Phenology from Landsat when data is scarce: Using MODIS and Dynamic Time-Warping to combine multi-year Landsat imagery to derive annual phenology curves

Green-leaf phenology describes the development of vegetation throughout a growing season and greatly affects the interaction between climate and the biosphere. Remote sensing is a valuable tool to characterize phenology over large areas but doing at fine- to medium resolution (e.g., with Landsat data) is difficult because of low numbers of cloud-free images in a single year. One way to overcome data availability limitations is to merge multi-year imagery into one time series, but this requires accounting for phenological differences among years. Here we present a new approach that employed a time series of a MODIS vegetation index data to quantify interannual differences in phenology, and Dynamic Time Warping (DTW) to re-align multi-year Landsat images to a common phenology that eliminates year-to-year phenological differences. This allowed us to estimate annual phenology curves from Landsat between 2002 and 2012 from which we extracted key phenological dates in a Monte-Carlo simulation design, including green-up (GU), start-of-season (SoS), maturity (Mat), senescence (Sen), end-of-season (EoS) and dormancy (Dorm). We tested our approach in eight locations across the United States that represented forests of different types and without signs of recent forest disturbance. We compared Landsat-based phenological transition dates to those derived from MODIS and ground-based camera data from the PhenoCam-network. The Landsat and MODIS comparison showed strong agreement. Dates of green-up, start-of-season and maturity were highly correlated (r 0.86-0.95), as were senescence and end-of-season dates (r > 0.85) and dormancy (r > 0.75). Agreement between the Landsat and PhenoCam was generally lower, but correlation coefficients still exceeded 0.8 for all dates. In addition, because of the high data density in the new Landsat time series, the confidence intervals of the estimated keydates were substantially lower than in case of MODIS and PhenoCam. Our study thus suggests that by exploiting multi-year Landsat imagery and calibrating it with MODIS data it is possible to describe green-leaf phenology at much finer spatial resolution than previously possible, highlighting the potential for fine scale phenology maps using the rich Landsat data archive over large areas.     

A Spatio-Temporal Enhancement Method for medium resolution LAI (STEM-LAI)

Satellite remote sensing has been used successfully to map leaf area index (LAI) across landscapes, but advances are still needed to exploit multi-scale data streams for producing LAI at both high spatial and temporal resolution. A multi-scale Spatio-Temporal Enhancement Method for medium resolution LAI (STEM-LAI) has been developed to generate 4-day time-series of Landsat-scale LAI from existing medium resolution LAI products. STEM-LAI has been designed to meet the demands of applications requiring frequent and spatially explicit information, such as effectively resolving rapidly evolving vegetation dynamics at sub-field (30 m) scales. In this study, STEM-LAI is applied to Moderate Resolution Imaging Spectroradiometer (MODIS) based LAI data and utilizes a reference-based regression tree approach for producing MODIS-consistent, but Landsat-based, LAI. The Spatial and Temporal Adaptive Reflectance Fusion Model (STARFM) is used to interpolate the downscaled LAI between Landsat acquisition dates, providing a high spatial and temporal resolution improvement over existing LAI products. STARFM predicts high resolution LAI by blending MODIS and Landsat based information from a common acquisition date, with MODIS data from a prediction date. To demonstrate its capacity to reproduce fine-scale spatial features observed in actual Landsat LAI, the STEM-LAI approach is tested over an agricultural region in Nebraska. The implementation of a 250 m resolution LAI product, derived from MODIS 1 km data and using a scale consistent approach based on the Normalized Difference Vegetation Index (NDVI), is found to significantly improve accuracies of spatial pattern prediction, with the coefficient of efficiency (E) ranging from 0.77–0.94 compared to 0.01–0.85 when using 1 km LAI inputs alone. Comparisons against an 11-year record of in-situ measured LAI over maize and soybean highlight the utility of STEM-LAI in reproducing observed LAI dynamics (both characterized by r² = 0.86) over a range of plant development stages. Overall, STEM-LAI represents an effective downscaling and temporal enhancement mechanism that predicts in-situ measured LAI better than estimates derived through linear interpolation between Landsat acquisitions. This is particularly true when the in-situ measurement date is greater than 10 days from the nearest Landsat acquisition, with prediction errors reduced by up to 50%. With a streamlined and completely automated processing interface, STEM-LAI represents a flexible tool for LAI disaggregation in space and time that is adaptable to different land cover types, landscape heterogeneities, and cloud cover conditions.     

