Spatial And Temporal Adaptive Reflectance Fusion Model Algorithm Research Articles

Quantitative Evaluation of Grassland SOS Estimation Accuracy Based on Different MODIS-Landsat Spatio-Temporal Fusion Datasets

Estimating the Start of Growing Season (SOS) of grassland on the global scale is an important scientific issue since it can reflect the response of the terrestrial ecosystem to environmental changes and determine the start time of grazing. However, most remote sensing data has coarse- temporal and spatial resolution, resulting in low accuracy of SOS retrieval based on remote sensing methods. In recent years, much research has focused on multi-source data fusion technology to improve the spatio-temporal resolution of remote sensing information, and to provide a feasible path for high-accuracy remote sensing inversion of SOS. Nevertheless, there is still a lack of quantitative evaluation for the accuracy of these data fusion methods in SOS estimation. Therefore, in this study, the SOS estimation accuracy is quantitatively evaluated based on the spatio-temporal fusion daily datasets through the Spatial and Temporal Adaptive Reflectance Fusion Model (STARFM) and other models in Xilinhot City, Inner Mongolia, China. The results show that: (1) the accuracy of SOS estimation based on spatio-temporal fusion daily datasets has been slightly improved, the average Root Mean Square Error (RMSE) of SOS based on 8d composite datasets is 11.1d, and the best is 9.7d (fstarfm8); (2) the estimation accuracy based on 8d composite datasets (RMSE¯ = 11.1d) is better than daily fusion datasets (RMSE¯ = 18.2d); (3) the lack of the Landsat data during the SOS would decrease the quality of the fusion datasets, which ultimately reduces the accuracy of the SOS estimation. The RMSE¯ of SOS based on all three models increases by 11.1d, and the STARFM is least affected, just increases 2.7d. The results highlight the potential of the spatio-temporal data fusion method in high-accuracy grassland SOS estimation. It also shows that the dataset fused by the STARFM algorithm and composed for 8 days is better for SOS estimation.

Daily Landsat-scale evapotranspiration estimation over a forested landscape in North Carolina, USA, using multi-satellite data fusion

Abstract. As a primary flux in the global water cycle, evapotranspiration (ET) connects hydrologic and biological processes and is directly affected by water and land management, land use change and climate variability. Satellite remote sensing provides an effective means for diagnosing ET patterns over heterogeneous landscapes; however, limitations on the spatial and temporal resolution of satellite data, combined with the effects of cloud contamination, constrain the amount of detail that a single satellite can provide. In this study, we describe an application of a multi-sensor ET data fusion system over a mixed forested/agricultural landscape in North Carolina, USA, during the growing season of 2013. The fusion system ingests ET estimates from the Two-Source Energy Balance Model (TSEB) applied to thermal infrared remote sensing retrievals of land surface temperature from multiple satellite platforms: hourly geostationary satellite data at 4 km resolution, daily 1 km imagery from the Moderate Resolution Imaging Spectroradiometer (MODIS) and biweekly Landsat thermal data sharpened to 30 m. These multiple ET data streams are combined using the Spatial and Temporal Adaptive Reflectance Fusion Model (STARFM) to estimate daily ET at 30 m resolution to investigate seasonal water use behavior at the level of individual forest stands and land cover patches. A new method, also exploiting the STARFM algorithm, is used to fill gaps in the Landsat ET retrievals due to cloud cover and/or the scan-line corrector (SLC) failure on Landsat 7. The retrieved daily ET time series agree well with observations at two AmeriFlux eddy covariance flux tower sites in a managed pine plantation within the modeling domain: US-NC2 located in a mid-rotation (20-year-old) loblolly pine stand and US-NC3 located in a recently clear-cut and replanted field site. Root mean square errors (RMSEs) for NC2 and NC3 were 0.99 and 1.02 mm day−1, respectively, with mean absolute errors of approximately 29 % at the daily time step, 12 % at the monthly time step and 0.7 % over the full study period at the two flux tower sites. Analyses of water use patterns over the plantation indicate increasing seasonal ET with stand age for young to mid-rotation stands up to 20 years, but little dependence on age for older stands. An accounting of consumptive water use by major land cover classes representative of the modeling domain is presented, as well as relative partitioning of ET between evaporation (E) and transpiration (T) components obtained with the TSEB. The study provides new insights about the effects of management and land use change on water yield over forested landscapes.

