Rasim Latifovic

Trends in vegetation NDVI from 1 km AVHRR data over Canada for the period 1985-2006

Long‐term changes in the Normalized Difference Vegetation Index (NDVI) have been evaluated in several studies but results have not been conclusive due to differences in data processing as well as the length and time of the analysed period. In this research a newly developed 1 km Advanced Very High Resolution Radiometer (AVHRR) satellite data record for the period 1985–2006 was used to rigorously evaluate NDVI trends over Canada. Furthermore, climate and land cover change as potential causes of observed trends were evaluated in eight sample regions. The AVHRR record was generated using improved geolocation, cloud screening, correction for sun‐sensor viewing geometry, atmospheric correction, and compositing. Results from both AVHRR and Landsat revealed an increasing NDVI trend over northern regions where comparison was possible. Overall, 22% of the vegetated area in Canada was found to have a positive NDVI trend based on the Mann–Kendal test at the 95% confidence level. Of these, 40% were in northern ecozones. The mean absolute difference of NDVI measurements between AVHRR and Landsat data was <7%. When compared with results from other studies, similar trends were found over northern areas, while in southern regions the results were less consistent. Local assessment of potential causes of trends in each region revealed a stronger influence of climate in the north compared to the south. Southern regions with strong positive trends appeared to be most influenced by land cover change.

Classification by progressive generalization : a new automated methodology for remote sensing multichannel data

A new procedure for digital image classi® cation is described. The procedure, labelled Classi® cation by Progressive Generalization (CPG), was developed to avoid drawbacks associated with most supervised and unsupervised classi® cations. Using lessons from visual image interpretation and map making, non-recursive CPG aims to identify all signi® cant spectral clusters within the scene to be classi® ed. The basic principles are: (i) initial data compression using spectral and spatial techniques; (ii) identi® cation of all potentially signi® cant spectral clusters in the scene to be classi® ed; (iii) minimum distance classi® cation; and (iv) the use of spectral, spatial and large-scale pattern information in the progressive merging of the increasingly dissimilar clusters. The procedure was tested with high- (Landsat Thematic Mapper (TM)) and medium- (Advanced Very High Resolution Radiometer (AVHRR) 1km composites) resolution data. It was found that the CPG yields classi® cation accuracies comparable to, or better than, current unsupervised classi® cation methods, is less sensitive to control parameters than a commonly used unsupervised classi® er, and works well with both TM and AVHRR data. The CPG requires only three parameters to be speci® ed at the outset, all specifying sizes of clusters that can be neglected at certain stages in the process. Although the procedure can be run automatically until the desired number of classes is reached, it has been designed to provide information to the analyst at the last stage so that ® nal cluster merging decisions can be made with the analysts input. It is concluded that the strategy on which the CPG is based provides an eA ective approach to the classi® cation of remote sensing data. The CPG also appears to have a considerable capacity for data compression.

Radiometric normalization, compositing, and quality control for satellite high resolution image mosaics over large areas

An objective normalization procedure has been developed to create image mosaics of radiometric equalization radiometric normalization for image mosaics (RNIM). The procedure employs a band-specific principal component analysis for overlap areas to achieve accurate and consistent radiometric transforms in each spectral band. It is demonstrated that the result of radiometric equalization is independent of the order of images to be mosaicked after the radiometric normalization adjustment is made. The selection of corresponding pixel pairs in the overlap area is controlled by using band-specific linear correlation coefficients, and the criteria for rejecting the cloudy and land-cover changed pixels. The final result is controlled quantitatively by employing the first and second principal components for the input data, which in turn depend on the selection of corresponding pixel pairs in the overlap area. In general, the radiometric resolution of input images can be conserved as long as gain /spl ges/1 and offset /spl ges/0 because of the stored format of the unsigned integer. The RNIM procedure accommodates these conditions. To take the best advantage of the data in the overlap areas, a pixel-based composite technique is employed in the production of the final mosaic. The selection of corresponding pixel pairs and the final result can be controlled and assessed with quantitative criteria. Therefore, this approach produces an objective, analyst-independent result and can be automated. The method has been successfully applied to six Landsat TM images of the BOREAS transect in Saskatchewan and Manitoba, Canada.

