Wildfire Damage Assessment over Australia Using Sentinel-2 Imagery and MODIS Land Cover Product within the Google Earth Engine Cloud Platform

Advancements in remote sensing for active fire detection: A review of datasets and methods

This study comprehensively and critically reviews active fire detection advancements in remote sensing from 1975 to the present, focusing on two main perspectives: datasets and corresponding instruments, and detection algorithms. The study highlights the increasing role of machine learning, particularly deep learning techniques, in active fire detection. Looking forward, the review outlines current challenges and future research opportunities in remote sensing for active fire detection. These include exploring data quality management and multi-modal learning, developing spatiotemporally explicit models, investigating self-supervised learning models, improving explainable and interpretable models, integrating physical-process based models with machine learning, and building digital twins to replicate wildfire dynamics and perform what-if scenario analysis. The review aims to serve as a valuable resource for informing natural resource management and enhancing environmental protection efforts through the application of remote sensing technology.

Temporal assessment of forest cover dynamics in response to forest fires and other environmental impacts using AI

The rapid reduction of forests due to environmental impacts such as deforestation, global warming, natural disasters such as forest fires as well as various human activities is an escalating concern. The increasing frequency and severity of forest fires are causing significant harm to the ecosystem, economy, wildlife, and human safety. During dry and hot seasons, the likelihood of forest fires also increases. It is crucial to accurately monitor and analyze the large-scale changes in the forest cover to ensure sustainable forest management. Remote sensing technology helps to precisely study such changes in forest cover over a wide area over time. This research analyzes the impact of forest fires over time, identifies hotspots, and explores the environmental factors that affect forest cover change. Sentinel-2 imagery was utilized to study changes in Brunei Darussalam's forest cover area over five years from 2017 to 2022. An object-based approach, Simple Non-Iterative Clustering (SNIC), is employed to cluster the region using NDVI values and analyze the changes per cluster. The results indicate that the area of the clusters reduced where fire incidence occurred as well as the precipitation dropped. Between 2017 and 2022, the increased forest fires and decreased precipitation levels resulted in the change in cluster areas as follows: 66.11%, 69.46%, 68.32%, 73.88%, 77.27%, and 78.70%, respectively. Additionally, hotspots in response to forest fires each year were identified in the Belait district. This study will help forest managers assess the causes of forest cover loss and develop conservation and afforestation strategies.

Remote sensing technology for rapid extraction of burned areas and ecosystem environmental assessment

Forest fires are one of the significant disturbances in forest ecosystems. It is essential to extract burned areas rapidly and accurately to formulate forest restoration strategies and plan restoration plans. In this work, we constructed decision trees and used a combination of differential normalized burn ratio (dNBR) index and OTSU threshold method to extract the heavily and mildly burned areas. The applicability of this method was evaluated with three fires in Muli County, Sichuan, China, and we concluded that the extraction accuracy of this method could reach 97.69% and 96.37% for small area forest fires, while the extraction accuracy was lower for large area fires, only 89.32%. In addition, the remote sensing environment index (RSEI) was used to evaluate the ecological environment changes. It analyzed the change of the RSEI level through the transition matrix, and all three fires showed that the changes in RSEI were stronger for heavily burned areas than for mildly burned areas, after the forest fire the ecological environment (RSEI) was reduced from good to moderate. These results realized the quantitative evaluation and dynamic evaluation of the ecological environment condition, providing an essential basis for the restoration, decision making and management of the affected forests.

Mapping burn severity and monitoring CO content in Türkiye's 2021 Wildfires, using Sentinel-2 and Sentinel-5P satellite data on the GEE platform

This study investigated forest fires in the Mediterranean of Türkiye between July 28, 2021, and August 11, 2021. Burn severity maps were produced with the difference normalised burned ratio index (dNBR) and difference normalised difference vegetation index (dNDVI) using Sentinel-2 images on the Google Earth Engine (GEE) cloud platform. The burned areas were estimated based on the determined burning severity degrees. Vegetation density losses in burned areas were analysed using the normalised difference vegetation index (NDVI) time series. At the same time, the post-fire Carbon Monoxide (CO) column number densities were determined using the Sentinel-5P satellite data. According to the burn severity maps obtained with dNBR, the sum of high and moderate severity areas constitutes 34.64%, 20.57%, 46.43%, 51.50% and 18.88% of the entire area in Manavgat, Gündoğmuş, Marmaris, Bodrum and Köyceğiz districts, respectively. Likewise, according to the burn severity maps obtained with dNDVI, the sum of the areas of very high severity and high severity constitutes 41.17%, 30.16%, 30.50%, 42.35%, and 10.40% of the entire region, respectively. In post-fire NDVI time series analyses, sharp decreases were observed in NDVI values from 0.8 to 0.1 in all burned areas. While the Tropospheric CO column number density was 0.03 mol/m² in all regions burned before the fire, it was observed that this value increased to 0.14 mol/m² after the fire. Moreover, when the area was examined more broadly with Sentinel 5P data, it was observed that the amount of CO increased up to a maximum value of 0.333 mol/m². The results of this study present significant information in terms of determining the severity of forest fires in the Mediterranean region in 2021 and the determination of the CO column number density after the fire. In addition, monitoring polluting gases with RS techniques after forest fires is essential in understanding the extent of the damage they can cause to the environment.

