Transcript Slide 1
Fine-scale comparisons of satellite estimates Chris Kidd School of Geography, Earth and Environmental Sciences University of Birmingham Rationale for finescale comparisons Daily and monthly estimates hide algorithm problems: • Rain areas/occurrence • Rain intensities - Temporal and spatial smoothing reduces irregularities Daily products also have sampling issues – which can cause strobe-like effects with rain movement 3rd IPWG workshop, Melbourne, Australia. 23-28 October 2006 Which UK validation data set? Gauges 'Ideal' choice – representing 'true' 'at surface' rainfall, but: • daily coverage good – hourly sparse (even in the UK) • poor immediacy (~1-2 months delay) • higher-temporal resolution available, but poor intensity resolution (tips/min logging = 6 mm/h min rain rate) Radar Temporally and spatially superior (down to 5min, 2km), available within an hour of collection: but, • ground clutter & bright band (despite corrections applied) • range dependency (ditto) 3rd IPWG workshop, Melbourne, Australia. 23-28 October 2006 Daily vs hourly gauge data Daily gauge network 06-06Z Hourly gauge network 3rd IPWG workshop, Melbourne, Australia. 23-28 October 2006 Radar: advantages/disadvantages IR:radar matching Daily total (mm) 14 Sept 2006 Blue = radar rain / IR no-rain Red = IR rain / radar no-rain 3rd IPWG workshop, Melbourne, Australia. 23-28 October 2006 Time skill scores of rain retrievals Radar PMW IR Rainfall is temporally very fickle 3rd IPWG workshop, Melbourne, Australia. 23-28 October 2006 Finescale Comparisons Instantaneous comparisons: • Results at instantaneous / 5 km resolutions • AMSR L2 rainfall product (GPROF) • PCT (thresholds set – Kidd 1998 → dT×0.04+dT2×0.005) • data remapped and processed on European IPWG polar-stereographic projection Future comparisons… 3rd IPWG workshop, Melbourne, Australia. 23-28 October 2006 SSMI PCT 06-09-02 06:36 SSMI PCT 06-09-02 07:12 SSMI PCT 06-09-02 09:18 AMSR PCT 06-09-02 03:31 AMSR-L2 06-09-02 13:30 3rd IPWG workshop, Melbourne, Australia. 23-28 October 2006 20060930-1230 20060930-0245 20060929-1315 20060929-0215 20060928-1230 20060927-1330 20060927-0215 20060926-1245 20060926-0130 20060925-0230 20060924-1300 20060924-0145 20060923-0245 20060922-1315 20060922-0200 20060921-1230 20060920-1330 20060920-0215 20060919-1245 20060918-1330 20060918-0230 20060917-1300 20060917-0145 20060916-0245 20060915-1300 20060914-1230 20060914-0245 20060913-0215 20060912-1230 20060911-1330 20060911-0215 20060909-0230 20060908-1300 20060907-0245 20060906-1315 20060905-1230 20060902-1330 20060901-1300 0.9 20060901-0145 correlation Correlations : instantaneous cases AMSR PCT & GPROF 0.8 0.7 0.6 0.5 PCT L2 0.4 0.3 0.2 0.1 0.0 3rd IPWG workshop, Melbourne, Australia. 23-28 October 2006 20060930-1230 20060930-0245 20060929-1315 20060929-0215 20060928-1230 20060927-1330 20060927-0215 20060926-1245 20060926-0130 20060925-0230 20060924-1300 20060924-0145 20060923-0245 20060922-1315 20060922-0200 20060921-1230 20060920-1330 20060920-0215 20060919-1245 20060918-1330 20060918-0230 20060917-1300 20060917-0145 20060916-0245 20060915-1300 20060914-1230 20060914-0245 20060913-0215 20060912-1230 20060911-1330 20060911-0215 20060909-0230 20060908-1300 20060907-0245 20060906-1315 20060905-1230 20060902-1330 20060901-1300 20060901-0145 rain total ratio Ratio – accumulation : instan. cases 2.5 2.0 1.5 PCT ratio L2 ratio 1.0 0.5 0.0 3rd IPWG workshop, Melbourne, Australia. 