Efficient and automatic object recognition in road scenes plays an essential role in smart city applications such as autonomous driving and intelligent infrastructure. As an important component of road scenes, pole-like facilities (PLFs) have been challenging to recognize high-definition road mapping. To achieve the automatic recognition of PLFs from cluttered mobile laser scanning (MLS) data, a novel instance segmentation method is proposed. First, candidate poles are detected by a supervoxel-based histogram analysis from partitioned off-ground point clouds. Then, instance segmentation of PLFs is achieved through a constrained hierarchical region-growing algorithm based on voxelized point clouds. A pyramid cascaded Fisher vector (FV) model and a random forest (RF) classifier are applied to classify the delineated pole-like road facilities into six predefined categories: trees, traffic signs, traffic lights, lamps, bare poles, and other objects. The proposed method is tested on three datasets collected in street scenes with different types of road facilities and point densities. Results demonstrate that our method can effectively achieve instance segmentation of PLFs in complex road environments. The proposed method outperformed state of the art for PLF detection in correctness (93.5%), completeness (95.3%), and quality (88.81%). Besides, the proposed method achieved satisfactory results for instance-level semantic segmentation with an average F1 score of 90.2%, demonstrating the effectiveness of geometric information enhancement in the designed FV coding approach.

Deep learning networks have become a new paradigm for semantic segmentation of remote sensing images. However, their great achievements heavily rely on a large number of high-quality labels, which costs a lot in acquisition. To cope with this issue, this paper introduced point labels, which provided only the categories of a small number of pixels. To fully exploit point labels for semantic segmentation, this paper proposed a simple but effective end-to-end point-supervised method, which combined a dense energy loss for learning the spatial relations of unlabeled pixels and a partial cross-entropy loss for learning labeled pixels. The Vaihingen and Zurich summer datasets were tested for the experiments. Results show that the proposed method can improve the mean F1 values by 1.57%-5.12%, compared to the baseline, and it can also retain well the boundary information of small objects.

Fine-resolution land surface temperature (LST) data is highly desired for comprehensively assessing the ecological and societal impacts of extreme climate events, such as heat waves. However, due to technical constraints, current single satellite sensor cannot provide LST data with both high temporal and spatial resolution simultaneously. To address this limitation, various downscaling and data fusion methods have been proposed to enhance the spatial details of geostationary satellites with high temporal resolution (e.g., hourly or sub-hourly). This study compares the performance of the statistical downscaling method and deep learning model in disaggregating LST observations of the new-generation geostationary satellite, GOES-R, from a resolution of 2 km to 70 m. The findings can contribute to the advancement of downscaling techniques and have practical implications for better analysis of fine-scale LST variations to support decision-making processes related to urban sustainability, land management, and extreme heat mitigation.

Floods pose severe global risks. While Earth observation satellites offer extensive flood monitoring, cloud cover limits the effectiveness of optical satellite imagery. This paper develops a novel reconstruction method for spatially seamless time series flood extent mapping. Utilizing a fine-tuned large foundation model, the proposed method identifies water bodies and then reconstructs cloud-covered flooding areas based on water occurrence data in the Global Surface Water dataset. The reconstructed water maps are finally refined by spatiotemporal Markov random field modeling to delineate flood areas. Evaluated with Harmonized Landsat and Sentinel-2 datasets, the developed method achieves seamless flood extent mapping at 2–3-day intervals and 30-m resolution. This paper offers an effective approach for flood monitoring under cloudy and rainy conditions, supporting emergency response and disaster management.

