Txomin Hermosilla

Pixel-Based Image Compositing for Large-Area Dense Time Series Applications and Science

Abstract Free and open access to the more than 40 years of data captured in the Landsat archive, combined with improvements in standardized image products and increasing computer processing and storage capabilities, have enabled the production of large-area, cloud-free, surface reflectance pixel-based image composites. Best-available-pixel (BAP) composites represent a new paradigm in remote sensing that is no longer reliant on scene-based analysis. A time series of these BAP image composites affords novel opportunities to generate information products characterizing land cover, land cover change, and forest structural attributes in a manner that is dynamic, transparent, systematic, repeatable, and spatially exhaustive. Herein, we articulate the information needs associated with forest ecosystem science and monitoring in a Canadian context, and indicate how these new image compositing approaches and subsequent derived products can enable us to address these needs. We highlight some of the issues and opportunities associated with an image compositing approach and demonstrate annual composite products at a national-scale for a single year, with more detailed analyses for two prototype areas using 15 years of Landsat data. Recommendations concerning how to best link compositing decisions to the desired use of the composite (and the information need) are presented, along with future research directions. Résumé L’accès libre et gratuit à plus de 40 ans de données dans l’archive Landsat combiné à l’amélioration des produits d’imagerie standardisés et l’augmentation des capacités de traitement et de stockage informatiques ont permis la production d’images composites basées sur les pixels de réflectance de surface de grande superficie sans nuages. Les composites du « meilleur pixel disponible » (best-available-pixel; BAP) représentent un nouveau paradigme en matière de télédétection qui ne dépend plus de l’analyse par scène. Une série chronologique de ces images composites BAP offre de nouvelles occasions de générer des produits d’information qui caractérisent la couverture terrestre, le changement de la couverture terrestre et les attributs structurels de la forêt d’une manière dynamique, transparente, systématique, répétable et spatialement exhaustive. Ici, nous articulons les besoins d’information liés à la science et à la surveillance des écosystèmes forestiers dans un contexte canadien, et nous indiquons comment ces nouvelles approches de composition d’image et les produits qui en découlent peuvent nous permettre de répondre à ces besoins. Nous soulignons quelques-uns des problèmes et des possibilités associés à une approche de composition d’image et nous démontrons des produits composites annuels à l’échelle nationale pour une année, avec des analyses plus détaillées pour deux zones prototypes utilisant 15 ans de données Landsat. Des recommandations concernant la meilleure façon de lier des décisions de composition d’images à l’utilisation souhaitée du composite (et le besoin d’information) ainsi que les orientations futures de la recherche sont présentées.

Definition of a comprehensive set of texture semivariogram features and their evaluation for object-oriented image classification

In this paper, a comprehensive set of texture features extracted from the experimental semivariogram of specific image objects is proposed and described, and their usefulness for land use classification of high resolution images is evaluated. Fourteen features are defined and categorized into three different groups, according to the location of their respective parameters in the semivariogram curve: (i) features that use parameters close to the origin of the semivariogram, (ii) the parameters employed extend to the first maximum, and (iii) the parameters employed are extracted from the first to the second maximum. A selection of the most relevant features has been performed, combining the analysis and interpretation of redundancies, and using statistical discriminant analysis methods. The suitability of the proposed features for object-based image classification has been evaluated using digital aerial images from an agricultural area on the Mediterranean coast of Spain. The performance of the selected semivariogram features has been compared with two different sets of texture features: those derived from the grey level co-occurrence matrix, and the values of raw semivariance directly extracted from the semivariogram at different positions. As a result of the tests, the classification accuracies obtained using the proposed semivariogram features are, in general, higher and more balanced than those obtained using the other two sets of standard texture features.

Evaluation of Automatic Building Detection Approaches Combining High Resolution Images and LiDAR Data

In this paper, two main approaches for automatic building detection and localization using high spatial resolution imagery and LiDAR data are compared and evaluated: thresholding-based and object-based classification. The thresholding-based approach is founded on the establishment of two threshold values: one refers to the minimum height to be considered as building, defined using the LiDAR data, and the other refers to the presence of vegetation, which is defined according to the spectral response. The other approach follows the standard scheme of object-based image classification: segmentation, feature extraction and selection, and classification, here performed using decision trees. In addition, the effect of the inclusion in the building detection process of contextual relations with the shadows is evaluated. Quality assessment is performed at two different levels: area and object. Area-level evaluates the building delineation performance, whereas object-level assesses the accuracy in the spatial location of individual buildings. The results obtained show a high efficiency of the evaluated methods for building detection techniques, in particular the thresholding-based approach, when the parameters are properly adjusted and adapted to the type of urban landscape considered.