Evaluating MODIS snow products for modelling snowmelt runoff: Case study of the Rio Grande headwaters

Snow-covered area (SCA) is a key variable in the Snowmelt-Runoff Model (SRM) and in other models for simulating discharge from snowmelt. Landsat Thematic Mapper (TM), Enhanced Thematic Mapper (ETM + ) or Operational Land Imager (OLI) provide remotely sensed data at an appropriate spatial resolution for mapping SCA in small headwater basins, but the temporal resolution of the data is low and may not always provide sufficient cloud-free dates. The coarser spatial resolution Moderate Resolution Imaging Spectroradiometer (MODIS) offers better temporal resolution and in cloudy years, MODIS data offer the best alternative for mapping snow cover when finer spatial resolution data are unavailable. However, MODIS’ coarse spatial resolution (500 m) can obscure fine spatial patterning in snow cover and some MODIS products are not sensitive to end-of-season snow cover. In this study, we aimed to test MODIS snow products for use in simulating snowmelt runoff from smaller headwater basins by a) comparing maps of TM and MODIS-based SCA and b) determining how SRM streamflow simulations are changed by the different estimates of seasonal snow depletion. We compared gridded MODIS snow products (Collection 5 MOD10A1 fractional and binary SCA; SCA derived from Collection 6 MOD10A1 Normalised Difference Snow Index (NDSI) Snow Cover), and the MODIS Snow Covered-Area and Grain size retrieval (MODSCAG) canopy-corrected fractional SCA (SCA_MG), with reference SCA maps (SCA_REF) generated from binary classification of TM imagery. SCA_MG showed strong agreement with SCA_REF; excluding true negatives (where both methods agreed no snow was present) the median percent difference between SCA_REF and SCA_MG ranged between −2.4% and 4.7%. We simulated runoff for each of the four study years using SRM populated with and calibrated for snow depletion curves derived from SCA_REF. We then substituted in each of the MODIS-derived depletion curves. With efficiency coefficients ranging between 0.73 and 0.93, SRM simulation results from the SCA_MG runs yielded the best results of all the MODIS products and only slightly underestimated discharge volume (between 7 and 11% of measured annual discharge). SRM simulations that used SCA derived from Collection 6 NDSI Snow Cover also yielded promising results, with efficiency coefficients ranging between 0.73 and 0.91.In conclusion, we recommend that when simulating snowmelt runoff from small basins (<4000 km²) with SRM, we recommend that users select either canopy-corrected MODSCAG or create their own site-specific products from the Collection 6 MOD10A1 NDSI.     

Testing estimation of water surface in Italian rice district from MODIS satellite data

Recent changes in rice crop management within Northern Italy rice district led to a reduction of seeding in flooding condition, which may have an impact on reservoir water management and on the animal and plant communities that depend on the flooded paddies. Therefore, monitoring and quantifying the spatial and temporal variability of water presence in paddy fields is becoming important. In this study we present a method to estimate dynamics of presence of standing water (i.e. fraction of flooded area) in rice fields using MODIS data. First, we produced high resolution water presence maps from Landsat by thresholding the Normalised Difference Flood Index (NDFI) made: we made it by comparing five Landsat 8 images with field-obtained information about rice field status and water presence. Using these data we developed an empirical model to estimate the flooding fraction of each MODIS cell. Finally we validated the MODIS-based flooding maps with both Landsat and ground information. Results showed a good predictability of water surface from Landsat (OA = 92%) and a robust usability of MODIS data to predict water fraction (R² = 0.73, EF = 0.57, RMSE = 0.13 at 1 × 1 km resolution). Analysis showed that the predictive ability of the model decreases with the greening up of rice, so we used NDVI to automatically discriminate estimations for inaccurate cells in order to provide the water maps with a reliability flag. Results demonstrate that it is possible to monitor water dynamics in rice paddies using moderate resolution multispectral satellite data. The achievement is a proof of concept for the analysis of MODIS archives to investigate irrigation dynamics in the last 15 years to retrieve information for ecological and hydrological studies.     

Intercomparison and validation of MODIS and GLASS leaf area index (LAI) products over mountain areas: A case study in southwestern China

