474
This illustrates the importance of carefully considering all input variables at different
steps of the data fusion and assimilation process before using a RS product for
further analyses.
A range of data assimilation techniques is available, from univariate (scalar) and
multivariate (vector) 3D and 4D Kalman filter to ensemble methods that are particularly suitable where large sets of parameters are required and models are complex
(for details see, e.g., Bouttier and Courtier 2002; Evensen 2002).
17.5.3 Validation
Like any other source of data, spaceborne remotely sensed products have errors and
uncertainties associated with them. Validation is thus an ongoing challenge, although
a number of guidelines and recommendations for best practice exist. For example,
the NASA Land Product Validation Subgroup has published a framework for product validation and inter-comparison, as well as a “guide to the expression of uncertainty in measurement” (Schaepman-Strub et al. 2014). In the case of remotely
sensed LAI, the list of auxiliary parameters with associated uncertainties that should
be considered when validating LAI measures is long. It includes input data [land
cover, radiometric calibration error, geometry, aerosol optical depth at 550 nm, canopy condition (chlorophyll, dry matter, and moisture content), understory reflectance and geolocation (sensitivity to terrain slope), sensor noise (especially for dark
targets such as dense vegetation), clear sky top-of-atmosphere radiance, bidirectional reflectance distribution function (BRDF) modeling uncertainty, canopy and
understory modeling uncertainty, and geometric considerations (where products are
gridded in map projection systems of varying shape and area, Fernandes et al. 2014).
A detailed overview of validation techniques used across different levels of RS data
is given in Zeng et al. (2015), and guidelines on terminology, unified satellite validation metrics, and strategies, as well as explicit examples of RS validation techniques, including their mathematical basis, are provided in Loew et al. (2017).
Luckily for the end user, many of these validation steps are performed by the respective satellite agencies (e.g., the European Space Agency (Dorigo et al. 2017) and
NASA (Justice et al. 2013). However, being aware of the complexity of this endeavor
and the importance of considering both the target variable (e.g., LAI) and its associated quality measure (e.g., uncertainty) as provided by the space agencies is of
utmost importance to ensure appropriate use of RS products.
Apart from validating spaceborne RS data, validation techniques are also used to
evaluate the quality of modeled secondary indices, as well as to assess uncertainty
propagation after data fusion [e.g., accounting for uncertainty due to variable data
quality of in-situ or crowdsourced data (see, e.g., Comber et al. 2016) or to validate
downscaled RS products and airborne RS products]. For instance, crowdsourced
data have been used to validate a high-resolution global land cover map (Fritz et al.
2017), and in-situ measurements of LAI collected simultaneously with airborne
hyperspectral images were used to validate canopy radiative transfer models in
F. Schrodt et al.
This illustrates the importance of carefully considering all input variables at different
steps of the data fusion and assimilation process before using a RS product for
further analyses.
A range of data assimilation techniques is available, from univariate (scalar) and
multivariate (vector) 3D and 4D Kalman filter to ensemble methods that are particularly suitable where large sets of parameters are required and models are complex
(for details see, e.g., Bouttier and Courtier 2002; Evensen 2002).
17.5.3 Validation
Like any other source of data, spaceborne remotely sensed products have errors and
uncertainties associated with them. Validation is thus an ongoing challenge, although
a number of guidelines and recommendations for best practice exist. For example,
the NASA Land Product Validation Subgroup has published a framework for product validation and inter-comparison, as well as a “guide to the expression of uncertainty in measurement” (Schaepman-Strub et al. 2014). In the case of remotely
sensed LAI, the list of auxiliary parameters with associated uncertainties that should
be considered when validating LAI measures is long. It includes input data [land
cover, radiometric calibration error, geometry, aerosol optical depth at 550 nm, canopy condition (chlorophyll, dry matter, and moisture content), understory reflectance and geolocation (sensitivity to terrain slope), sensor noise (especially for dark
targets such as dense vegetation), clear sky top-of-atmosphere radiance, bidirectional reflectance distribution function (BRDF) modeling uncertainty, canopy and
understory modeling uncertainty, and geometric considerations (where products are
gridded in map projection systems of varying shape and area, Fernandes et al. 2014).
A detailed overview of validation techniques used across different levels of RS data
is given in Zeng et al. (2015), and guidelines on terminology, unified satellite validation metrics, and strategies, as well as explicit examples of RS validation techniques, including their mathematical basis, are provided in Loew et al. (2017).
Luckily for the end user, many of these validation steps are performed by the respective satellite agencies (e.g., the European Space Agency (Dorigo et al. 2017) and
NASA (Justice et al. 2013). However, being aware of the complexity of this endeavor
and the importance of considering both the target variable (e.g., LAI) and its associated quality measure (e.g., uncertainty) as provided by the space agencies is of
utmost importance to ensure appropriate use of RS products.
Apart from validating spaceborne RS data, validation techniques are also used to
evaluate the quality of modeled secondary indices, as well as to assess uncertainty
propagation after data fusion [e.g., accounting for uncertainty due to variable data
quality of in-situ or crowdsourced data (see, e.g., Comber et al. 2016) or to validate
downscaled RS products and airborne RS products]. For instance, crowdsourced
data have been used to validate a high-resolution global land cover map (Fritz et al.
2017), and in-situ measurements of LAI collected simultaneously with airborne
hyperspectral images were used to validate canopy radiative transfer models in
F. Schrodt et al.