The requirements for data records and information products derived from Earth Observation technology are consistent quality, precision and accuracy. Besides, in order to produce reliable measurements as input to the geophysical parameter retrieval, a basic requirement is that the spaceborne instruments are properly calibrated. This ensures that the retrieval algorithms are independent of instrument errors and establishes confidence in the geophysical parameter estimates. For climate measurements, for example, changes of only a few percent in geophysical parameters are of critical interest.
The activity that endeavours to ensure that remote sensing products are highly consistent and reproducible is known as calibration and validation or "cal/val". This is an evolving discipline that is becoming increasingly important as more long-term studies on global change are undertaken and new satellite missions are launched.
Calibration is the process of quantitatively defining the system responses to controlled signal inputs. Validation is the process of assessing, by independent means, the quality of the data products derived from the system outputs. These definitions are internationally accepted and are most often used in the remote sensing context to refer specifically to sensor radiometric calibration and geophysical data product validation. Agencies usually undertake the calibration of their respective mission satellite systems; however to extend this beyond the commissioning phase is potentially very difficult. Therefore, well-instrumented benchmark test sites and data sets for calibration are needed to supplement or substitute for on-board calibration. These sites could well receive rotating support from agencies with the agency (or company) with the most recent satellite launch funding the site during the commissioning phase.
Besides having a history of long data records, the ideal requirements for test sites include that they are large, flat homogenous areas, with nearly Lambertian and nearly flat spectral reflectance, and with high percentage of clear skies and dry conditions. They should primarily be bright targets, including a few low reflectance targets as well. Besides having year-round availability, the test site should somehow have “controlled” human disturbance as well.
The primary objective of validation is to assess the quality, and as far as possible to quantify the accuracy of remote sensing data products. Ideally, validation activities seek to compare the data products with more accurate independent measurements of the same quantity over a statistically significant number of samples and wide variety of situations. The problem is that the space and time scales of the in-situ data and the satellite data are rarely directly comparable. Typically, there are insufficient in-situ measurements to cover a satellite field of view, whether the field of view is some meters (high resolution imagers, for example), or several km (broadband radiometers, L-band radiometers, etc). Thus, even a perfect remote sensing measurement will be expected to differ from an in-situ verification measurement because of the inability to match the observations in time or space. In-situ measurements are invariably taken at small space and time scales, while a satellite overpass is a few second snapshot of a large area. For broadband radiometers, a common example is the comparison of a surface flux radiometer to a satellite based estimate for the radiometer field of view. The matching error can be reduced (but not eliminated) by increasing the number of surface observations within the field of view (level 2 data) or grid box (level 3 data). This error can also be minimized by using very large ensembles of matched data.
Field campaigns are best for process studies and hypothesis formation. They provide the most complete case study data, but usually provide very limited statistical significance. Dedicated sites, BSRN (Baseline Surface Radiation Network), surface sites, and other long-term sites are considered the best strategy for validation, in this case, of cloud and radiation data.
In many geophysical fields, the region of the earth viewed by a satellite observation rarely matches exactly that viewed by the surface radiometer. Recourse must be made to statistical intercomparisons in order to reduce the sampling noise induced by the different space and time sampling characteristics of satellite and surface data. The special character of remote sensing measurements to correspond to area integrated values, obliges independent in situ measurements to be representative of zones of a minimum number of pixels of the sensor under consideration. The large pixel size of some satellite missions such as GERB, EarthCARE (Earth Clouds, Aerosols and Radiation Explorer), CERES (Clouds and the Earth’s Radiant Energy System), SMOS (Soil Moisture and Ocean Salinity), etc., introduces a number of scientific issues that make it relevant and even necessary to develop a specific methodology and carry out specific measurements over large extended areas. These regions should be well controlled from the viewpoint of other complementary measurements, also at a large scale.
The University of Valencia has recently set up a robust automatic meteorological station located towards the North-West part of the Valencia region, in the Utiel-Requena Plateau, at about 80 km from the city of Valencia. The main objective of the VALENCIA Anchor Station is to define and characterise a large, reasonably homogeneous and flat area, mainly dedicated to vineyards, as reference for Cal/Val activities in low-resolution large-scale pixel size satellite sensors, as those corresponding to the missions mentioned above. The area is well documented and has previously been used in other projects. It is desirable that the VALENCIA Anchor Station area, together with the Central Spain area where the University of Castilla-La Mancha has three other twin stations operational, could define a still larger and reasonably homogeneous region of about 300 x 200 km2, in order to be able to count with a minimum number of large size pixels of the order of 50 x 50 km2.