23-28 October 2006 20060930-1230 20060930-0245 20060929-1315 20060929-0215 20060928-1230 20060927-1330 20060927-0215 20060926-1245 20060926-0130 20060925-0230 20060924-1300 20060924-0145 20060923-0245 20060922-1315 20060922-0200 20060921-1230 20060920-1330 20060920-0215 20060919-1245 20060918-1330 20060918-0230 20060917-1300 20060917-0145 20060916-0245 20060915-1300 20060914-1230 20060914-0245 20060913-0215 20060912-1230 20060911-1330 20060911-0215 20060909-0230 20060908-1300 20060907-0245 20060906-1315 20060905-1230 20060902-1330 20060901-1300 20060901-0145 Area ratio Ratio – occurrence : instan. cases 3.0 2.5 2.0 1.5 PCT L2 1.0 0.5 0.0 0 400 300 Need for case-classification - rather than the wholesale 'lumping' all data into large temporal results – need to look at the component meteorology associated with the estimates: 200 100 270 400 300 200 100 100 100 200 300 400 180 200 300 400 90 Statistics: blame it on the weather! Movement: Is the movement perpendicular or along the rain band? Intensity What is the range of values within the rain area? Size/variability What is the size and variability of the rain area(s)? Statistical success has as much to do with meteorology as the algorithms ability… So… what now? i) we must remember that PM instantaneous results are better than Vis/IR-based techniques – including merged techniques ii) high temporal and spatial data can produce very good statistics – if the data is of good quality iii) prescribed temporal and spatial sampling is not always ideal – are these applicable to applications? • At present, comparisons at fixed regions and time scales • Need for flexibility – to match user requirements • Initial step at thinking about user-defined spatial and temporal time scales 3rd IPWG workshop, Melbourne, Australia. 23-28 October 2006 Current 'interactive' comparison The User FTP User text Info checks e-mail; date range; time range Why FTP? Simple to use and set up batch jobs… User data QC checks file size; byte order; data range Radar data generate time slots; copy radar files; accumulate data E-mail User Why e-mail? Puts the results on the User's desktop… Disk-store Maybe a Java version too? Statistics: bias, ratio, RMSE, CC, HSS etc Graphics 'Standard' IPWG EU region 3rd IPWG workshop, Melbourne, Australia. 23-28 October 2006 Conclusions Finescale – instantaneous / ~5km important: it allows us • to disentangle algorithm performance • to assess performance under different conditions • address issues of rain occurrence and intensities But, issues over: • data integrity (data reliability – flagging of bad pixels) • instrument noise (e.g. AMSR – and RFI) Need for fine-resolution test cases: particularly with common input data sets. 3rd IPWG workshop, Melbourne, Australia. 23-28 October 2006 Freezing levels “Only one thing we do know is that the freezing level is relatively stable” Tom Wilheit Surface Variability Effects and contribution of surface variability to precipitation retrievals. V19 stddev V37 V85 4th International GPM Planning Meeting, DC : 15-17 June 2004 Rain/no-rain induced biases -0.5 -1.0 • Differences in rain/no-rain boundaries reveal regional variations that do not exist in reality • Further complicated since rain/no-rain boundaries tend to differ over land/sea areas Trends in Global Water Cycle Variables, UNESCO, Paris. 3-5 November 2004 Conclusions PMW estimates are capable of retrieving light rainfall Statistics often confuse the issue: more light rain tends to produce poorer statistics Instrument noise can be problematic (e.g. AMSR) Surface screening - potential problems with 'false alarms' over cold/snow surfaces Lack of 'common' data sets – different algorithms use different source data – different Q.C. global quick-look images Raw Data Algorithms Products Remapping to polar stereographic projection Gauges Radar Statistical analysis & image generation User-defined periods (& resolutions) Daily 00-24Z results