Continuous Change Detection And Classification Research Articles

BATSCCD: a new change detection method for mapping building age in rapidly changing urban areas using Landsat time series data

ABSTRACT The building age is an important feature in the vulnerability assessment of disasters and building energy consumption modeling. It provides indispensable information for the study of built environment. However, obtaining high-precision maps of building ages is challenging, which impedes the development of relevant research and applications. In this study, a new building age mapping method based on Landsat time series classification and change detection (BATSCCD) was proposed, which integrates Landsat time series data with random forest and LandTrendr algorithm to reflect non-building to building changes, building reconstruction, and corresponding change characteristics. The proposed method achieved overall building age mapping accuracies of 80.45% within a 2-year tolerance and 87.66% within a 3-year tolerance in Guangzhou and accuracies of 78.08% and 87.52%, respectively, in Beijing. The improvement was notable, especially compared with the continuous change detection and classification (CCDC) method, with corresponding accuracy values of 36.22% and 46.31%, respectively, in Guangzhou and 49.7% and 60.19%, respectively, in Beijing. The results suggest that the BATSCCD method can be used to accurately estimate the building age and provide reliable mapping results. Thanks to the accessibility of required data and its high accuracy of building age estimation, the proposed method exhibits great application potential.

Broad-area-search of new construction using time series analysis of Landsat and Sentinel-2 data

New construction activities can alter surface albedo and structure, which then affect surface temperature and roughness, and hence have a significant impact on urban climate. Construction activity is also an important indicator of human development and movement and is of high interest to the intelligence community. A new approach for Broad-Area-Search of New Construction activities (BASC) by combining time series analysis and rule-based filters using Landsat data was developed and tested in five selected cities (Boston, Shanghai, São Paulo, Dubai, and Ho Chi Minh City). The algorithm transforms Landsat images into fractions of a set of four endmembers using Linear Spectral Mixture Analysis (LSMA) and then applies the Continuous Change Detection and Classification (CCDC) algorithm for change detection. A set of rule-based filters and spatial processing was then applied to narrow the search to changes related to construction activities. Overall, BASC reached a recall of 0.83, a precision of 0.58, and an F1-Score of 0.68. Among the five cities, Dubai had the highest recall of 1.0 and the highest F1-score of 0.75, while Boston had the highest precision of 0.63. BASC performed worst in Shanghai with an F1-Score of 0.6, mainly due to it having the lowest recall of 0.62, while São Paulo has the lowest precision of 0.5. Common sources of omission errors include low-density, redevelopment, and small sites, while common commission errors include roofing, land clearing, water level changes, and re-surfacing projects. For comparison, BASC using Sentinel-2 Top-of-Atmosphere (TOA) Reflectance data recorded an overall F1-Score of 0.63, but with higher recall and lower precision. Integration of Sentinel-2 Surface Reflectance and Sentinel-1 SAR data has the potential to further improve the performance of BASC. The new algorithm provided a method for routine monitoring of construction activities over large areas. The result of such monitoring can be used as a baseline to narrow down the candidate targets of construction activities, where very high-resolution imagery can then be requested to perform further examination.

Is the change deforestation? Using time-series analysis of satellite data to disentangle deforestation from other forest degradation causes

