Assessing the Natural Recovery of Mangroves after Human Disturbance Using Neural Network Classification and Sentinel-2 Imagery in Wunbaik Mangrove Forest, Myanmar

In this study, we examined the natural recovery of mangroves in abandoned shrimp ponds located in the Wunbaik Mangrove Forest (WMF) in Myanmar using artificial neural network (ANN) classification and a change detection approach with Sentinel-2 satellite images. In 2020, we conducted various experiments related to mangrove classification by tuning input features and hyper-parameters. The selected ANN model was used with a transfer learning approach to predict the mangrove distribution in 2015. Changes were detected using classification results from 2015 and 2020. Naturally recovering mangroves were identified by extracting the change detection results of three abandoned shrimp ponds selected during field investigation. The proposed method yielded an overall accuracy of 95.98%, a kappa coefficient of 0.92, mangrove and non-mangrove precisions of 0.95 and 0.98, respectively, recalls of 0.96, and F1 scores of 0.96 for the 2020 classification. For the 2015 prediction, transfer learning improved model performance, resulting in an overall accuracy of 97.20%, a kappa coefficient of 0.94, mangrove and non-mangrove precisions of 0.98 and 0.96, respectively, recalls of 0.98 and 0.97, and F1 scores of 0.96. The change detection results showed that mangrove forests in the WMF slightly decreased between 2015 and 2020. Naturally recovering mangroves were detected at approximately 50% of each abandoned site within a short abandonment period. This study demonstrates that the ANN method using Sentinel-2 imagery and topographic and canopy height data can produce reliable results for mangrove classification. The natural recovery of mangroves presents a valuable opportunity for mangrove rehabilitation at human-disturbed sites in the WMF.

Download Full-text

National-scale mangrove forest mapping by using Sentinel-1 SAR and Sentinel-2 MSI imagery on the Google Earth Engine Platform

10.5194/egusphere-egu2020-5305 ◽

2020 ◽

Author(s):

Luojia Hu ◽

Wei Yao ◽

Zhitong Yu ◽

Lei Wang

Keyword(s):

Spatial Resolution ◽

Mangrove Forest ◽

Large Scale ◽

Landsat Imagery ◽

Google Earth ◽

Accuracy Evaluation ◽

Mangrove Forests ◽

Forest Mapping ◽

Google Earth Engine ◽

Sentinel 2

Mangrove forest is considered as one of the pivotal ecosystems to near-shore environment health, adjacent terrestrial ecosystems and even global climate change migration. However, for past two decades, they are declining rapidly. In order to take effective steps to prevent the extinction of mangroves, high spatial resolution information of large-scale mangrove distribution is urgent. Recent study has indicated that a suitable pixel size for extracting mangroves should be at least equal to 10 m. Hence, Sentinel imagery (Sentinel-1 C-band synthetic aperture radar (SAR) and Sentinel-2 Multi-Spectral Instrument (MSI) imagery) whose spatial resolution is 10 m may hold great potentials to achieve this goal, but there are limited researches investigating it. Therefore, in this study, we will explore the potential of Sentinel imagery to extract mangrove forests in China on the Google Earth Engine platform. Specifically, our study was mainly conducted around 3 questions: (1) Which Sentinel imagery provides a higher accuracy for mangrove forest mapping, Sentinel-1 SAR data or Sentinel-2 multi-spectral data? (2) which combination of features from Sentinel imagery provides the most accurate mangrove forest map? (3) Compared to 30-m resolution mangrove products derived from Landsat imagery, how does 10-m resolution map improve our knowledge about the distribution of mangrove forest in China?&#160;Our results show that: (1) The highest producer&#8217;s accuracies (the reason why using producer&#8217;s accuracy as an accuracy evaluation indicator here is that the omission errors in mangrove forest extent map are much larger than commission errors) of mangrove forest maps derived from Sentinel-1 and Sentinel-2 imagery are 91.76% and 90.39%, respectively, which means that the contributions of Sentinel-1 SAR and Sentinel-2 MSI imagery to mangrove mapping are similar; (2) The highest producer&#8217;s accuracy of mangrove forest map at 10-m resolution is 95.4%. The mangrove forest map with the highest accuracy is obtained by combining quantiles of spectral and backscatter bands, spectral index, and texture index derived from time series of Sentinel-1 and Sentinel-2 imagery, indicating that the combination of Sentinel-1 SAR and Sentinel-2 MSI imagery is more useful in mangrove forest mapping than using them separately; (3) In China, the total area of mangrove forest extent at 10-m resolution is similar to that at 30-m resolution (20003 ha vs. 19220 ha). However, compared to 30-m resolution mangrove products, the 10-m resolution mangrove map identifies 1741 ha (occupying 8.7% of total mangrove forest area in China) mangrove forests in size smaller than 1 ha, which are especially important to low-lying coastal zone. This study demonstrates the feasibility of Sentinel imagery in large-scale mangrove forest mapping and gives guidance to map global mangrove forest at 10-m resolution in the future. &#160;&#160;

