Estimating Bias of SatelliteBased Precipitation Estimates Relative to
Estimating Bias of Satellite-Based Precipitation Estimates Relative to In Situ Measurements Thomas Smith 1 Phillip A. Arkin 2 George J. Huffman 3 John J. Bates 1 1. NOAA/National Climatic Data Center/NESDIS 2. ESSIC, Univ. of Maryland, College Park 3. Science Systems and Applications, Inc. and NASA Goddard Space Flight Center NOAA’s National. Data Climatic Data Center 1 NOAA’s National Climatic Center
Precipitation Bias • Bias adjustments are needed for satellitebased precipitation. • Bias Evaluations: estimates of satellite bias relative to gauge data. • Bias Uncertainty: how good are the bias estimates given the available gauges. • Oceanic Precipitation Bias: can satellite estimates be adjusted over oceans to minimize the bias. NOAA’s National. Data Climatic Data Center 2 NOAA’s National Climatic Center
Bias Evaluations • Bias = Satellite – Gauge difference. • Analysis of monthly differences using a large-scale optimum interpolation (OI). • Bias spatial scales are needed for analysis. NOAA’s National. Data Climatic Data Center 3 NOAA’s National Climatic Center
Bias Correlation Scales • Zonal Averages at each latitude. • Red: satellite bias w. r. t. gauges (land only). • Black: satellite to satellite difference scales (land sea). • Scales are similar, mostly 1000 km +. • Use constant scales of 750 km in both directions, to minimize excessive spreading of bias. NOAA’s National. Data Climatic Data Center 4 NOAA’s National Climatic Center
Examples: Zonal Averages • 1996 -2003 Bias typically a few mm/day. • OPI bias smallest. • SSMI combined and GPI biases have opposite signs, some annual cycle. NOAA’s National. Data Climatic Data Center 5 NOAA’s National Climatic Center
Examples: 1996 -2003 Averages • Positive bias over central Africa. • Negative over east Asia and Hawaii. • Different sign biases may cancel when data are combined. NOAA’s National. Data Climatic Data Center 6 NOAA’s National Climatic Center
Differences that Bias Corrections Make • 1996 -2003 Average uncorrected and corrected SSMIc. • Most regions show little difference. • Large differences over central Africa, N. W. North America. NOAA’s National. Data Climatic Data Center 7 NOAA’s National Climatic Center
Bias Uncertainty • How well can the OI analysis reflect the bias? (Input data 2. 5 o and monthly; OI uses data from 12. 5 o centered region. ) • How does bias uncertainty change with the number of raw biases analyzed? • What is the bias uncertainty when there are no gauges in a region? NOAA’s National. Data Climatic Data Center 8 NOAA’s National Climatic Center
OI Relative Error • Set all raw biases = 1 and analyze, compute error change with n = number of raw biases. • An exponential function estimates the error as a function of n. n NOAA’s National. Data Climatic Data Center 9 NOAA’s National Climatic Center
RMS Bias • RMS Bias (RMSB) is the typical bias magnitude (no correction), from only near gauges, global. • RMS difference between each satellite and all other satellites (RMSDs, land + sea) is mostly comparable to RMSB. OPI and TOVS have lower bias than the others, which have similar magnitudes. RMSB (mm/day) computed using n ≥ 20 for each satellite, 1996 -2003 and globally. SMMI RMSB is the same for all. The RMSD between each satellite and all other satellites is also shown. Satellite OPI AGPI SSMI/TOVS RMSB 0. 9 1. 5 2. 0 1. 8 1. 5 1. 2 RMSDs 2. 0 1. 7 2. 1 2. 0 1. 8 2. 0 NOAA’s National. Data Climatic Data Center 10 NOAA’s National Climatic Center
Bias Error from a Merged Analysis • Merged error depends on how correlated the biases are. • All uncorrelated: • All correlated: • Part correlated: NOAA’s National. Data Climatic Data Center 11 NOAA’s National Climatic Center
Bias Correlations • Measured spatial correlations from wellsampled regions (n ≥ 20). • Negative correlations reduce bias error by canceling positive and negative errors. • Most have weak positive correlations (r usually < 0. 5, some bias error reduction from use of multiple satellites). NOAA’s National. Data Climatic Data Center 12 NOAA’s National Climatic Center
Test: January 2000, all Satellites • Observed correlations, equal weighting for all 8 satellite estimates. • No bias analysis → no corrections (oceans). • More satellite coverage → less error (tropics). NOAA’s National. Data Climatic Data Center 13 NOAA’s National Climatic Center
High-Latitude Gauge Biases: Work in Progress • Bias errors from evaporation and blowing snow. They may offset each other, but they can also be as large as 30% of the mean (Groisman et al. ). • Gauge adjustments are recommended for high latitudes. Uncertainty should be factored into bias uncertainty. NOAA’s National. Data Climatic Data Center 14 NOAA’s National Climatic Center
Oceanic Bias Adjustments: Work in Progress • No oceanic gauges except scattered islands, near coasts, and on some buoys. An oceanic low-bias estimate needed for adjustments. • TMI satellite estimates were found to have little bias compared to tropical Pacific buoys (Bowman et al. 2003). • Over oceans, Bias adjustment relative to TMI (or a satellite with least bias relative to TMI). Uncertainty from uncertainty estimate of TMI or of the satellite. • TMI limited in time (beginning 1998) and to between about 40ºS and 40ºN. NOAA’s National. Data Climatic Data Center 15 NOAA’s National Climatic Center
Summary and Conclusions • Using available data, satellite precipitation bias is evaluated near gauges. • Bias spatial scales are usually large (> 1000 km), but may be smaller near coasts. • Bias may be reduced by direct adjustment and by combining analyses with uncorrelated biases. • Oceanic estimates may be adjusted to a low-bias satellite estimate, such as TMI. • Because of uncertainties in the base analyses (gauges, TMI) the absolute bias is difficult to define. Adjustment is relative to a best estimate to minimize bias. NOAA’s National. Data Climatic Data Center 16 NOAA’s National Climatic Center
- Slides: 16