Mapping of heavy metal pollution in river water at daily time-scale using spatio-temporal fusion of MODIS-aqua and Landsat satellite imageries

For river water quality monitoring at 30m × 1-day spatio-temporal scales, a spatial and temporal adaptive reflectance fusion model (STARFM) is developed for estimating turbidity (Tu), total suspended solid (TSS), and six heavy metals (HV) of iron, zinc, copper, chromium, lead and cadmium, by blending the Moderate-Resolution Imaging Spectroradiometer (MODIS) and Landsat (Ls) spectral bands. A combination of regression analysis and genetic algorithm (GA) techniques are applied to develop spectral relationships between Tu-Ls, TSS-Tu, and each HV-TSS. The STARFM algorithm and all the developed relationship models are evaluated satisfactorily by various performance evaluation measures to develop heavy metal pollution index-based vulnerability maps at 1-km resolution in the Brahmani River in eastern India. The Monte-Carlo simulation based analysis of the developed formulations reveals that the uncertainty in estimating Zn and Cd is the minimum (1.04%) and the maximum (5.05%), respectively. Hence, the remote sensing based approach developed herein can effectively be used in many world rivers for real-time monitoring of heavy metal pollution.

Comparative Analysis on Two Schemes for Synthesizing the High Temporal Landsat-like NDVI Dataset Based on the STARFM Algorithm

The NDVI dataset with high temporal and spatial resolution (HTSN) is significant for extracting information about the phenological change of vegetation in regions with a complex earth surface. The Spatial and Temporal Adaptive Reflectance Fusion Model (STARFM) has been successfully applied to synthesize the HTSN by fusing the data with different characteristics. Based on the model, there are two different schemes for synthesizing the HTSN. One scheme is that red reflectance and near-infrared (NIR) reflectance are synthesized, respectively, and the HTSN is then obtained through algebraic operation (Scheme 1); the other scheme is that the red and NIR reflectance are used to calculate NDVI, which is directly taken as input data to synthesize the HTSN (Scheme 2). In this paper, taking the hill areas in eastern Sichuan China as a case, the two schemes were compared with each other. Seven Landsat images and time-series MOD13Q1 datasets spanning from October 2001 to February 2003 were used as the test data. The results showed the prediction accuracies of both derived HTSNs by the two different schemes were generally in good agreement, and Scheme 2 was slightly superior to Scheme 1 (R2: 0.14 < Scheme 1 < 0.53; 0.15 < Scheme 2 < 0.53). Although the two HTSNs showed high temporal and spatial consistence, the small spatiotemporal difference between them had a different influence on different applications. The coincidence rate of cropping intensity extracted from two derived HTSNs was fairly high, reaching up to 93.86%, while the coincidence rate of crop peak dates (i.e., the emerging dates of peaks in an annual time-series NDVI curve) was only 70.95%. Therefore, it is deemed that Scheme 2 can replace Scheme 1 in the application of extracting cropping intensity, so that more calculation time and memory space can be saved. For extracting more quantitative crop phenological information like crop peak dates, more tests are still needed in order to compare the absolute accuracy for both schemes.

Performance and effects of land cover type on synthetic surface reflectance data and NDVI estimates for assessment and monitoring of semi-arid rangeland

