加入加入 NASA：多卫星联合反演共性技术多卫星联合反演共性技术；(1700+引用引用)全球天地空标准产品系列全球天地空标准产品系列：TMPA30-day HQ coefficientsInstant-aneous S
8、SM/I TRMM AMSR AMSU3-hourly merged HQHourly IR TbHourly HQ-calib IR precip3-hourly multi-satellite(MS)Monthly gaugesMonthly SGRescale 3-hourly MS to monthly SGRescaled 3-hourly MSCalibrate High-Quality (HQ)Estimates to “Best”Space RadarMerge HQ EstimatesMatch IR and HQ,generate coeffsApply IR coeffi
9、cientsMerge IR,merged HQ estimatesCompute monthly satellite-gauge combination(SG)30-day IR coefficients26深深度度学习学习方方法研法研制制全球全球卫卫星产星产品品研研制制青藏西南部IR云图相应时段降水情况在深度学习中，我们可以将不同频段的可见光、红外、微波影像同时作 为训练数据输入模型，且不需要事先设定Feature，海量的遥感影像下，让 模型自己去寻找Feature。5-minute 250mRainfall Dataover USA2.mPING 美国版灾害Crowdsourcing移动
10、平台技术2.移动众包信息收集可视化云平台 mPING Crowd Sourcing Tool and Data750,000+App Downloads Since Dec 2013硅谷SF IoT/BigData Weather 2.0 Service Inc.Ensemble Coupled Hydro-Landslide Modeling SystemWater Balance ComponentCREST(Variable Infiltration Curve)SAC-SMARunoff RoutingCell-by-cell linear reservoirLandslide Mod
11、el EnsembleTRIGRSSLIDE+Surface Flow and InundationSoil Water ContentOther variablesOccurrence and Locations of landslidesRemote Sensing basedPrecipitation EstimatesTopographyLand cover/Land Use3.基于水利大数据的全球水洪泥石流灾害预测预报 National Flash Landslide System3.基于水利大数据的全球水洪泥石流灾害预测预报美国暴雨山洪泥石流灾害链业务化系统NFL:NMQ：Nati
12、onal Mosaic and Multi-Sensor QPE(NMQ)FLASH：Flooded Locations And Simulated HydrographsLANDSLIDE:SLope-Infiltration-Distributed Equilibrium ModelNMQ Radar Precipitation Observations 250 m/2.5 minFLASH Distributed CRESTHydrologic Models10-11 June 2010,Albert Pike RecArea,Arkansas250 mm150200Simulated
13、surface water flow20fatalitiesLANDSLIDELandslide Hotspot ModelsRed:ObservationsPink:PredictionsLandslide predictionIntegrated Hydrologic-Landslide Model iCRESLIDE=CREST+SLIDECoupled Routing and Excess STorage(CREST)Jointly developed by OU/NASARun operationally overglobeDistributed,fully coupled runo
14、ff generation and routingWang amnoddHe lo n g et al.2011 HSJIntegrated Hydrologic-Landslide Model:iCRESLIDEDevelopment and Application-CREST has been set up at both national and basin scales in China;-iCRESLIDE shows great capability in forecasting shallow landslides around the world;-More flood and
15、 landslide event data is needed.250m/5-min resolution of Q2 precipitation forcing and model outputsAddresses service needs in NWS;flash flooding is#1 weather-related killer6/11 12:30am-4am 20 deaths:Little Missouri River Crested from 3 ft to 23.5 ft within 2 hoursInclude data assimilation and probab
16、ilistic productsReadily incorporate dual-pol radar products(Q3)and stormscale ensemble forecastsNFL:Real-time,direct prediction of flash floods a realityPhoto source:National Geographic美国暴雨山洪泥石流灾害链耦合系统核心模型Physically-coupled iCRESTSLIDE(SLope Infiltration-Distributed Equilibrium)020408010012000.210.8
17、0.60.460Radius(m)POD FAR CSIValidation with inventory dataRed:Observations Pink:Predictions美国北卡州 梅肯县Within 18-m 120-meter buffer zonePOD 0.50.9CSI 0.10.8FAR 0.90.2(Liao et al.,2011,Nat.Hazards)16th hrFS Map vs.Time18th hr21st hrForecast Streamflow(2010)Recurrence Interval(2010)Inundation(2015)State-
18、Param Estimation DREAM(2010)Observed StreamflowGroundwater MODFLOWRouting Kinematic wave(2014)Linear reservoir(2010)4.基于概率洪水风险预报 EF5Ensemble Framework For Flash Flood ForecastingBest distributed hydrologic System yetPrecip Forcing1.MRMS2.TMPA RT3.WRR/HRRR QPFEvapotranspiration1.FEWS NET PET2.HRRR te
19、mp3.VIIRS?Surface Runoff CREST(2010)SAC-SMA(2013)Hydrophobic(2015)Snowmelt SNOW-17 (2015)-2m TempCurrent VersionFutureAdditionEF5:Probability of Flash Flood Forecast(PFFF)基于概率洪水风险预报100%50%0%PFFF(RP=5 yr)The New Features of uCREST Model 1-10 Meter DEM and Urban Drainage System Urban Canopy and High R
20、ise Building Impact on the RainfallInterception Enhanced Impervious(pavement,roof etc.)and Non-impervious surface infiltration and Surface Processes(runoff,ET etc)Urban Sewer/Pipeline Module included as a special InterflowProcess/reservoir Has been tested and implemented in Oklahoma City and Dallas