Evaluating the efficiency of coarser to finer resolution multispectral satellites in mapping paddy rice fields using GEE implementation

Timely and accurate estimation of rice-growing areas and forecasting of production can provide crucial information for governments, planners, and decision-makers in formulating policies. While there exists studies focusing on paddy rice mapping, only few have compared multi-scale datasets performance in rice classification. Furthermore, rice mapping of large geographical areas with sufficient accuracy for planning purposes has been a challenge in Pakistan, but recent advancements in Google Earth Engine make it possible to analyze spatial and temporal variations within these areas. The study was carried out over southern Punjab (Pakistan)-a region with 380,400 hectares devoted to rice production in year 2020. Previous studies support the individual capabilities of Sentinel-2, Landsat-8, and Moderate Resolution Imaging Spectroradiometer (MODIS) for paddy rice classification. However, to our knowledge, no study has compared the efficiencies of these three datasets in rice crop classification. Thus, this study primarily focuses on comparing these satellites’ data by estimating their potential in rice crop classification using accuracy assessment methods and area estimation. The overall accuracies were found to be 96% for Sentinel-2, 91.7% for Landsat-8, and 82.6% for MODIS. The F1-Scores for derived rice class were 83.8%, 75.5%, and 65.5% for Sentinel-2, Landsat-8, and MODIS, respectively. The rice estimated area corresponded relatively well with the crop statistics report provided by the Department of Agriculture, Punjab, with a mean percentage difference of less than 20% for Sentinel-2 and MODIS and 33% for Landsat-8. The outcomes of this study highlight three points; (a) Rice mapping accuracy improves with increase in spatial resolution, (b) Sentinel-2 efficiently differentiated individual farm level paddy fields while Landsat-8 was not able to do so, and lastly (c) Increase in rice cultivated area was observed using satellite images compared to the government provided statistics.

Google Earth Engine and Artificial Intelligence (AI): A Comprehensive Review

Remote sensing (RS) plays an important role gathering data in many critical domains (e.g., global climate change, risk assessment and vulnerability reduction of natural hazards, resilience of ecosystems, and urban planning). Retrieving, managing, and analyzing large amounts of RS imagery poses substantial challenges. Google Earth Engine (GEE) provides a scalable, cloud-based, geospatial retrieval and processing platform. GEE also provides access to the vast majority of freely available, public, multi-temporal RS data and offers free cloud-based computational power for geospatial data analysis. Artificial intelligence (AI) methods are a critical enabling technology to automating the interpretation of RS imagery, particularly on object-based domains, so the integration of AI methods into GEE represents a promising path towards operationalizing automated RS-based monitoring programs. In this article, we provide a systematic review of relevant literature to identify recent research that incorporates AI methods in GEE. We then discuss some of the major challenges of integrating GEE and AI and identify several priorities for future research. We developed an interactive web application designed to allow readers to intuitively and dynamically review the publications included in this literature review.

Evaluating Effects of Post-Fire Climate and Burn Severity on the Early-Term Regeneration of Forest and Shrub Communities in the San Gabriel Mountains of California from Sentinel-2(MSI) Images

Studying the early changes in post-fire vegetation communities may improve the overall resilience of forests. The necessity for doing so was demonstrated by the Bobcat Fire, which seriously threatened the central San Gabriel Mountains and the Angeles National Forest in California. This study aimed to monitor and quantify the effects of climatological and topographic conditions along with burn severity on early (within 1 year) post-fire forests and shrubs community regeneration. In this study, we used Sentinel-2(MSI) intensive time-series imagery (July 2020–October 2021) to make a confusion matrix combined with 389 vegetation sample points on Google Earth Pro. The overall accuracy (OA) and the Kappa coefficient, calculated from the confusion matrix, were used as evaluation parameters to validate the classification results. With multiple linear regression models and Environmental Systems Research Institute (ESRI) historical images, we analyzed the effects of climate and slope aspects on the regeneration of post-fire forest and shrub communities. We also quantitatively analyzed the regeneration rates based on five burn severity types. The results show that the normalized burning rate (NBR) was the most accurate vegetation classification indicator in this study (OA: 92.3–99.5%, Kappa: 0.88–0.98). The vegetation classification accuracy based on SVM is about 6.6% higher than K-Means. The overall accuracy of the burn area is 94.87%. Post-fire climate factors had a significant impact on the regeneration of the two vegetation communities (R2: 0.42–0.88); the optimal regeneration slope was 15–35°; and the fire severity changed the original competition relationship and regeneration rate. The results provide four main insights into the regeneration of post-fire vegetation communities: (1) climate factors in the first regenerating season have important impacts on the regeneration of forest and shrub communities; (2) daytime duration and rainfall are the most significant factors for forests and shrubs regeneration; (3) tolerable low burn severity promotes forests regeneration; and (4) forests have a certain ability to resist fires, while shrubs can better tolerate high-intensity fire ecology. This study could support the implementation of strategies for regionalized forest management and the targeted enhancement of post-fire vegetation community resilience.