Federal land management agencies provide stewardship over much of the rangelands in the arid and semi-arid western United States, but they often lack data of the proper spatiotemporal resolution and extent needed to assess range conditions and monitor trends. Recent advances in the blending of complementary, remotely sensed data could provide public lands managers with the needed information. We applied the Spatial and Temporal Adaptive Reflectance Fusion Model (STARFM) to five Landsat TM and concurrent Terra MODIS scenes, and used pixel-based regression and difference image analyses to evaluate the quality of synthetic reflectance and NDVI products associated with semi-arid rangeland. Predicted red reflectance data consistently demonstrated higher accuracy, less bias, and stronger correlation with observed data than did analogous near-infrared (NIR) data. The accuracy of both bands tended to decline as the lag between base and prediction dates increased; however, mean absolute errors (MAE) were typically ≤10%. The quality of area-wide NDVI estimates was less consistent than either spectral band, although the MAE of estimates predicted using early season base pairs were ≤10% throughout the growing season. Correlation between known and predicted NDVI values and agreement with the 1:1 regression line tended to decline as the prediction lag increased. Further analyses of NDVI predictions, based on a 22 June base pair and stratified by land cover/land use (LCLU), revealed accurate estimates through the growing season; however, inter-class performance varied. This work demonstrates the successful application of the STARFM algorithm to semi-arid rangeland; however, we encourage evaluation of STARFM’s performance on a per product basis, stratified by LCLU, with attention given to the influence of base pair selection and the impact of the time lag.

Dryland vegetation phenology across an elevation gradient in Arizona, USA, investigated with fused MODIS and Landsat data

The patchy and heterogeneous arrangement of vegetation in dryland areas complicates the extraction of phenological signals using existing remote sensing data. This study examined whether the phenological analysis of a range of dryland land cover classes would benefit from the availability of synthetic images at Landsat spatial resolution and MODIS time intervals. We assembled a series of 500m MODIS and Landsat-5 TM datasets from April to November, 2005–2009, over a study site in central Arizona that encompasses diverse dryland vegetation classes along an elevation gradient of 2000m. We applied the spatial and temporal adaptive reflectance fusion model (STARFM) to each MODIS image to create a time series of synthetic images at 30m resolution. We subjected a subset of the synthetic imagery to a pixel-based regression analysis with temporally coincident Landsat images to analyze the effect of the underlying vegetation class on the accuracy of the STARFM results. To evaluate the usefulness of the increased spatial resolution compared to a MODIS product, we analyzed the variability of the date of peak greenness values of all 30mpixels within unmixed MODIS pixels. Finally, we examined differences in the temporal distributions of peak greenness extracted from both the Enhanced Vegetation Index (EVI) and Normalized Difference Vegetation Index (NDVI) synthetic imagery time series. Our results indicate that characteristics of the vegetation classes strongly influence STARFM algorithm performance, with Pearson correlation coefficient values ranging from 0.72 to 0.96 depending on the Landsat band and the land cover class. Responses in the near-infrared (NIR) spectrum yielded the lowest correlations, particularly for the Ponderosa Pine class. The phenological variability exhibited by each land cover class was dependent on the precipitation patterns of each growing season, but was sufficiently high to make the application of STARFM imagery at this scale uniformly beneficial. The peak greenness dates extracted from EVI and NDVI time series were temporally synchronized for the Grassland class but diverged for the classes of mixed woody and herbaceous vegetation types.

Preparing Landsat Image Time Series (LITS) for Monitoring Changes in Vegetation Phenology in Queensland, Australia

Time series of images are required to extract and separate information on vegetation change due to phenological cycles, inter-annual climatic variability, and long-term trends. While images from the Landsat Thematic Mapper (TM) sensor have the spatial and spectral characteristics suited for mapping a range of vegetation structural and compositional properties, its 16-day revisit period combined with cloud cover problems and seasonally limited latitudinal range, limit the availability of images at intervals and durations suitable for time series analysis of vegetation in many parts of the world. Landsat Image Time Series (LITS) is defined here as a sequence of Landsat TM images with observations from every 16 days for a five-year period, commencing on July 2003, for a Eucalyptus woodland area in Queensland, Australia. Synthetic Landsat TM images were created using the Spatial and Temporal Adaptive Reflectance Fusion Model (STARFM) algorithm for all dates when images were either unavailable or too cloudy. This was done using cloud-free scenes and a MODIS Nadir BRDF Adjusted Reflectance (NBAR) product. The ability of the LITS to measure attributes of vegetation phenology was examined by: (1) assessing the accuracy of predicted image-derived Foliage Projective Cover (FPC) estimates using ground-measured values; and (2) comparing the LITS-generated normalized difference vegetation index (NDVI) and MODIS NDVI (MOD13Q1) time series. The predicted image-derived FPC products (value ranges from 0 to 100%) had an RMSE of 5.6. Comparison between vegetation phenology parameters estimated from LITS-generated NDVI and MODIS NDVI showed no significant difference in trend and less than 16 days (equal to the composite period of the MODIS data used) difference in key seasonal parameters, including start and end of season in most of the cases. In comparison to similar published work, this paper tested the STARFM algorithm in a new (broadleaf) forest environment and also demonstrated that the approach can be used to form a time series of Landsat TM images to study vegetation phenology over a number of years.

