Multispectral Aerial Imagery Research Articles

A comparison between multispectral aerial and satellite imagery in precision viticulture

In this work we tested consistency and reliability of satellite-derived Prescription Maps (PMs) respect to those that can be obtained by aerial imagery. Test design considered a vineyard of Moscato Reale sited in Apulia (South-Eastern Italy) and two growing seasons (2013 and 2014). Comparisons concerned Landsat 8 OLI images and aerial datasets from airborne RedLake MS4100 multispectral camera. We firstly investigated the role of spatial resolution in radiometric features of data and, in particular, of NDVI maps and consequently of vigour maps. We first measured the maximum expected correlation between satellite- and aerial-derived maps. We found that, without any pixel selection and spatial interpolation, correlation ranges between 0.35 and 0.60 depending on the degree of heterogeneity of the vineyard. We also found that this result can be improved by operating a selection of those pixels representing vines canopy in aerial imagery and spatially interpolating them. In this way correlation coefficient can be improved up to 0.85 (minimum 0.60) suggesting an excellent capability of satellite data to approximate aerial ones at vineyard level. Prescription maps derived from vigour one demonstrated to be spatially consistent; but we also found that the quantitative interpretation of mapped vigour was changing in strength according to datasets and time of acquisition. Therefore, in spite of a satisfying consistency of spatial distribution, results showed that vigour strength at vineyard level from aerial and satellite datasets is generally not consistent, partially for the presence of a bias (that we modelled).

Urban landcover classification from multispectral image data using optimized AdaBoosted random forests

ABSTRACTWith an ever growing need to classify multispectral images, the accuracy of the classification becomes a matter of concern, especially when mapping heterogeneous environments such as urban areas. Not all of the features of a high-resolution multispectral image would equally contribute to the formulation of an ideal hypothesis, some of which might be redundant and might hinder the classification itself. Random forests (RF), an ensemble of decision trees, would assist in the removal of such extraneous features. While RF has proved to be one of the most efficient classifiers for spectral imagery, it holds untapped potential which could be honed to yield much more accurate prediction rates. This article discusses the application of AdaBoosted random forest (ABRF) to classify landcover segments from multispectral satellite or aerial imagery. A stage-wise optimization procedure is applied to achieve the best possible performance. It involves genetic feature selection and strategic empirical calibration to determine the ideal set of parameters. The application of boosting to RF resulted in the increase in the overall accuracy from 84.42% to 88.8% with an increase in kappa coefficient from 0.804 to 0.859.

Channel bar feature extraction for a mining-contaminated river using high-spatial multispectral remote-sensing imagery

Mapping and monitoring changes of geomorphological features over time are important for understanding fluvial process and effects of its controlling factors. Using high spatial resolution multispectral images has become common practice in the mapping as these images become widely available. Traditional pixel-based classification relies on statistical characteristics of single pixels and performs poorly in detailed mapping using high resolution multispectral images. In this work, we developed a hybrid method that detects and maps channel bars, one of the most important geomorphological features, from high resolution multispectral aerial imagery. This study focuses on the Big River which drains the Ozarks Plateaus region in southeast Missouri and the Old Lead Belt Mining District which was one of the largest producers of lead worldwide in the early and middle 1900s. Mapping and monitoring channel bars in the Big River is essential for evaluating the fate of contaminated mining sediment released to the Big River. The dataset in this study is 1 m spatial resolution and is composed of four bands: Red (Band 3), Green (Band 2), Blue (Band 1) and Near-Infrared (Band 4). The proposed hybrid method takes into account both spectral and spatial characteristics of single pixels, those of their surrounding contextual pixels and spatial relationships of objects. We evaluated its performance by comparing it with two traditional pixel-based classifications including Maximum Likelihood (MLC) and Support Vector Machine (SVM). The findings indicate that derived characteristics from segmentation and human knowledge can highly improve the accuracy of extraction and our proposed method was successful in extracting channel bars from high spatial resolution images.