Mapping urban man-made objects, such as roads and buildings, from high-resolution remote sensing imagery is an essential need for monitoring global urbanization. However, the generalization ability of most existing models is limited due to the inconsistent data distribution of images across different regions. The emergence of the segment anything model (SAM) has significantly advanced image segmentation, primarily attributed to its strong zero-shot segmentation ability. However, SAM tends to underperform in various remote sensing tasks, such as road and building extraction, primarily due to the complexity of remote sensing imagery. This article introduced the multi-LoRA fine-tuned SAM (SAM_MLoRAF) framework, a simple yet effective network designed to extract urban man-made objects, which injected multiple parallel low-rank LoRA structures into the SAM encoder to approximate a high-rank LoRA, effectively mitigating the overfitting problem. In addition, it adopted a pyramid decoder to integrate multilevel information. For model optimization, supervised and unsupervised fine-tuning strategies were employed. Initially, the SAM_MLoRA was trained on publicly available datasets in a supervised manner to adapt to the task of urban man-made object extraction. In the second step, based on the idea of consistency regularization, unsupervised fine-tuning was employed to adapt the model to the target region by leveraging unlabeled images from the target region. Extensive experiments conducted on five continents have demonstrated that the proposed SAM_MLoRAF framework can efficiently leverage the robust segmentation capabilities of the SAM foundation model with a few trainable parameters, and most intersections over union (IoUs) of the mapping performance improved by over 10% compared to previous segmentation models. The code and datasets will be released at:https://github.com/xiaoyan07/SAM_MLoRA.

Cloud detection is a crucial procedure in remote sensing preprocessing. However, cloud detection is challenging in cloud–snow coexisting areas because cloud and snow have a similar spectral characteristic in visible spectrum. To overcome this challenge, we presented an automatic cloud detection neural network (ACD net) integrated remote sensing imagery with geospatial data and aimed to improve the accuracy of cloud detection from high-resolution imagery under cloud–snow coexistence. The proposed ACD net consisted of two parts: 1) feature extraction networks and 2) cloud boundary refinement module. The feature extraction networks module was designed to extract the spectral–spatial and geographic semantic information of cloud from remote sensing imagery and geospatial data. The cloud boundary refinement module is used to further improve the accuracy of cloud detection. The results showed that the proposed ACD net can provide a reliably cloud detection result in cloud–snow coexistence scene. Compared with the state-of-the-art deep learning algorithms, the proposed ACD net yielded substantially higher overall accuracy of 97.92%. This letter provides a new approach to how remote sensing imagery and geospatial big data can be integrated to obtain high accuracy of cloud detection in the circumstance of cloud–snow coexistence.

Landsat images have played an important role in the field of Earth observation and geoinformatics. However, optical Landsat images are frequently contaminated by cloud cover, especially in tropical and subtropical regions, which limits the utilization of these images. To improve the utilization of Landsat images, in this study, we propose a novel spatiotemporal neural network with four modules: a cloud detection module, a spatial–temporal learning module, a spatial–temporal feature fusion module, and a reconstruction module. The results of the experiments demonstrate that the proposed method is quantitatively effective (root mean square error < 0.0179) and can achieve a better result for reconstructing Landsat images than some of the widely used existing deep learning methods and multitemporal methods. The proposed neural network method provides an effective tool for the removal of contiguous, thick clouds from satellite images, so as to improve the quality of subsequent remote sensing mapping and geoinformation extraction.

Wetlands are highly sensitive ecosystems that have experienced largely undocumented loss across Canada. Accurate statistics of historic loss of wetlands across many provinces is vague at best or non-existent at worst, as exemplified in Newfoundland and Labrador (NL). Thus, NL represents a perfect candidate for implementing historical remote sensing data sets and change detection methods. Given recent advancements in earth observation technology, it is now feasible to implement remote sensing-based change detection methods at scales never previously possible. As such, the goal of this work is to develop a methodology to assess wetland class change across the island of Newfoundland between 1985 and 2015 using historic and current Landsat imagery, Random Forest classification, and the Google Earth Engine (GEE) platform. The resulting accuracies ranged from 84.37% to 88.96%. The analysis reveals that wetland classes over the last 30 years have been unstable, and the biggest loss of wetlands to anthropogenic land cover occurred between the 1980's and the 1990's. Index Terms - Wetlands, Change Detection, Landsat, Geo big data

