Geolocation

A misregistration relative to coastlines and other land features was recognized in the AMSR-E brightness temperature imagery. The error in geolocation was about 5 - 7 km and was due to a misalignment of the AMSR-E sensor relative to the spacecraft. The problem was fixed by a trial-and-error method in which various roll, pitch, and yaw corrections were applied to the AMSR-E alignment until proper registration was obtained.

A separate geolocation analysis was done for each channel, which showed that for a given frequency the v-pol and h-pol channels are well aligned. Furthermore, the 19, 23, and 37 GHz Channels are also well aligned with each other. Corrections to the geolocation for Channels 19 GHz through 89 GHz were implemented in Version B06 algorithm.

In Version B07, the 7 and 11 GHz Channels were resampled to match the locations of the higher frequencies since it was determined that the 7-GHz and 11-GHz horns are pointing in a slightly different direction than the 19, 23, and 37 GHz horns. This required re deriving the sampling weights with the center of the target cell positioned at the location of the 19 GHz Channel rather than at the 7 or 11 GHz footprint position. One drawback is that the un-resampled 7 and 11 GHz observations are missing an exact specification of latitudes and longitudes.

Calibration

AMSR-E's calibration system has a cold mirror that provides a clear view of deep space (a known temperature of 2.7 K) and a hot reference load that acts as a blackbody emitter; its temperature is measured by eight precision thermostats. After launch, large thermal gradients due to solar heating developed within the hot load, making it difficult to determine from the thermostat readings the average effective temperature, or the temperature the radiometer sees.

Before 07 January 2005 (B01 and B02 algorithms)

Remote Sensing Systems (RSS) used coincident Tropical Rainfall Measuring Mission (TRMM) Microwave Imager (TMI) and Special Sensor Microwave/Imager (SSM/I) satellite data over oceans to estimate the effective hot load temperature. A radiative transfer model used these data to compute the intensity of radiation entering the feedhorns. The model took into account different view geometries and channel differences between AMSR-E, SSM/I, and TMI. This process essentially provided earth-target calibration points, which were combined with the cold mirror temperature to compute a two-point linear extrapolation that yielded the hot load effective temperature for 10.7 to 89 GHz.

Before March 2006 (B03 through B06)

To estimate the effective hot load temperature, RSS assumed that the effective temperature is independent of channel polarization and then used climatic cloud and water vapor data along with daily SST and NCEP winds, to define the effective temperature as a function of the observed polarization differences.

After March 2006 (B07 and newer algorithms)

Computing the Effective Temperature of the AMSR-E Hot Load

An on-orbit calibration is required for AMSR-E because of a design flaw in the AMSR-E hot load. For the initial method, the effective temperature of the AMSR-E hot load was estimated using SSM/I and TMI, sea-surface temperature (SST), wind, vapor, and cloud water observations. Although the old method of computing T_eff worked well, it created a dependency on the SMM/I and TMI observations which could delay processing. Thus, a hot load calibration procedure that did not rely on other satellites was desirable.

The primary reason SSM/I and TMI were required was to specify the water vapor and cloud water. The other two relevant parameters, SST and wind, can be obtained with sufficient accuracy from NCEP Final Analysis fields. To remove the dependence on the SSM/I and TMI vapor and cloud retrievals, modifications were made to the way in which the effective temperature is computed. The new method is based on the assumption that the effective temperature does not depend on polarization, for example, it is the same for v-pol and h–pol. This assumption seems to be valid based on both empirical data (v-pol and h-pol effective temperatures coming from the SSM/I-TMI method are similar) and physical considerations (hot-load is an unpolarized source). Under this assumption, the method can be modified to make it insensitive to the specification of vapor and cloud.

The first step of the process is to estimate the value of Effective Temperature (T_eff) for each AMSR-E observation. The following expression is used to estimate the effective temperature

Where:

T_C is the brightness temperature for the cold space observation

T_Ai,rtm is the antenna temperature computed from the RTM given the ancillary information such as SST, wind, vapor, and cloud.

The subscript i denotes AMSR channel.

The term ρ_i is the following ratio of the radiometer counts

Where:

subscripts C, H, and E denote cold-space counts, hot-load counts, and earth-viewing counts, respectively.

Equation 1 is a linear extrapolation based on the cold-calibration point and the earth-calibration point.

