Search

Satellite-based forest alert systems are an important tool for ecosystem monitoring, planning conservation, and increasing public awareness of forest cover change. Continuous monitoring in tropical regions, such as those experiencing pronounced monsoon seasons, can be complicated by spatially extensive and persistent cloud cover. One solution is to use Synthetic Aperture Radar (SAR) imagery acquired by the European Space Agency’s Sentinel-1A and B satellites. The Sentinel 1A and B satellites acquire C-band radar data that penetrates cloud cover and can be acquired during the day or night. One challenge associated with operational use of radar imagery is that the speckle associated with the backscatter values can complicate traditional pixel-based analysis approaches. A potential solution is to use deep learning semantic segmentation models that can capture predictive features that are more robust to pixel-level noise. In this analysis, we present a prototype SAR-based forest alert system that utilizes deep learning classifiers, deployed using the Google Earth Engine cloud computing platform, to identify forest cover change with near real-time classification over two Cambodian wildlife sanctuaries. By leveraging a pre-existing forest cover change dataset derived from multispectral Landsat imagery, we present a method for efficiently developing a SAR-based semantic segmentation dataset. In practice, the proposed framework achieved good performance comparable to an existing forest alert system while offering more flexibility and ease of development from an operational standpoint.

Supporting successful global mangrove conservation and policy requires accurate identification of anthropogenic and biophysical drivers of mangrove extent, yet such studies are scarce. We apply a hybrid methodology, combining existing remote sensing mangrove maps with local expert knowledge of vegetation and land use dynamics. We conducted stratified random sampling in eight subregions, and local experts visually interpreted over 20,900 plots using high-resolution imagery in Collect Earth Online. Similar to previous estimates, we found 147,771 km2 (±1.4%) of mangroves globally in 2020 and that rates of mangrove loss have decreased from 2000–2010 to 2010–2020, largely driven by South and Southeast Asia. Anthropogenic drivers of loss have shifted across subregions, with oil palm cultivation emerging in South and Southeast Asia and aquaculture in South America and Western and Central Africa, highlighting the need for ongoing monitoring and adaptable conservation efforts. Natural expansion outpaced natural retraction in both periods. This is the first global study uncovering land use drivers of mangrove decline and recovery, only made possible by collaboration with local experts. Key breakthroughs include successfully discerning spectrally similar anthropogenic from biophysical drivers, such as aquaculture from natural retraction, and creating data collection approaches that streamline visual interpretation efforts.

Air pollution from burning sugarcane is an important environmental issue in Thailand. Knowing the location and extent of sugarcane plantations would help in formulating effective strategies to reduce burning. High resolution satellite imagery combined with deep-learning technologies can be effective to map sugarcane with high precision. However, land cover mapping using high resolution data and computationally intensive deep-learning networks can be computationally costly. In this study, we used high resolution satellite imagery from Planet that has been made available to the public through the Norway's International Climate and Forest Initiative (NICFI). We tested a U-Net deep-learning algorithm with a lightweight MobileNetV2 network as the encoder branch using the Google Earth Engine computational platform. We trained a model using the RGB channels with pre-trained network (RGBt), a RGB model with randomly initialized weights (RGBr) and a model with randomly initialized weights including the NIR channel (RGBN). We found an F1-score of 0.9550, 0.9262 and 0.9297 for the RGBt, RGBr and RGBN models, respectively. For an independent model evaluation we found F1-scores of 0.9141, 0.8681 and 0.8911. We also found a discrepancy in the recall values reported by the model and those from the independent validation. We found that lightweight deep-learning models produce satisfactory results while providing effective means to apply mapping efforts at scale with reduced computational costs. We highlight the importance of central data repositories with labeled data as pre-trained networks were found to be effective in improving the accuracy.

Land cover maps play an integral role in environmental management. However, countries and institutes encounter many challenges with producing timely, efficient, and temporally harmonized updates to their land cover maps. To address these issues we present a modular Regional Land Cover Monitoring System (RLCMS) architecture that is easily customized to create land cover products using primitive map layers. Primitive map layers are a suite of biophysical and end member maps, with land cover primitives representing the raw information needed to make decisions in a dichotomous key for land cover classification. We present best practices to create and assemble primitives from optical satellite using computing technologies, decision tree logic and Monte Carlo simulations to integrate their uncertainties. The concept is presented in the context of a regional land cover map based on a shared regional typology with 18 land cover classes agreed on by stakeholders from Cambodia, Laos PDR, Myanmar, Thailand, and Vietnam. We created annual map and uncertainty layers for the period 2000–2017. We found an overall accuracy of 94% when taking uncertainties into account. RLCMS produces consistent time series products using free long term historical Landsat and MODIS data. The customizable architecture can include a variety of sensors and machine learning algorithms to create primitives and the best suited smoothing can be applied on a primitive level. The system is transferable to all regions around the globe because of its use of publicly available global data (Landsat and MODIS) and easily adaptable architecture that allows for the incorporation of a customizable assembly logic to map different land cover typologies based on the user's landscape monitoring objectives

