Accuracy Assessment Chapter 14. Significance Accuracy of information is surprisingly difficult to address We may define accuracy, in a working sense,

Slides:



Advertisements
Similar presentations
Professor Gary Merlo Westfield State College
Advertisements

Major Operations of Digital Image Processing (DIP) Image Quality Assessment Radiometric Correction Geometric Correction Image Classification Introduction.
Enhancing Data Quality of Distributive Trade Statistics Workshop for African countries on the Implementation of International Recommendations for Distributive.
Accuracy Assessment of Thematic Maps
Selection of Research Participants: Sampling Procedures
Multiple Criteria for Evaluating Land Cover Classification Algorithms Summary of a paper by R.S. DeFries and Jonathan Cheung-Wai Chan April, 2000 Remote.
Beginning the Research Design
Lecture 14: Classification Thursday 18 February 2010 Reading: Ch – 7.19 Last lecture: Spectral Mixture Analysis.
The Calibration Process
Lecture 14: Classification Thursday 19 February Reading: “Estimating Sub-pixel Surface Roughness Using Remotely Sensed Stereoscopic Data” pdf preprint.
February 15, 2006 Geog 458: Map Sources and Errors
Global Land Cover: Approaches to Validation Alan Strahler GLC2000 Meeting JRC Ispra 3/02.
Sampling Designs Avery and Burkhart, Chapter 3 Source: J. Hollenbeck.
Trieschmann, Hoyt & Sommer Risk Identification and Evaluation Chapter 2 ©2005, Thomson/South-Western.
Cross Tabulation and Chi-Square Testing. Cross-Tabulation While a frequency distribution describes one variable at a time, a cross-tabulation describes.
Accuracy Assessment. 2 Because it is not practical to test every pixel in the classification image, a representative sample of reference points in the.
Chapter 9 Accuracy assessment in remotely sensed categorical information 遥感类别信息精度评估 Jingxiong ZHANG 张景雄 Chapter 9 Accuracy assessment in remotely sensed.
Methods of Validating Maps of Deforestation and Selective Logging Carlos Souza Jr. Instituto do Homem e Meio Ambiente da Amazônia—Imazon.
Aaker, Kumar, Day Ninth Edition Instructor’s Presentation Slides
Hypothesis Testing.
Fundamentals of Data Analysis Lecture 7 ANOVA. Program for today F Analysis of variance; F One factor design; F Many factors design; F Latin square scheme.
Rsensing6_khairul 1 Image Classification Image Classification uses the spectral information represented by the digital numbers in one or more spectral.
RESEARCH A systematic quest for undiscovered truth A way of thinking
Copyright © 2012 Wolters Kluwer Health | Lippincott Williams & Wilkins Chapter 14 Measurement and Data Quality.
Copyright © 2008 by Nelson, a division of Thomson Canada Limited Chapter 11 Part 3 Measurement Concepts MEASUREMENT.
Environmental Remote Sensing Lecture 5: Image Classification " Purpose: – categorising data – data abstraction / simplification – data interpretation –
Classification & Vegetation Indices
Lecture 3 Forestry 3218 Avery and Burkhart, Chapter 3 Shiver and Borders, Chapter 2 Forest Mensuration II Lecture 3 Elementary Sampling Methods: Selective,
Fundamentals of Data Analysis Lecture 10 Management of data sets and improving the precision of measurement pt. 2.
Texture. Texture is an innate property of all surfaces (clouds, trees, bricks, hair etc…). It refers to visual patterns of homogeneity and does not result.
Image Classification 영상분류
Chapter 1 Introduction to Statistics. Statistical Methods Were developed to serve a purpose Were developed to serve a purpose The purpose for each statistical.
Chapter 16 The Chi-Square Statistic
Accuracy Assessment Having produced a map with classification is only 50% of the work, we need to quantify how good the map is. This step is called the.
The Semivariogram in Remote Sensing: An Introduction P. J. Curran, Remote Sensing of Environment 24: (1988). Presented by Dahl Winters Geog 577,
QUANTITATIVE RESEARCH AND BASIC STATISTICS. TODAYS AGENDA Progress, challenges and support needed Response to TAP Check-in, Warm-up responses and TAP.
Measures of central tendency are statistics that express the most typical or average scores in a distribution These measures are: The Mode The Median.
Managerial Economics Demand Estimation & Forecasting.
Digital Image Processing Definition: Computer-based manipulation and interpretation of digital images.
Accuracy of Land Cover Products Why is it important and what does it all mean Note: The figures and tables in this presentation were derived from work.
1 Joint Research Centre (JRC) Using remote sensing for crop and land cover area estimation
Digital Image Processing
Change Detection The process of identifying differences in the state of an object or phenomenon by observing it at different times. Change detection applications:
Chuvieco and Huete (2009): Fundamentals of Satellite Remote Sensing, Taylor and Francis Emilio Chuvieco and Alfredo Huete Fundamentals of Satellite Remote.
Remote Sensing Classification Accuracy
Error & Uncertainty: II CE / ENVE 424/524. Handling Error Methods for measuring and visualizing error and uncertainty vary for nominal/ordinal and interval/ratio.
BOT / GEOG / GEOL 4111 / Field data collection Visiting and characterizing representative sites Used for classification (training data), information.
Spatial Smoothing and Multiple Comparisons Correction for Dummies Alexa Morcom, Matthew Brett Acknowledgements.
Chapter 7 Measuring of data Reliability of measuring instruments The reliability* of instrument is the consistency with which it measures the target attribute.
Chapter 15 The Chi-Square Statistic: Tests for Goodness of Fit and Independence PowerPoint Lecture Slides Essentials of Statistics for the Behavioral.
Semantic Alignment Spring 2009 Ben-Gurion University of the Negev.
Housekeeping –5 sets of aerial photo stereo pairs on reserve at SLC under FOR 420/520 –June 1993 photography.
Fundamentals of Data Analysis Lecture 4 Testing of statistical hypotheses pt.1.
Lesson 3 Measurement and Scaling. Case: “What is performance?” brandesign.co.za.
Accuracy Assessment Accuracy Assessment Error Matrix Sampling Method
McGraw-Hill/Irwin Copyright © 2009 by The McGraw-Hill Companies, Inc. All Rights Reserved. Chapter 3 Forecasting.
Class Seven Turn In: Chapter 18: 32, 34, 36 Chapter 19: 26, 34, 44 Quiz 3 For Class Eight: Chapter 20: 18, 20, 24 Chapter 22: 34, 36 Read Chapters 23 &
Accuracy Assessment of Thematic Maps THEMATIC ACCURACY.
Measurement and Scaling Concepts
Risk Identification and Evaluation Chapter 2
26. Classification Accuracy Assessment
26. Classification Accuracy Assessment
Accuracy Assessment of Thematic Maps
Incorporating Ancillary Data for Classification
REMOTE SENSING Multispectral Image Classification
REMOTE SENSING Multispectral Image Classification
2. Stratified Random Sampling.
Housekeeping 5 sets of aerial photo stereo pairs on reserve at SLC under FOR 420/520 June 1993 photography.
Assessment of data quality
ALI assignment – see amended instructions
Presentation transcript:

