Methodologies and Quality 3

The validation of satellite-derived terrestrial observations has perennially faced the challenge of finding consistent in-situ measurements that cover both a wide range of surface conditions and provide timely traceable product accuracy and uncertainty information. The Committee on Earth Observation Satellites (CEOS), the space arm of the Group on Earth Observations (GEO), plays a key role in coordinating the land product validation process. The Land Product Validation (LPV) subgroup of the CEOS Working Group on Calibration and Validation (WGCV) aims to address the challenges associated with the validation of global land products. This paper will provide the status of the LPV subgroup focus area activities.  The LPV focus areas cover seven terrestrial Global Climate Observing System (GCOS) Essential Climate Variables (ECVs): (1) Snow Cover, (2) Surface Albedo, (3) Land Cover, (4) Leaf Area Index, (5) Fraction of Absorbed Photosynthetically Active Radiation (FAPAR), (6) Fire, and (7) Soil Moisture; as well as two additional variables (Land Surface Phenology and Land Surface Temperature), that are key parameters of high priority for the LPV community. A primary focus of the LPV subgroup is the implementation of a global validation framework for product intercomparison and validation. This framework is based on a citable protocol, fiducial reference data, and automated subsetting. Ideally, each of these parts will be integrated into an online platform where quantitative tests are run, and standardized intercomparison and validation results reported for all products in a validation exercise. The establishment of consensus guidelines for in situ measurements along with the intercomparison of trends derived from independently-obtained reference data and derived products will enhance coordination of the scientific needs of Earth system communities with global LPV activities (http://lpvs.gsfc.nasa.gov/).

Spaceborne remote sensing plays an important role in collecting long-term consistent global data to underpin climate change research efficiently. These observations are used to derive biophysical essential climate variables (ECVs) that enable large-scale evaluation of the response of vegetation to changes in climate. ECVs such as leaf/plant area index (LAI/PAI), fraction of absorbed photosynthetically active radiation (fAPAR) and albedo are good indicators of small fluctuations in terrestrial vegetation over time. The validation of spaceborne ECV products is generally based on comparison either with in situ estimates or with reference products from different earth observation missions. However, most of these methods estimate biophysical quantities indirectly using a variety of assumptions and hypotheses. Therefore, direct comparison (true ‘validation’) and end-to-end traceability of in situ measurements and satellite-derived ECVs observations are difficult.

We suggest a framework for understanding, calibrating and validating in situ and satellite fAPAR and PAI estimates by generating a set of reference measurements using forest structural models to drive radiative transfer (RT) models. If these models can be used to simulate ground-based and EO data, we can control all aspects of sensor and environment properties, which would not be possible using measured data. Subsequently, biases resulting from differing algorithm assumptions can be quantified. RT models require canopy structure to be quantified. The introduction of terrestrial laser scanning (TLS), also referred to as terrestrial LiDAR, has enabled the possibility of generating tree models based on 3D scans. We present a processing chain that uses terrestrial LiDAR data and tree reconstruction to represent the explicit 3D forest structure used in radiative transfer models.

Our study area is a 6 ha deciduous forest at Wytham, Oxford, UK. TLS and PAI data were collected representing full leaf-on conditions during the summer of 2015 and leaf-off during winter 2015/16. TLS data were collected with a RIEGL VZ-400 terrestrial laser scanner at 176 scan locations within a 20 x 20 m regular grid pattern across the study area. PAI data were collected using several techniques and instruments including digital hemispherical photography (DHP), Li-Cor LAI-2000 and LAI-2200. We employed the VALERI sampling design for these optical PAI measurements (1800 measurements in total). The librat 3D Monte Carlo ray tracing (MCRT) RT model is employed to simulate DHPs, LAI-2000 and LAI-2200 measurements. This model requires a 3D explicit description of the forest structure, and spectral information about the canopy constituents used to represent the forest structure. We use TLS data to reconstruct tree models using the quantitative structure model (QSM) approach in [1-2] to model the stem and branching structure. Addition of the leaves is based on the derived light availability extracted from the point cloud and comparison of the leaf on and leaf-off scan data. Field measurements from the FieldSpec (ASD Inc.) were collected to provide the spectral characteristics of the leaves, bark and understory.

The various field instruments used here have been characterised radiometrically and geometrically, enabling calibration information to feed directly into the simulation framework process. As a result, this work provides for the first time an explicit framework for quantifying end-to-end traceability of various in situ fAPAR and PAI estimates RT modelling, stand reconstruction from TLS data and comparison of field estimates with simulations. To our knowledge, this study is the first reconstruction of a large (> 1 ha) study site directly from TLS data that can be used in a RT model to calibrate and validate ground-based, airborne and spaceborne sensors. Furthermore, the intensive sampling design that was deployed in the field will allow us to address issues of spatial variance and quantify the effect of different sampling strategies on the inferred ECVs.