Download Full-text

Large-Scale High-Resolution Coastal Mangrove Forests Mapping Across West Africa With Machine Learning Ensemble and Satellite Big Data

Frontiers in Earth Science ◽

10.3389/feart.2020.560933 ◽

2021 ◽

Vol 8 ◽

Author(s):

Xue Liu ◽

Temilola E. Fatoyinbo ◽

Nathan M. Thomas ◽

Weihe Wendy Guan ◽

Yanni Zhan ◽

...

Keyword(s):

Machine Learning ◽

High Resolution ◽

West Africa ◽

Spatial Resolution ◽

Mangrove Forest ◽

Large Scale ◽

Google Earth ◽

Mangrove Forests ◽

Data Products ◽

Sentinel 2

Coastal mangrove forests provide important ecosystem goods and services, including carbon sequestration, biodiversity conservation, and hazard mitigation. However, they are being destroyed at an alarming rate by human activities. To characterize mangrove forest changes, evaluate their impacts, and support relevant protection and restoration decision making, accurate and up-to-date mangrove extent mapping at large spatial scales is essential. Available large-scale mangrove extent data products use a single machine learning method commonly with 30 m Landsat imagery, and significant inconsistencies remain among these data products. With huge amounts of satellite data involved and the heterogeneity of land surface characteristics across large geographic areas, finding the most suitable method for large-scale high-resolution mangrove mapping is a challenge. The objective of this study is to evaluate the performance of a machine learning ensemble for mangrove forest mapping at 20 m spatial resolution across West Africa using Sentinel-2 (optical) and Sentinel-1 (radar) imagery. The machine learning ensemble integrates three commonly used machine learning methods in land cover and land use mapping, including Random Forest (RF), Gradient Boosting Machine (GBM), and Neural Network (NN). The cloud-based big geospatial data processing platform Google Earth Engine (GEE) was used for pre-processing Sentinel-2 and Sentinel-1 data. Extensive validation has demonstrated that the machine learning ensemble can generate mangrove extent maps at high accuracies for all study regions in West Africa (92%–99% Producer’s Accuracy, 98%–100% User’s Accuracy, 95%–99% Overall Accuracy). This is the first-time that mangrove extent has been mapped at a 20 m spatial resolution across West Africa. The machine learning ensemble has the potential to be applied to other regions of the world and is therefore capable of producing high-resolution mangrove extent maps at global scales periodically.

Download Full-text

COLONIZATION OF MANGROVE FOREST AT ABANDONED SHRIMP-POND OF SEGARA ANAKAN-CILACAP

Jurnal Teknosains ◽

10.22146/teknosains.6050 ◽

2015 ◽

Vol 4 (1) ◽

Author(s):

Tjut Sugandawaty Djohan

Keyword(s):

