most commonly applied mapping approach. The implication is that the majority of
classification methods cannot be applied completely unsupervised, indicating that
field knowledge is required for calibration as much as for validation.
As a caveat, the review itself has limitations, which need to be taken into
account when assessing the results. Firstly, the 80 papers represent just the published scientific literature and not the gray literature. The gray literature could
reveal additional findings since these are often reporting operational mapping
applications in comparison to scientific research applications. Secondly, papers
were assigned to just a single category in each of the tables assessed. However,
some of these papers could be assigned to more than one category since they
compared habitat mapping for a variety of environments, image data types, calibration and validation field methods, mapping approaches, or accuracy measures.
Thirdly, the review focused only on the most common accuracy measures and
sampling designs. It did not discuss less commonly used, alternative approaches,
such as regression analysis (Mumby et al. 1999) and the hold-out method (Andréfouët et al. 2004). Although the reported accuracy information provided in the
reviewed literature is limited, it is surmised that in most instances the authors have
more information than what was reported, and did not include the information due
for example to length restrictions of scientific papers.
14.4 Conclusions and Recommendations
This chapter highlights the need for all coral reef mapping applications to contain
detailed information on accuracy assessment methods. Missing information on the
sampling design and accuracy assessment measures can lead to misinterpretation
of the results, inability to compare with other mapping projects, and potentially
incorrect conclusions by scientists and/or managers. To achieve better communication and better results, the accuracy reporting should include at a minimum:
• Description of the study area: extent, depth, and range of water clarity
conditions.
• Sampling design: sample unit size, number of samples, spatial distribution of
samples (map of study site with sample locations), and data collection process at
each sample unit.
• Accuracy measure used for continuous data: RMSE, correlation coefficient or
other.
• Accuracy measure used for thematic data: error matrix, individual thematic
class accuracies (user/producer accuracy) and the map accuracy.
• Methodology: An explanation of the chosen sampling design, accuracy measure
and references samples.
The chapter indicated that the costs of collecting independent validation data
were relatively high to the total cost of habitat mapping project and a major factor
in reducing the number of samples collected (Green et al. 2000; Stehman 2001;
392
C. M. Roelfsema and S. R. Phinn
classification methods cannot be applied completely unsupervised, indicating that
field knowledge is required for calibration as much as for validation.
As a caveat, the review itself has limitations, which need to be taken into
account when assessing the results. Firstly, the 80 papers represent just the published scientific literature and not the gray literature. The gray literature could
reveal additional findings since these are often reporting operational mapping
applications in comparison to scientific research applications. Secondly, papers
were assigned to just a single category in each of the tables assessed. However,
some of these papers could be assigned to more than one category since they
compared habitat mapping for a variety of environments, image data types, calibration and validation field methods, mapping approaches, or accuracy measures.
Thirdly, the review focused only on the most common accuracy measures and
sampling designs. It did not discuss less commonly used, alternative approaches,
such as regression analysis (Mumby et al. 1999) and the hold-out method (Andréfouët et al. 2004). Although the reported accuracy information provided in the
reviewed literature is limited, it is surmised that in most instances the authors have
more information than what was reported, and did not include the information due
for example to length restrictions of scientific papers.
14.4 Conclusions and Recommendations
This chapter highlights the need for all coral reef mapping applications to contain
detailed information on accuracy assessment methods. Missing information on the
sampling design and accuracy assessment measures can lead to misinterpretation
of the results, inability to compare with other mapping projects, and potentially
incorrect conclusions by scientists and/or managers. To achieve better communication and better results, the accuracy reporting should include at a minimum:
• Description of the study area: extent, depth, and range of water clarity
conditions.
• Sampling design: sample unit size, number of samples, spatial distribution of
samples (map of study site with sample locations), and data collection process at
each sample unit.
• Accuracy measure used for continuous data: RMSE, correlation coefficient or
other.
• Accuracy measure used for thematic data: error matrix, individual thematic
class accuracies (user/producer accuracy) and the map accuracy.
• Methodology: An explanation of the chosen sampling design, accuracy measure
and references samples.
The chapter indicated that the costs of collecting independent validation data
were relatively high to the total cost of habitat mapping project and a major factor
in reducing the number of samples collected (Green et al. 2000; Stehman 2001;
392
C. M. Roelfsema and S. R. Phinn