Protecting natural ecosystems requires monitoring approaches that work as early warning systems to avoid degradation and protect biodiversity. However, separating forest disturbance causes in change-detection pipelines is challenging due to the complex interplay of multiple drivers affecting vegetation. This study aims to detect deforestation in highly heterogeneous ecosystems. We used Landsat NDVI time-series data for testing three unsupervised change detection methods: 1) the non-parametric phenological anomaly detection (npphen), 2) the continuous change detection and classification (CCDC), and 3) the pruned exact linear time (PELT) algorithms. We used visual interpretation of Google Earth Pro high-resolution data (< 10 m) to depict deforestation, and natural-induced changes, like forest browning and fires, evaluating the performance of the unsupervised methods. Additionally, a Random Forest model trained with the outputs from detection algorithms together with elevation and radar vegetation index data were utilised to depict deforestation in a second step. While PELT slightly outperformed other methods for tracking general vegetation changes, with overall accuracies (OA) ranging from 0.78 to 0.99, depending on the vegetation type, it also showed the slowest deforestation tracking response. CCDC presented the fastest response and an OA between 0.78 and 0.95. Additionally, we observed a mean OA of only 0.47 when separating deforestation from other changes using only the unsupervised models. On the other hand, deforestation was accurately detected (OA = 0.93; kappa = 0.83) when using CCDC outputs within a secondary supervised classification, agreeing with selected citizen-based complaints from the Environmental Superintendence. The relatively fast response in deforestation tracking using CCDC makes it a viable alternative for near real-time monitoring. Commonly used unsupervised detection methods may be coupled with supervised techniques to depict vegetation change sources robustly. This application constitutes a step forward for managing and monitoring vegetation areas in highly complex and dynamic landscapes, like Mediterranean ecosystems.

SSEBop Evapotranspiration Estimates Using Synthetically Derived Landsat Data from the Continuous Change Detection and Classification Algorithm

The operational Simplified Surface Energy Balance (SSEBop) model has been utilized to generate gridded evapotranspiration data from Landsat images. These estimates are primarily driven by two sources of information: reference evapotranspiration and Landsat land surface temperature (LST) values. Hence, SSEBop is limited by the availability of Landsat data. Here, in this proof-of-concept paper, we utilize the Continuous Change Detection and Classification (CCDC) algorithm to generate synthetic Landsat data, which are then used as input for SSEBop to generate evapotranspiration estimates for six target areas in the continental United States, representing forests, shrublands, and irrigated agriculture. These synthetic land cover data are then used to generate the LST data required for SSEBop evapotranspiration estimates. The synthetic LST, evaporative fractions, and evapotranspiration data from CCDC closely mirror the phenological cycles in the observed Landsat data. Across the six sites, the median correlation in seasonal LST was 0.79, and the median correlation in seasonal evapotranspiration was 0.8. The median root mean squared error (RMSE) values were 2.82 °C for LST and 0.50 mm/day for actual evapotranspiration. CCDC predictions typically underestimate the average evapotranspiration by less than 1 mm/day. The average performance of the CCDC evaporative fractions, and corresponding evapotranspiration estimates, were much better than the initial LST estimates and, therefore, promising. Future work could include bias correction to improve CCDC’s ability to accurately reproduce synthetic Landsat data during the summer, allowing for more accurate evapotranspiration estimates, and determining the ability of SSEBop to predict regional evapotranspiration at seasonal timescales based on projected land cover change from CCDC.

Tracking the Effects of Mangrove Changes and Spartina alterniflora Invasion on Soil Carbon Storage: A Case Study of the Beibu Gulf of Guangxi, China

In order to clarify the long-term changes in mangroves in the Beibu Gulf of Guangxi and the carbon storage changes after the invasion of Spartina alterniflora (S. alterniflora) in the Dandou Sea area, the Continuous Change Detection and Classification (CCDC) algorithm combined with feature indices was first used to track the changes. Subsequently, the random forest algorithm was applied to classify each change segment, and then sampling was conducted based on the distribution of S. alterniflora in different invasion years. The results showed that the Kappa coefficient of the classification result of the latest change segment was 0.78. The rapid expansion of S. alterniflora, aquaculture pond construction, and land reclamation activities have led to changes in mangroves, causing a decrease in the area of the mangrove region. A total of 814.57 hectares of mangroves has been converted into other land-cover types, with most pixels undergoing one to two changes, and many of these changes were expected to continue until 2022. An analysis of the distribution characteristics and influencing factors of soil organic carbon (SOC) and soil organic carbon storage (SOCS) at different invasion stages revealed that SOC and SOCS were mainly influenced by soil bulk density, soil moisture content, and electrical conductivity. It was found that S. alterniflora had higher SOC content compared to the mudflats. With the increase in invasion years, S. alterniflora continuously increased the SOC and SOCS content in coastal wetlands.