Land cover change and its impact on food security is a topic that has major implications for development in population-dense Southeast Asia. The main drivers of forest loss include the expansion of agriculture and plantation estates, growth of urban centers, extraction of natural resources, and water infrastructure development. The design and implementation of appropriate land use policies requires accurate and timely information on land cover dynamics to account for potential political, economical, and agricultural consequences. Therefore, SERVIR-Mekong led the collaborative development of a Regional Land Cover Monitoring System (RLCMS) with key regional stakeholders across the greater Mekong region. Through this effort, a modular system was used to create yearly land cover maps for the period 1988–2017. In this study, we compared this 30-year land cover time-series with Vietnam national forest resources and agricultural productivity statistics. We used remote sensing-derived land cover products to quantify landscape changes and linked those with food availability, one of food security dimension, from a landscape approach perspective. We found that agricultural production has soared while the coverage of agricultural areas has remained relatively stable. Land cover change dynamics coincide with important legislation regarding environmental management and sustainable development strategies in Vietnam. Our findings indicate that Vietnam has made major steps toward improving its' food security. We demonstrate that RLCMS is a valuable tool for evaluating the relationship between policies and their impacts on food security, ecosystem services and natural capital.

While deforestation has traditionally been the focus for forest canopy disturbance detection, forest degradation must not be overlooked. Both deforestation and forest degradation influence carbon loss and greenhouse gas emissions and thus must be included in activity data reporting estimates, such as for the Reduced Emissions from Deforestation and Degradation (REDD+) program. Here, we report on efforts to develop forest degradation mapping capacity in Nepal based on a pilot project in the country’s Terai region, an ecologically complex physiographic area. To strengthen Nepal’s estimates of deforestation and forest degradation, we applied the Continuous Degradation Detection (CODED) algorithm, which uses a time series of the Normalized Degradation Fraction Index (NDFI) to monitor forest canopy disturbances. CODED can detect low-grade degradation events and provides an easy-to-use graphical user interface in Google Earth Engine (GEE). Using an iterative process, we were able to create a model that provided acceptable accuracy and area estimates of forest degradation and deforestation in Terai that can be applied to the whole country. We found that between 2010 and 2020, the area affected by disturbance was substantially larger than the deforested area, over 105,650 hectares compared to 2753 hectares, respectively. Iterating across multiple parameters using the CODED algorithm in the Terai region has provided a wealth of insights not only for detecting forest degradation and deforestation in Nepal in support of activity data estimation but also for the process of using tools like CODED in applied settings. We found that model performance, measured using producer’s and user’s accuracy, varied dramatically based on the model parameters specified. We determined which parameters most altered the results through an iterative process; those parameters are described here in depth. Once CODED is combined with the description of each parameter and how it affects disturbance monitoring in a complex environment, this degradation-sensitive detection process has the potential to be highly attractive to other developing countries in the REDD+ program seeking to accurately monitor their forests.

Land cover maps are a critical component to make informed policy, development, planning, and resource management decisions. However, technical, capacity, and institutional challenges inhibit the creation of consistent and relevant land cover maps for use in developing regions. Many developing regions lack coordinated capacity, infrastructure, and technologies to produce a robust land cover monitoring system that meets land management needs. Local capacity may be replaced by external consultants or methods which lack long-term sustainability. In this study, we characterize and respond to the key land cover mapping gaps and challenges encountered in the Lower Mekong (LMR) and Hindu Kush-Himalaya (HKH) region through a needs assessment exercise and a collaborative system design. Needs were assessed using multiple approaches, including focus groups, user engagement workshops, and online surveys. Efforts to understand existing limitations and stakeholder needs resulted in a co-developed and modular land cover monitoring system which utilizes state-of-the-art cloud computing and machine learning which leverages freely available Earth observations. This approach meets the needs of diverse actors and is a model for transnational cooperation.

Time series land cover data statistics often fluctuate abruptly due to seasonal impact and other noise in the input image. Temporal smoothing techniques are used to reduce the noise in time series data used in land cover mapping. The effects of smoothing may vary based on the smoothing method and land cover category. In this study, we compared the performance of Fourier transformation smoothing, Whittaker smoother and Linear-Fit averaging smoother on Landsat 5, 7 and 8 based yearly composites to classify land cover in Province No. 1 of Nepal. The performance of each smoother was tested based on whether it was applied on image composites or on land cover primitives generated using the random forest machine learning method. The land cover data used in the study was from the years 2000 to 2018. Probability distribution was examined to check the quality of primitives and accuracy of the final land cover maps were accessed. The best results were found for the Whittaker smoothing for stable classes and Fourier smoothing for other classes. The results also show that classification using a properly selected smoothing algorithm outperforms a classification based on its unsmoothed data set. The final land cover generated by combining the best results obtained from different smoothing approaches increased our overall land cover map accuracy from 79.18% to 83.44%. This study shows that smoothing can result in a substantial increase in the quality of the results and that the smoothing approach should be carefully considered for each land cover class.