The applications of remote sensing technology have been focused on environmental issues and natural resources. Coarse- and medium-resolution optical and radar imagery has limited usage in urban areas. At the turn of the 21 st century, we have witnessed great advances in remote sensing and imaging science. Commercial satellites acquire imagery at a spatial resolution previously only possible to aerial platforms, but these satellites have advantages over aerial imageries including their capacity for synoptic coverage, inherently digital format, short revisit time, and capability to produce stereo image pairs conveniently for high-accuracy 3D mapping thanks to their flexible pointing mechanism. Hyperspectral imaging affords the potential for detailed identification of materials and better estimates of their abundance in the Earth's surface, enabling the use of remote sensing data collection to replace data collection that was formerly limited to laboratory testing or expensive field surveys. Lidar (Light Detection and Ranging) technology can provide high-accuracy height and other geometric information for urban structures and vegetation. In addition, radar technology has been re-inventing itself since the 1990s, due largely to the increase of spaceborne radar programs. These technologies are not isolated at all. In fact, their integrated uses with more established aerial photography and multispectral remote sensing techniques have been the main stream of current remote sensing research and applications. With the advent of the new sensor technology, the reinvention of “old” technology, and more capable computational techniques, the field of remote sensing and Earth observation is rapidly gaining, or regaining, interest in the geospatial technology community, governments, industries and the general public. The integration of the internet technology with remote sensing imaging science and GIS have led to the emergence of geo-referenced information over the web, such as Google Earth and Virtual Globe. These new geo-referenced “worlds”, in conjunction with GPS, mobile mapping, and modern telecommunication technologies, have sparked much interest in the public for remote sensing and imaging science (Weng, 2012). Within this context, urban remote sensing has become a new frontier in geospatial technology. This trend has been demonstrated by rapidly increasing publications on the topic and its widespread applications.

The applications of remote sensing technology have been focused on environmental issues and natural resources. Coarse- and medium-resolution optical and radar imagery has limited usage in urban areas. At the turn of the 21st century, we have witnessed great advances in remote sensing and imaging science. Commercial satellites acquire imagery at a spatial resolution previously only possible to aerial platforms, but these satellites have advantages over aerial imageries including their capacity for synoptic coverage, inherently digital format, short revisit time, and capability to produce stereo image pairs conveniently for high-accuracy 3D mapping thanks to their flexible pointing mechanism. Hyperspectral imaging affords the potential for detailed identification of materials and better estimates of their abundance in the Earth's surface, enabling the use of remote sensing data collection to replace data collection that was formerly limited to laboratory testing or expensive field surveys. Lidar (Light Detection and Rangng) technology can provide high-accuracy height and other geometric information for urban structures and vegetation. In addition, radar technology has been re-inventing itself since the 1990s, due largely to the increase of spaceborne radar programs. These technologies are not isolated at all. In fact, their integrated uses with more established aerial photography and multispectral remote sensing techniques have been the main stream of current remote sensing research and applications. With the advent of the new sensor technology, the reinvention of “old” technology, and more capable computational techniques, the field of remote sensing and Earth observation is rapidly gaining, or regaining, interest in the geospatial technology community, governments, industries and the general public. The integration of the internet technology with remote sensing imaging science and GIS have led to the emergence of geo-referenced information over the web, such as Google Earth and Virtual Globe. These new geo-referenced “worlds”, in conjunction with GPS, mobile mapping, and modern telecommunication technologies, have sparked much interest in the public for remote sensing and imaging science (Weng, 2012). Within this context, urban remote sensing has become a new frontier in geospatial technology. This trend has been demonstrated by rapidly increasing publications on the topic and its widespread applications.