Accuracy Assessment Chapter 14

Significance Accuracy of information is surprisingly difficult to address We may define accuracy, in a working sense, as the degree (often as a percentage) of correspondence between observation and reality.  We usually judge accuracy against existing maps, large scale aerial photos, or field checks.

Significance We can pose two fundamental questions about accuracy:  Is each category in a classification really present at the points specified on a map?  Are the boundaries separating categories valid as located?

Various types of errors diminish the accuracy of feature identification and category distribution.  We make most of the errors either in measuring or in sampling. Three types of errors dominate

Data Acquisition Errors: These include sensor performance, stability of the platform, and conditions of viewing.  Reduce or compensate for them by making systematic corrections (e.g., by calibrating detector response with on-board light sources generating known radiances).  Make corrections using ancillary data such as known atmospheric conditions, during the initial processing of the raw data.

Data Processing Errors: An example is misregistration of equivalent pixels in the different bands of the Landsat Thematic Mapper.  Geometric correction should keep the mismatch to a displacement to one pixel. Under ideal conditions, and with as many as 25 ground control points (GCP) spread around a scene, we can realize this goal.  Misregistrations of several pixels significantly compromise accuracy.

Scene-dependent Errors: One such error relates to how we define and establish the class, which, in turn, is sensitive to the resolution of the observing system and the reference map or photo. Mixed pixels fall into this category.

Ancillary data An often overlooked point about maps as reference standards is their intrinsic or absolute accuracy.  Maps require an independent frame of reference to establish their own validity.  For centuries, most maps were constructed without regard to assessment of their inherent accuracy.

Ancillary data In recent years, some maps come with a statement of confidence level. The U.S. Geological Survey has reported results of accuracy assessments of the 1:250,000 and 1:1,000,000 land use maps of Level 1 classifications, based on aerial photos, that meets the 85% accuracy criterion at the 95% confidence level.

Landsat ETM+ image Rand McNally map

Obtainable Accuracy Level of accuracy obtainable depends on diverse factors, such as  the suitability of training sites,  the size, shape, distribution, and frequency of occurrence of individual areas assigned to each class which together determine the degree to which pixels are mixed,

Obtainable Accuracy  Sensor performance and resolution  The methods involved in classifying (visual photointerpreting versus computer-aided statistical classifying). A quantitative measure of the mutual role of improved spatial resolution and size of target on decreasing errors appears in this plot:

The dramatic improvement in reducing errors ensues for resolutions of 30 m (98 ft) or better.  This relates, in part, to the nature of the target classes.  Coarse resolution is ineffective in distinguishing crop types, but high resolution (< 20 m) adds little in recognizing these other than perhaps identifying species

As the size of crop fields increases, the error decreases further. The anomalous trend for forests (maximum error at high resolution) may be the consequence of the dictum: "Can't see the forest for the trees". Here high resolution begins to display individual species and breaks in the canopy that can confuse the integrity of the class "forest". Two opposing trends influence the behavior of these error curves:  1) statistical variance of the spectral response values decreases whereas  2) the proportion of mixed pixels increases with poorer resolution.