Mapping of secondary forest age in China using stacked generalization and Landsat time series

A national distribution of secondary forest age (SFA) is essential for understanding the forest ecosystem and carbon stock in China. While past studies have mainly used various change detection algorithms to detect forest disturbance, which cannot adequately characterize the entire forest landscape. This study developed a data-driven approach for improving performances of the Vegetation Change Tracker (VCT) and Continuous Change Detection and Classification (CCDC) algorithms for detecting the establishment of forest stands. An ensemble method for mapping national-scale SFA by determining the establishment time of secondary forest stands using change detection algorithms and dense Landsat time series was proposed. A dataset of national secondary forest age for China (SFAC) for 1 to 34 and with a 30-m spatial resolution was produced from the optimal ensemble model. This dataset provides national, continuous spatial SFA information and can improve understanding of secondary forests and the estimation of forest carbon storage in China.

Mapping Forest Tree Species Using Sentinel-2 Time Series by Taking into Account Tree Age

Accurate classification of forest tree species holds great significance in the context of forest biodiversity assessment and the management of forest resources. In this study, we utilized Sentinel-2 time series data with high temporal and spatial resolution for tree species classification. To address potential classification errors stemming from spectral differences due to tree age variations, we implemented the Continuous Change Detection and Classification (CCDC) algorithm to estimate tree ages, which were integrated as additional features into our classification models. Four different combinations of classification features were created for both the random forest (RF) algorithm and extreme gradient boosting (XGB) algorithm: spectral band (Spec), spectral band combined with tree age feature (SpecAge), spectral band combined with spectral index (SpecVI), and spectral band combined with spectral index and tree age feature (SpecVIAge). The results demonstrated that the XGB-based models outperformed the RF-based ones, with the SpecVIAge model achieving the highest accuracy at 78.8%. The incorporation of tree age as a classification feature led to an improvement in accuracy by 2% to 3%. The improvement effect on classification accuracy varies across tree species, due to the varying uniformity of tree age among different tree species. These results also showed it is feasible to accurately map regional tree species based on a time-series multi-feature tree species classification model which takes into account tree age.

Characterization of Change in Tree Cover State and Condition over the Conterminous United States

Variability in the effects of disturbances and extreme climate events can lead to changes in tree cover over time, including partial or complete loss, with diverse ecological consequences. It is therefore critical to identify in space and time the change processes that lead to tree cover change. Studies of change are often hampered by the lack of data capable of consistently detecting different types of change. Using the Landsat satellite record to create a long time-series of land cover and land cover change, the U.S. Geological Survey Land Change Monitoring Assessment and Projection (LCMAP) project has made an annual time series of land cover across the conterminous United States for the period 1985 to 2018. Multiple LCMAP products analyzed together with map validation reference plots provide a robust basis for understanding tree cover change. In LCMAP (Collection 1.2), annual change detection is based on harmonic model breaks calculated at each Landsat pixel from the Continuous Change Detection and Classification (CCDC) algorithm. The results showed that the majority of CCDC harmonic model breaks (signifying change) indicated partial tree cover loss (associated with management practices such as tree cover thinning) as compared to complete tree cover loss (associated with practices like clearcut harvest or fire disturbance). Substantially fewer occurrences of complete tree cover loss were associated with change in land cover state. The area of annual tree cover change increased after the late 1990s and stayed high for the rest of the study period. The reference data showed that tree harvest dominated across the conterminous United States. The majority of tree cover change occurred in evergreen forests. Large estimates of disturbance-related tree cover change indicated that tree cover loss may have previously been underreported due to omission of partial tree cover loss in prior studies. This has considerable implications for forest carbon accounting along with tracking ecosystem goods and services.