Understanding land cover change dynamics and potential pathways of change is of critical importance for sustainable resource management, to promote food security and resilience on a range of spatial scales. Data scarcity is a key concern, however, with the availability of free Earth Observation (EO) data, such challenges can be suitably addressed. In this research we have developed a robust machine learning (random forest) approach utilizing EO and Geographic Information System (GIS) data, which enables an innovative means for our simulations to be driven only by historical drivers of change and hotspot prediction based on probability to change. We used the Mekong region as a case study to generate a training and validation sample from historical land cover patterns of change and used this information to train a random forest machine learning model. Data samples were created from the SERVIR-Mekong land cover data series. Data sets were created for 2 categories both containing 8 classes. The 2 categories included—any generic class to change into a specific one and vice versa. Classes included the following: Aquaculture; Barren; Cropland; Flooded Forest; Mangroves; Forest; Plantations; Wetlands; and Urban. The training points were used to sample a series of satellite-derived surface reflectance products and other data layers such as information on slope, distance to road and census data, which represent the drivers of change. The classifier was trained in binary mode and showed a clear separation between change and no change. An independent validation dataset of historical change pixels show that all median change probabilities are greater than 80% and all lower quantiles, except one, are greater than 70%. The 2018 probability change maps show high probabilities for the Plantations and Forest classes in the ‘Generic to Specific’ and ’Specific to generic’ category, respectively. A time-series analysis of change probability shows that forests have become more likely to convert into other classes during the last two decades, across all countries. We successfully demonstrated that historical change patters combined with big data and machine learning technologies are powerful tools for predictive change analytics on a planetary scale.

During the last few decades, a large number of people have migrated to Kathmandu city from all parts of Nepal, resulting in rapid expansion of the city. The unplanned and accelerated growth is causing many environmental and population management issues. To manage urban growth efficiently, the city authorities need a means to be able to monitor urban expansion regularly. In this study, we introduced a novel approach to automatically detect urban expansion by leveraging state-of-the-art cloud computing technologies using the Google Earth Engine (GEE) platform. We proposed a new index named Normalized Difference and Distance Built-up Index (NDDBI) for identifying built-up areas by combining the LandSat-derived vegetation index with distances from the nearest roads and buildings analysed from OpenStreetMap (OSM). We also focused on logical consistencies of land-cover change to remove unreasonable transitions supported by the repeat photography. Our analysis of the historical urban growth patterns between 2000 and 2018 shows that the settlement areas were increased from 63.68 sq km in 2000 to 148.53 sq km in 2018. The overall accuracy of mapping the newly-built areas of urban expansion was 94.33%. We have demonstrated that the methodology and data generated in the study can be replicated to easily map built-up areas and support quicker and more efficient land management and land-use planning in rapidly growing cities worldwide.

Forests in Southeast Asia are experiencing some of the highest rates of deforestation and degradation in the world, with natural forest species being replaced by cropland and plantation monoculture. In this work, we have developed an innovative method to accurately map rubber and palm oil plantations using fusion of Landsat-8, Sentinel 1 and 2. We applied cloud and shadow masking, bidirectional reflectance distribution function (BRDF), atmospheric and topographic corrections to the optical imagery and a speckle filter and harmonics for Synthetic Aperture Radar (SAR) data. In this workflow, we created yearly composites for all sensors and combined the data into a single composite. A series of covariates were calculated from optical bands and sampled using reference data of the land cover classes including surface water, forest, urban and built-up, cropland, rubber, palm oil and mangrove. This training dataset was used to create biophysical probability layers (primitives) for each class. These primitives were then used to create land cover and probability maps in a decision tree logic and Monte-Carlo simulations. Validation showed good overall accuracy (84%) for the years 2017 and 2018. Filtering for validation points with high error estimates improved the accuracy up to 91%. We demonstrated and concluded that error quantification is an essential step in land cover classification and land cover change detection. Our overall analysis supports and presents a path for improving present assessments for sustainable supply chain analyses and associated recommendations.

Historical forest management practices in the southwestern US have left forests prone to high-severity, stand-replacement fires. Reducing the cost of forest-fire management and reintroducing fire to the landscape without negative impact depends on detailed knowledge of stand composition, in particular, above-ground biomass (AGB). Lidar-based modeling techniques provide opportunities to increase ability of managers to monitor AGB and other forest metrics at reduced cost. We developed a regional lidar-based statistical model to estimate AGB for Ponderosa pine and mixed conifer forest systems of the southwestern USA, using previously collected field data. Model selection was performed using Bayesian model averaging (BMA) to reduce researcher bias, fully explore the model space, and avoid overfitting. The selected model includes measures of canopy height, canopy density, and height distribution. The model selected with BMA explains 71% of the variability in field-estimates of AGB, and the RMSE of the two independent validation data sets are 23.25 and 32.82 Mg/ha. The regional model is structured in accordance with previously described local models, and performs equivalently to these smaller scale models. We have demonstrated the effectiveness of lidar for developing cost-effective, robust regional AGB models for monitoring and planning adaptively at the landscape scale.