This study proposed an object-based method to estimate the relationship of optimal thresholds in different The Defense Meteorological Satellite Program/Operational Linescan System (DMSP/OLS) nighttime light (NTL) images for the purpose of mapping large-scale urban dynamics. In the method, the optimal threshold for an urban object was predicted by delineating the reference threshold and relating it to that in a target year through object-based normalization. The process of threshold normalization assumed the stability or proportional change of NTL intensity over time for pseudo invariant pixels. A test in China to extend the thresholds derived for F152000 NTL data to F152005 and F182010 NTL images showed that the proposed method successfully connect thresholds between years. The estimated threshold was highly correlated with the one derived from the reference urban area, with R2of 0.96 and 0.90 and RMSE of 2.9 and 4.9 for 2005 and 2010, respectively. At the object level, R2and RMSE of the estimated vs. reference urban areas were 0.91 and 14 pixels and 0.85 and 30 pixels for 2005 and 2010, respectively. At provincial level, R2was 0.92 and 0.87 and RMSE was 316 and 451 pixels for 2005 and 2010, respectively.

Satellite-retrieved aerosol optical depth (AOD) has been increasingly utilized for the mapping of fine particulate matter (PM2.5) concentrations. An accurate estimation and mapping of PM2.5concentrations depends on the high-resolution AOD data and a robust mathematical model that takes into account the spatial nonstationary relationship between PM2.5and AOD. Take the core portion of the Beijing-Hebei-Tianjin (Jing-Jin-Ji) urban agglomeration as case study (the most seriously polluted region in China). Land use, population, meteorological variables, and simplified aerosol retrieval algorithm-retrieved AOD at 1-km resolution are employed as the predictors for the geographically weighted regression (GWR) and the ordinary least squares (OLS) model to map the spatial distribution of PM2.5concentrations. The GWR model shows significant spatial variations in PM2.5concentrations over the region than the traditional OLS model, which reveals relative homogeneous variations. Validation with ground-level PM2.5concentrations demonstrates that PM2.5concentrations predicted by the GWR model (R2= 0.75, RMSE = 10 μg/m3) correlate better than those by the OLS model (R2= 0.53, RMSE = 16 μg/m3). These results suggest that the GWR model offered a more reliable way for the prediction of spatial distribution of PM2.5concentrations over urban areas.

Image segmentation is a key step in geographic object-based image analysis. Numerous segmentation techniques, e.g., watershed segmentation, mean-shift segmentation, and fractal net evolution algorithm, have been proposed and applied for various types of image analysis tasks. The majority of the segmentation algorithms require a user-defined parameter, namely, the scale parameter, to control the sizes of segments, yet the automation of the scale parameter remains a great challenge. Over the past few years, several automated parameterization methods, such as the estimation of scale parameters (ESP) tool, have been developed. However, few of the existing methods are able to enhance both intrasegment homogeneity and intersegment heterogeneity. In this letter, we proposed an energy function method that aimed at enhancing the characteristics of intrasegment homogeneity and intersegment heterogeneity, simultaneously, to identify the optimal segmentation scale for image segmentation. The intersegment heterogeneity was calculated as the weighted gradient from a segment to its neighbors by spectral angle, whereas the intrasegment homogeneity was quantified by the mean spectral angle within a segment. The performance of the proposed method was evaluated by applying it to a WorldView-2 multispectral image of Toronto, Canada, and comparing it with the local-peak-based method, which considered only the intrasegment homogeneity of an image. The scale parameter identified by the proposed method can better characterize the reference geo-objects over the entire image. The accuracy assessment result shows that the proposed method outperformed the existing technique by reducing the discrepancy by 17.9%.