Land cover from multiple thematic mapper scenes using a new enhancement‐classification methodology

The purpose of this paper is to test a methodology for extracting land cover information from high-resolution satellite images over large areas. The use of multiple scenes brings additional complications which are related to atmospheric, phenological, spectral, and classification legend aspects of the data set. The approach is based on a supervised digital mosaicking of the input scenes and on a guided unsupervised classification of the mosaic. The goal of the development was to produce a methodology which is robust, minimally sensitive to the biases of the analyst, and capable of extracting all land cover type information contained in the satellite data. The mosaicking procedure aims to achieve radiometric consistency across the mosaic for key cover types. The classification procedure, named enhancement-classification method (ECM), operates on three (or less) input channels and produces a classification in which virtually the entire relevant spectral content has been extracted. The above procedures were employed to produce a land cover classification of the BOREAS transect from six Landsat thematic mapper images. Initial assessment indicated that the classification accuracy of the mosaics varies with cover type (44–82% for forest cover types). The variability appears related to residual radiometric effects, lack of unique spectral land cover signatures, and vegetation phenology. Thus nonremote sensing (or multitemporal) information is likely to be required for consistent accuracies. While the two-step procedure could be successfully automated under some conditions, the input of an analyst will remain essential until more experience is obtained, especially with classification across multiple scenes.

Mapping insect‐induced tree defoliation and mortality using coarse spatial resolution satellite imagery

Insect‐induced defoliation causes significant timber and carbon losses in many forested countries. The purpose of this investigation was to examine the potential use of coarse spatial resolution satellite imagery for mapping tree defoliation and mortality caused by a large insect infestation. We examined 1 km multi‐temporal SPOT Vegetation (VGT) data over a coniferous forest region in Quebec, Canada that was severely defoliated during 1998–2000 by the eastern hemlock looper. A logistic regression model based on satellite change metrics was developed to map defoliation and mortality. The results suggest that coarse imagery is effective for mapping large‐scale conifer forest mortality caused by insects, and could also be useful for near real‐time monitoring of severe defoliation, although with 2–3 times greater errors of commission.

Generating historical AVHRR 1 km baseline satellite data records over Canada suitable for climate change studies

Generating historical AVHRR 1 km baseline satellite data records over Canada suitable for climate change studies Rasim Latifovic, Alexander P. Trishchenko, Ji Chen, William B. Park, Konstantin V. Khlopenkov, Richard Fernandes, Darren Pouliot, Calin Ungureanu, Yi Luo, Shusen Wang, Andrew Davidson, and Josef Cihlar Pages 324-346 Abstract. Satellite data are an important component of the global climate observing system (GCOS). To serve the purpose of climate change monitoring, these data should satisfy certain criteria in terms of the length of observations and the continuity and consistency between different missions and instruments. Despite the great potential and obvious advantages of satellite observations, such as frequent repeat cycles and global coverage, their use in climate studies is hindered by substantial difficulties arising from large data volumes, complicated processing, and significant computer resources required for archiving and analysis. Successful examples of satellite earth observation (EO) data in climate studies include, among others, analyses of the earths radiation budget (Earth Radiation Budget Experiment (ERBE), Scanner for Radiation Budget (ScaRaB), and Cloud and the Earths Radiant Energy System (CERES)), cloudiness (International Satellite Cloud Climatology Project (ISCCP)), vegetation research (Global Inventory Modeling and Mapping Studies (GIMMS)), and the National Oceanic and Atmospheric Administration – National Aeronautics and Space Administration (NOAA–NASA) Pathfinder Program. Despite several attempts, the great potential of the advanced very high resolution radiometer (AVHRR) 1 km satellite data for climate research remains substantially underutilized. To address this issue, the generation of a comprehensive satellite data archive of AVHRR data and products at 1 km spatial resolution over Canada for 1981–2004 (24 years) has been initiated, and a new system for processing at level 1B has been developed. This processing system was employed to generate baseline 1 day and 10 day year-round clear-sky composites for a 5700 km × 4800 km area of North America. This region is centred over Canada but also includes the northern United States, Alaska, Greenland, and surrounding ocean regions. The baseline products include top-of-atmosphere (TOA) visible and near-infrared reflectance, TOA band 4 and band 5 brightness temperature, a cloud – clear – shadow – snow and ice mask, and viewing geometry. Details of the data processing system are presented in the paper. An evaluation of the system characteristics and comparison with previous results demonstrate important improvements in the quality and efficiency of the data processing. The system can process data in a highly automated manner, both for snow-covered and snow-free scenes, and for daytime and nighttime orbits, with high georeferencing accuracy and good radiometric consistency for all sensors from AVHRR NOAA-6 to AVHRR NOAA-17. Other processing improvements include the implementation of advanced algorithms for clear sky – cloud – shadow – snow and ice scene identification, as well as atmospheric correction and compositing. At the time of writing, the assembled dataset is the most comprehensive AVHRR archive at 1 km spatial resolution over Canada that includes all available observations from AVHRR between 1981 and 2004. The archive and the processing system are valuable assets for studying different aspects of land, oceans, and atmosphere related to climate variability and climate change.