The Spatial–Temporal Characteristics of Land De-Urbanization in Metropolises: A Case Study of the Guangdong–Hong Kong–Macao Greater Bay Area

With a series of redevelopment activities, such as land consolidation and urban renewal, many cities in China have experienced land de-urbanization phenomena. These include the conversion of construction land into green spaces (such as parks, forests, and lawns), blue spaces (such as rivers, lakes, and wetlands), and farmland. However, there is currently limited research on diverse land de-urbanization types and pathways. This study focuses on investigating the land de-urbanization in the Guangdong–Hong Kong–Macao Greater Bay Area (GBA) from 2014 to September 2023 using the Continuous Change Detection and Classification (CCDC) method. The results demonstrate that the GBA experienced 72.74 square kilometers of de-urbanization during the study period, primarily through the conversion of construction land to land with low plant coverage, including grassland and farmland. There were significant differences in the quantity and spatial agglomeration of de-urbanization between cities and within individual cities. Temporally, de-urbanization predominantly occurred in the period of 2016 to 2021, with a sharp decline in 2022. The temporal changes were significantly influenced by urban renewal policies and the impact of the COVID-19 pandemic. In terms of spatial clustering characteristics, the de-urbanization process in the GBA exhibited spatial agglomeration but was primarily characterized by low-level clustering. This study also examines the correlations between de-urbanization and factors including location and the stage of urbanization. The analysis showed that de-urbanization within cities tended to concentrate near the main urban roads within a range of 10–30 km from city centers. The trend of de-urbanization followed a pattern that is consistent with the Northam curve, where de-urbanization tends to increase during the rapid urbanization phase and decline as urbanization reaches a mature stage. Overall, this study provides valuable insights for the redevelopment of construction land within the context of ecological civilization construction. It also offers suggestions for urban land development and redevelopment in metropolitan areas.

Thirty-meter map of young forest age in China

Abstract. Young forest age mapping at a fine spatial resolution is important for increasing the accuracy of estimating land–atmosphere carbon fluxes and guiding forest management practices. In recent decades, China has actively conducted afforestation and forest protection projects, thereby laying the foundation for the realization of carbon neutrality. However, very few studies have been conducted which map the ages of young forests for the whole of China at a fine spatial resolution. In this research, a continuous change detection and classification (CCDC)-based method suitable for large-scale forest age mapping is proposed and used to estimate young forest ages across China in 2020 at a spatial resolution of 30 m. First, a 10 m spatial-resolution land cover dataset (WorldCover2020) from the European Space Agency (ESA) was used to determine the forest cover areas in 2020. Then, the CCDC algorithm was used to identify stand-replacing disturbances to determine the stand age based on 436 967 Landsat tiles across China from 1990 to 2020. A validation sample set composed of multiple land use and land cover (LULC) products was used to calculate the overall accuracy (OA) of the 2020 young forest age (1–31-year) map of China, and the OA was 90.28 %. The reliability and applicability of the proposed CCDC-based forest age mapping method were validated by comparing the forest age map with Hansen's forest change dataset, Max Planck Institute for Biogeochemistry (MPI-BGC) 1 km global forest age datasets, and field measurements. The CCDC-based method has strong application potential in real-time mapping of the age of young forests at the global scale. The produced forest age map provides a basic dataset for research on the forest carbon cycle and forest ecosystem services as well as important guidance for government departments, such as the National Forestry and Grassland Administration and the National Development and Reform Commission in China. Data presented in this study is available at https://doi.org/10.6084/m9.figshare.21627023.v7 (Xiao, 2022).