Estimation of Postfire Reforestation with SAR Polarimetry and NDVI Time Series

This communication is devoted to the methodology of remote complex analysis of forest restoration after strong wildfires. It is proposed to quantify the projective leaf/needles area index by multispectral optical images. The increase in dimensions of trunks and branches commensurate with a radar wavelength of 24 cm is estimated using radar polarimetric data. It is shown that the growth’s potential of aboveground biomass in different spots of test site ranges from 35 to 70% in the case under consideration. Such a new approach will make it possible to further consider more accurately the role of boreal forests as one of the largest carbon stocks.

Normalized Burn Ratio Plus (NBR+): A New Index for Sentinel-2 Imagery

The monitoring of burned areas can easily be performed using satellite multispectral images: several indices are available in the literature for highlighting the differences between healthy vegetation areas and burned areas, in consideration of their different signatures. However, these indices may have limitations determined, for example, by the presence of clouds or water bodies that produce false alarms. To avoid these inaccuracies and optimize the results, this work proposes a new index for detecting burned areas named Normalized Burn Ratio Plus (NBR+), based on the involvement of Sentinel-2 bands. The efficiency of this index is verified by comparing it with five other existing indices, all applied on an area with a surface of about 500 km2 and covering the north-eastern part of Sicily (Italy). To achieve this aim, both a uni-temporal approach (single date image) and a bi-temporal approach (two date images) are adopted. The maximum likelihood classifier (MLC) is applied to each resulting index map to define the threshold separating burned pixels from non-burned ones. To evaluate the efficiency of the indices, confusion matrices are constructed and compared with each other. The NBR+ shows excellent results, especially because it excludes a large part of the areas incorrectly classified as burned by other indices, despite being clouds or water bodies.

Extreme fire weather is the major driver of severe bushfires in southeast Australia

In Australia, the proportion of forest area that burns in a typical fire season is less than for other vegetation types. However, the 2019-2020 austral spring-summer was an exception, with over four times the previous maximum area burnt in southeast Australian temperate forests. Temperate forest fires have extensive socio-economic, human health, greenhouse gas emissions, and biodiversity impacts due to high fire intensities. A robust model that identifies driving factors of forest fires and relates impact thresholds to fire activity at regional scales would help land managers and fire-fighting agencies prepare for potentially hazardous fire in Australia. Here, we developed a machine-learning diagnostic model to quantify nonlinear relationships between monthly burnt area and biophysical factors in southeast Australian forests for 2001-2020 on a 0.25° grid based on several biophysical parameters, notably fire weather and vegetation productivity. Our model explained over 80% of the variation in the burnt area. We identified that burnt area dynamics in southeast Australian forest were primarily controlled by extreme fire weather, which mainly linked to fluctuations in the Southern Annular Mode (SAM) and Indian Ocean Dipole (IOD), with a relatively smaller contribution from the central Pacific El Niño Southern Oscillation (ENSO). Our fire diagnostic model and the non-linear relationships between burnt area and environmental covariates can provide useful guidance to decision-makers who manage preparations for an upcoming fire season, and model developers working on improved early warning systems for forest fires.

Active Fire Detection from Landsat-8 Imagery Using Deep Multiple Kernel Learning

Active fires are devastating natural disasters that cause socio-economical damage across the globe. The detection and mapping of these disasters require efficient tools, scientific methods, and reliable observations. Satellite images have been widely used for active fire detection (AFD) during the past years due to their nearly global coverage. However, accurate AFD and mapping in satellite imagery is still a challenging task in the remote sensing community, which mainly uses traditional methods. Deep learning (DL) methods have recently yielded outstanding results in remote sensing applications. Nevertheless, less attention has been given to them for AFD in satellite imagery. This study presented a deep convolutional neural network (CNN) “MultiScale-Net” for AFD in Landsat-8 datasets at the pixel level. The proposed network had two main characteristics: (1) several convolution kernels with multiple sizes, and (2) dilated convolution layers (DCLs) with various dilation rates. Moreover, this paper suggested an innovative Active Fire Index (AFI) for AFD. AFI was added to the network inputs consisting of the SWIR2, SWIR1, and Blue bands to improve the performance of the MultiScale-Net. In an ablation analysis, three different scenarios were designed for multi-size kernels, dilation rates, and input variables individually, resulting in 27 distinct models. The quantitative results indicated that the model with AFI-SWIR2-SWIR1-Blue as the input variables, using multiple kernels of sizes 3 × 3, 5 × 5, and 7 × 7 simultaneously, and a dilation rate of 2, achieved the highest F1-score and IoU of 91.62% and 84.54%, respectively. Stacking AFI with the three Landsat-8 bands led to fewer false negative (FN) pixels. Furthermore, our qualitative assessment revealed that these models could detect single fire pixels detached from the large fire zones by taking advantage of multi-size kernels. Overall, the MultiScale-Net met expectations in detecting fires of varying sizes and shapes over challenging test samples.

