Error Assessment and Validation of the IASI Temperature and Water Vapor Profile Retrievals Validation by Radiosonde Correlative Measurements N. Pougatchev, T. August, X. Calbet, T. Hultberg, P. Schlüssel, B. Stiller K. St. Germain and G. Bingham ITSC-16 Angra dos Reis May 7, 8
Objectives Scientific/Methodological Testing of NPP CrIS Validation Assessment Model by its application to validation of IASI L2 data against radiosondes Scientific/Utilitarian Assessment of the IASI Temperature and Water Vapor retrieval errors in the form that can be utilized by the community regionally specific Covariance and Bias
Validation by Correlative Measurements True Validated Sounder State x sat ŷ sat =y sat(x sat)+εsat Validation Target True State x cor Validation Goal: Estimate the Error ε sat Correlative Measurements yˆ y = ˆε Validation output sat val val y =F(y ˆ ) cor st a Validation Model ŷ cor =y cor(x cor)+εcor
Validation Issues Why do We Need Validation Model Why We Can NOT Use Correlative Data As Is Characteristic Difference validated sounder and correlative measurements sample atmosphere differently. State Non-Coincidence correlative measurements are at different time and location. Validation Model reconciles the issues by modeling best linear estimate of the satellite measurements and assessing the errors y =By ˆ =y (x )+ε val cor sat sat val
IASI Validation Study Validation Data Set radiosondes at Lindenberg (Germany, 52.21 o N, 14.12 o E, 112 m a.s.l ). Dedicated launches 1 hour prior and at the overpass time; and synoptic times (0, 12, 6, and 18 UTC) Validated parameters Atmospheric Temperature and Water Vapor Vertical Profiles. Validated System IASI characterized by averaging kernels. Validated Data Set EUMETSAT v. 4.2 retrievals; cloud clear; ± 1 O Lat. and Long. about Lindenberg
Overpasses and Sonde Launches Overpasses Lindenberg (Germany) site AIRS(AQUA) and IASI (METOP-A) 0/24 UST 6 h UST 18 h UST 12 h UST
Temperature Non-Coincidence Error Free Troposphere 3.0 RMS non-coincidence error Averaged between - mb Sodankyla, Lindenberg, and Southern Great Plane ARM 2.5 rms error (K) 2.0 1.5 1.0 Sodankyla Lindenberg SGP ARM 0.5 0.0 0 1 2 3 4 5 6 7 8 9 10 11 12 13 Time non-coincidence (h)
Averaging Kernels Temperature Averaging Kernels Temperature retrieval Lindenberg June-August 7 Sum of diagonals=14.18-0.1 0.0 0.1 0.2 0.3 0.4 0.5 Averaging Kernels AvKe_IASI_LND_7.JNB
Temperature Retrieval Error Variances/rms Temperature IASI Retrieval Errors and Variances Expected retrieval - xˆ x ˆ =x +A(x -x )+ε Expexted error - xˆ -x S exp a true a exp exp true = (I - A)S (I - A) + S T exp x ε T Expected covariance AS A + S x ε Errors Variances 0 1 2 3 4 5 6 rms (K) Total expected error Total actual error Repeatability/ retrieval noise Variance actual retrievals Variance sondes Variance expected retrievals
Temperature Retrieval Error Covariance Matrices Actual Total Error Temperature Retrieval Correlation Matrix Expected Total Error Temperature Retrieval Correlation Matrix Pressure Actual (mb) Total Error Temperature Retrieval Rows of Correlation Matrix 16 14 12-0.6-0.4-0.2 0.0 0.2 0.4 0.6 0.8 1.0 Expected Total Error Temperature Retrieval Rows of Correlation Matrix 16 14 12 Altitude (km) 10 8 6 Altitude (km) 10 8 6 4 2 4 2 0-0.6-0.4-0.2 0.0 0.2 0.4 0.6 0.8 1.0 0-0.6-0.4-0.2 0.0 0.2 0.4 0.6 0.8 1.0
Temperature Retrieval Error Bias Temperature Retrieval Bias against Radiosondes -2-1 0 1 2 Sonde launch time about IASI overpass 0 min 54 min 0-54 min 1.5-3.6 h (synoptic times) Bias (K)