Human activities dominate a staged degradation pattern of coastal tidal wetlands in Jiangsu province, China

The tidal wetlands in Jiangsu Province (JS) play a crucial role as a major blue carbon sink in the eastern coastal area of China. These wetlands also contribute significantly function to biodiversity and the provision of ecosystem services. However, due to the escalating impact of anthropogenic activities and climatic changes, the ecological scope, structure, and function of tidal wetlands have changed drastically. This study aims to depict the spatiotemporal dynamics of these wetlands and explore the driving forces behind these changes. Based on the Google Earth Engine (GEE) platform, we conducted a comprehensive analysis of all the available Landsat imagery (5096 images, 30-m spatial resolution) using the Continuous Change Detection and Classification (CCDC) algorithm. The outputs consist of annual land cover maps of JS coastal zone, from 1990 to 2020. Results indicate that the generated coastal maps generally exhibit an exceptionally high average accuracy (92.62%) and there were 2666.54 km2 of tidal wetlands (Spartina alterniflora: 240.86 km2, Suaeda salsa: 37.31 km2, Phragmites australis: 150.63 km2, intertidal flat: 2158.18 km2, supratidal barren flat: 79.53 km2) in 2020, amounting to a 23.43% decrease of the initial year 1990. The tidal wetlands landscape had a staged decline during 1990–2020, and the hydrological vegetation habitat was the area with the greatest variability. Over the past three decades, the losses of tidal wetlands amounted to 1,466.22 km2, partially offset by gains of 650.05 km2. The two most substantial sources of losses were intertidal flat and Suaeda salsa with the most significant ratio of loss to gain (4.80, 10.31, respectively). Generally, we categorized tidal wetlands degradation into three transformation processes: human activities, species invasion, and natural succession. Human activities, particularly the conversion of intertidal flat into aquafarm, dominated the entire degradation process. Meanwhile, in the main driving analysis, we discovered that 60% of these losses and gains were directly tied to human activities. The indirect driving process includes natural coastal movements, climate change, unobservable human activities, and their aggregate effects, which gradually assist tidal wetlands ecosystems to re-establish and recover. The annual coastal wetland maps and detailed information on the evolutionary patterns and the corresponding drivers are valuable for identifying blue carbon distribution and enhancing the ecological services of the coastal zone.

Detecting subpixel human settlements in mountains using deep learning: A case of the Hindu Kush Himalaya 1990–2020

The majority of future population growth in mountains will occur in small- and medium-sized cities and towns and affect vulnerable ecosystems. However, mountain settlements are often omitted from global land cover analyses due to the low spatial resolution of satellite images, which cannot resolve the small scale of mountains settlements. This study demonstrates, for the first time, the potential of deep learning to detect human settlements in mountains at the sub-pixel level, based on Landsat satellite imagery. We hypothesized that adding spatial and temporal features could improve the detection of mountain settlements since spectral information alone led to inaccurate results. For spatial features, we compared a U-shaped neural network (U-Net), a deep learning algorithm that automatically learns spatial features, with a simple random forest (RF) algorithm. Then, we assessed whether temporal features would increase accuracy by comparing two input datasets, multispectral imagery and temporal features from the Continuous Change Detection and Classification (CCDC) algorithm. We evaluated each method by calculating the accuracies of (1) the binary settlement footprint, (2) the subpixel estimates of impervious surfaces, and (3) urban growth. We tested the accuracies using visually interpreted datasets from time-series Google Earth images across the Hindu Kush Himalaya that were not used for training to evaluate model transferability. The U-Net successfully improved mountain settlement mapping compared to the random forest, with a substantial discrepancy in small settlements. The time-series results from the U-Net successfully captured long-term urban growth but fewer short-term changes. Contrary to expectations, the CCDC temporal features reduced the accuracy of mountain settlement mapping due to frequent cloud cover in hilly areas. Our subpixel analysis reveals that the built-up area of the Hindu Kush Himalaya has expanded at a rate of 61 km2 per year from 1990 to 2020, which is about twice the estimate of the Global Human Settlement Layer using binary urban/non-urban classifications.