- Definition of a large scale validation area for low-spatial resolution missions
- Definition and characterisation of a large scale reference pixel
- Time Interpolation and Spatial Averaging. Study of scaling issues. Criteria for aggregation and disaggregation
Definition of a large scale validation area for low-spatial resolution missions
Valuable requirements for a test site are the information on the area, basic documentation, availability of retrospective measurements and maintenance and attention to the site. These conditions are especially accomplished in this case, both for the University of Valencia site, that is, the VALENCIA Anchor Station, and for the University of Castilla-La Mancha sites, that is the Tomelloso, El Bonillo and Barrax Anchor Stations.
On the one hand, the Regional Ministry for Public Works, Land Planning and Transports of the Regional Government of Valencia scientifically documented the territory from the viewpoint of soil as a natural resource, and provided physiography maps, maps of soils, land capability, potential erosion and actual erosion (Antolin, 1998). These maps may me obtained in a Geographical Information System environment, perfectly documented.
On the other hand, the Castilla-La Mancha region has thoroughly been studied in previous well-known international projects such as:
EFEDA (ECHIVAL Field Experiment
in a Desertification-Threatened Area)
ECHIVAL-EFEDA PHASE II Project no.
5: Remote Sensing and Radiometric Properties of the Surface: Assessment of
Desertification From Space
RISMOP (Radiometric Impact of
Surface Moisture on Precipitation)
RESMEDES (Remote Sensing of
Mediterranean Desertification and Environmental Stability)
RESYSMED (RESMEDES Synthesis of
Change Detection Parameters Into a Land-Surface Change Indicator for Long-Term
Desertification Studies)
The overall area proposed may be defined from the available information mentioned above, taking into account the characteristics of soils, climate, physiography, etc. It does then result in total a large extended area, more than 300 km wide, reasonably homogeneous taking into account the low spatial resolution of the remote sensing instruments under consideration. The first principal task to be undertaken should be to homogenize the large quantity of information available and complete intermediate zones by means of interpolation methods based on spatial statistical techniques (kriging and co-kriging) and by using remote sensing techniques. It may be stated that there does not exist in an area of similar characteristics, especially as large and homogeneous, and located in a climatic area of so much scientific interest from many different viewpoints
Definition and characterisation of a large scale reference pixel
Complementarily to the previous objective, it is also planned to carry out specific tasks to facilitate the preparation of future Cal/Val activities of the missions mentioned earlier. This is a valuable objective that both the GIST (GERB International Science Team) and the SMOS Science Team have positively valued by accepting the VALENCIA Anchor Station site a one the basic areas to carry out this kind of activities.
This objective includes several planned tasks starting from the classification of different homogeneous environmental units in the large pixel area.In the study, an area of about 50 km wide around the actual Valencia Anchor Station site will be selected, approximately of the size of a GERB or a SMOS pixel, where to design and carry out a number of distributed measurements of soil moisture content, soil temperature, surface temperature, reflectance, albedo, and net radiation and together with some other meteorological parameters.
Non-homogeneities within every unit may be characterised by defining transects and using mobiles instrumentation. As far as surface reflectance is concerned, the Climatology from Satellites Group may count on the collaboration of the Field Radiometry Group, both included in the Remote Sensing Research Unit of the University of Valencia. It would be desirable to be able to also count on the collaboration of other groups to carry out angular measurements by using L-band radiometers as well.
It is planned that this characterisation be made firstly in an area of about 5 km diameter, and then scaling up to 10 km, 20 km, etc, until covering the whole zone. It will also be carried out under different meteorological and climatological conditions along the year, minimally during a dry period and immediately after a significant rain event, and accounting for spatial variability and topographic effects.
Study of scaling issues. Criteria for aggregation and disaggregation. Time Interpolation and Spatial Averaging
With no doubt, the large size, not only of the whole area, but also of the reference pixel, makes it necessary to study in detail the change of scale processes to be able to compare measurements proceeding from so different sources at different scales, namely point measurements, aircraft observations, different spatial resolution acquisitions from different satellite platforms, etc., and establish criteria for aggregation and disaggregation to get different area averages and validate large scale pixels.
Similarly, time interpolation, that is, handling the time dependent diurnal cycles or getting monthly averages of surface temperature, solar zenith angle albedo dependence, and cloud property diurnal cycles, to name a few examples is also a difficult task except for geostationary satellites.