14.1 Introduction
Spatial information and spatial modeling are increasingly used by scientists and
managers to evaluate options for making decisions in research and conservation
activities. Previous chapters discussed how various passive and active remote
sensing sensors and their related applications are used to derive environmental data
for coral reef environments, specifically information on benthic or substrate
composition (Phinn et al. 2012), biophysical processes (Weeks et al. 2006; Pittman
et al. 2009; Scopélitis et al. 2010), characteristics of coral reef assemblages
(Knudby et al. 2011), and/or efficiency of marine park management (Newman
et al. 2007). Measuring and understanding the sources of associated errors contained within each of these image-based map or spatial products is essential to
determine the error levels and reliability of the output products. As a result,
scientists, technicians, and managers need to understand the commonly used
validation processes, resulting accuracy measures and their relative comparability
for remote sensing products (Congalton and Green 1999; Foody 2002; Andréfouët
2008; Foody 2011).
This chapter focuses on the process of validating remote sensing outputs, where
we define validation as measuring the difference between a mapped feature type or
value from a remotely sensed image and an accepted reference feature or value at
the same location and time. The reference data can be in the form of an existing
map or data collected from the field. We focus on three types of biotic and abiotic
environmental information: (1) composition of coral reefs at varying levels of
detail (e.g., substrate or benthic cover type, geomorphic zones, biotopes, or reef
community composition); (2) the form or shape of the coral reef (e.g., bathymetry
and topographic complexity); and (3) the biophysical properties of the environment around the coral reef (e.g., sea surface temperature, sea surface winds,
concentrations of organic material, underwater light fields).
Each of these environmental variables exhibit variations over specific spatial
scales, in terms of minimum size and maximum extent, and in terms of structure or
concentration over a range of temporal scales (Hatcher 1997). As a result, mapping
each feature and measuring appropriate ‘‘reference’’ values for validation requires
recognition and consideration of the spatial and temporal scales associated with
the variable being mapped (Fig. 14.1). For example, coral reef composition and
biophysical properties can change at different time and spatial scales, from minutes
over small areas (e.g., chlorophyll concentration in water column) to months over
larger areas (e.g., macro algae cover). Additionally, the available extent and scale
of different environmental variables do not always coincide. For instance, the
extent to which water properties are mapped (e.g., SST, chlorophyll concentrations) are often significantly larger than the extent to which reef properties are
mapped (e.g., benthic composition, depth).
The discrete versus continuous nature, and the temporal and spatial variability,
of the coral reef environmental variables being mapped and monitored influence
several aspects of the validation process. For example, different validation
376
C. M. Roelfsema and S. R. Phinn
Spatial information and spatial modeling are increasingly used by scientists and
managers to evaluate options for making decisions in research and conservation
activities. Previous chapters discussed how various passive and active remote
sensing sensors and their related applications are used to derive environmental data
for coral reef environments, specifically information on benthic or substrate
composition (Phinn et al. 2012), biophysical processes (Weeks et al. 2006; Pittman
et al. 2009; Scopélitis et al. 2010), characteristics of coral reef assemblages
(Knudby et al. 2011), and/or efficiency of marine park management (Newman
et al. 2007). Measuring and understanding the sources of associated errors contained within each of these image-based map or spatial products is essential to
determine the error levels and reliability of the output products. As a result,
scientists, technicians, and managers need to understand the commonly used
validation processes, resulting accuracy measures and their relative comparability
for remote sensing products (Congalton and Green 1999; Foody 2002; Andréfouët
2008; Foody 2011).
This chapter focuses on the process of validating remote sensing outputs, where
we define validation as measuring the difference between a mapped feature type or
value from a remotely sensed image and an accepted reference feature or value at
the same location and time. The reference data can be in the form of an existing
map or data collected from the field. We focus on three types of biotic and abiotic
environmental information: (1) composition of coral reefs at varying levels of
detail (e.g., substrate or benthic cover type, geomorphic zones, biotopes, or reef
community composition); (2) the form or shape of the coral reef (e.g., bathymetry
and topographic complexity); and (3) the biophysical properties of the environment around the coral reef (e.g., sea surface temperature, sea surface winds,
concentrations of organic material, underwater light fields).
Each of these environmental variables exhibit variations over specific spatial
scales, in terms of minimum size and maximum extent, and in terms of structure or
concentration over a range of temporal scales (Hatcher 1997). As a result, mapping
each feature and measuring appropriate ‘‘reference’’ values for validation requires
recognition and consideration of the spatial and temporal scales associated with
the variable being mapped (Fig. 14.1). For example, coral reef composition and
biophysical properties can change at different time and spatial scales, from minutes
over small areas (e.g., chlorophyll concentration in water column) to months over
larger areas (e.g., macro algae cover). Additionally, the available extent and scale
of different environmental variables do not always coincide. For instance, the
extent to which water properties are mapped (e.g., SST, chlorophyll concentrations) are often significantly larger than the extent to which reef properties are
mapped (e.g., benthic composition, depth).
The discrete versus continuous nature, and the temporal and spatial variability,
of the coral reef environmental variables being mapped and monitored influence
several aspects of the validation process. For example, different validation
376
C. M. Roelfsema and S. R. Phinn