A new algorithm for Continuous Change Detection and Classification (CCDC) of land cover using all available Landsat data is developed. It is capable of detecting many kinds of land cover change ...continuously as new images are collected and providing land cover maps for any given time. A two-step cloud, cloud shadow, and snow masking algorithm is used for eliminating “noisy” observations. A time series model that has components of seasonality, trend, and break estimates surface reflectance and brightness temperature. The time series model is updated dynamically with newly acquired observations. Due to the differences in spectral response for various kinds of land cover change, the CCDC algorithm uses a threshold derived from all seven Landsat bands. When the difference between observed and predicted images exceeds a threshold three consecutive times, a pixel is identified as land surface change. Land cover classification is done after change detection. Coefficients from the time series models and the Root Mean Square Error (RMSE) from model estimation are used as input to the Random Forest Classifier (RFC). We applied the CCDC algorithm to one Landsat scene in New England (WRS Path 12 and Row 31). All available (a total of 519) Landsat images acquired between 1982 and 2011 were used. A random stratified sample design was used for assessing the change detection accuracy, with 250pixels selected within areas of persistent land cover and 250pixels selected within areas of change identified by the CCDC algorithm. The accuracy assessment shows that CCDC results were accurate for detecting land surface change, with producer's accuracy of 98% and user's accuracies of 86% in the spatial domain and temporal accuracy of 80%. Land cover reference data were used as the basis for assessing the accuracy of the land cover classification. The land cover map with 16 categories resulting from the CCDC algorithm had an overall accuracy of 90%.
•A new algorithm for Continuous Change Detection and Classification of land cover•All available (a total of 519) Landsat images from Path 12 Row 31 were used.•It can detect many kinds of land cover change continuously.•It can provide land cover maps for any given time.•The results were accurate both in change detection and classification.
A new method called Fmask (Function of mask) for cloud and cloud shadow detection in Landsat imagery is provided. Landsat Top of Atmosphere (TOA) reflectance and Brightness Temperature (BT) are used ...as inputs. Fmask first uses rules based on cloud physical properties to separate Potential Cloud Pixels (PCPs) and clear-sky pixels. Next, a normalized temperature probability, spectral variability probability, and brightness probability are combined to produce a probability mask for clouds over land and water separately. Then, the PCPs and the cloud probability mask are used together to derive the potential cloud layer. The darkening effect of the cloud shadows in the Near Infrared (NIR) Band is used to generate a potential shadow layer by applying the flood-fill transformation. Subsequently, 3D cloud objects are determined via segmentation of the potential cloud layer and assumption of a constant temperature lapse rate within each cloud object. The view angle of the satellite sensor and the illuminating angle are used to predict possible cloud shadow locations and select the one that has the maximum similarity with the potential cloud shadow mask. If the scene has snow, a snow mask is also produced. For a globally distributed set of reference data, the average Fmask overall cloud accuracy is as high as 96.4%. The goal is development of a cloud and cloud shadow detection algorithm suitable for routine usage with Landsat images.
► A new method for automated cloud and cloud shadow detection in Landsat images. ► It is the result of combining past approaches and a new object-based approach. ► It is an improvement over the traditional ACCA cloud algorithm. ► The average Fmask cloud overall accuracy is 96.4%.
Identification of clouds, cloud shadows and snow in optical images is often a necessary step toward their use. Recently a new program (named Fmask) designed to accomplish these tasks was introduced ...for use with images from Landsats 4–7 (Zhu & Woodcock, 2012). In this paper, there are the following: (1) improvements in the Fmask algorithm for Landsats 4–7; (2) a new version for use with Landsat 8 that takes advantage of the new cirrus band; and (3) a prototype algorithm for Sentinel 2 images. Though Sentinel 2 images do not have a thermal band to help with cloud detection, the new cirrus band is found to be useful for detecting clouds, especially for thin cirrus clouds. By adding a new cirrus cloud probability and removing the steps that use the thermal band, the Sentinel 2 scenario achieves significantly better results than the Landsats 4–7 scenario for all 7 images tested. For Landsat 8, almost all the Fmask algorithm components are the same as for Landsats 4–7, except a new cirrus cloud probability is calculated using the new cirrus band, which improves detection of thin cirrus clouds. Landsat 8 results are better than the Sentinel 2 scenario, with 6 out of 7 test images showing higher accuracies.
•Cloud, cloud shadow, and snow detection for Landsats 4–8 and simulated Sentinel 2.•This algorithm improves the original Fmask algorithm for Landsats 4–7 images.•A new version developed for Landsat 8 that takes advantage of the new cirrus band.•A prototype algorithm designed for Sentinel 2 that does not have a thermal band.•For cloud detection, the cirrus band is more helpful than the thermal band.