ACKNOWLEDGMENTS

We thank A. Barker, T. Jackson and D. Fox for their assistance with fieldwork.

REFERENCES

[1] Calders, K., Newnham, G., Burt, A., Murphy, S., Raumonen, P. et al. (2015). Nondestructive estimates of above-ground biomass using terrestrial laser scanning. Methods Ecol Evol, 6, 198–208.

[2] Raumonen, P., Kaasalainen, M., Åkerblom, M., Kaasalainen, S., Kaartinen, H. et al. (2013). Fast Automatic Precision Tree Models from Terrestrial Laser Scanner Data. Remote Sensing, 5, 491–520.

European Space Agency’s (ESA) Soil Moisture and Ocean Salinity (SMOS) mission has been launched in November 2009 and has successfully spent almost six years in-orbit so far. SMOS is the second Earth Explorer Opportunity mission within the  ESA’s Living Planet Programme and it is the first mission providing global measurements of L-band brightness temperatures from space. ESA and the Centre National d'Etudes Spatiales (CNES) are jointly operating the SMOS mission.

The payload of SMOS consists of the Microwave Imaging Radiometer using Aperture Synthesis (MIRAS) instrument, a passive microwave 2-D interferometric full polarization radiometer, operating  at 1.413 GHz (wavelength of 21 cm) within the protected 1400-1427 MHz band. The interferometry technology has been developed for radio-astronomy and provides the opportunity to remove the spatial resolution constraint in the measurements from space, mainly due to the antenna size and the wavelength used to study the Earth’s water cycle. The MIRAS payload comprises a central structure and three deployable arms holding the equally distributed 69 antenna elements. The SMOS mission is based on a sun-synchronous orbit (dusk-dawn 6am/6pm) with a mean altitude of 758 km and an inclination of 98.44°. SMOS has a 149-days repeat cycle with a 18-days sub-cycle and a revisit time of 3 days. Due to the orbit geometry and the size of the MIRAS’s antennae the Sun appears in the antenna field of view close to the unit circle (e.g. about 60-90 degrees away from the antenna boresight) as a source of highly variable contamination for the image retrieved from the interferometric measurements. For this reason the SMOS level 1 data processor includes specific algorithm to estimate the Sun brightness temperature from the data itself and compensate it from the measured interferometric measurements. Such estimation of the Sun brightness temperature at L-band is archived in the level 1B product as well as the Sun position in the antenna unit circle. The estimated Sun brightness temperature is available for each SMOS image (i.e. every 1.2 s.) and for both H and V polarization on the antenna plane.

The paper presents the results of a validation study to assess the potentiality of the SMOS data: i) to support the research activities at L-band of the solar scientists community and  ii) to develop a space weather products derived from SMOS data.

The validation exercise had focused on SMOS data availability, coverage and statistical analysis for the SMOS derived Sun brightness temperature versus the Unites State Air Force (USAF) Radio Solar Telescope Network (RSTN). The RSTN  is a collection of data from four solar radio observatories operated by US Air Force located at Sagamore Hill (Hamilton, Massachusetts, USA), Palehua (Hawaii, USA), Learmonth (Australia) and San Vito dei Normanni (Italy).  The solar telescope network acquires data at eight frequencies ranging from 245 MHz to 15400 MHz. The acquisition frequency of 1415 MHz is very well suitable for comparison with the SMOS data set.

The validation has been done for different Sun condition, both eruptive Sun and quite/active Sun.  By analysing several Coral Mass Ejections (CMEs) events we checked the capability of the SMOS data set to track the evolution of the burst flux in the radio band. On the other side the quite Sun three years of data (2010 – 2012) has been analysed to assess the capability of the SMOS data to follow the increase of the mean solar activity during a period of transition from quite to active Sun. In both cases the two data sets (SMOS vs RSTN) have shown a strong statistical correlation. The results encourage to pursue further studies on the SMOS level 1 processing algorithm refinement and on the usage of SMOS data set as an additional source of information for space weather applications.

The paper also presents discussion about the main advantages in the usage of SMOS data like: i) the data set availability almost in near real time, ii) the homogeneity of the data set which does not require cross-calibration among different on Earth observatories, iii) an independent source of information for the interpretation of the on-ground observatories measurement that from time to time can be affected by Radio Frequency Interference (RFI).

By the way we also present some caveats in the use of SMOS dataset: i) image with strong contrast (i.e. Land-Sea transition) could impact the estimated Sun brightness temperature,  ii) RFI flagging should be considered (RFI also impacts SMOS measurements), iii) data availability does not cover full 24h (in the current algorithm version 620 the Sun is estimated only when it is in front of the antenna plane) and depends on the Earth position along its orbit around the Sun (this limitation will be removed in the future version of the SMOS processor which will include the estimation of the Sun brightness temperature also when the Sun is in the back of the antenna plane).  