Mangrove Forest ◽

Mangrove Ecosystem ◽

Mangrove Forests ◽

Shrimp Ponds ◽

Shrimp Pond ◽

Mangrove Tree ◽

Rhizophora Apiculata ◽

Acanthus Ilicifolius ◽

Avicennia Alba ◽

Mangrove Trees

During the 1996 to 1997, large areas of mangrove forest in the Segara Anakan were cleared and converted into intensive shrimp-ponds. After one to two years, these shrimp-ponds failed and were abandoned. These abandoned ponds created large gap areas and canopy gaps, which were colonized by mangrove shrub and liana. The Segara Anakan mangrove also experienced heavy siltation, and there were tree cuttings from the remnant of the mangrove trees. This research aimed to study the colonization of mangrove vegetation at the abandoned-shrimp pond. Vegetation data were collected using rectangular plots of 25 m x 25 m with 4 replicates. The water qualities were also studied. The results revealed that the mangrove forests were composed of two layers: canopy tree and floor-vegetation. The gap areas triggered the pioneer species of mangrove shrubs and liana, Acanthus ilicifolius and Derris heterophylla, to colonize and dominate 100% of the mangrove forest floor. The mangrove trees consisted of natural and planted tree species. The natural trees were Sonneratia alba, Avicennia alba, and saplings of Aegiceras corniculatum, which varied between 56 – 136, 4, and 4 individuals per ha, respectively. The planted trees were Rhizophora apiculata, which amounted to 4 – 12 individuals per ha, at the island of the ponds. These trees and saplings were entangled by the liana mangrove, which disturbed their growth. The A. ilicifolius and D. heterophylla prevented the mangrove tree propagules to grow, and they colonized and characterized those abandoned shrimp-ponds, which threatened the Segara Anakan mangrove ecosystem.

Download Full-text

Using Continuous Change Detection and Classification of Landsat Data to Investigate Long-Term Mangrove Dynamics in the Sundarbans Region

Remote Sensing ◽

10.3390/rs11232833 ◽

2019 ◽

Vol 11 (23) ◽

pp. 2833 ◽

Cited By ~ 3

Author(s):

Katie Awty-Carroll ◽

Pete Bunting ◽

Andy Hardy ◽

Gemma Bell

Keyword(s):

Climate Change ◽

Change Detection ◽

Mangrove Forest ◽

Mangrove Forests ◽

Landsat Data ◽

Continuous Change ◽

Sea Levels ◽

Heritiera Fomes ◽

The Sundarbans

Mangrove forests play a global role in providing ecosystem goods and services in addition to acting as carbon sinks, and are particularly vulnerable to climate change effects such as rising sea levels and increased salinity. For this reason, accurate long-term monitoring of mangrove ecosystems is vital. However, these ecosystems are extremely dynamic and data frequency is often reduced by cloud cover. The Continuous Change Detection and Classification (CCDC) method has the potential to overcome this by utilising every available observation on a per-pixel basis to build stable season-trend models of the underlying phenology. These models can then be used for land cover classification and to determine greening and browning trends. To demonstrate the utility of this approach, CCDC was applied to a 30-year time series of Landsat data covering an area of mangrove forest known as the Sundarbans. Spanning the delta formed by the confluence of the Ganges, Brahmaputra and Meghna river systems, the Sundarbans is the largest contiguous mangrove forest in the world. CCDC achieved an overall classification accuracy of 94.5% with a 99% confidence of being between 94.2% and 94.8%. Results showed that while mangrove extent in the Sundarbans has remained stable, around 25% of the area experienced an overall negative trend, probably due to the effect of die-back on Heritiera fomes. In addition, dates and magnitudes of change derived from CCDC were used to investigate damage and recovery from a major cyclone; 11% of the Sundarbans was found to have been affected by Cyclone Sidr in 2007, 47.6% of which had not recovered by mid-2018. The results indicate that while the Sundarbans forest is resilient to cyclone events, the long-term degrading effects of climate change could reduce this resilience to critical levels. The proposed methodology, while computationally expensive, also offers means by which the full Landsat archive can be analyzed and interpreted and should be considered for global application to mangrove monitoring.

Download Full-text

Deep Learning and Transfer Learning applied to Sentinel-1 DInSAR and Sentinel-2 optical satellite imagery for change detection

2020 International SAUPEC/RobMech/PRASA Conference ◽

10.1109/saupec/robmech/prasa48453.2020.9041139 ◽

2020 ◽

Author(s):

Zainoolabadien Karim ◽

Terence van Zyl

Keyword(s):

Deep Learning ◽

Change Detection ◽

Transfer Learning ◽

Satellite Imagery ◽

Sentinel 2

Download Full-text

Destruction of mangroves for shrimp ponds.