An Investigation of a Multidimensional CNN Combined with an Attention Mechanism Model to Resolve Small-Sample Problems in Hyperspectral Image Classification

The convolutional neural network (CNN) method has been widely used in the classification of hyperspectral images (HSIs). However, the efficiency and accuracy of the HSI classification are inevitably degraded when small samples are available. This study proposes a multidimensional CNN model named MDAN, which is constructed with an attention mechanism, to achieve an ideal classification performance of CNN within the framework of few-shot learning. In this model, a three-dimensional (3D) convolutional layer is carried out for obtaining spatial–spectral features from the 3D volumetric data of HSI. Subsequently, the two-dimensional (2D) and one-dimensional (1D) convolutional layers further learn spatial and spectral features efficiently at an abstract level. Based on the most widely used convolutional block attention module (CBAM), this study investigates a convolutional block self-attention module (CBSM) to improve accuracy by changing the connection ways of attention blocks. The CBSM model is used with the 2D convolutional layer for better performance of HSI classification purposes. The MDAN model is applied for classification applications using HSI, and its performance is evaluated by comparing the results with the support vector machine (SVM), 2D CNN, 3D CNN, 3D–2D–1D CNN, and CBAM. The findings of this study indicate that classification results from the MADN model show overall classification accuracies of 97.34%, 96.43%, and 92.23% for Salinas, WHU-Hi-HanChuan, and Pavia University datasets, respectively, when only 1% HSI data were used for training. The training and testing times of the MDAN model are close to those of the 3D–2D–1D CNN, which has the highest efficiency among all comparative CNN models. The attention model CBSM is introduced into MDAN, which achieves an overall accuracy of about 1% higher than that of the CBAM model. The performance of the two proposed methods is superior to the other models in terms of both efficiency and accuracy. The results show that the combination of multidimensional CNNs and attention mechanisms has the best ability for small-sample problems in HSI classification.

A Dual Attention Convolutional Neural Network for Crop Classification Using Time-Series Sentinel-2 Imagery

Accurate and timely mapping of crop types and having reliable information about the cultivation pattern/area play a key role in various applications, including food security and sustainable agriculture management. Remote sensing (RS) has extensively been employed for crop type classification. However, accurate mapping of crop types and extents is still a challenge, especially using traditional machine learning methods. Therefore, in this study, a novel framework based on a deep convolutional neural network (CNN) and a dual attention module (DAM) and using Sentinel-2 time-series datasets was proposed to classify crops. A new DAM was implemented to extract informative deep features by taking advantage of both spectral and spatial characteristics of Sentinel-2 datasets. The spectral and spatial attention modules (AMs) were respectively applied to investigate the behavior of crops during the growing season and their neighborhood properties (e.g., textural characteristics and spatial relation to surrounding crops). The proposed network contained two streams: (1) convolution blocks for deep feature extraction and (2) several DAMs, which were employed after each convolution block. The first stream included three multi-scale residual convolution blocks, where the spectral attention blocks were mainly applied to extract deep spectral features. The second stream was built using four multi-scale convolution blocks with a spatial AM. In this study, over 200,000 samples from six different crop types (i.e., alfalfa, broad bean, wheat, barley, canola, and garden) and three non-crop classes (i.e., built-up, barren, and water) were collected to train and validate the proposed framework. The results demonstrated that the proposed method achieved high overall accuracy and a Kappa coefficient of 98.54% and 0.981, respectively. It also outperformed other state-of-the-art classification methods, including RF, XGBOOST, R-CNN, 2D-CNN, 3D-CNN, and CBAM, indicating its high potential to discriminate different crop types.

DSMNN-Net: A Deep Siamese Morphological Neural Network Model for Burned Area Mapping Using Multispectral Sentinel-2 and Hyperspectral PRISMA Images

Wildfires are one of the most destructive natural disasters that can affect our environment, with significant effects also on wildlife. Recently, climate change and human activities have resulted in higher frequencies of wildfires throughout the world. Timely and accurate detection of the burned areas can help to make decisions for their management. Remote sensing satellite imagery can have a key role in mapping burned areas due to its wide coverage, high-resolution data collection, and low capture times. However, although many studies have reported on burned area mapping based on remote sensing imagery in recent decades, accurate burned area mapping remains a major challenge due to the complexity of the background and the diversity of the burned areas. This paper presents a novel framework for burned area mapping based on Deep Siamese Morphological Neural Network (DSMNN-Net) and heterogeneous datasets. The DSMNN-Net framework is based on change detection through proposing a pre/post-fire method that is compatible with heterogeneous remote sensing datasets. The proposed network combines multiscale convolution layers and morphological layers (erosion and dilation) to generate deep features. To evaluate the performance of the method proposed here, two case study areas in Australian forests were selected. The framework used can better detect burned areas compared to other state-of-the-art burned area mapping procedures, with a performance of >98% for overall accuracy index, and a kappa coefficient of >0.9, using multispectral Sentinel-2 and hyperspectral PRISMA image datasets. The analyses of the two datasets illustrate that the DSMNN-Net is sufficiently valid and robust for burned area mapping, and especially for complex areas.