INTRODUCTION

The TanDEM-X system is an innovative synthetic aperture radar (SAR) mission, which is comprised of two formation flying satellites, with the primary goal of generating a global digital elevation model (DEM) of unprecedented accuracy. TanDEM-X, being a large single-pass radar interferometer, achieves this accuracy through a flexible baseline selection enabling the acquisition of highly accurate cross-track interferograms that are not impacted by temporal decorrelation or atmospheric disturbances.

The global DEM data acquisition within the TanDEM-X mission, started in late 2010, has been concluded in autumn 2014. Following two full global coverages in 2011 and 2012, large areas have been additionally acquired with different SAR geometries in 2013 and 2014 in order to overcome the difficulties posed by e.g. mountains or deserts. Two dedicated campaigns have been performed to acquire Antarctica in the local winter for stable ground conditions. The interferometric processing and the generation of DEM products have been taking place parallel to the data acquisition.

In total around 500,000 single scenes are being processed into input DEMs for the subsequent calibration and mosaicking steps. The final DEM product will consist of about 20,000 geocells with an extension of about 110 km by 110 km, corresponding to 1° latitude by 1° longitude at the equator. Currently for more than 75% of the total land area TanDEM-X DEMs are available. By autumn 2016 the processing of the global DEM will be finished.

This presentation provides a status summary of the TanDEM-X global DEM with respect to the absolute and relative height accuracy as well as data completeness. These quality parameters will be presented on a global scale.

ABSOLUTE HEIGHT ACCURACY

The absolute height accuracy of the TanDEM-X data is globally validated using selected ICESat points. As the ICESat data is laser-based, there can be an offset to the radar-based TanDEM-X measured height, especially over vegetation or ice where the signal penetration of the two systems can differ.

The mean of the height deviation between these validation points and the currently available DEM data is 15 centimeters. The corresponding 90% linear error is 1.07 meters. The DEM product requirement of an absolute global height accuracy of at most 10 meters with a 90% linear error is thus met and far exceeded. See the attached figure “Final DEM – Absolute Height Accuracy" for a global representation.

RELATIVE HEIGHT ACCURACY

The DEM relative height accuracy is important for derivative products that make use of the local differences between adjacent elevation values, such as slope, aspect calculations, and drainage networks. The relative height accuracy per pixel can be estimated from the coherence (and number of looks) between the two SAR channels of the corresponding interferogram. As the coherence is a measure for the amount of noise in the interferogram, the respective relative height accuracy is given as the standard deviation of the corresponding error.

The relative height error specification describes the point-to-point error within a 1° x 1° geocell and requires a height accuracy of 2 m for flat terrain and 4 m for steep terrain at a confidence level of 90%.

Currently 9,099 out of 9,691 geocells have a relative height accuracy of more than 90% for the specified 2 m (4 m) of flat (steep) terrain or are not evaluated due to too few data points (e.g. small islands) or sea ice coverage. Furthermore, 514 geocells with lower relative height accuracy are dominated by highly forested areas. Due to volume decorrelation, the coherence estimation is deteriorated and consequently the relative height error is also increased. Hence, up to now only 143 geocells, or 1.5% of the produced geocells, do not meet the relative height accuracy specification. See the attached figure “Final DEM – Relative Height Accuracy" for a global representation.

As part of the Quality Assurance for Essential Climate Variable (QA44ECV*) project a new albedo Climate Data Record (CDR) is being developed. This CDR will be unique in that it will fuse together a number of European and US sensor data at level-2 (surface spectral and broadband BRFs) to retrieve a fused BRDF using an optimal estimation framework. The European level-1 data will be processed using the schema developed in the ESA GlobAlbedo project (Muller et al., 2013) whilst the US data will take existing level-2 products from MODIS and MISR. Both of the US products include uncertainty estimates at the per pixel level.

In addition, the CDR will be developed with the principles underpinning metrological traceability in mind. In terms of ECV CDR generation, metrological traceability is defined as “for each step of the ECV processing chain, the result of that step (and the associated uncertainties) are demonstrably derived from the output of the previous processing step”. The product will aim to be compliant with this principle as far as reasonably practicable.

As part of this drive towards full metrological traceability, the uncertainties associated with each of the input datasets needs to be understood, characterised and included within the processing chain and hence the final product uncertainties. The first step in achieving this is consideration of the Level 1 satellite data which is input into the processing chain.

The current work considers, and aims to characterise the uncertainties associated with VEGETATION, MERIS, (A)ATSR(2), PROBA-V Level 1 data through use of published data within the literature. The aim is to provide a set of uncertainties, or a generic uncertainty, which can be utilised within the re-processing of the albedo product due to be undertaken in the Summer of 2016.

* QA4ECV has received funding from the European Union’s Seventh Framework Programme (FP7/2007-2013) under grant agreement no. 607405.