Chuvieco and Huete (2009): Fundamentals of Satellite Remote Sensing, Taylor and Francis Emilio Chuvieco and Alfredo Huete Fundamentals of Satellite Remote Sensing – Chapter 8
Chuvieco and Huete (2009): Fundamentals of Satellite Remote Sensing, Taylor and Francis Accuracy assessment Need for accuracy assessment: Check whether the accuracy goals have been met. Demonstrate the quality of a new methodology Inconvenients: It requires additional cost and effort. Difficult to obtain an unbiased sample.
Chuvieco and Huete (2009): Fundamentals of Satellite Remote Sensing, Taylor and Francis Sampling for accuracy assessment Sampling design: how many samples?, where? Collect reference data. Compare ground truth with classification results (confusion matrix). Error analysis.
Chuvieco and Huete (2009): Fundamentals of Satellite Remote Sensing, Taylor and Francis Problems caused by non-spatial verification. The total number of gray pixels is the same but the spatial distribution is totally different Spatial/non spatial assessment
Chuvieco and Huete (2009): Fundamentals of Satellite Remote Sensing, Taylor and Francis Border Pixels Consequences of border errors. The border pixels will have an intermediate radiometric signal between two or more cover types Errors and landscape fragmentation
Chuvieco and Huete (2009): Fundamentals of Satellite Remote Sensing, Taylor and Francis DL The effect of radiometric contrast between neighboring covers can be observed in the profiles of areas with high and low contrast Errors and landscape contrast
Chuvieco and Huete (2009): Fundamentals of Satellite Remote Sensing, Taylor and Francis Spatial distribution of the error in digital classifications. The upper part corresponds to an image of an agricultural zone; the lower portion is a forested area (after Congalton 1988b) Errors and land use
Chuvieco and Huete (2009): Fundamentals of Satellite Remote Sensing, Taylor and Francis (i) (ii) (iii) (iv) (v) (i) Random (ii) Stratified random (iii) Systematic (iv) Unaligned systematic (v) Cluster Sampling approaches for accuracy assessment
Chuvieco and Huete (2009): Fundamentals of Satellite Remote Sensing, Taylor and Francis Field work for assessment Precise spatial location Well calibrated sensors Update information
Chuvieco and Huete (2009): Fundamentals of Satellite Remote Sensing, Taylor and Francis Comparing ground-truth and image data Interval-scale variables: Determination coefficient (r²) Residuals. Classified variables: Area covered at different spatial resolutions (binary variables). Confusion matrix.
Chuvieco and Huete (2009): Fundamentals of Satellite Remote Sensing, Taylor and Francis r²= r² = Impact of changing the observations used to compute the explained variance in the regression model. The white circles are points not used in computing the coefficient of determination coefficient (r²). Assessment of quantitative variables
Chuvieco and Huete (2009): Fundamentals of Satellite Remote Sensing, Taylor and Francis Metrics for quantitative variables Mean quadratic error: Cross-validation:
Chuvieco and Huete (2009): Fundamentals of Satellite Remote Sensing, Taylor and Francis Structure of a confusion matrix
Chuvieco and Huete (2009): Fundamentals of Satellite Remote Sensing, Taylor and Francis Example of confusion matrix After Salas and Chuvieco, 1994
Chuvieco and Huete (2009): Fundamentals of Satellite Remote Sensing, Taylor and Francis Accuracy metrics (1/2) Global accuracy: Class accuracy: User’s: Commission error: Producer’s: Omission error: A = A z*SE
Chuvieco and Huete (2009): Fundamentals of Satellite Remote Sensing, Taylor and Francis Accuracy metrics (2/2) Kappa index of agreement - Significance:
Chuvieco and Huete (2009): Fundamentals of Satellite Remote Sensing, Taylor and Francis Normalization of the confusion matrix
Chuvieco and Huete (2009): Fundamentals of Satellite Remote Sensing, Taylor and Francis Commission B:1 - (A/(A+B))= B/(A+B) Omission B: 1- (A/(A+C))= C/(A+C) U B DC BA UB Landsat MODIS C: B L -U M B: U L - B M A: B L -B M D: U L -U M B: Burned U: Unburned An example of cross-tabulation of a burned land map produced by two images with different spatial resolution Assessment by cross-tabulation
Chuvieco and Huete (2009): Fundamentals of Satellite Remote Sensing, Taylor and Francis Validation of a burned land mapping project of Latin America (after Chuvieco et al. 2008a). Measuring accuracy from scattergraphs
Chuvieco and Huete (2009): Fundamentals of Satellite Remote Sensing, Taylor and Francis Multitemporal confusion matrix A,B,C are different thematic categories. See text for explanation. Adapted from Congalton and Green (1999) and Biging et al. (1998).