We developed a new algorithm called Tmask (multiTemporal mask) for automated masking of cloud, cloud shadow, and snow for multitemporal Landsat images. This algorithm consists of two steps. The first ...step is based on a single-date algorithm called Fmask (Function of mask) that initially screens most of the clouds, cloud shadows, and snow. The second step benefits from the extra temporal information from the remaining “clear” pixels and further improves the cloud, cloud shadow, and snow mask. Three Top Of Atmosphere (TOA) reflectance bands (Bands 2, 4, and 5 — Landsat-7 band numbering) are used in a Robust Iteratively Reweighted Least Squares (RIRLS) method to estimate a time series model for each pixel. By comparing model estimates with Landsat observations for the three spectral bands, the Tmask algorithm is capable of detecting any remaining clouds, cloud shadows, and snow for an entire stack of Landsat images. Generally, this algorithm will not falsely identify land cover changes as clouds, cloud shadows, or snow, as it is capable of modeling land cover change. The multitemporal images also provide extra information for better discrimination of cloud and snow, which is difficult for single-date algorithm. A snow threshold is derived for Band 5 TOA reflectance for each pixel at each specific time based on a modified Norwegian Linear Reflectance-to-Snow-Cover (NLR) algorithm. By comparing the results of Tmask with a single-date algorithm (Fmask) for multitemporal Landsat images located at Path 12 Row 31, significant improvements are observed for identification of clouds, cloud shadows, and snow. The most significant improvement is observed for cloud shadow detection. Many of the errors in cloud, cloud shadow, and snow detection observed in Fmask are corrected by the Tmask algorithm. The goal is development of a cloud, cloud shadow, and snow detection algorithm that results in a multitemporal stack of images that is free of “noise” factors and thus suitable for detection of land cover change.
•A cloud, cloud shadow, and snow detection algorithm for multitemporal Landsat data•A modified Norwegian Linear Reflectance-to-Snow-Cover algorithm is proposed.•The results are more accurate than those of the single-date Fmask algorithm.•This algorithm is suitable for preparing Landsat data for change detection.
Tropical forest loss currently contributes 5 to 15% of anthropogenic carbon emissions to the atmosphere. The large uncertainty in emissions estimates is a consequence of many factors, including ...differences in definitions of forests and degradation, as well as estimation methodologies. However, a primary factor driving uncertainty is an inability to properly account for forest degradation. While remote sensing offers the only practical way of monitoring forest disturbances over large areas, and despite recent improvements in data quality and quantity and processing techniques, remote sensing approaches are still limited in their ability to detect forest degradation. In this paper, a system is presented that uses time series of Landsat data and spectral mixture analysis to detect both degradation and deforestation in forested landscapes. The Landsat data are transformed into spectral endmember fractions and are used to calculate the Normalized Degradation Fraction Index (NDFI; Souza et al., 2005). The spectrally unmixed data are used for disturbance monitoring and land cover classification via time series analysis. To assess the performance of the system, maps of deforestation and degradation were used to stratify the study area for collection of sample data to which unbiased estimators were applied to produce accuracy and area estimates of degradation and deforestation from 1990 to 2013. The approach extends previous research in spectral mixture analysis for identifying forest degradation to the temporal domain. The method was applied using the Google Earth Engine and tested in the Brazilian State of Rondônia. Degradation and deforestation were mapped with 88.0% and 93.3% User's Accuracy, and 68.1% and 85.3% Producer's Accuracy. Area estimates of degradation and deforestation were produced with margins of error of 13.9% and 5.3%, respectively, over the 24 year time period. These results indicate that for Rondônia a decreasing trend in deforestation after 2004 corresponds to an increase in degradation during the same time period.
•A methodology is presented for detecting forest degradation and deforestation.•The approach is based on spectral unmixing and time series analysis using Landsat.•Unbiased area estimates are calculated from 1990 to 2013 for Rondônia, Brazil.•High precision is achieved on the area estimates of the disturbance classes.•A decrease in deforestation in 2004 corresponded to an increase in degradation.
