Change Detection Techniques D. LU, P. MAUSEL, E. BRONDIZIO and E. MORAN Presented by Dahl Winters Geog 577, March 6, 2007

Background This paper summarizes change detection techniques, reviews their applications, and provides recommendations for selection of suitable change detection methods. This paper is organized into eight sections as follows: 1. a brief introduction to applications of change detection techniques 2. considerations before implementing change detection analyses 3. a summary and review of seven categories of change detection techniques 4. a review of comparative analyses among the different techniques 5. a brief review of global change analyses using coarse resolution satellite data 6. threshold selection 7. accuracy assessment 8. summary and recommendations

Introduction Good change detection research should provide the following information: 1. area change and change rate; 2. spatial distribution of changed types; 3. change trajectories of land-cover types; and 4. accuracy assessment of change detection results. When implementing a change detection project, three major steps are involved: 1. image preprocessing including geometrical rectification and image registration, radiometric and atmospheric correction, and topographic correction if the study area is in mountainous regions; 2. selection of suitable techniques to implement change detection analyses; and 3. accuracy assessment.

Introduction The accuracies of change detection results depend on many factors, including: 1. precise geometric registration between multi-temporal images, 2. calibration or normalization between multi-temporal images, 3. availability of quality ground truth data, 4. the complexity of landscape and environments of the study area, 5. change detection methods or algorithms used, 6. classification and change detection schemes, 7. analyst’s skills and experience, 8. knowledge and familiarity of the study area, and 9. time and cost restrictions.

Applications of change detection techniques land-use and land-cover (LULC) change forest or vegetation change forest mortality, defoliation and damage assessment deforestation, regeneration and selective logging wetland change forest fire and fire-affected area detection landscape change urban change environmental change, drought monitoring, flood monitoring, monitoring coastal marine environments, desertification, and detection of landslide areas other applications such as crop monitoring, shifting cultivation monitoring, road segments, and change in glacier mass balance and facies.

Considerations before doing change detection analyses The temporal, spatial, spectral, and radiometric resolutions of remotely-sensed images can significantly impact the success of change detection. Important environmental factors include atmospheric conditions, soil moisture conditions, and phenological characteristics. The following conditions must be met: 1. precise registration of multi-temporal images 2. precise radiometric and atmospheric calibration or normalization between multi- temporal images 3. similar phenological states between multi-temporal images to eliminate seasonal and phenological differences, and 4. selection of the same spatial and spectral resolution images if possible A change detection method should be selected that is appropriate to the data and study area. Some techniques can only provide change/no change information, while others can provide a complete matrix of change directions.

The seven change detection technique categories 1. algebra image differencing image regression image ratioing vegetation index differencing change vector analysis background subtraction 2. transformation PCA Tasseled Cap (KT) Gramm-Schmidt (GS) Chi-Square 3. classification Post-Classification Comparison Spectral-Temporal Combined Analysis EM Transformation Unsupervised Change Detection Hybrid Change Detection Artificial Neural Networks (ANN) 4. advanced models Li-Strahler Reflectance Model Spectral Mixture Model Biophysical Parameter Method 5. GIS Integrated GIS and RS Method GIS Approach 6. visual analysis Visual Interpretation 7. other change detection techniques Measures of spatial dependence Knowledge-based vision system Area production method Combination of three indicators: vegetation indices, land surface temperature, and spatial structure Change curves Generalized linear models Curve-theorem-based approach Structure-based approach Spatial statistics-based method

Category 1: Algebra

Category 2: Transformation

Category 3: Classification

Category 4: Advanced Models

Category 5: GIS

Categories 6 and 7

Comparing the Different Techniques Two types of change detection methods exist: those that detect binary change/non- change, and those that detect detailed “from-to” change between different classes. Different change detection techniques are often tested and compared to see which gives the best result, based on an accuracy assessment or qualitative assessment. After comparing several studies that used multiple change detection methods to identify the best method, the authors conclude that no single method is suitable for all cases. A combination of two change detection techniques can improve the change detection results (image differencing/PCA, NDVI/PCA, PCA/CVA). The most common change detection methods: image differencing, PCA, CVA, and post-classification comparison.

Global change analyses and image resolution For change detection at high or moderate spatial resolution: use Landsat TM, SPOT, or radar. For change detection at the continental or global scale, use coarse resolution data such as MODIS and AVHRR. AVHRR has daily availability at low cost; it is the best source of data for large area change detection. NDVI and land surface temperatures derived from MODIS or AVHRR thermal bands are especially useful in large area change detection.

Threshold Selection Many change detection algorithms (in the algebra and transformation categories) require threshold selection to determine whether a pixel has changed. Thresholds can be adjusted manually until the resulting image is satisfactory, or they can be selected statistically using a suitable standard deviation from a class mean. Both are highly subjective methods. Other methods exist for improving the change detection results, such as using fuzzy set and fuzzy membership functions to replace the thresholds. However, threshold selection is simple and intuitive, so it is still the most extensively applied method for detecting binary change/no-change information.

Accuracy Assessment Accuracy assessments are important for understanding the change detection results and using these results in decision-making. However, they are difficult to do because reliable temporal field-based datasets are often problematic to collect. The error matrix is the most common method for accuracy assessment. To properly generate one, the following factors must be considered: 1. ground truth data collection, 2. classification scheme, 3. sampling scheme, 4. spatial autocorrelation, and 5. sample size and sample unit.

Summary and Recommendations Before any change detection project, there must be precise geometrical registration and atmospheric correction or normalization between multi-temporal images. Also, suitable image acquisition dates and sensor data must be chosen, change categories must be selected, and appropriate change detection techniques must be used. The binary change/no-change threshold techniques all have difficulties in distinguishing true changed areas from the detected change areas. Single-band image differencing and PCA are the recommended methods. Classification-based change detection methods can avoid such problems, but requires more effort to implement. Post-classification comparison is a suitable method when sufficient training data is available. When multi-source data is available, GIS techniques can be helpful. Advanced techniques such as LSMA, ANN, or a combination of change detection methods can produce higher quality change detection results.