A Vegetation Mapping Strategy for Conifer Forests by Combining Airborne LiDAR Data and Aerial Imagery

. Accurate vegetation mapping is critical for natural resources management, ecological analysis, and hydrological modeling, among other tasks. Remotely sensed multispectral and hyperspectral imageries have proved to be valuable inputs to the vegetation mapping process, but they can provide only limited vegetation structure characteristics, which are critical for differentiating vegetation communities in compositionally homogeneous forests. Light detection and ranging (LiDAR) can accurately measure the forest vertical and horizontal structures and provide a great opportunity for solving this problem. This study introduces a strategy using both multispectral aerial imagery and LiDAR data to map vegetation composition and structure over large spatial scales. Our approach included the use of a Bayesian information criterion algorithm to determine the optimized number of vegetation groups within mixed conifer forests in two study areas in the Sierra Nevada, California, and an unsupervised classification technique and post hoc analysis to map these vegetation groups across both study areas. The results show that the proposed strategy can recognize four and seven vegetation groups at the two study areas, respectively. Each vegetation group has its unique vegetation structure characteristics or vegetation species composition. The overall accuracy and kappa coefficient of the vegetation mapping results are over 78% and 0.64 for both study sites.Résumé. La cartographie précise de la végétation est essentielle entre autres pour la gestion des ressources naturelles, l’analyse écologique, et la modélisation hydrologique. Les approches d’imagerie multispectrale et hyperspectrale par télédétection se sont avérées de précieuses contributions au processus de la cartographie de la végétation, mais elles ne peuvent fournir qu’un nombre limité de caractéristiques sur la structure de la végétation, qui sont essentielles pour différencier les communautés végétales dans les forêts de composition homogènes. La télédétection par laser «light detection and ranging» (LiDAR) peut mesurer avec précision les structures verticales et horizontales de la forêt, et fournit une formidable opportunité de résoudre ce problème. Cette étude présente une stratégie qui utilise à la fois l’imagerie multispectrale aérienne et des données LiDAR pour cartographier la composition et la structure de la végétation à grandes échelles spatiales. Notre approche comprenait l’utilisation d’un algorithme du critère d’information Bayésien pour déterminer le nombre optimal de groupes de végétation dans les forêts mixtes de conifères sur deux zones d’étude dans les Sierra Nevada, en Californie, ainsi qu’une technique de classification non supervisée et une analyse post hoc pour cartographier ces groupes de végétation dans les deux zones d’étude. Les résultats montrent que la stratégie proposée peut reconnaitre quatre et sept groupes de végétation dans les deux zones d’étude respectivement. Chaque groupe de végétation a des caractéristiques uniques de structure de la végétation ou de composition des espèces de la végétation. La précision globale et le coefficient kappa des résultats de la cartographie de la végétation sont de plus de 78% et 0,64 pour les deux sites d’étude.

ESTIMATION OF CORK PRODUCTION USINGAERIAL IMAGERY1

ABSTRACT Inventory and prediction of cork harvest over time and space is important to forest managers who must plan and organize harvest logistics (transport, storage, etc.). Common field inventory methods including the stem density, diameter and height structure are costly and generally point (plot) based. Furthermore, the irregular horizontal structure of cork oak stands makes it difficult, if not impossible, to interpolate between points. We propose a new method to estimate cork production using digital multispectral aerial imagery. We study the spectral response of individual trees in visible and near infrared spectra and then correlate that response with cork production prior to harvest. We use ground measurements of individual trees production to evaluate the model’s predictive capacity. We propose 14 candidate variables to predict cork production based on crown size in combination with different NDVI index derivates. We use Akaike Information Criteria to choose the best among them. The best model is composed of combinations of different NDVI derivates that include red, green, and blue channels. The proposed model is 15% more accurate than a model that includes only a crown projection without any spectral information.

WILDLIFE MULTISPECIES REMOTE SENSING USING VISIBLE AND THERMAL INFRARED IMAGERY ACQUIRED FROM AN UNMANNED AERIAL VEHICLE (UAV)

Abstract. Wildlife aerial surveys require time and significant resources. Multispecies detection could reduce costs to a single census for species that coexist spatially. Traditional methods are demanding for observers in terms of concentration and are not adapted to multispecies censuses. The processing of multispectral aerial imagery acquired from an unmanned aerial vehicle (UAV) represents a potential solution for multispecies detection. The method used in this study is based on a multicriteria object-based image analysis applied on visible and thermal infrared imagery acquired from a UAV. This project aimed to detect American bison, fallow deer, gray wolves, and elks located in separate enclosures with a known number of individuals. Results showed that all bison and elks were detected without errors, while for deer and wolves, 0–2 individuals per flight line were mistaken with ground elements or undetected. This approach also detected simultaneously and separately the four targeted species even in the presence of other untargeted ones. These results confirm the potential of multispectral imagery acquired from UAV for wildlife census. Its operational application remains limited to small areas related to the current regulations and available technology. Standardization of the workflow will help to reduce time and expertise requirements for such technology.

