Issue 8: Impact of observations scattered across space and time
Most meteorological observations are performed at regular times of day (e.g., 00Z and 12Z) in specific locations. In situ SST obs are major exception to this rule, but SST changes slowly compared to the atmosphere. Most satellite observations are for wide areas of the globe. GPS-RO is nearly unique because it observes point measurements that change in location and time of observation in a not quite random way. As a result, utilizing them for climate change analyses presents new challenges. Past analyses have averaged the observations into large grid boxes and then into large time segments (e.g., 10x10 degree boxes averaged over a month). An analysis of the errors this technique causes would be helpful in putting the use of GPS-RO for climate into perspective. Perhaps the potential impacts of the errors could be quantified by subsampling a very high resolution model output at the GPS-RO observation locations in time and space (by latitude band and over land and ocean and elevation) and comparing the results to the full model field results.
Response from Kevin Trenberth:
GPSRO is not actually a point measurement but samples a finite size footprint and this actually makes it more useful to climate and less useful for mesoscale meteorology. Many errors are likely to be random and thus average out, although this needs to be quantified. The exercise suggested may be useful
Also to the statement: Most meteorological observations are performed at regular times of day (e.g., 00Z and 12Z) in specific locations. Response: Not true: all satellite soundings are asynoptic.
6 Comments
Tom Yunck
The comprehensive diurnal and seasonal coverage of a large, well-distributed RO constellation may solve more sampling problems than it creates. Indeed, sampling error may be reduced to insignificance by this approach of nearly concurrent, randomized sampling of the globe.
Seth Gutman
Consolidated Comments by Gutman, Yoe and Reale
Seth Gutman's Response
As discussed previously, GPS-RO measurements are not point measurements. While the challenges of assimilating discrete observations made at random places and times have been successfully dealt with using variational data assimilation techniques, the challenges of assimilating path integrated observations remain daunting, requiring the estimation of horizontal refractivity gradients from a 1-D observation. Clearly ancillary information is needed here. The suggested method of evaluating errors is complicated by the fact that GPS-RO soundings typically contain detail in the upper atmosphere that cannot be realistically captured or analyzed by any current multi-dimensional data assimilation system and smear-out the effects of important wet and dry refractivity variations in the lower atmosphere over a relatively long (more or less horizontal) path.
Jim Yoe's Response
I would add that while we are accustomed to the temporal (and for that matter spatial) regularity of radiosondes and polar satellite soundings with regular local equator crossing times, this is not necessarily advantageous for climate analysis. We are accustomed to accounting for diurnal variations, etc., from these data sets, but in fact we might well do better with globally distributed RO soundings provided by a constellation of sensors.
Tony Reale's Response
... particularly in the context of characterizing respective satellite, sonde, etc characteristics at varying global locations.
Ben Ho
Response from Ben Ho
This is a very good suggestion and will be followed in a future study. Now we have around 10-year of outputs from a high-resolution model. We will conduct several well-designed OSSE using these model outputs and COSMIC RO data with predicted location and time to quantify the sampling errors related to the distributions of RO data. Potential impacts of COSMIC RO data in term of temporal and spatial sampling issue will be addressed.
Ulrich Foelsche
Response from Ulrich Foelsche
We are well aware of problems that could arise, e.g. through aliasing of diurnal effects in long-term satellite observations. We performed simulation studies and we routinely estimate the sampling error of RO climatologies by comparing climatologies derived from vertical ECMWF profiles at the RO times and locations with climatologies derived from the complete 4D ECMWF field. For climate applications potential systematic components are most important, e.g. through undersampling of the diurnal cycle. The worst case would be a very slow drift in local time (a few hours over many years). No current or planned RO mission is in this unfavorable situation. Most RO missions have precessing orbits. With a precession rate of ~3°/day, the RO measurements from a single COSMIC satellite (in final orbit) drift through all local times within ~60 days, and for the entire constellation (with 30° orbit plane separation) it takes about 10 days to sample the diurnal cycle (at Equator). The European satellite MetOp, on the other hand, is in a sun-synchronous orbit - the measurements are stationary in local time and the diurnal cycle is never fully sampled. We found a constant bias of about 0.04 K in MetOp RO climatologies, which should remain stationary, as long as the shape of the diurnal cycle does not change. For COSMIC we found systematic (but very small) oscillatory local time component of the sampling error in monthly mean climatologies in the extratropics (hemispherically antisymmetric, half cycle ~60 days, ± 0.03 K amplitude) that disappears to < 0.01 K when building seasonal means.
A detailed response can be found in the attached pdf file:
Response by Ulrich Foelsche
Jens Wickert
Response from Jens Wickert
The distribution of RO observations in space and time is not uniform there are some special features, which depend on the LEO and GPS orbit characteristics (e.g., preferred latitude bands, local time distribution of the measurements). This should be taken into account when defining new and optimal RO missions. E.g. a near equatorial LEO orbit would increase the amount of low latitude occultations. Here we have less occultations in case of (near) polar orbiting LEOs.
Rob Kursinski
One of the major features of a constellation of at least 6 GPSRO receivers in orbit is the unique ability to sample the entire diurnal cycle globally while profiling the atmospheric state with very high vertical resolution in clear and cloudy conditions. While in some sense this does create a challenge, this is a really unique observational combination and offers a real opportunity to achieve a fundamental requirement in observing and understanding how our climate is changing.
The Graz group has looked at the sampling far more closely than I have (See Ulrich's response above) but 6 satellites acquiring occultations within +/- 45 degrees of the velocity vector samples the entire equator statistically every orbit. The density of occultations varies with the number of orbiting satellites but there are no sampling gaps at the equator. Using several days worth of occultations provides denser occultation sampling. As Jens indicates, one or more GPSRO receivers in low inclination orbits would improve the density of occultations at low latitudes. There is no other way to provide the combination of high vertical resolution sampling spanning the entire diurnal cycle across the globe. This is a critical feature for measuring climate analogous to meeting the Nyquist rate in sampling theory. A passive microwave sounder in geosynchronous orbit will add diurnal sampling at low and mid latitudes but will be limited by coarse vertical weighting functions and surface emissivity ambiguities over land. These limitations will limit the ability of such a sounder to isolate the planetary boundary layer despite claims made in the NRC decadal study.
Full sampling of the diurnal cycle is critical as the diurnal cycle is almost certainly evolving as the climate evolves. Sun synchronous orbits will alias diurnal cycle changes in with long term warming trends. One of the reasons for the discrepancy between the RSS and UAH interpretations of the MSU temperature trends is the RSS group used a model to estimate the diurnal temperature cycle in accounting for the effects of orbital drift whereas the UAH group used the observations themselves. It appears that the model diurnal cycle is wrong and adversely affected the RSS interpretation of the MSU radiances. It is critical that the entire diurnal cycle be sampled all the time, not just two times per day with a local time that varies with longitude like the radiosondes do.