Land surface temperature (LST) is a valuable parameter in studies of surface energy balance, landscape thermal patterns, and human–environment interactions. An effective way to quantify the LST dynamics over spatial and temporal domains is to utilize the consistent Landsat thermal infrared (TIR) data since 1982. Currently, only a small proportion of studies utilized the Landsat TIR data for investigating both the intra- and interannual LST variations. The objectives of the study are to provide statistical evidence for the existence of the annual temperature cycle (ATC) and to develop a decomposition technique to explore landscape thermal patterns by land cover. Eighty-two cloud-free TIR images of Los Angeles County from Landsat TM between 2000 and 2010 were collected and consistently calibrated to the LSTs. The LSTs were then analyzed by the Lomb–Scargle periodogram technique to test whether the time series LSTs showed rhythmic patterns and by a decomposition model to analyze the intra- and interannual landscape thermal patterns. The periodogram analysis confirmed that ATC was statistically significant with the periodic time of 362 days. Furthermore, sensitivity analysis showed that the Lomb–Scargle technique can still discover the ATC with the difference of up to five days, even when the number of images decreased to 60. Based on the periodogram analysis, a decomposition model was initialized to disassemble the time series LSTs into seasonality and trend components for comparisons among land covers. Results suggested that the developed areas exhibited relatively low seasonal amplitude of 11.7 K, while largest mean annual LST value is 302.8 K. The difference of the averaged trend component between urban and other land covers reached 1.1 K over the decade. Future research may be directed in dealing with the time-varying seasonality component for better quantifying the thermal patterns.

Recent years have witnessed an emerging concern of the health impact of heat waves. A common approach to investigate heat waves is to resort to the geostationary thermal infrared imagery, such as those from the Geostationary Operational Environmental Satellite (GOES) and Meteosat Second Generation. However, coarse spatial resolutions of geostationary images cannot meet the need of assessing and monitoring heat waves in complex urban settings. To address the spatial and temporal variability of heat waves in urban areas, this letter presented a study of analyzing heat wave risk in Los Angeles, USA, by the synergistic use of GOES land surface temperature (LST), auxiliary geospatial, and census data within the framework of Crichton's Risk Triangle (i.e., hazard, exposure, and vulnerability). Principal component analysis and regression analysis were employed to downscale the original GOES LST imagery from 4 to 1 km. The resultant subhourly 1-km LST data was used to characterize and quantify heat hazard. The census population represented the exposure, while existing health, socioeconomic, and physical environmental conditions were used to describe the vulnerabilities. The risk map of heat wave was computed using the weighted indices of hazard, exposure, and vulnerability. The map was further overlaid with a zip-code data layer to generate statistics. The derived risk map showed that areas with high risk were identified in the central city, part of western LA County, and the desert area, based on a 10-point scale rank.

Anthropogenic heat is the heat flux generated by human activities and is a major contributor to the formation of an urban heat island. In a city such as Hong Kong, obtaining pure pixels from medium- or coarse-resolution remote sensing images is challenging. Considering the completely different thermal properties of vegetation and impervious surfaces, this letter developed a novel algorithm to estimate anthropogenic heat fluxes by decomposing image pixels into fractions of impervious surfaces and vegetation, and by estimating the total heat flux for the mixed pixel. The Chinese small satellite HJ-1B images with a spatial resolution of 30 and 300 m for visible and thermal wavebands, respectively, and the temporal resolution of four days were used for the heat flux modeling. Results show that anthropogenic heat fluxes in Hong Kong are correlated to the building density and the building height, with${r}^{2} = \mbox{0.92}\ \text{and}\ \mbox{0.58} $on October 11, 2012 and${r}^{2} = \mbox{0.94}\ \text{and}\ \mbox{0.62} $on January 13, 2013, respectively. The average anthropogenic heat fluxes in urban areas are 289.16 and 283.17 W/m2on October 11, 2012 and on January 13, 2013, respectively, and the commercial areas emit the largest anthropogenic heat fluxes around 500–600 W/m2compared with other land-use types. The derived anthropogenic heat fluxes can help in planning and environmental authorities to pinpoint “hot-spot” areas, and they can be used for compliance monitoring.