Relative Humidity Non-Coincidence Error Free Troposphere 30 RMS non-coincidence error Averaged between - mb Sodankyla, Lindenberg, and Southern Great Plane ARM 25 rms error (% RH) 20 15 10 Lindenberg Sodankyla SGP ARM 5 0 0 1 2 3 4 5 6 7 8 9 10 11 12 13 Time non-coincidence (h)
Averaging Kernels Relative Humidity Averaging Kernels RH retrieval Lindenberg June-August 7 Sum of diagonals=10.14-0.1 0.0 0.1 0.2 0.3 0.4 Averaging Kernel AvKe_IASI_LND_7.JNB
Relative Humidity Retrieval Error Variances/rms Relative Humidity IASI Retrieval Errors Relative Humidity IASI Retrieval and Sonde Variances 0 10 20 30 rms (% RH) 0 10 20 30 rms (% RH) Repeatability Total expected error Total actual error Variance RH Lindenberg Actual retrieval variance Expected retrieval variance Sonde profile variance
Relative Humidity Retrieval Error Covariance Matrices Actual Total Error RH Retrieval Correlation Matrix Expected Total Error RH Retrieval Correlation Matrix Actual Total Error RH Retrieval Rows of Correlation Matrix 16 14 12-0.4-0.2 0.0 0.2 0.4 0.6 0.8 1.0 16 14 12 Expected Total Error RH Retrieval Rows of Correlation Matrix Altitude (km) 10 8 6 Altitude (km) 10 8 6 4 4 2 2 0-0.6-0.4-0.2 0.0 0.2 0.4 0.6 0.8 1.0 0-0.6-0.4-0.2 0.0 0.2 0.4 0.6 0.8 1.0
Relative Humidity Retrieval Error Bias RH Retrieval Bias against Radiosondes -20-10 0 10 20 Sonde launch time about IASI overpass 0 min 54 min 0-54 min Inter-sondes interpolation Bias (% RH)
IASI and AIRS Performance Temperature Validation T retrievals vs. radiosondes IASI (Lindenberg) and AIRS (ARM SGP) IASI - red lines; AIRS - black lines (AIRS - Tobin et al., JGR 6, 111, D09S14, doi:10.1029/5jd006103) bias rms Variance of the sondes -2-1 0 1 2 3 4 5 6 T (K)
IASI and AIRS Performance Relative Humidity Validation Water Vapor retrievals vs. radiosondes IASI (Lindenberg) and AIRS (ARM SGP) IASI - red lines; AIRS - black lines (AIRS - Tobin et al., JGR 6, 111, D09S14, doi:10.1029/5jd006103) bias rms Variance of the sondes -20 0 20 40 60 80 H 2 O % This is NOT Relative Humidity!!!
Conclusions At Lindenberg state dependent error dominates the total error for both for Temperature and Relative Humidity. That implies that retrieval errors should be characterized regionally and seasonally specifically.
Temperature Conclusions Our Validation Assessment Model and radiosondes allow to account for non-coincidence error and finite vertical resolution of IASI and assess retrieval errors accurately: Variances/rms of a single FOV retrieval are 1K between mb with increase to 2 K in tropopause and at the surface Bias against radiosondes oscillates within ±0.5K between 950 mb. Actual vertical resolution apparently lower than expected in midtroposphere and tropopause regions
Relative Humidity Conclusions Complex spatial structure of highly variable RH field does not allow to assess retrieval errors accurately: Variances/rms of a single FOV retrieval are between 6-18 % RH in mb with best performance at mb No statistically significant bias was detected between mb. Actual vertical resolution apparently lower than expected in midtroposphere region Techniques other than radiosondes, e. g. NAST-I in combination with drop-sondes, are needed for accurate RH retrieval error assessment.
Conclusions IASI and AIRS retrieval performance are in reasonable agreement with direct comparison needed for more definitive conclusion The Validation Assessment Model analysis of temporal variations of T and RH at Lindenberg and ARM SGP sites reveals that Lindenberg is better for T validation (synoptic time launches can be used) and ARM is better for RH validation. This approach can be used for Cal/Val planning through optimal validation tools and sites selection and characterization.