The next step is to form a linear combination of effective temperatures that are relatively insensitive to variations in vapor, cloud, and to some degree wind. For a given frequency, this linear combination of T_eff is represented by the following equation

Where:

subscript j denotes channel frequency, p_ij and q_ij are static regression coefficients

W is wind speed. The summations are over the 10 AMSR channels from 7 to 37 GHz. However, a slightly modified version is used for the 89 GHz Channels. The function ∧(W) smoothly goes from 0 to 1 as the wind speed goes from 3 to 11 m/s. Thus, the first term in Equation 3 corresponds to low-to-moderate winds, and the second term corresponds to moderate-to-high winds. Two wind speed terms are used because the wind speed response of the RTM is highly non-linear. The wind speed value comes from NCEP Final Analysis fields.

The regression coefficients p_ij and q_ij are found from computer simulations so as to minimize the error in the estimation of effective temperature due to variations in wind, vapor, and cloud. The method for finding the coefficients is similar to that used to find the regression coefficients for the geophysical retrieval algorithms. In essence, Equation 3 is an algorithm for estimating the hot load effective temperature.

The next step is to correlate with the thermistor temperatures and orbit position. This is done by using the following expression

Where:

subscript k denotes the eight hot-load thermistors

T_k are the thermistor temperatures

α - 90 degrees is the orbit position angle relative to the ascending node crossing of the ecliptic plane.

There are 27 regression coefficients to be found for each frequency: a_kj, b_kj, and c_k. These regression coefficients are allowed to vary in time. A ±150 orbit moving time window is used to specify the coefficients. The weighting function for the time window is triangular, going to zero at 150 orbits before the current orbit and 150 orbits after the current orbit. Thus, the time scale for the variation in the a_kj, b_kj, and c_kj coefficients is approximately 10 days.

There are three additional improvements that were implemented. The first relates to the SSTs that were used to specify T_Ai,rtm. The SST values come from Reynolds' weekly Optimum Interpolation (OI) SST product. Since this SST product is a weekly value, it does not include any diurnal components. For the AMSR-E early afternoon orbit, diurnal variations in SST can be large when there are light winds. Accordingly, a diurnal adjustment is applied to the OI SST model the diurnal variability of SST.

The last improvement is to apply a smoothing function to the estimation of the radiometer gain, which is given by the following equation

The gain is expected to vary smoothly over the orbit. Random errors in the estimation of shows up as rapid variations in G. A smoothing function is applied to G to remove the variability on short time scales (minutes), thereby further reducing the error in the estimation.

Recalibration of AMSR-E Hot Load and Along-Scan Adjustments

The correction of the AMSR-E geolocation problems required that the AMSR-E sensor alignment be rolled relative to the spacecraft frame by 0.09 degrees. Changing the roll of the sensor results in a change in the incidence angle. With this roll adjustment, the true incidence angle was found to vary across the swath contrary to the assumption that it was nearly constant. This modification of the incidence angle has small but significant effects on the AMSR-E calibration. The calibration is based on measurements of the ocean surface, and the emission from the ocean surface varies with incidence angle. To address the change in incidence angle, AMSR-E was completely recalibrated and a new hot load effective temperature table was found. The along-scan antenna temperature adjustment was also recalculated using the new roll angle Thus, the along-scan errors in the antenna temperatures are significantly reduced except near the swath edge where the cold mirror interferes with the field of view.

Earth incidence angle is computed and included in Level-2A files in data field Earth_Incidence. These variable earth incidence angle data should be used appropriately, especially for geophysical retrievals depending upon incidence angle. Using the previously assumed constant incidence angle will likely result in along scan bias. The Level-2A Brightness Temperature data set is intended to be a faithful representation of the brightness temperatures observed by the instrument. Thus, due to the sensor-to-spacecraft roll, incidence angle is variable along scan, and brightness temperatures are correspondingly variable along scan. This along scan variability in observed brightness temperatures is not removed and should be accounted for by using the computed Earth_Incidence.

Implementation of Flagging Algorithm for RFI from Geostationary TV Satellites

As part of the SST and wind validation activity, anomalous retrievals were found off the West Coast of Europe and in the Mediterranean Sea. After some investigation, these erroneous retrievals were determined to be due to Radio Frequency Interference (RFI) from a European satellite TV service. ASMR-E is receiving the broadcast from two European geostationary satellites that operate near the 10.7 GHz band. The satellite TV signal is reflecting off the ocean surface into the AMSR-E field of view. AMSR-E bandwidth at 10.7 GHz is 100 MHz, whereas the protected band is only 20 MHz. The TV signal is coming from the unprotected part of the 100-MHz band.