The use of shadows in high spatial resolution, remotely sensed, imagery to estimate the height of individual Eucalyptus trees on undulating land

The shadows cast by 180 individual Eucalyptus trees, of varying canopy condition, on undulating land in south-eastern Australia were used to infer their heights from 50-cm spatial resolution, multispectral aerial imagery (blue = 0.4–0.5 μm; green = 0.5–0.6 μm; red = 0.6–0.7 μm; near infrared = 0.7–1 μm). A geometrical shadow model was developed incorporating the local slope and aspect of the ground from a digital elevation model at each tree location. A method of deriving ‘local tree time’ to infer the solar elevation angle, in situations where the image acquisition time is not available, was also developed. Based on a measurement of the shadow length from the geometric centre of the tree crowns, and ignoring the role of the crown periphery in distorting the shadow shape, the tree heights were estimated with a root mean square error of ±5.6 m (~±27%) with some overestimated by as much as 50%. A geometric correction for shadow distortion assuming spherical crown geometry provided an improved estimate with a root mean square error of ±4.8 m (~±23%).

Assessment of soybean injury from glyphosate using airborne multispectral remote sensing.

Glyphosate drift onto off-target sensitive crops can reduce growth and yield and is of great concern to growers and pesticide applicators. Detection of herbicide injury using biological responses is tedious, so more convenient and rapid detection methods are needed. The objective of this research was to determine the effects of glyphosate on biological responses of non-glyphosate-resistant (non-GR) soybean and to correlate vegetation indices (VIs) derived from aerial multispectral imagery. Plant height, shoot dry weight and chlorophyll (CHL) content decreased gradually with increasing glyphosate rate, regardless of weeks after application (WAA). Accordingly, soybean yield decreased by 25% with increased rate from 0 to 0.866 kg AI ha(-1) . Similarly to biological responses, the VIs derived from aerial imagery - normalized difference vegetation index, soil adjusted vegetation index, ratio vegetation index and green NDVI - also decreased gradually with increasing glyphosate rate, regardless of WAA. The VIs were highly correlated with plant height and yield but poorly correlated with CHL, regardless of WAA. This indicated that indices could be used to determine soybean injury from glyphosate, as indicated by the difference in plant height, and to predict the yield reduction due to crop injury from glyphosate.

Remote sensing can locate and assess the changing abundance of hollow-bearing trees for wildlife in Australian native forests

Context Hollow-bearing trees are an important breeding and shelter resource for wildlife in Australian native forests and hollow availability can influence species abundance and diversity in forest ecosystems. A persistent problem for forest managers is the ability to locate and survey hollow-bearing trees with a high level of accuracy at low cost over large areas of forest. Aims The aim of this study was to determine whether remote-sensing techniques could identify key variables useful in classifying the likelihood of a tree to contain hollows suitable for wildlife. Methods The data were high-resolution, multispectral aerial imagery and light detection and ranging (Lidar). A ground-based survey of 194 trees, 96 Eucalyptus crebra and 98 E. chloroclada and E. blakelyi, were used to train and validate tree-senescence classification models. Key results We found that trees in the youngest stage of tree senescence, which had a very low probability of hollow occurrence, could be distinguished using multispectral aerial imagery from trees in the later stages of tree senescence, which had a high probability of hollow occurrence. Independently, the canopy-height model used to estimate crown foliage density demonstrated the potential of Lidar-derived structural parameters as predictors of senescence and the hollow-bearing status of individual trees. Conclusions This study demonstrated a ‘proof of concept’ that remotely sensed tree parameters are suitable predictor variables for the hollow-bearing status of an individual tree. Implications Distinguishing early stage senescence trees from later-stage senescence trees using remote sensing offers potential as an efficient, repeatable and cost-effective way to map the distribution and abundance of hollow-bearing trees across the landscape. Further development is required to automate this process across the landscape, particularly the delineation of tree crowns. Further improvements may be obtained using a combination of these remote-sensing techniques. This information has important applications in commercial forest inventory and in biodiversity monitoring programs.