10.1079/ac.85780.20203483576 ◽

2020 ◽

Author(s):

Md. Abdus Shahid

Keyword(s):

Mangrove Forest ◽

Soil Acidity ◽

Coastal Region ◽

Shrimp Farming ◽

Mangrove Forests ◽

Salt Tolerant ◽

Shrimp Culture ◽

Shrimp Ponds ◽

Plant Life ◽

Vast Area

Abstract The Bangladesh coastal region located at the apex of the Bay of Bengal is endowed with a vast area of mangrove forests covering about 6,900 km2. Mangroves are a group of salt-tolerant trees and other plant species which thrive in the inter-tidal zones of sheltered tropical shores, islands and estuaries. They have specially adapted aerial and salt-filtering roots and salt-excreting leaves that enable them to occupy the saline wetlands where other plant life cannot survive. This forest ecosystem plays an important role for coastal inshore as well as offshore fisheries as it provides food and nursery grounds. Unfortunately, mangrove forests are being denuded for shrimp culture. An important example of mangrove denudation is the Chakaria Sundarban where 8,540 ha of mangrove forest has been encroached upon by shrimp farming. The production of shrimp in the mangrove-cleared area has gradually decreased due to increasing soil acidity and decreasing mangrove litter fall in the area. During the period 1975-2001, a total of 9,734 ha of mangrove forest (less than 2% of the total mangrove forest area of the country) was found to be denuded due to shrimp farming along the Bangladesh coast. Although denudation of mangrove forests is negligible compared to other countries of the region, it is creating environmental problems in the coastal region of Bangladesh.

Download Full-text

Semi-Automatic Methodology for Fire Break Maintenance Operations Detection with Sentinel-2 Imagery and Artificial Neural Network

Remote Sensing ◽

10.3390/rs12060909 ◽

2020 ◽

Vol 12 (6) ◽

pp. 909 ◽

Cited By ~ 4

Author(s):

João E. Pereira-Pires ◽

Valentine Aubard ◽

Rita A. Ribeiro ◽

José M. Fonseca ◽

João M. N. Silva ◽

...

Keyword(s):

Neural Network ◽

Artificial Neural Network ◽

Change Detection ◽

Large Network ◽

Spectral Indices ◽

Burnt Area ◽

Object Based ◽

Artificial Neural ◽

Definition Of ◽

Sentinel 2

The difficult job of fighting fires and the nearly impossible task to stop a wildfire without great casualties requires an imperative implementation of proactive strategies. These strategies must decrease the number of fires, the burnt area and create better conditions for the firefighting. In this line of action, the Portuguese Institute of Nature and Forest Conservation defined a fire break network (FBN), which helps controlling wildfires. However, these fire breaks are efficient only if they are correctly maintained, which should be ensured by the local authorities and requires verification from the national authorities. This is a fastidious task since they have a large network of thousands of hectares to monitor over a full year. With the increasing quality and frequency of the Earth Observation Satellite imagery with Sentinel-2 and the definition of the FBN, a semi-automatic remote sensing methodology is proposed in this article for the detection of maintenance operations in a fire break. The proposed methodology is based on a time-series analysis, an object-based classification and a change detection process. The change detection is ensured by an artificial neural network, with reflectance bands and spectral indices as features. Additionally, an analysis of several bands and spectral indices is presented to show the behaviour of the data during a full year and in the presence of a maintenance operation. The proposed methodology achieved a relative error lower than 4% and a recall higher than 75% on the detection of maintenance operations.

Download Full-text

A New Vegetation Index to Detect Periodically Submerged Mangrove Forest Using Single-Tide Sentinel-2 Imagery

Remote Sensing ◽

10.3390/rs11172043 ◽

2019 ◽

Vol 11 (17) ◽

pp. 2043 ◽

Cited By ~ 26

Author(s):

Jia ◽

Wang ◽

Mao ◽

Zhang

Keyword(s):