The TV RFI was determined to be coming from two satellites: one positioned at a longitude of 13 degrees E above the equator and the other at 19 degrees E. An algorithm was developed that computes the angle between AMSR-E look vector and the specular reflection vector for the TV RFI. This angle is called the RFI angle. Small RFI angles correspond to cases in which the TV RFI is being reflected off the ocean surface directly towards AMSR-E.

When the RFI angle is less than 12 degrees, in general the observation should be flagged as RFI contaminated. However in the North Sea, the RFI is particularly strong. For this region, an RFI angle of 17 degrees is the threshold.

Correction for August-September 2003 Aqua Pitch Error

An anomaly occurred between 12 August 12 and 6 September 2003 due to an error in knowledge of the spacecraft pitch. In the Level-1A data, the definitive ephemeris was used for this period so the geolocation values are correct. However, the fact that the anomaly caused Aqua to be inadvertently pitched during this time period was not accounted for in the Level-1A data. In the Level-2A algorithm, the satellite pitch anomaly during this period was modeled as a ramp function and corrected.

Correction for Lunar Radiation Entering the AMSR-E Cold Mirror

Twice each month the moon enters the Field of View (FOV) of the AMSR-E cold mirror. The moon's surface temperature varies from 120 K night to 370 K day and has a relatively high emissivity. As a result, the moon acts as a source of contamination to the cold sky measurement.

A correction was applied to remove the lunar contamination. The correction depends upon the following factors:

1. The angle between the vector going from the satellite to the moon and the boresight vector of the cold mirror. This is the dominant term. When this angle becomes small, a few degrees or less, lunar contamination becomes significant. This angle is called the lunar angle.
2. The phase of the moon. A full moon is hotter than a new moon and hence has a higher brightness temperature.
3. The distance from the satellite to the moon. Radiation intensity falls off as the inverse of the square of the distance.

The lunar antenna temperature contribution to the cold sky observations is computed and then is scaled in terms of cold counts. The lunar cold counts are subtracted from the AMSR-E cold count observations to obtain a cold count value free of lunar contamination. For the case of 89 GHz, when the lunar angle is less than one degree, the lunar contamination is too large to perform the correction and these observations are flagged as bad and are not processed. The excluded observations are extremely rare. The accuracy of this correction is estimated to be in the order of 0.1 K.

Adjustment to Match the 89A and 89B Observations During Resampling

When the 89 GHz Channels are resampled to lower spatial resolutions, the observations from the A-horn and the B-horn are combined. However, the incidence angles for these two horns are different with the B-horn incidence angle being about 0.6 degrees smaller than the A-horn. To compensate for the difference in incidence angle, the following adjustments were made to the A-horn measurements before resampling

These expressions were found from doing linear regression of actual A-horn and B-horn observations over the first two mission years of AMSR-E. The application of these equations normalizes the A-horn measurements to the B-horn incidence angle.

Implementation of using UT1 time to compute the Earth rotation angle.

Aqua's position, velocity, and attitude vectors are given in terms of the J2000 inertial coordinate system. To compute Earth latitudes and particularly longitudes, it is necessary to compute the Earth rotation relative to the J2000 systems. The proper calculation requires using the UT1 time, which can be as much as one second different from UTC time. To obtain UT1, the Level-2A algorithm accesses the U.S. Naval Observatory database each day to obtain the current UT1. One advantage of this procedure is that it is independent of leap seconds; therefore, there is no discontinuity in the geolocation parameters when a leap second occurs.

Intercalibration with SSM/I F13 and WindSat

F13 and WindSat provide a backbone of consistent long-term observations, relatively free from diurnal effects. Local observation times for F10, F11, F14, F15, and F16 all drift by two to six hours over the years, so these instruments have been measuring a continuously differing viewpoint in Earth's diurnal cycle. The larger the diurnal drift, the more challenging it is to distinguish from sensor drift. AMSR-E is in a stable orbit on Aqua, so diurnal drift is not a problem. However, AMSR-E observes local times of 1:30 a.m./p.m.; AMSR-E day and night passes observe significantly more pronounced diurnal variations than F13 and Windsat passes at approximately 6:00 a.m./p.m.. For more information about intercalibration choices and methodology, please refer to the RSS Crossing Times Web page.