The validation study of leaf area index (LAI) products over rugged surfaces not only gives additional insights into data quality of LAI products, but deepens understanding of uncertainties regarding land surface process models depended on LAI data over complex terrain. This study evaluated the performance of MODIS and GLASS LAI products using the intercomparison and direct validation methods over southwestern China. The spatio-temporal consistencies, such as the spatial distributions of LAI products and their statistical relationship as a function of topographic indices, time, and vegetation types, respectively, were investigated through intercomparison between MODIS and GLASS products during the period 2011–2013. The accuracies and change ranges of these two products were evaluated against available LAI reference maps over 10 sampling regions which standed for typical vegetation types and topographic gradients in southwestern China.The results show that GLASS LAI exhibits higher percentage of good quality data (i.e. successful retrievals) and smoother temporal profiles than MODIS LAI. The percentage of successful retrievals for MODIS and GLASS is vulnerable to topographic indices, especially to relief amplitude. Besides, the two products do not capture seasonal dynamics of crop, especially in spring over heterogeneously hilly regions. The yearly mean LAI differences between MODIS and GLASS are within ±0.5 for 64.70% of the total retrieval pixels over southwestern China. The spatial distribution of mean differences and temporal profiles of these two products are inclined to be dominated by vegetation types other than topographic indices. The spatial and temporal consistency of these two products is good over most area of grasses/cereal crops; however, it is poor for evergreen broadleaf forest. MODIS presents more reliable change range of LAI than GLASS through comparison with fine resolution reference maps over most of sampling regions. The accuracies of direct validation are obtained for GLASS LAI (r = 0.35, RMSE = 1.72, mean bias = −0.71) and MODIS LAI (r = 0.49, RMSE = 1.75, mean bias = −0.67). GLASS performs similarly to MODIS, but may be marginally inferior to MODIS based on our direct validation results. The validation experience demonstrates the necessity and importance of topographic consideration for LAI estimation over mountain areas. Considerable attention will be paid to the improvements of surface reflectance, retrieval algorithm and land cover types so as to enhance the quality of LAI products in topographically complex terrain.     

Spatially detailed retrievals of spring phenology from single-season high-resolution image time series

Vegetation indices derived from satellite image time series have been extensively used to estimate the timing of phenological events like season onset. Medium spatial resolution (≥250 m) satellite sensors with daily revisit capability are typically employed for this purpose. In recent years, phenology is being retrieved at higher resolution (≤30 m) in response to increasing availability of high-resolution satellite data. To overcome the reduced acquisition frequency of such data, previous attempts involved fusion between high- and medium-resolution data, or combinations of multi-year acquisitions in a single phenological reconstruction. The objectives of this study are to demonstrate that phenological parameters can now be retrieved from single-season high-resolution time series, and to compare these retrievals against those derived from multi-year high-resolution and single-season medium-resolution satellite data. The study focuses on the island of Schiermonnikoog, the Netherlands, which comprises a highly-dynamic saltmarsh, dune vegetation, and agricultural land. Combining NDVI series derived from atmospherically-corrected images from RapidEye (5 m-resolution) and the SPOT5 Take5 experiment (10m-resolution) acquired between March and August 2015, phenological parameters were estimated using a function fitting approach. We then compared results with phenology retrieved from four years of 30 m Landsat 8 OLI data, and single-year 100 m Proba-V and 250 m MODIS temporal composites of the same period. Retrieved phenological parameters from combined RapidEye/SPOT5 displayed spatially consistent results and a large spatial variability, providing complementary information to existing vegetation community maps. Retrievals that combined four years of Landsat observations into a single synthetic year were affected by the inclusion of years with warmer spring temperatures, whereas adjustment of the average phenology to 2015 observations was only feasible for a few pixels due to cloud cover around phenological transition dates. The Proba-V and MODIS phenology retrievals scaled poorly relative to their high-resolution equivalents, indicating that medium-resolution phenology retrievals need to be interpreted with care, particularly in landscapes with fine-scale land cover variability.     

Resolution vs. image quality in pre-tsunami imagery used for tsunami impact models in Aceh,Indonesia

Land cover roughness coefficients (LCRs) have been used in multivariate spatial models to test the mitigation potential of coastal vegetation to reduce impacts of the 2004 tsunami in Aceh, Indonesia. Previously, a Landsat 2002 satellite imagery was employed to derive land cover maps, which were then combined with vegetation characteristics, i.e., stand height, stem diameter and planting density to obtain LCRs. The present study tested LCRs extracted from 2003 and 2004 Landsat (30 m) images as well as a combination of 2003 and 2004 higher spatial resolution SPOT (10 m) imagery, while keeping the previous vegetation characteristics. Transects along the coast were used to extract land cover, whenever availability and visibility allowed. These new LCRs applied in previously developed tsunami impact models on wave outreach, casualties and damages confirmed previous findings regarding distance to the shoreline as a main factor reducing tsunami impacts. Nevertheless, the models using the new LCRs did not perform better than the original one. Particularly casualties models using 2002 LCRs performed better (δAIC > 2) than the more recent Landsat and SPOT counterparts. Cloud cover at image acquisition for Landsat and low area coverage for SPOT images decreased statistical predictive power (fewer observations). Due to the large spatial heterogeneity of tsunami characteristics as well as topographic and land-use features, it was more important to cover a larger area. Nevertheless, if more land cover classes would be referenced and high resolution imagery with low cloud cover would be available, the full benefits of higher spatial resolution imagery used to extract more precise land use roughness coefficients could be exploited.     