Can interannual land surface signal be discerned in composite AVHRR data

The ability to make repeated measurements of the changing Earths surface is the principal advantage of satellite remote sensing. To realize its potential, it is necessary that true surface changes be isolated in the satellite signal from other effects which also influence the signal. In this study, we explore the magnitude of such effects in composite NOAA advanced very high resolution radiometer (AVHRR) images with a pixel spacing of 1 km. A compositing procedure is frequently used in the preparation of data sets for land biosphere studies to minimize the effect of clouds. However, the composite images contain residual artifacts which make it difficult to compare measurements at various times. We have employed a 4-year (1993–1996) AVHRR data set from NOAA 11 and 14 covering the Canadian landmass and corrected these data for the influence of the remaining clouds (full pixel or subpixel), atmospheric attenuation, and bidirectional reflectance. We have found that such corrections are essential for studies of interannual variations. The magnitude of the interannual signal varied with the AVHRR channel, land cover type, and satellite sensor but it was reduced by a factor of 2 to 8 between top of the atmosphere and the normalized surface reflectance. The remaining variations consisted of true interannual signal and the residual noise in the data (including sensor calibration) which was not removed by the correction process. Assuming that barren or sparsely vegetated land in northern Canada has not changed over the 4-year period, the mean residual uncertainty in surface reflectance of the selected sites was 0.012 for AVHRR channel 1, 0.042 for channel 2, and 0.068 for the normalized difference vegetation index (NDVI). These values decreased to 0.011, 0.024 and 0.038, respectively, when excluding 1994 data because their atmospheric and bidirectional corrections were hampered by high solar zenith angles (mean values above 55° in all 1994 composite periods). The errors could be further reduced by more refined corrections for bidirectional and atmospheric effects. The impact of the uncertainty of channel 1 and 2 measurements is also significantly diminished by using ratio indices such as the NDVI. It is concluded that interannual variability exceeding 0.015–0.038 in NDVI (averaged over multiple pixels) can be detected for similar data sets and conditions, provided that sensor calibration does not introduce additional errors. Since such errors can be large for some conditions and applications, the importance of accurate sensor calibration cannot be overemphasized.

A comparison of BRDF models for the normalization of satellite optical data to a standard Sun-target-sensor geometry