Long term data fusion for a dense time series analysis with MODIS and Landsat imagery in an Australian Savanna

The spatial resolution of Landsat imagery has proven to be well suited for the analysis of vegetation patterns and dynamics at regional scale; however, the low temporal frequency is often a limitation for the quantification of vegetation dynamics. The spatial and temporal adaptive reflectance fusion model (STARFM) combines moderate resolution imaging spectrometer (MODIS) and Landsat thematic mapper/enhanced thematic mapper plus (TM/ETM+) imagery to a high spatiotemporal resolution dataset. A time series of 333 STARFM images was generated between February 2000 and September 2007 (8-day interval) at Landsat spatial and spectral resolution for a 12×10 km heterogeneous test area within the North Queensland Savannas. Time series of observed Landsat and predicted STARFM images correlated high for each spectral band (0.89 to 0.99). The STARFM algorithm was tested in a regionalization study where sudden change events were analyzed for a pallustrine wetland. A MODIS subpixel analysis showed a very close relationship between STARFM normalized difference vegetation index (NDVI) data and MODIS NDVI data (root mean square error of 0.027). A phenological description of the major vegetation classes within the region revealed distinct differences and lag times within the ecosystem. The 2004 dry season NDVI minimum-map correlated highly with the validated 2004 foliage projective cover product (r2 = 0.92) from the Queensland Department of Environment and Resource Management.

Evaluation of Landsat and MODIS data fusion products for analysis of dryland forest phenology

Current satellite sensors provide data of insufficient spatial and temporal resolutions to fully characterize the patchy phenology patterns of dryland forests. The spatial and temporal adaptive reflectance fusion model (STARFM) is an algorithm that fuses Landsat 30 m data with MODIS 500 m data to produce synthetic imagery at Landsat spatial resolution and MODIS time steps. In this study, we evaluated the feasibility of using STARFM to produce synthetic imagery over a dryland vegetation study site for the purpose of tracking phenological changes. We assembled subsets of six Landsat-5 TM scenes and temporally-coincident MODIS datasets spanning the 2006 April–October growing season in central-northern Arizona, which is characterized by large tracts of dryland forest. To investigate the effects of temporal compositing, BRDF-adjustment, and base pair timing on the accuracy of the resulting synthetic imagery, we employed a range of MODIS 500 m surface reflectance datasets (daily, 8-day composite, and 16-day Nadir BRDF-Adjusted Reflectance (NBAR)) as well as initial Landsat/MODIS imagery pairs from opposite ends of the growing season. The STARFM algorithm was applied to each MODIS data series to produce up to twelve synthetic images corresponding to each Landsat image. We evaluated the accuracy of the synthetic images by comparing the reflectance values of a random sample of the vegetation pixels with the corresponding pixel values of the reference Landsat image on a band-by-band basis. Our results indicate that the NBAR imagery is the optimal dataset for use with Landsat-5 TM data in this area. The NBAR dataset consistently returned the lowest absolute difference values and the highest correlations. A comparison of landscape-scale maps of the timing and value of the peak NDVI derived from STARFM, Landsat, and MODIS (NBAR) time series across the full 2006 growing season shows the effect of the heightened spatial and temporal resolution offered by a STARFM-based dataset. This work demonstrates the feasibility of using the STARFM algorithm to assemble an imagery time series at Landsat spatial resolution and MODIS temporal resolution in vegetated dryland ecosystems.