Forest cover trends from time series Landsat data for the Australian continent

In perennial and natural vegetation systems, monitoring changes in vegetation over time is of fundamental interest for identifying and quantifying impacts of management and natural processes. Subtle changes in vegetation cover can be identified by calculating the trends of a vegetation density index over time. In this paper, we apply such an index-trends approach, which has been developed and applied to time series Landsat imagery in rangeland and woodland environments, to continental-scale monitoring of disturbances within forested regions of Australia. This paper describes the operational methods used for the generation of National Forest Trend (NFT) information, which is a time-series summary providing visual indication of within-forest vegetation changes (disturbance and recovery) over time at 25 m resolution. This result is based on a national archive of calibrated Landsat TM/ETM+ data from 1989 to 2006 produced for Australia's National Carbon Accounting System (NCAS). The NCAS was designed in 1999 initially to provide consistent fine-scale classifications for monitoring forest cover extent and changes (i.e. land use change) over the Australian continent using time series Landsat imagery. NFT information identifies more subtle changes within forested areas and provides a capacity to identify processes affecting forests which are of primary interest to ecologists and land managers. The NFT product relies on the identification of an appropriate Landsat-based vegetation cover index (defined as a linear combination of spectral image bands) that is sensitive to changes in forest density. The time series of index values at a location, derived from calibrated imagery, represents a consistent surrogate to track density changes. To produce the trends summary information, statistical summaries of the index response over time (such as slope and quadratic curvature) are calculated. These calculated index responses of woody vegetation cover are then displayed as maps where the different colours indicate the approximate timing, direction (decline or increase), magnitude and spatial extent of the changes in vegetation cover. These trend images provide a self-contained and easily interpretable summary of vegetation change at scales that are relevant for natural resource management (NRM) and environmental reporting.     

Monitoring land surface albedo and vegetation dynamics using high spatial and temporal resolution synthetic time series from Landsat and the MODIS BRDF/NBAR/albedo product

Seasonal vegetation phenology can significantly alter surface albedo which in turn affects the global energy balance and the albedo warming/cooling feedbacks that impact climate change. To monitor and quantify the surface dynamics of heterogeneous landscapes, high temporal and spatial resolution synthetic time series of albedo and the enhanced vegetation index (EVI) were generated from the 500 m Moderate Resolution Imaging Spectroradiometer (MODIS) operational Collection V006 daily BRDF/NBAR/albedo products and 30 m Landsat 5 albedo and near-nadir reflectance data through the use of the Spatial and Temporal Adaptive Reflectance Fusion Model (STARFM). The traditional Landsat Albedo (Shuai et al., 2011) makes use of the MODIS BRDF/Albedo products (MCD43) by assigning appropriate BRDFs from coincident MODIS products to each Landsat image to generate a 30 m Landsat albedo product for that acquisition date. The available cloud free Landsat 5 albedos (due to clouds, generated every 16 days at best) were used in conjunction with the daily MODIS albedos to determine the appropriate 30 m albedos for the intervening daily time steps in this study. These enhanced daily 30 m spatial resolution synthetic time series were then used to track albedo and vegetation phenology dynamics over three Ameriflux tower sites (Harvard Forest in 2007, Santa Rita in 2011 and Walker Branch in 2005). These Ameriflux sites were chosen as they are all quite nearby new towers coming on line for the National Ecological Observatory Network (NEON), and thus represent locations which will be served by spatially paired albedo measures in the near future. The availability of data from the NEON towers will greatly expand the sources of tower albedometer data available for evaluation of satellite products. At these three Ameriflux tower sites the synthetic time series of broadband shortwave albedos were evaluated using the tower albedo measurements with a Root Mean Square Error (RMSE) less than 0.013 and a bias within the range of ±0.006. These synthetic time series provide much greater spatial detail than the 500 m gridded MODIS data, especially over more heterogeneous surfaces, which improves the efforts to characterize and monitor the spatial variation across species and communities. The mean of the difference between maximum and minimum synthetic time series of albedo within the MODIS pixels over a subset of satellite data of Harvard Forest (16 km by 14 km) was as high as 0.2 during the snow-covered period and reduced to around 0.1 during the snow-free period. Similarly, we have used STARFM to also couple MODIS Nadir BRDF Adjusted Reflectances (NBAR) values with Landsat 5 reflectances to generate daily synthetic times series of NBAR and thus Enhanced Vegetation Index (NBAR-EVI) at a 30 m resolution. While normally STARFM is used with directional reflectances, the use of the view angle corrected daily MODIS NBAR values will provide more consistent time series. These synthetic times series of EVI are shown to capture seasonal vegetation dynamics with finer spatial and temporal details, especially over heterogeneous land surfaces.     