Mangrove Forest ◽

Vegetation Index ◽

Near Infrared ◽

Aquatic Vegetation ◽

Vegetation Indices ◽

Short Wave ◽

Mangrove Forests ◽

Red Edge ◽

Jensen Shannon Divergence ◽

Sentinel 2

Mangrove forests are tropical trees and shrubs that grow in sheltered intertidal zones. Accurate mapping of mangrove forests is a great challenge for remote sensing because mangroves are periodically submerged by tidal floods. Traditionally, multi-tides images were needed to remove the influence of water; however, such images are often unavailable due to rainy climates and uncertain local tidal conditions. Therefore, extracting mangrove forests from a single-tide imagery is of great importance. In this study, reflectance of red-edge bands in Sentinel-2 imagery were utilized to establish a new vegetation index that is sensitive to submerged mangrove forests. Specifically, red and short-wave near infrared bands were used to build a linear baseline; the average reflectance value of four red-edge bands above the baseline is defined as the Mangrove Forest Index (MFI). To evaluate MFI, capabilities of detecting mangrove forests were quantitatively assessed between MFI and four widely used vegetation indices (VIs). Additionally, the practical roles of MFI were validated by applying it to three mangrove forest sites globally. Results showed that: (1) theoretically, Jensen–Shannon divergence demonstrated that a submerged mangrove forest and water pixels have the largest distance in MFI compared to other VIs. In addition, the boxplot showed that all submerged mangrove forests could be separated from the water background in the MFI image. Furthermore, in the MFI image, to separate mangrove forests and water, the threshold is a constant that is equal to zero. (2) Practically, after applying the MFI to three global sites, 99–102% of submerged mangrove forests were successfully extracted by MFI. Although there are still some uncertainties and limitations, the MFI offers great benefits in accurately mapping mangrove forests as well as other coastal and aquatic vegetation worldwide.

Download Full-text

Deep/Transfer Learning with Feature Space Ensemble Networks (FeatSpaceEnsNets) and Average Ensemble Networks (AvgEnsNets) for Change Detection Using DInSAR Sentinel-1 and Optical Sentinel-2 Satellite Data Fusion

Remote Sensing ◽

10.3390/rs13214394 ◽

2021 ◽

Vol 13 (21) ◽

pp. 4394

Author(s):

Zainoolabadien Karim ◽

Terence L. van Zyl

Keyword(s):

Deep Learning ◽

Change Detection ◽

Transfer Learning ◽

Binary Classification ◽

Feature Space ◽

Fine Tuning ◽

Urban Change ◽

Random Initialization ◽

Optical Imagery ◽

Sentinel 2

Differential interferometric synthetic aperture radar (DInSAR), coherence, phase, and displacement are derived from processing SAR images to monitor geological phenomena and urban change. Previously, Sentinel-1 SAR data combined with Sentinel-2 optical imagery has improved classification accuracy in various domains. However, the fusing of Sentinel-1 DInSAR processed imagery with Sentinel-2 optical imagery has not been thoroughly investigated. Thus, we explored this fusion in urban change detection by creating a verified balanced binary classification dataset comprising 1440 blobs. Machine learning models using feature descriptors and non-deep learning classifiers, including a two-layer convolutional neural network (ConvNet2), were used as baselines. Transfer learning by feature extraction (TLFE) using various pre-trained models, deep learning from random initialization, and transfer learning by fine-tuning (TLFT) were all evaluated. We introduce a feature space ensemble family (FeatSpaceEnsNet), an average ensemble family (AvgEnsNet), and a hybrid ensemble family (HybridEnsNet) of TLFE neural networks. The FeatSpaceEnsNets combine TLFE features directly in the feature space using logistic regression. AvgEnsNets combine TLFEs at the decision level by aggregation. HybridEnsNets are a combination of FeatSpaceEnsNets and AvgEnsNets. Several FeatSpaceEnsNets, AvgEnsNets, and HybridEnsNets, comprising a heterogeneous mixture of different depth and architecture models, are defined and evaluated. We show that, in general, TLFE outperforms both TLFT and classic deep learning for the small dataset used and that larger ensembles of TLFE models do not always improve accuracy. The best performing ensemble is an AvgEnsNet (84.862%) comprised of a ResNet50, ResNeXt50, and EfficientNet B4. This was matched by a similarly composed FeatSpaceEnsNet with an F1 score of 0.001 and variance of 0.266 less. The best performing HybridEnsNet had an accuracy of 84.775%. All of the ensembles evaluated outperform the best performing single model, ResNet50 with TLFE (83.751%), except for AvgEnsNet 3, AvgEnsNet 6, and FeatSpaceEnsNet 5. Five of the seven similarly composed FeatSpaceEnsNets outperform the corresponding AvgEnsNet.

Download Full-text