Fine-scale population estimation in urban areas provides information useful in such fields as emergency response, epidemiological applications, and urban management. It is however a challenge because of lack of detailed building morphologic information. This research investigated the capability of LiDAR data for extraction of residential buildings and used the results for population estimation in heterogeneous environments in Indianapolis, USA. A morphological building detection algorithm was applied, to extract buildings from LiDAR point cloud, and yielded an overall detection accuracy of 95%. Extracted buildings were then categorized into nonresidential buildings, apartments, single-family houses, and other buildings based on selected geometric features (e.g., area, height, and volume) and background characteristics (vegetation and impervious cover) by a random forest classifier. Linear regression modeling, based on area, volume, and housing units, was applied to examine the relationship between census population and LiDAR-derived residential variables. The results show that morphological metrics extracted from LiDAR can be applied to classify buildings with relatively high accuracy, with an overall accuracy of 81.67%. The shape indexes contributed mostly to the residential building extraction followed by building background metrics. By excluding nonresidential buildings, the accuracy of population estimation increased from an RMSE of 20 and a mean absolute relative error (MARE) of 61.38% to an RMSE of 13 and a MARE of 33.52%. The differentiation between single-family houses and apartments contributed to the improved estimation. Additionally, the introduction of building height resulted in relatively accurate unit-based estimation. This study provides important insights into fine-scale population estimation in heterogeneous urban regions, when detailed building information is unavailable.

Data-driven and model-driven strategies are two basic approaches for building reconstruction based on LiDAR data. Due to the data noise and limitations in existing algorithms, the data-driven approach can neither construct a complete roof plane nor construct irregular planes. This letter proposes a model-driven approach to reconstruct 3-D building structures by developing prototypical roofs for commercial and residential buildings. The experiment was conducted in the City of Indianapolis, IN, USA, using the LiDAR data and building footprints provided by the city government. Irregular building footprints were first decomposed into nonintersecting and mostly quadrangular blocks for the identification of the most probable prototypical roofs. A decision tree classifier was applied to classify all building footprints into seven subtypes based on the physical and morphological parameters of buildings. The modeling of prototypical roofs was finished with the parameters including the length, the width, and the orientation of the principal axis of each building block that were computed from the LiDAR data using the static moment equations. Additional parameters, including the mean height, the top height, and the gutter height, were also considered as needed for some subtypes. Complex roofs were reconstructed by assembling adjacent prototypical roofs. The decision tree classification method was finally applied to 268 building blocks and achieved an overall accuracy of 82.1% for seven classes. A 3-D geographic information system building database that includes commercial and residential buildings in two chosen city blocks was created for further applications. This letter created a more completed building roof structure than the existing data-driven approach and demonstrated the reliability of a decision tree classifier in categorizing building roofs.

Vegetation-cover-based thermal sharpening techniques have mostly been developed and tested in agricultural areas. Overlooking the impact of soil moisture on surface temperatures is a common problem in these algorithms. This letter developed a vegetation thermal sharpening method for the City of Indianapolis, Indiana, USA, and estimated land surface temperature by disaggregated Landsat Thematic Mapper thermal infrared data from 120 to 30 m. The root-mean-square error was yielded at 1.90°C and 1.91°C using NDVI and fractional vegetation cover as predictors, respectively. The error of the estimation was overlaid with a soil moisture map, which was derived based on the surface energy balance modeling. The pixels with large errors were largely distributed in the areas with low soil moisture. These areas were covered by impervious surfaces such as major roads, commercial land, and the airport. This result suggested that in the urban areas, besides vegetation cover and soil moisture, impervious surfaces must be incorporated in developing any future thermal sharpening techniques. The incorporation of population density and per capita consumption of energy may provide further improvements in the estimation.

Although Earth observation data have been used in urban thermal applications extensively, these studies are often limited by the choices made in data selection, i.e., either using data with high spatial and low temporal resolution, or data with high temporal and low spatial resolution. The challenge of advancing the low spatial (3-5 km) resolution of geostationary land surface temperature (LST) images to 1 km-while maintaining the excellent temporal resolution of 15 min-is approached in this letter. The downscaling was performed using different advanced regression algorithms, such as support vector regression machines, neural networks, and regression trees, and its performance was improved using gradient boosting. The methodologies were tested on Meteosat Second Generation (MSG) SEVIRI LST images over an area of 19 600 km2centered in Athens, Greece. The output 1-km downscaled LST images were assessed against coincident LST maps derived from the thermal infrared imagery of the Moderate Resolution Imaging Spectroradiometer, the Advanced Very High Resolution Radiometer, and the Advanced Along Track Scanning Radiometer. The results showed that support vector machines coupled with gradient boosting proved to be a robust high-performance methodology reaching correlation coefficients from 0.69 to 0.81 when compared with the other satellite-derived LST maps.