An integrated remote sensing and GIS approach for monitoring areas affected by selective logging: A case study in northern Mato Grosso,Brazilian Amazon

Forest cover disturbances due to processes such as logging and forest fires are a widespread issue especially in the tropics, and have heavily affected forest biomass and functioning in the Brazilian Amazon in the past decades. Satellite remote sensing has played a key role for assessing logging activities in this region; however, there are still remaining challenges regarding the quantification and monitoring of these processes affecting forested lands. In this study, we propose a new method for monitoring areas affected by selective logging in one of the hotspots of Mato Grosso state in the Brazilian Amazon, based on a combination of object-based and pixel-based classification approaches applied on remote sensing data. Logging intensity and changes over time are assessed within grid cells of 300 m × 300 m spatial resolution. Our method encompassed three main steps: (1) mapping forest/non-forest areas through an object-based classification approach applied to a temporal series of Landsat images during the period 2000–2015, (2) mapping yearly logging activities from soil fraction images on the same Landsat data series, and (3) integrating information from previous steps within a regular grid-cell of 300 m × 300 m in order to monitor disturbance intensities over this 15-years period. The overall accuracy of the baseline forest/non-forest mask (year 2000) and of the undisturbed vs disturbed forest (for selected years) were 93% and 84% respectively. Our results indicate that annual forest disturbance rates, mainly due to logging activities, were higher than annual deforestation rates during the whole period of study. The deforested areas correspond to circa 25% of the areas affected by forest disturbances. Deforestation rates were highest from 2001 to 2005 and then decreased considerably after 2006. In contrast, the annual forest disturbance rates show high temporal variability with a slow decrease over the 15-year period, resulting in a significant increase of the ratio between disturbed and deforested areas. Although the majority of the areas, which have been affected by selective logging during the period 2000–2014, were not deforested by 2015, more than 70% of the deforested areas in 2015 had been at least once identified as disturbed forest during that period.     

Global, 30-m resolution continuous fields of tree cover: Landsat-based rescaling of MODIS vegetation continuous fields with lidar-based estimates of error

Abstract

We developed a global, 30-m resolution dataset of percent tree cover by rescaling the 250-m MOderate-resolution Imaging Spectroradiometer (MODIS) Vegetation Continuous Fields (VCF) Tree Cover layer using circa- 2000 and 2005 Landsat images, incorporating the MODIS Cropland Layer to improve accuracy in agricultural areas. Resulting Landsat-based estimates maintained consistency with the MODIS VCF in both epochs (RMSE =8.6% in 2000 and 11.9% in 2005), but showed improved accuracy in agricultural areas and increased discrimination of small forest patches. Against lidar measurements, the Landsat-based estimates exhibited accuracy slightly less than that of the MODIS VCF (RMSE=16.8% for MODIS-based vs. 17.4% for Landsat-based estimates), but RMSE of Landsat estimates was 3.3 percentage points lower than that of the MODIS data in an agricultural region. The Landsat data retained the saturation artifact of the MODIS VCF at greater than or equal to 80% tree cover but showed greater potential for removal of errors through calibration to lidar, with post-calibration RMSE of 9.4% compared to 13.5% in MODIS estimates. Provided for free download at the Global Land Cover Facility (GLCF) website (www.landcover.org), the 30-m resolution GLCF tree cover dataset is the highest-resolution multi-temporal depiction of Earth's tree cover available to the Earth science community.     

Satellite-based automated burned area detection: A performance assessment of the MODIS MCD45A1 in the Brazilian savanna

Burnings, which cause major changes to the environment, can be effectively monitored via satellite data, regarding both the identification of active fires and the estimation of burned areas. Among the many orbital sensors suitable for mapping burned areas on global and regional scales, the moderate resolution imaging spectroradiometer (MODIS), on board the Terra and Aqua platforms, has been the most widely utilized. In this study, the performance of the MODIS MCD45A1 burned area product was thoroughly evaluated in the Brazilian savanna, the second largest biome in South America and a global biodiversity hotspot, characterized by a conspicuous climatic seasonality and the systematic occurrence of natural and anthropogenic fires. Overall, September MCD45A1 polygons (2000–2012) compared well to the Landsat-based reference mapping (r² = 0.92) and were closely accompanied, on a monthly basis, by MOD14 and MYD14 hotspots (r² = 0.89), although large omissions errors, linked to landscape patterns, structures, and overall conditions depicted in each reference image, were observed. In spite of its spatial and temporal limitations, the MCD45A1 product proved instrumental for mapping and understanding fire behavior and impacts on the Cerrado landscapes.     

