Andréfouët 2008). The validation sampling in coral reef environment can take a
long time and result in high costs due to a combination of boat travel time, boating,
diving and/or snorkeling requirements, variable sea state and weather conditions.
To reduce the cost and increase the number of validation samples, alternative
approaches for acquiring validation information can be utilized, such as community-based validation of land cover maps using the GeoWikki portal (Fritz et al.
2009), or validation of ocean color products through the SeaWiFS Bio-optical
Archive and Storage System (SEABASS), which is populated with both voluntary
and funded data (Fargion et al. 2004; Bailey and Werdell 2006).
This chapter also discussed validation processes, where mapped data sets were
compared with reference data sets for a specific location and time. However, it did
not discuss alternative approaches to assess the accuracy of a map in case no or
insufficient reference data is available. Congalton (1991) suggested that in place of
a quantitative accuracy assessment, accuracy could be tested through visual
inspection, non site-specific analysis, difference image creation, or error budgeting.
For example, validation of ocean color products is not only based on matching
image data with field observations (Bailey and Werdell 2006), but also on algorithm
assessment using long time-series, single location reference data sets (Werdell et al.
2007), assessing temporal trends (Campbell et al. 1995), and comparison of trends
derived with different processing algorithms (Campbell et al. 1995).
Although the chapter focused predominantly on validation of coral reef and
seagrass habitat mapping, the results and conclusions are also applicable to other
image-derived maps, such as water depth, rugosity and water properties. Obviously, there are differences in regards to field data collection for the different
environmental information types. Some of these differences include:
• Depth and rugosity require specialized equipment and processing techniques,
such as single or multi-beam SONAR, which are boat based.
• Water properties require sampling strategies that take into account the large
extent of the water surrounding the coral reef and the continuous three
dimensional movements and mixing of water molecules and particles in the
water column through tides, currents and wind.
• Maps presenting continuous properties (e.g., cholorphyll concentration) require
sampling of the properties such that measurements represent a normalized
distribution of the continuous property.
• Maps presenting thematic properties (e.g., habitat classis, geomorphic zones)
require sampling in the field for each thematic class to be mapped at the full
extent of the coral reef assessed.
To better understand the accuracy of coral reef maps and to improve approaches
in the future, additional research should create models to help predict and interpret
these accuracies for specific coral reef environments and particular image types.
For habitat information, this model should combine prediction of habitat map
accuracy in relation to the sensor type (Andréfouët et al. 2003), benthic spatial
complexity (Gustafson 1998), detail of mapping categories required (Mumby et al.
14 Validation
393
long time and result in high costs due to a combination of boat travel time, boating,
diving and/or snorkeling requirements, variable sea state and weather conditions.
To reduce the cost and increase the number of validation samples, alternative
approaches for acquiring validation information can be utilized, such as community-based validation of land cover maps using the GeoWikki portal (Fritz et al.
2009), or validation of ocean color products through the SeaWiFS Bio-optical
Archive and Storage System (SEABASS), which is populated with both voluntary
and funded data (Fargion et al. 2004; Bailey and Werdell 2006).
This chapter also discussed validation processes, where mapped data sets were
compared with reference data sets for a specific location and time. However, it did
not discuss alternative approaches to assess the accuracy of a map in case no or
insufficient reference data is available. Congalton (1991) suggested that in place of
a quantitative accuracy assessment, accuracy could be tested through visual
inspection, non site-specific analysis, difference image creation, or error budgeting.
For example, validation of ocean color products is not only based on matching
image data with field observations (Bailey and Werdell 2006), but also on algorithm
assessment using long time-series, single location reference data sets (Werdell et al.
2007), assessing temporal trends (Campbell et al. 1995), and comparison of trends
derived with different processing algorithms (Campbell et al. 1995).
Although the chapter focused predominantly on validation of coral reef and
seagrass habitat mapping, the results and conclusions are also applicable to other
image-derived maps, such as water depth, rugosity and water properties. Obviously, there are differences in regards to field data collection for the different
environmental information types. Some of these differences include:
• Depth and rugosity require specialized equipment and processing techniques,
such as single or multi-beam SONAR, which are boat based.
• Water properties require sampling strategies that take into account the large
extent of the water surrounding the coral reef and the continuous three
dimensional movements and mixing of water molecules and particles in the
water column through tides, currents and wind.
• Maps presenting continuous properties (e.g., cholorphyll concentration) require
sampling of the properties such that measurements represent a normalized
distribution of the continuous property.
• Maps presenting thematic properties (e.g., habitat classis, geomorphic zones)
require sampling in the field for each thematic class to be mapped at the full
extent of the coral reef assessed.
To better understand the accuracy of coral reef maps and to improve approaches
in the future, additional research should create models to help predict and interpret
these accuracies for specific coral reef environments and particular image types.
For habitat information, this model should combine prediction of habitat map
accuracy in relation to the sensor type (Andréfouët et al. 2003), benthic spatial
complexity (Gustafson 1998), detail of mapping categories required (Mumby et al.
14 Validation
393