21、Metropolitan at spatial resolution5.城市洪水模型Urban CREST介绍A High-Resolution UrbanCREST Flood Modeling and Mapping System For Urban and Built-up Environments2010 June 14,OKC Flash Flood101 km1Return Period(years)210200+No FloodingFloodingSevereFloodingUrban-CREST Flood Model Implemented at Oklahoma City
22、&Dallas Metropolitan137 km6.全球风暴数据库及CI-FLOW Global Storms(2000-2010)*Sellars et al.(2013),Computational Earth Science:Big Data Transformed Into Insight,EOS Trans.AGU,94(32),277Nov 2011 BAMSThe CI-FLOW Project:A System for Total Water Level Prediction From The Summit To The SeaCI-FLOW summary paper w
23、ith Hurricane Isabel,Hurricane Earl,&Tropical Storm Nicole resultsVolume#Number#November 2011BAMSAmerican Meteorological SocietySuzanne Van Cooten,Yang Hong,et al.,2011:The ci-flow project:a system for total water level prediction from the summit to the sea.Bull.Amer.Meteor.Soc.,92,14271442.已应用到美国北卡
24、罗来纳州、墨西哥湾等易受飓风和风暴潮影响的海岸带地区海洋风暴潮与内陆洪水监测预警系统(CI-FLOW)CI-FLOWCoastal and Inland Flooding Observation and WarningTracking the raindrops and disasters from the SKY and the SUMMIT to the seaCI-FLOW:HL-RDHM/SWAN/ADCIRC Coupled ModelPrecipitationSig.Wave HeightsTotal Water LevelsRiver BCsDischargeSurface BC
25、sPressure Wind ForcingSurface BCsWave ForcingHydrodynamic Model(ADCIRC)HydrologicModelAtmospheric ModelWave ModelPrecipitation Source:QPE/QPF Atmospheric Model:NAM or NHC trackHydrologic Model:HL-RDHM,Vflo or CRESTWave Model:unstructured SWAN7.中国区域多尺度洪水模拟及预警系统的建立中国的山洪预警系统量融合，驱动CREST模型，模拟径流分布 与气象局以及国
26、家气象中心合作开发 多源降水产品和地面台站数据进行雨 地貌水动力学模型模拟洪水淹没情景的时空演进，实时动态提取洪水淹没范围、水深分布和淹没时间分布，实现对洪水的模拟洪水模拟拟的时时间间：199806280501001502002503000500010000150002000025000Date 3/5/19975/8/19977/11/19979/13/199711/16/19971/19/19983/24/19985/27/19987/30/199810/2/199812/5/19982/7/19994/12/19996/15/19998/18/199910/21/199912/24/19
27、992/26/20004/30/20007/3/20009/5/200011/8/20001/11/20013/16/20015/19/20017/22/20019/24/200111/27/20011/30/20024/4/20026/7/20028/10/200210/13/200212/16/20022/18/20034/23/20036/26/20038/29/200311/1/20031/4/20043/8/20045/11/20047/14/20049/16/200411/19/20041/22/20053/27/20055/30/20058/2/200510/5/200512/8
28、/2005R_Obs in(m3/s)R(v2.1)in(m3/s)rain率定期验证期NSCE=0.897CC=0.947Bias=-1.57%20 年、10 年、5年、2年、1年 一遇洪水外州站CREST模型率定/模拟效果：气象台站数据驱动7.中国区域多尺度洪水模拟及预警系统的建立114114.5115115.5116116.51172525.52626.52727.52828.529iMAP 在嘉陵江流域的应用结果7.中国区域多尺度洪水模拟及预警系统的建立9.基于ArcGIS平台开发的ArcCREST介绍 ArcCREST UIPrecip ThiessenEvap ThiessenGe
29、o DataUsed for rainfall sites(Cell-based data need some effort)Parameters distribution need more advanced methodBugs in code,the results are not correctGeo and Hydro data management and operationParameters distribution settingModel running and results showArcCREST运行结果分析ArcCRESTv1.0(Uncalib)ArcCRESTv
30、1.0Nash-Sutliffe-0.415460.8121Bias(%)-99.999915.25CC0.79630.8382300200100040050011325374961738597109121133145157169181193205217229241253265277289301313325337349361Discharge(m3)Time(24h)Discharge:ArcCREST vs GageCalibUnCalibActualR=0.7025501001502002503000050100150200250300350ArcCRESTGageDischarge:Ar
31、cCREST vs GageR2=0.7025ArcCREST tends to overestimate dischargeUncalibrated results indicate no model sensitivity and unreliable estimations Flash Flood Guidance:FFG is the amount of rainfall required in a given period of time to produce bank full conditions on small basinsfrom Flash Flood Guidance
32、1970toHydrologic Flash Flood Guidance 201227-4343-5454-6868-8282-9898-115115-139139-192192-3051h FFG(level 1)CMAunit：mm12-278.基于ArcGIS平台的FFG1h FFG(level 1)CMA FFG(Flash Flood Guidance)Distributed FFG（0.189）in South China采用ArcGIS插 值模块得到面 临界雨量分布 单位：mmFlash Flood Potential Index (FFPI)：Developed by hyd
33、rologist Greg Smith,CBRFC(2003).Geographical features playa key role in flash floodingDeveloped as background information to be incorporated into production of better gridded Flash Flood GuidanceUsing the FFPI,the roles of soil,slope,vegetation and urbanization can be visualized基于ArcGIS平台的中国洪水风险潜在指标FFPI