Estimation of cotton yield with varied irrigation and nitrogen treatments using aerial multispectral imagery

Cotton yield varies spatially within a field. The variability can be caused by various production inputs such as soil properties, water management, and fertilizer application. Airborne multispectral imaging is capable of providing data and information to study effects of the inputs on yield qualitatively and quantitatively in a timely and cost-effective fashion. A 10-ha cotton field with irrigation and non-irrigation 2×2 blocks was used in this study. Six nitrogen application treatments were randomized with two replications within each block. As plant canopy was closed, airborne multispectral images of the field were acquired using a 3-CCD MS4100 camera. The images were processed to generate various vegetation indices. The vegetation indices were evaluated for the best performance to characterize yield. The effect of irrigation on vegetation indices was significant. Models for yield estimation were developed and verified by comparing the estimated and actual yields. Results indicated that ratio of vegetation index (RVI) had a close relationship with yield (R2=0.47). Better yield estimation could be obtained using a model with RVI and soil electrical conductivity (EC) measurements of the field as explanatory variables (R2=0.53). This research demonstrates the capability of aerial multispectral remote sensing in estimating cotton yield variation and considering soil properties and nitrogen.

Fusion of remotely sensed data from airborne and ground-based sensors to enhance detection of cotton plants

The study investigated the use of aerial multispectral imagery and ground-based hyperspectral data for the discrimination of different crop types and timely detection of cotton plants over large areas. Airborne multispectral imagery and ground-based spectral reflectance data were acquired at the same time over three large agricultural fields in Burleson Co., Texas during the 2010 growing season. The discrimination accuracy of aerial- and ground-based data was examined individually; then a multi-sensor data fusion technique was applied on both datasets in order to improve the accuracy of discrimination. The individual classification accuracy of data taken with the aerial- and ground-based sensors were 90% and 93.3%, respectively. In comparison, the accuracy of discriminating crop types with fused data was 100% in the calibration and only 3.33% misclassification in the cross-validation. These results suggest that data fusion techniques could greatly enhance our ability to detect cotton from other plants.

SPECIES-SPECIFIC FOREST VARIABLE ESTIMATION USING NON-PARAMETRIC MODELING OF MULTI-SPECTRAL PHOTOGRAMMETRIC POINT CLOUD DATA

Abstract. The recent development in software for automatic photogrammetric processing of multispectral aerial imagery, and the growing nation-wide availability of Digital Elevation Model (DEM) data, are about to revolutionize data capture for forest management planning in Scandinavia. Using only already available aerial imagery and ALS-assessed DEM data, raster estimates of the forest variables mean tree height, basal area, total stem volume, and species-specific stem volumes were produced and evaluated. The study was conducted at a coniferous hemi-boreal test site in southern Sweden (lat. 58° N, long. 13° E). Digital aerial images from the Zeiss/Intergraph Digital Mapping Camera system were used to produce 3D point-cloud data with spectral information. Metrics were calculated for 696 field plots (10 m radius) from point-cloud data and used in k-MSN to estimate forest variables. For these stands, the tree height ranged from 1.4 to 33.0 m (18.1 m mean), stem volume from 0 to 829 m3 ha-1 (249 m3 ha-1 mean) and basal area from 0 to 62.2 m2 ha-1 (26.1 m2 ha-1 mean), with mean stand size of 2.8 ha. Estimates made using digital aerial images corresponding to the standard acquisition of the Swedish National Land Survey (Lantmäteriet) showed RMSEs (in percent of the surveyed stand mean) of 7.5% for tree height, 11.4% for basal area, 13.2% for total stem volume, 90.6% for pine stem volume, 26.4 for spruce stem volume, and 72.6% for deciduous stem volume. The results imply that photogrammetric matching of digital aerial images has significant potential for operational use in forestry.

GEOSPATIAL DATA INTEGRATION FOR ASSESSING LANDSLIDE HAZARD ON ENGINEERED SLOPES

Abstract. Road and rail networks are essential components of national infrastructures, underpinning the economy, and facilitating the mobility of goods and the human workforce. Earthwork slopes such as cuttings and embankments are primary components, and their reliability is of fundamental importance. However, instability and failure can occur, through processes such as landslides. Monitoring the condition of earthworks is a costly and continuous process for network operators, and currently, geospatial data is largely underutilised. The research presented here addresses this by combining airborne laser scanning and multispectral aerial imagery to develop a methodology for assessing landslide hazard. This is based on the extraction of key slope stability variables from the remotely sensed data. The methodology is implemented through numerical modelling, which is parameterised with the slope stability information, simulated climate conditions, and geotechnical properties. This allows determination of slope stability (expressed through the factor of safety) for a range of simulated scenarios. Regression analysis is then performed in order to develop a functional model relating slope stability to the input variables. The remotely sensed raster datasets are robustly re-sampled to two-dimensional cross-sections to facilitate meaningful interpretation of slope behaviour and mapping of landslide hazard. Results are stored in a geodatabase for spatial analysis within a GIS environment. For a test site located in England, UK, results have shown the utility of the approach in deriving practical hazard assessment information. Outcomes were compared to the network operator’s hazard grading data, and show general agreement. The utility of the slope information was also assessed with respect to auto-population of slope geometry, and found to deliver significant improvements over the network operator’s existing field-based approaches.