The REDD+ mechanism of UNFCCC was established to reduce greenhouse gases emissions by means of financial incentives. Of importance to the success of REDD+ and similar initiatives is the provision of ...credible evidence of reductions in the extent of land change activities that release carbon to the atmosphere (e.g. deforestation). The criteria for reporting land change areas and associated emissions within REDD+ stipulate the use of sampling-based approaches, which allow for unbiased estimation and uncertainty quantification. But for economic compensation for emission reductions to be feasible, agreements between participating countries and donors often require reporting every year or every second year. With the rates of land change typically being very small relative to the total study area, sampling-based approaches for estimation of annual or bi-annual areas have proven problematic, especially when comparing area estimates over time. In this paper, we present a methodology for monitoring and estimating areas of land change activity at high temporal resolution that is compliant with international guidelines. The methodology is based on a break detection algorithm applied to time series of Landsat data in the Colombian Amazon between 2001 and 2016. A biennial stratified sampling approach was implemented to (1) remove the bias introduced by the change detection and classification algorithm in mapped areas derived from pixel-counting; and (2) provide confidence intervals for area estimates obtained from the reference data collected for the sample. Our results show that estimating the area of land change, like deforestation, at annual or bi-annual resolution is inherently challenging and associated with high degrees of uncertainty. We found that better precision was achieved if independent sample datasets of reference observations were collected for each time interval for which area estimates are required. The alternative of selecting one sample of continuous reference observations analyzed for inference of area for each time interval did not yield area estimates significantly different from zero. Also, when large stable land covers (primary forest in this case, occupying almost 90% of the study area) are present in the study area in combination with small rates of land change activity, the impact of omission errors in the map used for stratifying the study area will be substantial and potentially detrimental to usefulness of land change studies. The introduction of a buffer stratum around areas of mapped land change reduced the uncertainty in area estimates by up to 98%. Results indicate that the Colombian Amazon has experienced a small but steady decrease in primary forest due to establishment of pastures, with forest-to-pasture conversion reaching 103 ± 30 kha (95% confidence interval) in the period between 2013 and 2015, corresponding to 0.22% of the study area. Around 29 ± 17 kha (95% CI) of pastureland that had been abandoned shortly after establishment reverted to secondary forest within the same period. Other gains of secondary forest from more permanent pastures averaged about 12 ± 11 kha (95% CI), while losses of secondary forest averaged 20 ± 12 kha (95% CI).
•Presented methods allow for monitoring activities and post-disturbance landscapes.•Deforestation, driven by conversion to pastures, is increasing at very small rate.•Less than a fifth of the area of deforestation was abandoned and left to regenerate.•Using a buffer stratum around change areas increased precision in area estimates.•Samples representing each period for which area estimates are desired were required.
Anthropogenic and natural forest disturbance cause ecological damage and carbon emissions. Forest disturbance in the Amazon occurs in the form of deforestation (conversion of forest to non‐forest ...land covers), degradation from the extraction of forest resources, and destruction from natural events. The crucial role of the Amazon rainforest in the hydrologic cycle has even led to the speculation of a disturbance “tipping point” leading to a collapse of the tropical ecosystem. Here we use time series analysis of Landsat data to map deforestation, degradation, and natural disturbance in the Amazon Ecoregion from 1995 to 2017. The map was used to stratify the study area for selection of sample units that were assigned reference labels based on their land cover and disturbance history. An unbiased statistical estimator was applied to the sample of reference observations to obtain estimates of area and uncertainty at biennial time intervals. We show that degradation and natural disturbance, largely during periods of severe drought, have affected as much of the forest area in the Amazon Ecoregion as deforestation from 1995 to 2017. Consequently, an estimated 17% (1,036,800 ± 24,800 km2, 95% confidence interval) of the original forest area has been disturbed as of 2017. Our results suggest that the area of disturbed forest in the Amazon is 44%–60% more than previously realized, indicating an unaccounted for source of carbon emissions and pervasive damage to forest ecosystems.
Using remote sensing analysis, we estimated the area affected by deforestation, forest degradation, and natural disturbance in the Amazon Ecoregion from 1995 to 2017. Our results show that degradation and natural disturbance have affected as much forest as deforestation during the study period, largely during periods of severe drought.
We developed an algorithm called Cmask (Cirrus cloud mask) for cirrus cloud detection in Landsat 8 imagery using time series of Cirrus Band (1.36–1.39 μm) observations. For each pixel, a harmonic ...model, which includes a water vapor regressor, based on all available Cirrus Band observations is estimated using the Robust Iteratively Reweighted Least Squares (RIRLS) regression approach, and pixels affected by cirrus cloud are identified by comparing model predictions and actual satellite observations of the Cirrus Band Top-Of-Atmosphere (TOA) reflectance. Furthermore, we analyzed the effect of increasing Cirrus Band TOA reflectance on the surface reflectance of the Blue, Green, Red, Near Infrared (NIR), and two Shortwave Infrared (SWIR) (SWIR 1 and SWIR 2) Bands based on a set of globally distributed random samples. The goal of this study is to answer the question of what are cirrus clouds in the context of a Landsat observation, or more specifically, when should we identify a pixel as cirrus cloud such that we know the reflectance in the other spectral bands has been seriously affected by cirrus clouds. The challenge is to then develop a simple and operational algorithm for accurate detection of cirrus clouds in Landsat 8 images. The Cmask algorithm reduced almost by half the errors found in the U.S. Geological Survey (USGS) Quality Assessment (QA) Band for distinguishing cirrus cloud and clear observations (8% versus 15% error).