Non-linear fourth-order image interpolation for subpixel edge detection and localization

A fourth-order non-linear interpolation procedure based on the ENO (Essentially Non-Oscillatory) methodology is presented and evaluated, with the purpose of increasing the geometric accuracy of edge detection in digital images. Two possible cases are considered one that considers that each pixel of the image represents a point value, the other that the pixel is an average value of a function. After image interpolation to obtain a finer grid of pixels, the Canny edge detection algorithm is applied, with the objective of improving the localization and geometry of the edges at a subpixel level. The results are compared with other schemes based on fourth order two-dimensional interpolation methods, such as a centered scheme based on a cubic convolution, a fourth order non-centered lineal scheme and a centered cubic convolution based on local gradient features. The evaluation is performed using visual and analytical techniques applied over aerial and satellite images, analyzing the positional errors of the detected edges, as well as the errors due to changes in scale and orientation. In addition to the subpixel edge detection, the quality of the interpolated images is tested. We conclude that the proposed methodology based on ENO interpolation improves the detection of edges in images as compared to other fourth-order methods.

Mass data processing of time series Landsat imagery: pixels to data products for forest monitoring

ABSTRACT Free and open access to the Landsat archive has enabled the implementation of national and global terrestrial monitoring projects. Herein, we summarize a project characterizing the change history of Canada’s forested ecosystems with a time series of data representing 1984–2012. Using the Composite2Change approach, we applied spectral trend analysis to annual best-available-pixel (BAP) surface reflectance image composites produced from Landsat TM and ETM+ imagery. A total of 73,544 images were used to produce 29 annual image composites, generating ∼400 TB of interim data products and resulting in ∼25 TB of annual gap-free reflectance composites and change products. On average, 10% of pixels in the annual BAP composites were missing data, with 86% of pixels having data gaps in two consecutive years or fewer. Change detection overall accuracy was 89%. Change attribution overall accuracy was 92%, with higher accuracy for stand-replacing wildfire and harvest. Changes were assigned to the correct year with an accuracy of 89%. Outcomes of this project provide baseline information and nationally consistent data source to quantify and characterize changes in forested ecosystems. The methods applied and lessons learned build confidence in the products generated and empower others to develop or refine similar satellite-based monitoring projects.

Particulate matter emitted from poultry and pig houses: source identification and quantification

There is need to identify and quantify the contribution of different sources to airborne particulate matter (PM) emissions from animal houses. To this end, we compared the chemical and morphological characteristics of fine and coarse PM from known sources collected from animal houses with the characteristics of on-farm fine and coarse airborne PM using two methods: classification rules based on decision trees and multiple linear regression. Fourteen different farms corresponding to seven different housing systems for poultry and pigs were sampled during winter. A total of 28 fine and 28 coarse on-farm airborne PM samples were collected, together with a representative sample of each known source per farm (56 known source samples in total). Source contributions were calculated as relative percentage contributions in particle numbers and then estimated in particle mass. Based on particle numbers, results showed that in poultry houses, most on-farm airborne PM originated from feathers (ranging from 4% to 43% in fine PM and from 6% to 35% in coarse PM) and manure (ranging from 9% to 85% in fine PM and from 30% to 94% in coarse PM). For pigs, most on-farm airborne PM originated from manure (ranging from 70% to 98% in fine PM and from 41% to 94% in coarse PM). Based on particle mass, for poultry most on-farm airborne PM still originated from feathers and manure; for pigs, however, most PM originated from skin and manure. Feed had a negligible contribution to on-farm airborne PM compared with other sources. Results presented in this study improve the understanding of sources of PM in different animal housing systems, which may be valuable when choosing optimal PM reduction techniques.

Using Small-Footprint Discrete and Full-Waveform Airborne LiDAR Metrics to Estimate Total Biomass and Biomass Components in Subtropical Forests

An accurate estimation of total biomass and its components is critical for understanding the carbon cycle in forest ecosystems. The objectives of this study were to explore the performances of forest canopy structure characterization from a single small-footprint Light Detection and Ranging (LiDAR) dataset using two different techniques focusing on (i) 3-D canopy structural information by discrete (XYZ) LiDAR metrics (DR-metrics), and (ii) the detailed geometric and radiometric information of the returned waveform by full-waveform LiDAR metrics (FW-metrics), and to evaluate the capacity of these metrics in predicting biomass and its components in subtropical forest ecosystems. This study was undertaken in a mixed subtropical forest in Yushan Mountain National Park, Jiangsu, China. LiDAR metrics derived from DR and FW LiDAR data were used alone, and in combination, in stepwise regression models to estimate total as well as above-ground, root, foliage, branch and trunk biomass. Overall, the results indicated that three sets of predictive models performed well across the different subtropical forest types (Adj-R2 = 0.42–0.93, excluding foliage biomass). Forest type-specific models (Adj-R2 = 0.18–0.93) were generally more accurate than the general model (Adj-R2 = 0.07–0.79) with the most accurate results obtained for coniferous stands (Adj-R2 = 0.50–0.93). In addition, LiDAR metrics related to vegetation heights were the strongest predictors of total biomass and its components. This research also illustrates the potential for the synergistic use of DR and FW LiDAR metrics to accurately assess biomass stocks in subtropical forests, which suggest significant potential in research and decision support in sustainable forest management, such as timber harvesting, biofuel characterization and fire hazard analyses.