A Preliminary Global Automatic Burned-Area Algorithm at Medium Resolution in Google Earth Engine

A preliminary version of a global automatic burned-area (BA) algorithm at medium spatial resolution was developed in Google Earth Engine (GEE), based on Landsat or Sentinel-2 reflectance images. The algorithm involves two main steps: initial burned candidates are identified by analyzing spectral changes around MODIS hotspots, and those candidates are then used to estimate the burn probability for each scene. The burning dates are identified by analyzing the temporal evolution of burn probabilities. The algorithm was processed, and its quality assessed globally using reference data from 2019 derived from Sentinel-2 data at 10 m, which involved 369 pairs of consecutive images in total located in 50 20 × 20 km2 areas selected by stratified random sampling. Commissions were around 10% with both satellites, although omissions ranged between 27 (Sentinel-2) and 35% (Landsat), depending on the selected resolution and dataset, with highest omissions being in croplands and forests; for their part, BA from Sentinel-2 data at 20 m were the most accurate and fastest to process. In addition, three 5 × 5 degree regions were randomly selected from the biomes where most fires occur, and BA were detected from Sentinel-2 images at 20 m. Comparison with global products at coarse resolution FireCCI51 and MCD64A1 would seem to show to a reliable extent that the algorithm is procuring spatially and temporally coherent results, improving detection of smaller fires as a consequence of higher-spatial-resolution data. The proposed automatic algorithm has shown the potential to map BA globally using medium-spatial-resolution data (Sentinel-2 and Landsat) from 2000 onwards, when MODIS satellites were launched.

A Training Sample Migration Method for Wetland Mapping and Monitoring Using Sentinel Data in Google Earth Engine

Wetlands are one of the most important ecosystems due to their critical services to both humans and the environment. Therefore, wetland mapping and monitoring are essential for their conservation. In this regard, remote sensing offers efficient solutions due to the availability of cost-efficient archived images over different spatial scales. However, a lack of sufficient consistent training samples at different times is a significant limitation of multi-temporal wetland monitoring. In this study, a new training sample migration method was developed to identify unchanged training samples to be used in wetland classification and change analyses over the International Shadegan Wetland (ISW) areas of southwestern Iran. To this end, we first produced the wetland map of a reference year (2020), for which we had training samples, by combining Sentinel-1 and Sentinel-2 images and the Random Forest (RF) classifier in Google Earth Engine (GEE). The Overall Accuracy (OA) and Kappa coefficient (KC) of this reference map were 97.93% and 0.97, respectively. Then, an automatic change detection method was developed to migrate unchanged training samples from the reference year to the target years of 2018, 2019, and 2021. Within the proposed method, three indices of the Normalized Difference Vegetation Index (NDVI), Normalized Difference Water Index (NDWI), and the mean Standard Deviation (SD) of the spectral bands, along with two similarity measures of the Euclidean Distance (ED) and Spectral Angle Distance (SAD), were computed for each pair of reference–target years. The optimum threshold for unchanged samples was also derived using a histogram thresholding approach, which led to selecting the samples that were most likely unchanged based on the highest OA and KC for classifying the test dataset. The proposed migration sample method resulted in high OAs of 95.89%, 96.83%, and 97.06% and KCs of 0.95, 0.96, and 0.96 for the target years of 2018, 2019, and 2021, respectively. Finally, the migrated samples were used to generate the wetland map for the target years. Overall, our proposed method showed high potential for wetland mapping and monitoring when no training samples existed for a target year.

A Remote Sensing and Machine Learning-Based Approach to Forecast the Onset of Harmful Algal Bloom

In the last few decades, harmful algal blooms (HABs, also known as “red tides”) have become one of the most detrimental natural phenomena in Florida’s coastal areas. Karenia brevis produces toxins that have harmful effects on humans, fisheries, and ecosystems. In this study, we developed and compared the efficiency of state-of-the-art machine learning models (e.g., XGBoost, Random Forest, and Support Vector Machine) in predicting the occurrence of HABs. In the proposed models the K. brevis abundance is used as the target, and 10 level-02 ocean color products extracted from daily archival MODIS satellite data are used as controlling factors. The adopted approach addresses two main shortcomings of earlier models: (1) the paucity of satellite data due to cloudy scenes and (2) the lag time between the period at which a variable reaches its highest correlation with the target and the time the bloom occurs. Eleven spatio-temporal models were generated, each from 3 consecutive day satellite datasets, with a forecasting span from 1 to 11 days. The 3-day models addressed the potential variations in lag time for some of the temporal variables. One or more of the generated 11 models could be used to predict HAB occurrences depending on availability of the cloud-free consecutive days. Findings indicate that XGBoost outperformed the other methods, and the forecasting models of 5–9 days achieved the best results. The most reliable model can forecast eight days ahead of time with balanced overall accuracy, Kappa coefficient, F-Score, and AUC of 96%, 0.93, 0.97, and 0.98 respectively. The euphotic depth, sea surface temperature, and chlorophyll-a are always among the most significant controlling factors. The proposed models could potentially be used to develop an “early warning system” for HABs in southwest Florida.