Urban Tree Cover Mapping with Relief-corrected Aerial Imagery and Lidar

This article discusses how urban tree canopy cover is often mapped by classifying high-resolution multispectral imagery. However, it can be difficult to differentiate low-lying vegetation from tree cover using optical data alone. Combining a lidar-derived normalized digital surface model (NDSM) improves classification accuracy, but the optical imagery is often imperfectly aligned with the NDSM. Aerial imagery is normally orthorectified using the ground elevation. However, tall objects in the orthorectified imagery still suffer from relief displacement and this can cause classification errors when lidar and the aerial imagery are combined. This article presents an approach for urban tree cover mapping that is composed of two parts. The first part is a method for correcting the relief displacement of trees in previously orthrectified aerial imagery, and the second part is an object-based classification method that combines relief-corrected multispectral aerial imagery with a lidar-derived NDSM. Using these methods, tree cover was mapped for a 1,600 ha region of London, Ontario, Canada with improved positional and classification accuracy.

A Remote Sensing Approach for Landslide Hazard Assessment on Engineered Slopes

Earthworks such as embankments and cuttings are integral to road and rail networks but can be prone to instability, necessitating rigorous and continual monitoring. To date, the potential of remote sensing for earthwork hazard assessment has been largely overlooked. However, techniques such as airborne laser scanning (ALS) are now ripe for addressing these challenges. This research presents the development of a novel hazard assessment strategy, combining high-resolution remote sensing with a numerical modeling approach. The research was implemented at a railway test site located in northern England, U.K.; ALS data and multispectral aerial imagery facilitated the determination of key slope stability variables, which were then used to parameterize a coupled hydrological-geotechnical model, in order to simulate slope behavior under current and future climates. A software toolset was developed to integrate the core elements of the methodology and determine resultant slope failure hazard which could then be mapped and queried within a geographical information system environment. Results indicate that the earthworks are largely stable, which is in broad agreement with the management company's slope hazard grading data, and in terms of morphological analysis, the remote methodology was able to correctly identify 99% of earthworks classed as embankments and 100% of cuttings. The developed approach provides an effective and practicable method for remotely quantifying slope failure hazard at fine spatial scales (0.5 m) and for prioritizing and reducing on-site inspection.

A Biologically Inspired Object Spectral-Texture Descriptor and Its Application to Vegetation Classification in Power-Line Corridors

The use of appropriate features to represent an output class or object is critical for all classification problems. In this letter, we propose a biologically inspired object descriptor to represent the spectral-texture patterns of images or objects. The proposed feature descriptor is generated from the pulse spectral frequencies (PSFs) of a pulse-coupled neural network, which is invariant to rotation, translation, and small scale changes. The proposed method is first evaluated in a rotation- and scale-invariant texture classification using the University of Southern California Signal and Image Processing Institute texture database. It is further evaluated in an application of vegetation species classification in power-line corridor monitoring using airborne multispectral aerial imagery. The results from the two experiments demonstrate that the PSF feature is effective in representing the spectral-texture patterns of objects, and it shows better results than classic color histogram and texture features.

Combining multispectral aerial imagery and digital surface models to extract urban buildings

Please click here to download the map associated with this article. This paper presents an automated classification of buildings in Coleraine, Northern Ireland. The classification was generated using very high spatial resolution data (10 cm) from a Digital Mapping Camera (DMC) for March 2009. The visible to near infrared (VNIR) bands of the DMC enabled a supervised classification to be performed to extract buildings from vegetation. A Digital Surface Model (DSM) was also created from the image to differentiate between buildings and other land classes with similar spectral profiles, such as roads. The supervised classification had the lowest classification accuracy (50%) while the DSM had an accuracy of 81%. The combination of the DSM and the supervised classification achieved an overall classification accuracy of 95%. Two spatial metrics (percentage of the landscape and number of patches) were also used to test the level of agreement between the classification and digitised building data. The results suggest that fine resolution multispectral aerial imagery can automatically detect buildings to a very high level of accuracy. Current space borne sensors, such as IKONOS and QuickBird, lag behind airborne sensors with VNIR bands provided at a much coarser spatial resolution (4m and 2.4m respectively). Techniques must be developed from current airborne sensors that can be applied to new space borne sensors in the future. The ability to generate DSMs from high resolution aerial imagery will afford new insights into the three-dimensional aspects of urban areas which will in turn inform future urban planning.