Landsat 8 vs. Landsat 5: A comparison based on urban and peri-urban land cover mapping

An image dataset from the Landsat OLI spaceborne sensor is compared with the Landsat TM in order to evaluate the excellence of the new imagery in urban landcover classification. Widely known pixel-based and object-based image analysis methods have been implemented in this work like Maximum Likelihood, Support Vector Machine, k-Nearest Neighbor, Feature Analyst and Sub-pixel. Classification results from Landsat OLI provide more accurate results comparing to the Landsat TM. Object-based classifications produced a more uniform result, but suffer from the absorption of small rare classes into large homogenous areas, as a consequence of the segmentation, merging and the spatial parameters in the spatial resolution (30 m) of Landsat images. Based exclusively on the overall accuracy reports, the SVM pixel-based classification from Landsat 8 proved to be the most accurate for the purpose of mapping urban land cover, using medium spatial resolution imagery.     

Species’ habitat use inferred from environmental variables at multiple scales: How much we gain from high-resolution vegetation data?

Spatial resolution of environmental data may influence the results of habitat selection models. As high-resolution data are usually expensive, an assessment of their contribution to the reliability of habitat models is of interest for both researchers and managers. We evaluated how vegetation cover datasets of different spatial resolutions influence the inferences and predictive power of multi-scale habitat selection models for the endangered brown bear populations in the Cantabrian Range (NW Spain). We quantified the relative performance of three types of datasets: (i) coarse resolution data from Corine Land Cover (minimum mapping unit of 25 ha), (ii) medium resolution data from the Forest Map of Spain (minimum mapping unit of 2.25 ha and information on forest canopy cover and tree species present in each polygon), and (iii) high-resolution Lidar data (about 0.5 points/m²) providing a much finer information on forest canopy cover and height. Despite all the models performed well (AUC > 0.80), the predictive ability of multi-scale models significantly increased with spatial resolution, particularly when other predictors of habitat suitability (e.g. human pressure) were not used to indirectly filter out areas with a more degraded vegetation cover. The addition of fine grain information on forest structure (LiDAR) led to a better understanding of landscape use and a more accurate spatial representation of habitat suitability, even for a species with large spatial requirements as the brown bear, which will result in the development of more effective measures to assist endangered species conservation.     

A spatial–temporal Hopfield neural network approach for super-resolution land cover mapping with multi-temporal different resolution remotely sensed images

The mixed pixel problem affects the extraction of land cover information from remotely sensed images. Super-resolution mapping (SRM) can produce land cover maps with a finer spatial resolution than the remotely sensed images, and reduce the mixed pixel problem to some extent. Traditional SRMs solely adopt a single coarse-resolution image as input. Uncertainty always exists in resultant fine-resolution land cover maps, due to the lack of information about detailed land cover spatial patterns. The development of remote sensing technology has enabled the storage of a great amount of fine spatial resolution remotely sensed images. These data can provide fine-resolution land cover spatial information and are promising in reducing the SRM uncertainty. This paper presents a spatial–temporal Hopfield neural network (STHNN) based SRM, by employing both a current coarse-resolution image and a previous fine-resolution land cover map as input. STHNN considers the spatial information, as well as the temporal information of sub-pixel pairs by distinguishing the unchanged, decreased and increased land cover fractions in each coarse-resolution pixel, and uses different rules in labeling these sub-pixels. The proposed STHNN method was tested using synthetic images with different class fraction errors and real Landsat images, by comparing with pixel-based classification method and several popular SRM methods including pixel-swapping algorithm, Hopfield neural network based method and sub-pixel land cover change mapping method. Results show that STHNN outperforms pixel-based classification method, pixel-swapping algorithm and Hopfield neural network based model in most cases. The weight parameters of different STHNN spatial constraints, temporal constraints and fraction constraint have important functions in the STHNN performance. The heterogeneity degree of the previous map and the fraction images errors affect the STHNN accuracy, and can be served as guidances of selecting the optimal STHNN weight parameters.     

Detection and mapping the spatial distribution of bracken fern weeds using the Landsat 8 OLI new generation sensor