Mapping Forest Burn Extent from Hyperspatial Imagery Using Machine Learning

Support vector machines are shown to be highly effective in mapping burn extent from hyperspatial imagery in grasslands. Unfortunately, this pixel-based method is hampered in forested environments that have experienced low-intensity fires because unburned tree crowns obstruct the view of the surface vegetation. This obstruction causes surface fires to be misclassified as unburned. To account for misclassifying areas under tree crowns, trees surrounded by surface burn can be assumed to have been burned underneath. This effort used a mask region-based convolutional neural network (MR-CNN) and support vector machine (SVM) to determine trees and burned pixels in a post-fire forest. The output classifications of the MR-CNN and SVM were used to identify tree crowns in the image surrounded by burned surface vegetation pixels. These classifications were also used to label the pixels under the tree as being within the fire’s extent. This approach results in higher burn extent mapping accuracy by eliminating burn extent false negatives from surface burns obscured by unburned tree crowns, achieving a nine percentage point increase in burn extent mapping accuracy.

A Fully Automatic, Interpretable and Adaptive Machine Learning Approach to Map Burned Area from Remote Sensing

The paper proposes a fully automatic algorithm approach to map burned areas from remote sensing characterized by human interpretable mapping criteria and explainable results. This approach is partially knowledge-driven and partially data-driven. It exploits active fire points to train the fusion function of factors deemed influential in determining the evidence of burned conditions from reflectance values of multispectral Sentinel-2 (S2) data. The fusion function is used to compute a map of seeds (burned pixels) that are adaptively expanded by applying a Region Growing (RG) algorithm to generate the final burned area map. The fusion function is an Ordered Weighted Averaging (OWA) operator, learnt through the application of a machine learning (ML) algorithm from a set of highly reliable fire points. Its semantics are characterized by two measures, the degrees of pessimism/optimism and democracy/monarchy. The former allows the prediction of the results of the fusion as affected by more false positives (commission errors) than false negatives (omission errors) in the case of pessimism, or vice versa; the latter foresees if there are only a few highly influential factors or many low influential ones that determine the result. The prediction on the degree of pessimism/optimism allows the expansion of the seeds to be appropriately tuned by selecting the most suited growing layer for the RG algorithm thus adapting the algorithm to the context. The paper illustrates the application of the automatic method in four study areas in southern Europe to map burned areas for the 2017 fire season. Thematic accuracy at each site was assessed by comparison to reference perimeters to prove the adaptability of the approach to the context; estimated average accuracy metrics are omission error = 0.057, commission error = 0.068, Dice coefficient = 0.94 and relative bias = 0.0046.

Burned Area Mapping over the Southern Cape Forestry Region, South Africa Using Sentinel Data within GEE Cloud Platform

Planted forests in South Africa have been affected by an increasing number of economically damaging fires over the past four decades. They constitute a major threat to the forestry industry and account for over 80% of the country’s commercial timber losses. Forest fires are more frequent and severe during the drier drought conditions that are typical in South Africa. For proper forest management, accurate detection and mapping of burned areas are required, yet the exercise is difficult to perform in the field because of time and expense. Now that ready-to-use satellite data are freely accessible in the cloud-based Google Earth Engine (GEE), in this study, we exploit the Sentinel-2-derived differenced normalized burned ratio (dNBR) to characterize burn severity areas, and also track carbon monoxide (CO) plumes using Sentinel-5 following a wildfire that broke over the southeastern coast of the Western Cape province in late October 2018. The results showed that 37.4% of the area was severely burned, and much of it occurred in forested land in the studied area. This was followed by 24.7% of the area that was burned at a moderate-high level. About 15.9% had moderate-low burned severity, whereas 21.9% was slightly burned. Random forests classifier was adopted to separate burned class from unburned and achieved an overall accuracy of over 97%. The most important variables in the classification included texture, NBR, and the NIR bands. The CO signal sharply increased during fire outbreaks and marked the intensity of black carbon over the affected area. Our study contributes to the understanding of forest fire in the dynamics over the Southern Cape forestry landscape. Furthermore, it also demonstrates the usefulness of Sentinel-5 for monitoring CO. Taken together, the Sentinel satellites and GEE offer an effective tool for mapping fires, even in data-poor countries.