•We analyzed cirrus cloud impacts on the surface reflectance of Landsat 8 data.•We developed a cirrus cloud detection algorithm called Cmask.•TOA reflectance from the Cirrus Band and water vapor data were used.•The combined use of Cmask and Fmask is recommended.
A new algorithm for generating synthetic Landsat images is developed based on all available Landsat data. This algorithm is capable of predicting Landsat surface reflectance for any desired date. It ...first excludes cloud, cloud shadow, and snow observations, and then uses the remaining clear observations to estimate time series models for each Landsat pixel. Three time series models (a simple model, advanced model, and full model) are used for estimating surface reflectance for each pixel, and the selection of a time series model is dependent on the number of clear observations available: the more clear observations, the more complex the model will be that is used. For each time series model there are three components (seasonality, trend, and breaks), that are used for modeling intra-annual and inter-annual differences and abrupt surface change. Abrupt surface changes are detected by differencing predicted and observed Landsat observations, and if the difference is larger than twice the Root Mean Square Error (RMSE) for six consecutive observations, it will be detected as a “break” in the time series model. The RMSE values are temporally adjusted to provide better threshold range. For each “synthetic” image, a Quality Assessment (QA) Band is provided that contains information on how the time series model was estimated and used for generating the synthetic data. We have applied this approach to six Landsat scenes within the United States. We visually compared the synthetic images with real Landsat images for different kinds of environments and they are similar for all image pairs. We also quantitatively assessed the accuracy of the synthetic data by calculating the RMSE value for all clear Landsat observations. The RMSE values for the three visible bands are the lowest (approximately 0.01), and the Short-wave Infrared (SWIR) bands are slightly higher in magnitude (between 0.01 and 0.02). The Near Infrared (NIR) band has the highest RMSE values (between 0.02 and 0.03). The goal of this paper is to provide Landsat images that are free of cloud, cloud shadow, snow, and Scan Line Corrector (SLC)-off gaps that can be used to derive land cover and bio-physical products.
•All available Landsat data are used for predicting surface reflectance.•Synthetic Landsat images can be generated for any given time.•Model selection, LASSO, and temporally-adjusted RMSE are used for better modeling.•Synthetic images are similar to real Landsat images for all image pairs.•The prediction accuracy is similar in magnitude to the noise levels in Landsat data.
Land use and land cover (LULC) change caused by human activities is a major source of anthropogenic carbon emissions and a driver of climate change. The Mekong Region is highly dynamic, experiencing ...extensive LULC change in recent decades. This study provides a spatially and temporally continuous estimate of LULC change for the Mekong River Basin for 2001–2019 using time series analysis of MODIS data coupled with a spatiotemporal carbon bookkeeping model to track carbon losses and recovery. The LULC change product has an overall accuracy of 74.4 ± 1.9% (82.1 ± 1.7% after consolidating tree-dominated classes), including an increase of 5.6% after combining with existing MODIS products (referred to as the M-CCDC process). Two of the largest components of LULC change in the region are the establishment of plantations and agricultural expansion, which were estimated to be 33,617 ± 7342 km2 and 14,915 ± 4682 km2 between 2003 and 2014. We found that 82% of the deforested area was converted to tree plantations. Among all the newly added plantations, 86% replaced natural forests and 12% replaced agricultural land. In addition, existing maps of annual tree canopy cover (TCC) were used to assess forest disturbances that do not result in LULC conversions. The M-CCDC results combined with the forest disturbances derived from TCC maps were coupled to a spatiotemporal carbon bookkeeping model to estimate carbon emissions and uptake. Carbon emissions were 72.9 ± 6.2 Tg C yr−1 during 2001–2017; emissions increase to 102.8 ± 8.6 Tg C yr−1 if including carbon not yet released to the atmosphere in the form of decomposing slash and wood products. Carbon uptake for the same period was −35.5 ± 4.9 Tg C yr−1, with carbon uptake from new plantations offsetting almost half of the emissions from deforestation in this area. Assessment of post-deforestation land use is crucial for quantifying the short- and longer- term carbon consequences of LULC change.
•A spatiotemporal assessment of LULC change and carbon fluxes for the Mekong Region•Combining time series- and annual composite-based approaches improved accuracy by 6%•The largest driver of LULC change in this area is conversion to new plantations•Carbon uptake from new plantations offsets almost half of the emissions in this area•Assessment of post-deforestation land use is crucial for quantifying carbon cycle.