Climate change studies require consistent, long time series, surface reflectance data. The characterization of the bidirectional reflectance distribution function (BRDF) is important for normalizing the solar radiation reflected from the earths surface. We evaluated four BRDF models to identify the preferred approach to the normalization of multiyear National Oceanic and Atmospheric Administration Advanced Very High Resolution Radiometer (AVHRR) and SPOT4-VEGETATION (VGT) composite images to a common illumination and viewing geometry. Four models by the following authors were included: Walthall, Roujean, Ross-Li, and a new nonlinear temporal angular model (NTAM). NTAM accounts for hotspot effects and also responds to seasonal changes in land cover properties (using vegetation indexes as surrogate temporal measures). We compared the performance of the models under different scenarios of coefficient derivation and model application including model ability to reproduce theoretical BRDF curves, model consistency in single, multiyear, and incomplete sampling schemes, and comparison of AVHRR and LANDSAT Thematic Mapper surface reflectance prior and after BRDF normalization. We found that in all the tests, NTAM yielded the best fits between the observed and estimated values. NTAM requires eight coefficients and a lengthier iterative procedure to derive the coefficients, but the resulting coefficients are applied to the entire growing season rather than one temporal window. NTAM also performed well for different sensors (AVHRR, VGT) and geographic areas (Canada, east Asia, southern United States). Our results contradict the often-encountered perception that semiempirical BRDF models for angular normalization are all similarly effective, and the research on this topic is mature. We also describe a procedure for routine normalization of satellite optical data. For northern ecosystems, the NTAM coefficients derived from AVHRR and VGT data for Canada are available via ftp://ccrs.nrcan.gc.ca.

Canada-wide foliage clumping index mapping from multiangular POLDER measurements

In this paper, vegetation canopy structural information is retrieved over Canada from multiangular Advanced Earth Observing Satellite (ADEOS-1) polarization and directionality of the earths reflectance (POLDER) data based on canopy radiative transfer simulations using the Five-Scale model. The retrieval methodology makes use of the angular signature of the reflectance at the hot spot, where the sun and view angles coincide, and at the dark spot, where the reflectance is at its minimum. The POLDER data show that the normalized difference hot spot dark spot (NDHD) constructed from the hot spot and dark spot reflectances has no correlation with the nadir-normalized normalized difference vegetation index (NDVI), from which vegetation properties are often inferred, indicating that this angular index has additional information. Five-Scale simulations are used to assess the effects of foliage distribution on this angular index for different crown sizes, spatial distribution of crowns and foliage inside crowns, and foliage density variations. The simulations show that the NDHD is related to canopy structure quantified using a clumping index. This latter relationship is further exploited to derive a Canada-wide clumping index map at 7 km by 7 km resolution using spaceborne POLDER data. This map provides a critical new source of information for advanced modelling of radiation interaction with vegetation and energy and mass (water and carbon) exchanges between the surface and the atmosphere.

Selecting Representative High Resolution Sample Images for Land Cover Studies. Part 1: Methodology

Abstract This is the first of two articles which explore the combined use of coarse and fine resolution data in land cover studies. It describes the development and evaluation of an objective procedure to select a representative sample of tiles of high resolution images that complements a coarse resolution coverage of an entire region of interest. The second article explores the use of the procedure for an accurate estimation of cover type composition at the regional scale. The Purposive Selection Algorithm (PSA) assumes that a relationship exists between land cover compositions at the two spatial scales. It selects one tile at a time, seeking the sample which most closely resembles the composition of the coarse resolution map. Two selection criteria were used, fraction of cover types and contagion index. PSA was evaluated using two land cover maps for a 288 km×165 km area in central Saskatchewan, Canada derived from Landsat Thematic Mapper images (30 m pixels) and Advanced Very High Resolution Radiometer (AVHRR, 1000 m pixels), each divided into 64 tiles. The performance of an intermediate sensor (480 m pixels) was assessed by resampling the TM map. When using cover type composition alone, it was found that the procedure rapidly converges on a representative set of tiles with land cover composition very similar to the full coverage. The match between the domain and sample cover type fractions was very close, with errors less than 0.002% once about 1/5 to 1/3 of the tiles were selected and no discernible bias in the selected sample. Compared to the TM whole area coverage, samples selected with AVHRR classification were as representative as those obtained using the TM map. The performance of samples selected by a combination of cover composition and contagion index responded to the characteristics of individual tiles in terms of the selection criteria. A rigorous application of the algorithm with spatial heterogeneity measures such as the contagion index is computationally very demanding. It is concluded that PSA provides an efficient and effective tool to select a representative sample for land cover studies in which both large area coverage and local detail are desired.