A Novel Spatial Simulation Method for Mapping the Urban Forest Carbon Density in Southern China by the Google Earth Engine

Urban forest is an important component of terrestrial ecosystems and is highly related to global climate change. However, because of complex city landscapes, deriving the spatial distribution of urban forest carbon density and conducting accuracy assessments are difficult. This study proposes a novel spatial simulation method, optimized geographically weighted logarithm regression (OGWLR), using Landsat 8 data acquired by the Google Earth Engine (GEE) and field survey data to map the forest carbon density of Shenzhen city in southern China. To verify the effectiveness of the novel method, multiple linear regression (MLR), k-nearest neighbors (kNN), random forest (RF) and geographically weighted regression (GWR) models were established for comparison. The results showed that OGWLR achieved the highest coefficient of determination (R2 = 0.54) and the lowest root mean square error (RMSE = 13.28 Mg/ha) among all estimation models. In addition, OGWLR achieved a more consistent spatial distribution of carbon density with the actual situation. The carbon density of the forests in the study area was large in the central and western regions and coastal areas and small in the building and road areas. Therefore, this method can provide a new reference for urban forest carbon density estimation and mapping.

Mangrove Ecosystem Mapping Using Sentinel-1 and Sentinel-2 Satellite Images and Random Forest Algorithm in Google Earth Engine

Mangroves are among the most productive ecosystems in existence, with many ecological benefits. Therefore, generating accurate thematic maps from mangrove ecosystems is crucial for protecting, conserving, and reforestation planning for these valuable natural resources. In this paper, Sentinel-1 and Sentinel-2 satellite images were used in synergy to produce a detailed mangrove ecosystem map of the Hara protected area, Qeshm, Iran, at 10 m spatial resolution within the Google Earth Engine (GEE) cloud computing platform. In this regard, 86 Sentinel-1 and 41 Sentinel-2 data, acquired in 2019, were employed to generate seasonal optical and synthetic aperture radar (SAR) features. Afterward, seasonal features were inserted into a pixel-based random forest (RF) classifier, resulting in an accurate mangrove ecosystem map with average overall accuracy (OA) and Kappa coefficient (KC) of 93.23% and 0.92, respectively, wherein all classes (except aerial roots) achieved high producer and user accuracies of over 90%. Furthermore, comprehensive quantitative and qualitative assessments were performed to investigate the robustness of the proposed approach, and the accurate and stable results achieved through cross-validation and consistency checks confirmed its robustness and applicability. It was revealed that seasonal features and the integration of multi-source remote sensing data contributed towards obtaining a more reliable mangrove ecosystem map. The proposed approach relies on a straightforward yet effective workflow for mangrove ecosystem mapping, with a high rate of automation that can be easily implemented for frequent and precise mapping in other parts of the world. Overall, the proposed workflow can further improve the conservation and sustainable management of these valuable natural resources.

A Burned Area Mapping Algorithm for Sentinel-2 Data Based on Approximate Reasoning and Region Growing

Sentinel-2 (S2) multi-spectral instrument (MSI) images are used in an automated approach built on fuzzy set theory and a region growing (RG) algorithm to identify areas affected by fires in Mediterranean regions. S2 spectral bands and their post- and pre-fire date (Dpost-pre) difference are interpreted as evidence of burn through soft constraints of membership functions defined from statistics of burned/unburned training regions; evidence of burn brought by the S2 spectral bands (partial evidence) is integrated using ordered weighted averaging (OWA) operators that provide synthetic score layers of likelihood of burn (global evidence of burn) that are combined in an RG algorithm. The algorithm is defined over a training site located in Italy, Vesuvius National Park, where membership functions are defined and OWA and RG algorithms are first tested. Over this site, validation is carried out by comparison with reference fire perimeters derived from supervised classification of very high-resolution (VHR) PlanetScope images leading to more than satisfactory results with Dice coefficient >0.84, commission error <0.22 and omission error <0.15. The algorithm is tested for exportability over five sites in Portugal (1), Spain (2) and Greece (2) to evaluate the performance by comparison with fire reference perimeters derived from the Copernicus Emergency Management Service (EMS) database. In these sites, we estimate commission error <0.15, omission error <0.1 and Dice coefficient >0.9 with accuracy in some cases greater than values obtained in the training site. Regression analysis confirmed the satisfactory accuracy levels achieved over all sites. The algorithm proposed offers the advantages of being least dependent on a priori/supervised selection for input bands (by building on the integration of redundant partial burn evidence) and for criteria/threshold to obtain segmentation into burned/unburned areas.

Mapping the Forest Canopy Height in Northern China by Synergizing ICESat-2 with Sentinel-2 Using a Stacking Algorithm

The forest canopy height (FCH) plays a critical role in forest quality evaluation and resource management. The accurate and rapid estimation and mapping of the regional forest canopy height is crucial for understanding vegetation growth processes and the internal structure of the ecosystem. A stacking algorithm consisting of multiple linear regression (MLR), support vector machine (SVM), k-nearest neighbor (kNN), and random forest (RF) was used in this paper and demonstrated optimal performance in predicting the forest canopy height by synergizing Sentinel-2 images acquired from the cloud-based computation platform Google Earth Engine (GEE) with data from ICESat-2 (Ice, Cloud, and Land Elevation Satellite-2). This research was conducted to achieve continuous mapping of the canopy height of plantations in Saihanba Mechanical Forest Plantation, which is located in Chengde City, northern Hebei province, China. The results show that stacking achieved the best prediction accuracy for the forest canopy height, with an R2 of 0.77 and a root mean square error (RMSE) of 1.96 m. Compared with MLR, SVM, kNN, and RF, the RMSE obtained by stacking was reduced by 25.2%, 24.9%, 22.8%, and 18.7%, respectively. Since Sentinel-2 images and ICESat-2 data are publicly available, this opens the door for the accurate mapping of the continuous distribution of the forest canopy height globally in the future.