Bracken fern is an invasive plant that presents serious environmental, ecological and economic problems around the world. An understanding of the spatial distribution of bracken fern weeds is therefore essential for providing appropriate management strategies at both local and regional scales. The aim of this study was to assess the utility of the freely available medium resolution Landsat 8 OLI sensor in the detection and mapping of bracken fern at the Cathedral Peak, South Africa. To achieve this objective, the results obtained from Landsat 8 OLI were compared with those derived using the costly, high spatial resolution WorldView-2 imagery. Since previous studies have already successfully mapped bracken fern using high spatial resolution WorldView-2 image, the comparison was done to investigate the magnitude of difference in accuracy between the two sensors in relation to their acquisition costs. To evaluate the performance of Landsat 8 OLI in discriminating bracken fern compared to that of Worldview-2, we tested the utility of (i) spectral bands; (ii) derived vegetation indices as well as (iii) the combination of spectral bands and vegetation indices based on discriminant analysis classification algorithm. After resampling the training and testing data and reclassifying several times (n = 100) based on the combined data sets, the overall accuracies for both Landsat 8 and WorldView-2 were tested for significant differences based on Mann-Whitney U test. The results showed that the integration of the spectral bands and derived vegetation indices yielded the best overall classification accuracy (80.08% and 87.80% for Landsat 8 OLI and WorldView-2 respectively). Additionally, the use of derived vegetation indices as a standalone data set produced the weakest overall accuracy results of 62.14% and 82.11% for both the Landsat 8 OLI and WorldView-2 images. There were significant differences {U (100) = 569.5, z = −10.8242, p < 0.01} between the classification accuracies derived based on Landsat OLI 8 and those derived using WorldView-2 sensor. Although there were significant differences between Landsat and WorldView-2 accuracies, the magnitude of variation (9%) between the two sensors was within an acceptable range. Therefore, the findings of this study demonstrated that the recently launched Landsat 8 OLI multispectral sensor provides valuable information that could aid in the long term continuous monitoring and formulation of effective bracken fern management with acceptable accuracies that are comparable to those obtained from the high resolution WorldView-2 commercial sensor.     

Mapping seagrass coverage and spatial patterns with high spatial resolution IKONOS imagery

Seagrass habitats in subtidal coastal waters provide a variety of ecosystem functions and services and there is an increasing need to acquire information on spatial and temporal dynamics of this resource. Here, we explored the capability of IKONOS (IKO) data of high resolution (4 m) for mapping seagrass cover [submerged aquatic vegetation (%SAV) cover] along the mid-western coast of Florida, USA. We also compared seagrass maps produced with IKO data with that obtained using the Landsat TM sensor with lower resolution (30 m). Both IKO and TM data, collected in October 2009, were preprocessed to calculate water depth invariant bands to normalize the effect of varying depth on bottom spectra recorded by the two satellite sensors and further the textural information was extracted from IKO data. Our results demonstrate that the high resolution IKO sensor produced a higher accuracy than the TM sensor in a three-class % SAV cover classification. Of note is that the OA of %SAV cover mapping at our study area created with IKO data was 5–20% higher than that from other studies published. We also examined the spatial distribution of seagrass over a spatial range of 4–240 m using the Ripley’s K function [L(d)] and IKO data that represented four different grain sizes [4 m (one IKO pixel), 8 m (2 × 2 IKO pixels), 12 m (3 × 3 IKO pixels), and 16 m (4 × 4 IKO pixels)] from moderate-dense seagrass cover along a set of six transects. The Ripley’s K metric repeatedly indicated that seagrass cover representing 4 m × 4 m pixels displayed a dispersed (or slightly dispersed) pattern over distances of <4–8 m, and a random or slightly clustered pattern of cover over 9–240 m. The spatial pattern of seagrass cover created with the three additional grain sizes (i.e., 2 × 24 m IKO pixels, 3 × 34 m IKO pixels, and 4 × 4 m IKO pixels) show a dispersed (or slightly dispersed) pattern across 4–32 m and a random or slightly clustered pattern across 33–240 m. Given the first report on using satellite observations to quantify seagrass spatial patterns at a spatial scale from 4 m to 240 m, our novel analyses of moderate-dense SAV cover utilizing Ripley’s K function illustrate how data obtained from the IKO sensor revealed seagrass spatial information that would be undetected by the TM sensor with a 30 m pixel size. Use of the seagrass classification scheme here, along with data from the IKO sensor with enhanced resolution, offers an opportunity to synoptically record seagrass cover dynamics at both small and large spatial scales.     

Improving the mapping of crop types in the Midwestern U.S. by fusing Landsat and MODIS satellite data