Analysis of the factors affecting LiDAR DTM accuracy in a steep shrub area

The creation of a quality Digital Terrain Model (DTM) is essential for representing and analyzing the Earth in a digital form. The continuous improvements in the acquisition and the potential of airborne Light Detection and Ranging (LiDAR) data are increasing the range of applications of this technique to the study of the Earth surface. The aim of this study was to determine the optimal parameters for calculating a DTM by using an iterative algorithm to select minimum elevations from LiDAR data in a steep mountain area with shrub vegetation. The parameters were: input data type, analysis window size, and height thresholds. The effects of slope, point density, and vegetation on DTM accuracy were also analyzed. The results showed that the lowest root mean square error (RMSE) was obtained with an analysis window size of 10 m, 5 m, and 2.5 m, rasterized data as input data, and height thresholds equal to or greater than 1.5 m. These parameters showed a RMSE of 0.19 m. When terrain slope varied from 0–10% to 50–60%, the RMSE increased by 0.11 m. The RMSE decreased by 0.06 m when point density was increased from 4 to 8 points/m2, and increased by 0.05 m in dense vegetation areas.

Estimation of forest structure and canopy fuel parameters from small-footprint full-waveform LiDAR data

Precise knowledge of fuel conditions is important for predicting fire hazards and simulating fire growth and intensity across the landscape. We present a methodology to retrieve and map forest canopy fuel and other forest structural parameters using small-footprint full-waveform airborne light detection and ranging (LiDAR) data. Full-waveform LiDAR sensors register the complete returned backscattered signal through time and can describe physical properties of the intercepted objects. This study was undertaken in a mixed forest dominated by Douglas-fir, occasionally mixed with other conifers, in north-west Oregon (United States). We extracted two sets of LiDAR metrics using pulse detection and waveform modelling and then constructed several predictive models using forward stepwise multiple linear regression. The resulting models explained ~80% of the variability for many of the canopy fuel and forest structure parameters: aboveground biomass (R2=0.84), quadratic mean diameter (R2=0.82), canopy height (R2=0.79), canopy base height (R2=0.78) and canopy fuel load (R2=0.79). The lowest performing models included basal area (R2=0.76), stand volume (R2=0.73), canopy bulk density (R2=0.67) and stand density index (R2=0.66). Our results indicate that full-waveform LiDAR systems show promise in systematically characterising the structure and canopy fuel loads of forests, which may enable accurate fire behaviour forecasting that in turn supports the development of prevention and planning policies.

Forest recovery trends derived from Landsat time series for North American boreal forests

ABSTRACT A critical component of landscape dynamics is the recovery of vegetation following disturbance. The objective of this research was to characterize the forest recovery trends associated with a range of spectral indicators and report their observed performance and identified limitations. Forest disturbances were mapped for a random sample of three major bioclimate zones of North American boreal forests. The mean number of years for forest to recover, defined as time required to for a pixel to attain 80% of the mean spectral value of the 2 years prior to disturbance, was estimated for each disturbed pixel. The majority of disturbed pixels recovered within the first 5 years regardless of the index ranging from approximately 78% with normalized burn ratio (NBR) to 95% with tasselled cap greenness (TCG) and after 10 years more than 93% of disturbed pixels had recovered. Recovery rates suggest that normalized differenced vegetation index (NDVI) and TCG saturate earlier than indices that emphasize longer wavelengths. Thus, indices such as NBR and the mid-infrared spectral band offer increased capacity to characterize different levels of forest recovery. The mean length of time for spectral indices to recover to 80% of the pre-disturbance value for pixels disturbed 10 or more years ago was highest for NBR, 5.6 years, and lowest for TCG, 1.7 years. The mid-infrared spectral band had the greatest difference in recovered pixels among bioclimate zones 1 year after disturbance, ranging from approximately 42% of disturbed pixels for the cold and mesic bioclimate zone to 60% for the extremely cold and mesic bioclimate zone. The cold and mesic bioclimate zone had the longest mean years to recover ranging from 1.9 years for TCG to 4.2 years for NBR, while the cool temperate and dry bioclimate zone had the shortest mean years to recover ranging from 1.6 years for TCG to 2.9 years for NBR suggesting differences in pre-disturbance conditions or successional processes. The results highlight the need for caution when selecting and interpreting a spectral index for recovery characterization, as spectral indices, based upon the constituent wavelengths, are sensitive to different vegetation conditions and will provide a variable representation of structural conditions of forests.