High-biomass sorghum yield estimate with aerial imagery

To reach the goals laid out by the U.S. Government for displacing fossil fuels with biofuels, high-biomass sorghum is well-suited to achieving this goal because it requires less water per unit dry biomass and can produce very high biomass yields. In order to make biofuels economically competitive with fossil fuels it is essential to maximize production efficiency throughout the system. The goal of this study was to use remote sensing technologies to optimize the yield and harvest logistics of high-biomass sorghum with respect to production costs based on spatial variability within and among fields. Specific objectives were to compare yield to aerial multispectral imagery and develop predictive relationships. A 19.2-ha high-biomass sorghum field was selected as a study site and aerial multispectral images were acquired with a four-camera imaging system on July 17, 2009. Sorghum plant samples were collected at predetermined geographic coordinates to determine biomass yield. Aerial images were processed to find relationships between image reflectance and yield of the biomass sorghum. Results showed that sorghum biomass yield in early August was closely related (R2 = 0.76) to spectral reflectance. However, in the late season the correlations between the biomass yield and spectral reflectance were not as positive as in the early season. The eventual outcome of this work could lead to predicted-yield maps based on remotely sensed images, which could be used in developing field management practices to optimize yield and harvest logistics.

Weed detection for site‐specific weed management: mapping and real‐time approaches

López‐Granados F (2011). Weed detection for site‐specific weed management: mapping and real‐time approaches. Weed Research 51, 1–11.SummaryThis work describes the current status of remote and proximal (on‐ground) weed detection systems for site‐specific weed management and discusses the limitations and opportunities of these technologies. Remote sensing based on multispectral aerial imagery can provide accurate weed maps, especially at late weed phenological stages, whereas images from high spatial resolution satellite and unmanned aerial vehicles must still be analysed. Hyperspectral images produce highly accurate maps at early and late phenological stages at a farm scale or medium spatial scale. However, this technology is not profitable, because of current operating costs, which are prohibitive. In studies of on‐ground weed seedling detection, accurate results can be obtained at a medium farm scale. Despite numerous efforts, a powerful and flexible classifier of soil, weeds and crops in a number of situations, remains the greatest challenge of this technology. The main limitations of remote and proximal sensing may be summarised in the following two points: (i) the time and education required for applying new technological advances and (ii) the high cost of the technology and the lack of compatibility of the machinery. Possible solutions might include: (i) offering an advisory service that provides technical support, agronomic knowledge and specific training courses, (ii) the development and implementation of uniform and cheaper standards, (iii) increased research of both high resolution satellite imagery exploring object‐based image analysis and pan‐sharpened imagery and unmanned aerial vehicles (UAV) and (iv) enabling the development of current prototypes of robotic weeding into commercial products. The general lack of multidisciplinary research groups can be a disadvantage when comparing the economic feasibility of site‐specific weed management with conventional systems.

Predicting Avian Abundance Within and Across Tidal Marshes Using Fine-Scale Vegetation and Geomorphic Metrics

Tidal marsh monitoring and restoration can benefit from the union of fine-scale remote sensing products and field-based survey data via spatial predictive models. As part of an interdisciplinary wetland monitoring project in San Francisco Bay, we developed a suite of 1-m pixel-level spatial metrics describing patterns in marsh vegetation and geomorphology for six sites across a large salinity gradient. These metrics, based on multi-spectral aerial imagery and derived vegetation maps, provided a basis for fine-scale spatial modeling of avian habitat potential. Using common yellowthroat (Geothlypis trichas), song sparrow (Melospiza melodia), and black rail (Later- allus jamaicensis) abundance data, we developed statistical models with relatively high explanatory power. In each case, models were improved by including vegetation-map variables, but variables directly extracted from aerial imagery were more reliable indicators of avian abundance. Although results varied by species, our models achieved reasonable within-site predictive success. When predicting to sites not used in the training set, however, validation results were inconsistent and often poor, suggesting that these models should be used with caution outside of the original study sites. As remotely sensed data become more readily available, our methods may be applied to a diverse range of sites, resulting in improved model generality and applicability.