Uni-Temporal Multispectral Imagery for Burned Area Mapping with Deep Learning

Accurate burned area information is needed to assess the impacts of wildfires on people, communities, and natural ecosystems. Various burned area detection methods have been developed using satellite remote sensing measurements with wide coverage and frequent revisits. Our study aims to expound on the capability of deep learning (DL) models for automatically mapping burned areas from uni-temporal multispectral imagery. Specifically, several semantic segmentation network architectures, i.e., U-Net, HRNet, Fast-SCNN, and DeepLabv3+, and machine learning (ML) algorithms were applied to Sentinel-2 imagery and Landsat-8 imagery in three wildfire sites in two different local climate zones. The validation results show that the DL algorithms outperform the ML methods in two of the three cases with the compact burned scars, while ML methods seem to be more suitable for mapping dispersed burn in boreal forests. Using Sentinel-2 images, U-Net and HRNet exhibit comparatively identical performance with higher kappa (around 0.9) in one heterogeneous Mediterranean fire site in Greece; Fast-SCNN performs better than others with kappa over 0.79 in one compact boreal forest fire with various burn severity in Sweden. Furthermore, directly transferring the trained models to corresponding Landsat-8 data, HRNet dominates in the three test sites among DL models and can preserve the high accuracy. The results demonstrated that DL models can make full use of contextual information and capture spatial details in multiple scales from fire-sensitive spectral bands to map burned areas. Using only a post-fire image, the DL methods not only provide automatic, accurate, and bias-free large-scale mapping option with cross-sensor applicability, but also have potential to be used for onboard processing in the next Earth observation satellites.

Earthquake Damage Region Detection by Multitemporal Coherence Map Analysis of Radar and Multispectral Imagery

Earth, as humans’ habitat, is constantly affected by natural events, such as floods, earthquakes, thunder, and drought among which earthquakes are considered one of the deadliest and most catastrophic natural disasters. The Iran-Iraq earthquake occurred in Kermanshah Province, Iran in November 2017. It was a 7.4-magnitude seismic event that caused immense damages and loss of life. The rapid detection of damages caused by earthquakes is of great importance for disaster management. Thanks to their wide coverage, high resolution, and low cost, remote-sensing images play an important role in environmental monitoring. This study presents a new damage detection method at the unsupervised level, using multitemporal optical and radar images acquired through Sentinel imagery. The proposed method is applied in two main phases: (1) automatic built-up extraction using spectral indices and active learning framework on Sentinel-2 imagery; (2) damage detection based on the multitemporal coherence map clustering and similarity measure analysis using Sentinel-1 imagery. The main advantage of the proposed method is that it is an unsupervised method with simple usage, a low computing burden, and using medium spatial resolution imagery that has good temporal resolution and is operative at any time and in any atmospheric conditions, with high accuracy for detecting deformations in buildings. The accuracy analysis of the proposed method found it visually and numerically comparable to other state-of-the-art methods for built-up area detection. The proposed method is capable of detecting built-up areas with an accuracy of more than 96% and a kappa of about 0.89 in overall comparison to other methods. Furthermore, the proposed method is also able to detect damaged regions compared to other state-of-the-art damage detection methods with an accuracy of more than 70%.

Landsat and Sentinel-2 Based Burned Area Mapping Tools in Google Earth Engine

Four burned area tools were implemented in Google Earth Engine (GEE), to obtain regular processes related to burned area (BA) mapping, using medium spatial resolution sensors (Landsat and Sentinel-2). The four tools are (i) the BA Cartography tool for supervised burned area over the user-selected extent and period, (ii) two tools implementing a BA stratified random sampling to select the scenes and dates for validation, and (iii) the BA Reference Perimeter tool to obtain highly accurate BA maps that focus on validating coarser BA products. Burned Area Mapping Tools (BAMTs) go beyond the previously implemented Burned Area Mapping Software (BAMS) because of GEE parallel processing capabilities and preloaded geospatial datasets. BAMT also allows temporal image composites to be exploited in order to obtain BA maps over a larger extent and longer temporal periods. The tools consist of four scripts executable from the GEE Code Editor. The tools’ performance was discussed in two case studies: in the 2019/2020 fire season in Southeast Australia, where the BA cartography detected more than 50,000 km2, using Landsat data with commission and omission errors below 12% when compared to Sentinel-2 imagery; and in the 2018 summer wildfires in Canada, where it was found that around 16,000 km2 had burned.