Automatically identifying the vegetation destruction and restoration of various open-pit mines utilizing remotely sensed images: Auto-VDR

The spatio-temporal pattern of vegetation disturbance by open-pit mining activities such as the disturbance area and magnitude are important for coordinated development of surface mining and environmental protection. This paper proposes an automatic method (Auto-VDR) for identifying vegetation destruction and restoration of various open-pit mines. The proposed method consists of three steps: construction of the sample dataset, identification of disturbance types, and extraction of disturbance time and magnitude. First, a sample dataset with three types was constructed according to the interannual trends of Normalized Difference Vegetation Index (NDVI) in mining areas: “Restoration after destruction (RAD)”, “Destruction without restoration (DWR)” and “No destruction (ND)”. Second, in order to accommodate to various open-pit mines with different background conditions (original vegetation cover prior to mining, loss rate and magnitude of vegetation due to mining, etc.), the “Shape-based distance” was used to match the NDVI curves of each pixel to the three disturbance types. Third, the disturbance time and magnitude in the destroyed regions were extracted based on the vegetated/bare ground threshold (α). The Shendong coal base in China was selected as the study area consisting of 133 open-pit coal mines. Annual NDVI was generated using Landsat time-series data. The maps of vegetation destruction and restoration caused by surface mining during 1990–2021 were drawn with accuracies of 0.82 and 0.87, respectively. The identification accuracy of the Auto-VDR is higher than Landsat-based detection of Trends in Disturbance and Recovery (LandTrendr) and Continuous Change Detection and Classification (CCDC). The results show that from 1990 to 2021, a total of 400.08 km2 of vegetation has been destroyed, accounting for 35.93% of the study area. Furthermore, restoration activities mainly began in 2010, lagging mining activities by approximately three years. The accumulated area of restoration was 177.91 km2 and accounts for 15.98% of the study area. The vegetation destruction at the Shendong coal base has gone through four stages, with 2008, 2015 and 2018 as splits, corresponding to the developing stage of the China coal industry. In particular, the area of vegetation destruction per year reached the maximum of 61.16 km2 in 2008, which was 1.38 times the total area mined over the past 17 years. There were 14% of the 133 open-pit mines in Shendong coal base with a restoration rate of 0.8 and 59% of the mines with a restoration rate of 0.5. This method has been demonstrated to be able to extract vegetation destruction and restoration processes accurately and rapidly for various open-pit mines at a scale of coal base, which provides an important reference for supervising the impact of mining on the environment and assessing the effectiveness of ecological restoration.

Mapping the Age of Subtropical Secondary Forest Using Dense Landsat Time Series Data: An Ensemble Model

Quantifying secondary forest age (SFA) is essential to evaluate the carbon processes of forest ecosystems at regional and global scales. However, the successional stages of secondary forests remain poorly understood due to low-frequency thematic maps. This study aimed to estimate SFA with higher frequency and more accuracy by using dense Landsat archives. The performances of four time-series change detection algorithms—moving average change detection (MACD), Continuous Change Detection and Classification (CCDC), LandTrendr (LT), and Vegetation Change Tracker (VCT)—for detecting forest regrowth were first evaluated. An ensemble model was then developed to determine more accurate timings for forest regrowth based on the evaluation results. Finally, after converting the forest regrowth year to the SFA, the spatiotemporal and topographical distributions of the SFA were analyzed. The proposed ensemble model was validated in Jiangxi province, China, which is located in a subtropical region and has experienced drastic forest disturbances, artificial afforestation, and natural regeneration. The results showed that: (1) the developed ensemble model effectively determined forest regrowth time with significantly decreased omission and commission rates compared to the direct use of the four single algorithms; (2) the optimal ensemble model combining the independent algorithms obtained the final SFA for Jiangxi province with the lowest omission and commission rates in the spatial domain (14.06% and 24.71%) and the highest accuracy in the temporal domain (R2 = 0.87 and root mean square error (RMSE) = 3.17 years); (3) the spatiotemporal and topographic distribution from 1 to 34 years in the 2021 SFA map was analyzed. This study demonstrated the feasibility of using change detection algorithms for estimating SFA at regional to national scales and provides a data foundation for forest ecosystem research.