Mapping crop types is of great importance for assessing agricultural production, land-use patterns, and the environmental effects of agriculture. Indeed, both radiometric and spatial resolution of Landsat’s sensors images are optimized for cropland monitoring. However, accurate mapping of crop types requires frequent cloud-free images during the growing season, which are often not available, and this raises the question of whether Landsat data can be combined with data from other satellites. Here, our goal is to evaluate to what degree fusing Landsat with MODIS Nadir Bidirectional Reflectance Distribution Function (BRDF)-Adjusted Reflectance (NBAR) data can improve crop-type classification. Choosing either one or two images from all cloud-free Landsat observations available for the Arlington Agricultural Research Station area in Wisconsin from 2010 to 2014, we generated 87 combinations of images, and used each combination as input into the Spatial and Temporal Adaptive Reflectance Fusion Model (STARFM) algorithm to predict Landsat-like images at the nominal dates of each 8-day MODIS NBAR product. Both the original Landsat and STARFM-predicted images were then classified with a support vector machine (SVM), and we compared the classification errors of three scenarios: 1) classifying the one or two original Landsat images of each combination only, 2) classifying the one or two original Landsat images plus all STARFM-predicted images, and 3) classifying the one or two original Landsat images together with STARFM-predicted images for key dates. Our results indicated that using two Landsat images as the input of STARFM did not significantly improve the STARFM predictions compared to using only one, and predictions using Landsat images between July and August as input were most accurate. Including all STARFM-predicted images together with the Landsat images significantly increased average classification error by 4% points (from 21% to 25%) compared to using only Landsat images. However, incorporating only STARFM-predicted images for key dates decreased average classification error by 2% points (from 21% to 19%) compared to using only Landsat images. In particular, if only a single Landsat image was available, adding STARFM predictions for key dates significantly decreased the average classification error by 4 percentage points from 30% to 26% (p < 0.05). We conclude that adding STARFM-predicted images can be effective for improving crop-type classification when only limited Landsat observations are available, but carefully selecting images from a full set of STARFM predictions is crucial. We developed an approach to identify the optimal subsets of all STARFM predictions, which gives an alternative method of feature selection for future research.     

Forest cover classification using Landsat ETM+ data and time series MODIS NDVI data

Forest cover plays a key role in climate change by influencing the carbon stocks, the hydrological cycle and the energy balance. Forest cover information can be determined from fine-resolution data, such as Landsat Enhanced Thematic Mapper Plus (ETM+). However, forest cover classification with fine-resolution data usually uses only one temporal data because successive data acquirement is difficult. It may achieve mis-classification result without involving vegetation growth information, because different vegetation types may have the similar spectral features in the fine-resolution data. To overcome these issues, a forest cover classification method using Landsat ETM+ data appending with time series Moderate-resolution Imaging Spectroradiometer (MODIS) Normalized Difference Vegetation Index (NDVI) data was proposed. The objective was to investigate the potential of temporal features extracted from coarse-resolution time series vegetation index data on improving the forest cover classification accuracy using fine-resolution remote sensing data. This method firstly fused Landsat ETM+ NDVI and MODIS NDVI data to obtain time series fine-resolution NDVI data, and then the temporal features were extracted from the fused NDVI data. Finally, temporal features combined with Landsat ETM+ spectral data was used to improve forest cover classification accuracy using supervised classifier. The study in North China region confirmed that time series NDVI features had significant effects on improving forest cover classification accuracy of fine resolution remote sensing data. The NDVI features extracted from time series fused NDVI data could improve the overall classification accuracy approximately 5% from 88.99% to 93.88% compared to only using single Landsat ETM+ data.     

Assessing the accuracy of hyperspectral and multispectral satellite imagery for categorical and Quantitative mapping of salinity stress in sugarcane fields

This study evaluates the feasibility of hyperspectral and multispectral satellite imagery for categorical and quantitative mapping of salinity stress in sugarcane fields located in the southwest of Iran. For this purpose a Hyperion image acquired on September 2, 2010 and a Landsat7 ETM+ image acquired on September 7, 2010 were used as hyperspectral and multispectral satellite imagery. Field data including soil salinity in the sugarcane root zone was collected at 191 locations in 25 fields during September 2010. In the first section of the paper, based on the yield potential of sugarcane as influenced by different soil salinity levels provided by FAO, soil salinity was classified into three classes, low salinity (1.7–3.4 dS/m), moderate salinity (3.5–5.9 dS/m) and high salinity (6–9.5) by applying different classification methods including Support Vector Machine (SVM), Spectral Angle Mapper (SAM), Minimum Distance (MD) and Maximum Likelihood (ML) on Hyperion and Landsat images. In the second part of the paper the performance of nine vegetation indices (eight indices from literature and a new developed index in this study) extracted from Hyperion and Landsat data was evaluated for quantitative mapping of salinity stress. The experimental results indicated that for categorical classification of salinity stress, Landsat data resulted in a higher overall accuracy (OA) and Kappa coefficient (KC) than Hyperion, of which the MD classifier using all bands or PCA (1–5) as an input performed best with an overall accuracy and kappa coefficient of 84.84% and 0.77 respectively. Vice versa for the quantitative estimation of salinity stress, Hyperion outperformed Landsat. In this case, the salinity and water stress index (SWSI) has the best prediction of salinity stress with an R² of 0.68 and RMSE of 1.15 dS/m for Hyperion followed by Landsat data with an R² and RMSE of 0.56 and 1.75 dS/m respectively. It was concluded that categorical mapping of salinity stress is the best option for monitoring agricultural fields and for this purpose Landsat data are most suitable.