Accuracy and Precision Accuracy is the “correctness” Precision is the detail We may increase accuracy by decreasing precision  If we define something as “forest” it could include pine, broadleaf, scrub, etc.

Significance Why is it important  Legal standing of maps and reports  Operational usefulness  Validity as basis of scientific research Should be evaluated through a well- defined, quantitative process  This needs to be supported by independent evidence

Sources of error Errors is exist in any classification  Misidentification  Excessive generalization  Misregistration  Etc Simplest error may be misassignment of informational categories to spectral categories

Sources of error Most errors probably caused by complex factors  Mixed pixels A simple landscape with large uniform parcels is the easiest to classify

Sources of error Important Landscape Variables  Parcel size  Variation in parcel size  Parcel type  Number of types  Arrangement of different types  Number of parcels per type  Shapes of parcels  Radiometric and spectral contrasts

Sources of error Errors change from region to region and date to date

Error characteristics Classification error – assignment of a pixel in one category to another category  Errors are not randomly distributed across the image  Errors are not random to various categories  Tend to show clumped distribution on space  Errors may have spatial correlation to parcels Occur at edges or in the interior

Map Accuracy Measurement The task is to compare a map prepared from RS data, with another map (reference map) created from different source material.  The reference map is assumed to be accurate  If seasonal changes are important, reference should also reflect this

Map Accuracy Measurement Both maps must register Both maps must use the same classifications Both maps must be mapped at same level of detail

Map Accuracy Measurement Simplest comparison is total area of each class  Called non-site-specific accuracy  Imperfect because underestimation in one are can be compensated by overestimation in another Called inventory error

Map Accuracy Measurement Site specific accuracy is based on detailed assessment between the two maps  In most cases pixels are the unit of comparison  Known as classification error This is misidentification of pixels There may also be boundary errors

Error Matrix In the evaluation of classification errors, a classification error matrix is typically formed.  This matrix is sometimes called confusion matrix or contingency table. In this table, classification is given as rows and verification (ground truth) is given as columns for each sample point.

Error Matrix The diagonal elements in this matrix indicate numbers of sample for which the classification results agree with the reference data. Off diagonal elements in each row present the numbers of sample that has been misclassified by the classifier,  i.e., the classifier is committing a label to those samples which actually belong to other labels. The misclassification error is called commission error. The off-diagonal elements in each column are those samples being omitted by the classifier.  Therefore, the misclassification error is also called omission error.

Error Matrix

The most common error estimate is the overall accuracy From the example of confusion matrix, we can obtain ω = ( )/100 = 63%.

Error Matrix More specific measures are needed because the overall accuracy does not indicate how the accuracy is distributed across the individual categories.  The categories could, and frequently do, exhibit drastically differing accuracies but overall accuracy method considers these categories as having equivalent or similar accuracies.

Error Matrix From the confusion matrix, it can be seen that at least two methods can be used to determine individual category accuracies. (1) The ratio between the number of correctly classified and the row total  the user's accuracy - because users are concerned about what percentage of the classes has been correctly classified. (2) The ratio between the number of correctly classified and the column total  is called the producer's accuracy.

Error Matrix A more appropriate way of presenting the individual classification accuracies.  Commission error = 1 - user's accuracy  Omission error = 1 - producer's accuracy Overall Accuracy Consumer’s Accuracy Producer’s Accuracy

Error Matrix

Accuracy – e.g. Forest class  

Producers accuracy = =0.933  0r 93.3 % Consumers Accuracy = =0.49  Or 49%

The Kappa coefficient The Kappa coefficient (K) measures the relationship between beyond chance agreement and expected disagreement.  This measure uses all elements in the matrix and not just the diagonal ones.  The estimate of Kappa is the proportion of agreement after chance agreement is removed from consideration:

The Kappa coefficient = (p o - p c )/(1 - p c ) = (obs – exp)/(1-exp) p o = proportion of units which agree, =Spii = overall accuracy p c = proportion of units for expected chance agreement =Sp i+ p +i p ij = e ij /NT p i+ = row subtotal of p ij for row i p +i = column subtotal of p ij for column i

Error Matrix Grand Total = 100, Total correct = 63, Observed correct = 63/100 = 0.63

P i+ = 0.3 x 0.57 =.171, 0.3 x 0.21 =.063, 0.4 x 0.22 = 0.88 P c = Exp correct = = P o = Obs correct = = 0.63 (Overall Accuracy)

Kappa Coefficient One of the advantages of using this method is that we can statistically compare two classification products.  For example, two classification maps can be made using different algorithms and we can use the same reference data to verify them.  Two s can be derived, 1, 2. For each, the variance can also be calculated.

Another Way The following shows an alternative way to do the error matrix  Errors of Omission and Commission are both calculated from the row totals in this technique