Mapping Cropland Abandonment in Mountainous Areas in China Using the Google Earth Engine Platform

Knowledge about the spatial-temporal pattern of cropland abandonment is the premise for the management of abandoned croplands. Traditional mapping approaches of abandoned croplands usually utilize a multi-date classification-based land cover change trajectory. It requires quality training samples for land cover classification at each epoch, which is challenging in regions of smallholder agriculture in the absence of high-resolution imagery. Facing these challenges, a theoretical model is proposed to recognize abandoned croplands based on post-abandonment secondary succession. It applies the continuous change detection and classification (CCDC) temporal segmentation algorithm to Landsat time series (1986~2021) to obtain disjoint segments, representing croplands’ status. The post-abandonment secondary succession showing a greening trend is recognized using NDVI-based harmonic analysis, so as to capture its preceding abandonment. This algorithm is applied to a mountainous area in southwest China, where cropland abandonments are widespread. Validation based on stratified random samples referenced by a vegetation index time series and satellite images shows that the detected abandoned croplands have user accuracy, producer accuracy and an F1 score ranging from 43% to 71%, with variation among abandonment year. The study area has a potential cropland extent of 22,294 km2, within which 9252 km2 of the cropland was abandoned. The three peak years of abandonment were 1994, 2000, and 2011. The algorithm is suitable to be applied to large-scale mapping due to its automatic manner.

A novel approach towards continuous monitoring of forest change dynamics in fragmented landscapes using time series Landsat imagery

Carbon emissions from forest ecosystems are greatly impacted by the acceleration of fragmentation and edge effects. Understanding these effects requires accurate monitoring of changes in fragmented forest landscapes. However, these changes are often low-intensity and small-scale, making it difficult to detect them using medium spatial resolution satellite images (e.g., Landsat). To address this challenge, this study developed the Pure Forest Index (PFI), which uses a combination of the existing vegetation index (VI) and spectral mixture analysis (SMA) to more effectively detect and characterize the contribution of forests to the observed spectral response of a pixel. The PFI was applied to detect forest changes in the Amazon rainforest from 1986 to 2020 using the Continuous Change Detection and Classification (CCDC) algorithm (hereafter referred to as the CCDC-PFI algorithm). The results showed reliable performance in mapping forest changes, with an overall accuracy of 0.94 (±0.03) at the spatial scale and a temporal accuracy of 91.1 % (within a two-year window). Comparison with other indices revealed that the PFI improves the ability to monitor forest dynamics with an increased overall accuracy of 0.02–0.35. The PFI also demonstrated advantages in enhancing sub-pixel forest information and suppressing non-forest backgrounds in various scenes compared to conventional VIs. The proposed approach is expected to benefit further research on forests and ecosystems.

Globally vs. Locally Trained Machine Learning Models for Landslide Detection: A Case Study of a Glacial Landscape

Landslide risk mitigation is limited by data scarcity; however, this could be improved using continuous landslide detection systems. To investigate which image types and machine learning models are most useful for landslide detection in a Norwegian setting, we compared the performance of five different machine learning models, for the Jølster case study (30 July 2019), in Western Norway. These included three globally pre-trained models; (i) the continuous change detection and classification (CCDC) algorithm, (ii) a combined k-means clustering and random forest classification model, and (iii) a convolutional neural network (CNN), and two locally trained models, including; (iv) classification and regression Trees and (v) a U-net CNN model. Images used included Sentinel-1, Sentinel-2, as well as digital elevation model (DEM) and slope. The globally trained models performed poorly in shadowed areas and were all outperformed by the locally trained models. A maximum Matthew’s correlation coefficient (MCC) score of 89% was achieved with a CNN U-net deep learning model, using combined Sentinel-1 and -2 images as input. This is one of the first attempts to apply deep learning to detect landslides with both Sentinel-1 and -2 images. Using Sentinel-1 images only, the locally-trained deep-learning model significantly outperformed the conventional machine learning model. These findings contribute to developing a national continuous monitoring system for landslides.