An important issue in urban thermal remote sensing is how to use pixel-based measurements of land surface temperature (LST) to characterize and quantify the urban heat island (UHI) observed at the mesoscale and macroscale. Characterization and modeling of UHIs must consider the inherent spatial nonstationarity within land surface variables. This study extended a kernel convolution modeling method for 2-D LST imagery to characterize and model the UHI in Indianapolis, IN, as a Gaussian process. To understand the UHI pattern over space and time, four Advanced Spaceborne Thermal Emission and Reflection Radiometer images of different seasons/years were acquired and analyzed. Furthermore, we employed linear spectral mixture analysis to extract subpixel urban biophysical variables [i.e., green vegetation (GV) and impervious surface (IS)] and developed new indexes of greenness and imperviousness based on the convoluted images of GV and IS fractions. These indexes were proposed to show the contrast in the urban-rural biophysical environmental conditions. Results indicate that the UHI intensity possessed a stronger correlation with both greenness and imperviousness indexes than with GV and IS abundance. Because this study utilized a smoothing kernel to characterize the local variability of urban biophysical parameters, including LST, characterized UHIs can therefore be examined as a scale-dependent process. To this end, we categorized the smoothing parameters into three groups, corresponding to the three scales that are suitable to studying the urban thermal landscape at the microscale, mesoscale, and regional scale, respectively. The identified scales can then be matched with various applications in urban planning and environmental management.

Remote sensing estimation of impervious surface is significant in monitoring urban development and determining the overall environmental health of a watershed, and it has therefore attracted more interest recently in the remote sensing community. The main objective of this paper is to examine and compare the effectiveness of two advanced algorithms for estimating impervious surfaces from medium spatial resolution satellite images, namely, linear spectral mixture analysis (LSMA) and artificial neural network (ANN). Terra's Advanced Spaceborne Thermal Emission and Reflection Radiometer [(ASTER); acquired on June 16,2001] and a Landsat Enhanced Thematic Mapper Plus (ETM+) image (acquired on June 22, 2000) of Indianapolis, IN, were used for the analysis. The LSMA was employed to generate high- and low-albedo, vegetation, and soil fraction images (endmembers), and an image of impervious surfaces was then estimated by adding high- and low-albedo fraction images. Furthermore, an ANN model, specifically the multilayer-perceptron feedforward network with the back-propagation learning algorithm, was employed as a subpixel image classifier to estimate impervious surfaces. Accuracy assessment was performed against a high- resolution digital orthophoto. The results show that ANN was more effective than LSMA in generating impervious surfaces with high statistical accuracy. For the ASTER image, the root-mean-square error (RMSE) of the impervious surface map with the ANN model was 12.3%, and the one that resulted from LSMA was 13.2%. For the ETM+ image, the RMSE with the ANN model was 16.7%, and the one from LSMA was 18.9%. The better performance of ANN over LSMA is mainly attributable to the ANN'S capability of handling the nonlinear mixing of image spectrum. In order to test the seasonal sensitivity of satellite images for estimating impervious surfaces, LSMA was applied to two additional ASTER images of the same area, which are acquired on April 5, 2004, and October 3, 2000, respectively. The results were then compared with the ASTER image acquired in June in terms of RMSE. The June image had the highest accuracy, whereas the October image was better than the one in April. Plant phenology caused changes in the variance partitioning and impacted the mixing-space characterization, leading to a less accurate estimation of impervious surfaces.