Toward consistent change detection across irregular remote sensing time series observations

The use of remote sensing in time series analysis enables wall-to-wall monitoring of the land surface and is critical for assessing and understanding land cover and land use change and for understanding the Earth system as a whole. However, variability in remote sensing observation frequency through time and across space presents challenges for producing consistent change detection results throughout the available satellite record using approaches such as the Continuous Change Detection and Classification (CCDC) change detection methodology. Here we investigate new modifications to this methodology with the goal of improving accuracy and consistency in results and increasing flexibility for operational usage and future development. The modified method (Band-First Probability, or CCD-BFP) change detection procedure works by calculating a test for each band through time before summarizing between bands. We evaluate the CCD-BFP method compared to an existing implementation of CCDC using a variety of approaches, including a validation dataset of human-interpreted locations, comparison with data from fire events, use of simulated remote sensing data, and qualitative inspection of areas of interest. We find CCD-BFP improves consistency across time and space compared to the existing implementation of CCDC, with more similarity in rates of change across Landsat swath boundaries and before and after the launch of Landsat 7. Also, we find that CCD-BFP detects more of the change events in the validation dataset while reducing the overall number of change detections, indicating that it is able to more accurately capture the most notable land surface change events.

Evaluation of Mangrove Wetlands Protection Patterns in the Guangdong–Hong Kong–Macao Greater Bay Area Using Time-Series Landsat Imageries

The protection of mangroves through nature reserves has been demonstrated to be effective. There were many studies evaluating the mangrove protection effect. However, the evaluation of mangrove growth quality with positive or negative growth trends, as well as restoration potential against disturbance in nature reserves, is still lacking. Thus, this study proposed a hierarchical evaluation framework for mangrove protection in nature reserves, which takes long-term metrics at three levels of loss and gain areas, patch pattern dynamics, and pixel growth trends into account. The continuous change detection and classification (CCDC) was utilized to identify the change condition of mangroves in six nature reserves of the Guangdong–Hong Kong–Macao Greater Bay Area. The Entropy Weight Technique for Order Preference by Similarity to Ideal Solution (TOPSIS) was utilized for scores evaluation of protection effort comparison from 2000 to 2020. The study results had the following three main findings. Firstly, the mangrove forest area increased by about 294.66 ha in four reserves and slightly decreased by about 58.86 ha in two. Most reserves showed an improved patches intact pattern and more positive growth trends. Secondly, the establishment of nature reserves and afforestation were the main causes of mangrove area gain. Until 2010, aquaculture, agriculture, and urban development were the biggest threats to mangroves. Finally, the protection of the reserves was successful in the early decades, but the general evaluation scores showed a decline in recent years once we considered the growth trends for quality. The proposed hierarchical evaluation methods provide a new sight to research the impacts of abrupt change and protection resilience status of the gradual restoration of nature reserves.

An ensemble method for monitoring land cover changes in urban areas using dense Landsat time series data

This study proposes a new method for monitoring land cover change in urban areas using all available Landsat time series data, named the Ensemble of Bidirectional Time Series Analysis (EBTSA). In this method, the bidirectional Continuous Change Detection and Classification (CCDC) and the Chow Test are combined to improve the robustness against data scarcity in earlier times and reduce break detection errors and refine classification results. There are three key stages in this method: break detection using bidirectional CCDs, break refinement using the Chow Test, and bidirectional model integration and classification. The EBTSA method was evaluated over the Tianjin metropolitan area in China using Landsat data from 1989 to 2018. Results show that the proposed method improved spatial and temporal accuracies of both land cover classification and change detection, by reducing the influence of sparser Landsat data in the earlier years and the break detection errors. Using the land cover change results in the Tianjin area obtained using the EBTSA method, we analyzed the spatio-temporal distribution of land cover classification and change detection. It is found from the results that the Tianjin area experienced dramatic urban land changes, characterized by rapid urban expansion and noticeable transition from vegetation to urban land, with diverse changes from urban land to various nonurban land cover types. Results of this study showcase the effectiveness of the proposed method in land cover change monitoring in urban areas, which facilitates a more comprehensive understanding of urban land change dynamics and diversity.