Accounting for Training Data Error in Machine Learning Applied to Earth Observations

Mapping the diversity of land uses following deforestation across Africa

African forest are increasingly in decline as a result of land-use conversion due to human activities. However, a consistent and detailed characterization and mapping of land-use change that results in forest loss is not available at the spatial-temporal resolution and thematic levels suitable for decision-making at the local and regional scales; so far they have only been provided on coarser scales and restricted to humid forests. Here we present the first high-resolution (5 m) and continental-scale mapping of land use following deforestation in Africa, which covers an estimated 13.85% of the global forest area, including humid and dry forests. We use reference data for 15 different land-use types from 30 countries and implement an active learning framework to train a deep learning model for predicting land-use following deforestation with an F1-score of [Formula: see text] for the whole of Africa. Our results show that the causes of forest loss vary by region. In general, small-scale cropland is the dominant driver of forest loss in Africa, with hotspots in Madagascar and DRC. In addition, commodity crops such as cacao, oil palm, and rubber are the dominant drivers of forest loss in the humid forests of western and central Africa, forming an "arc of commodity crops" in that region. At the same time, the hotspots for cashew are found to increasingly dominate in the dry forests of both western and south-eastern Africa, while larger hotspots for large-scale croplands were found in Nigeria and Zambia. The increased expansion of cacao, cashew, oil palm, rubber, and large-scale croplands observed in humid and dry forests of western and south-eastern Africa suggests they are vulnerable to future land-use changes by commodity crops, thus creating challenges for achieving the zero deforestation supply chains, support REDD+ initiatives, and towards sustainable development goals.

A global land cover training dataset from 1984 to 2020

State-of-the-art cloud computing platforms such as Google Earth Engine (GEE) enable regional-to-global land cover and land cover change mapping with machine learning algorithms. However, collection of high-quality training data, which is necessary for accurate land cover mapping, remains costly and labor-intensive. To address this need, we created a global database of nearly 2 million training units spanning the period from 1984 to 2020 for seven primary and nine secondary land cover classes. Our training data collection approach leveraged GEE and machine learning algorithms to ensure data quality and biogeographic representation. We sampled the spectral-temporal feature space from Landsat imagery to efficiently allocate training data across global ecoregions and incorporated publicly available and collaborator-provided datasets to our database. To reflect the underlying regional class distribution and post-disturbance landscapes, we strategically augmented the database. We used a machine learning-based cross-validation procedure to remove potentially mis-labeled training units. Our training database is relevant for a wide array of studies such as land cover change, agriculture, forestry, hydrology, urban development, among many others.

Climate change and coastal morphodynamics: Interactions on regional scales

Climate change and its impacts, combined with unchecked human activities, intensify pressures on coastal environments, resulting in modification of the coastal morphodynamics. Coastal zones are intricate and constantly changing areas, making the monitoring and interpretation of data a challenging task, especially in remote beaches and regions with limited historical data. Traditionally, remote sensing and numerical methods have played a vital role in analysing earth observation data and supporting the monitoring and modelling of complex coastal ecosystems. However, the emergence of artificial intelligence-based techniques has shown promising results, offering the additional advantage of filling data gaps, predicting data in data-scarce regions, and analysing multidimensional datasets collected over extended periods of time and larger spatial scales. The main objective of this study is to provide a comprehensive review of the existing literature, discussing both traditional methods and various emerging artificial intelligence-based approaches used in studying the coastal dynamics, shoreline change analysis, and coastal monitoring. Ultimately, the study proposes a climate resilience framework to enhance coastal zone management practices and policies, fostering resilience among coastal communities. The outcome of this study aligns with and supports particularly SDG 13 of the UN (Climate Action) and advances it by identifying relevant methods in coastal erosion studies and proposing integrated management plans informed by real-time data collection and analysis/modelling using physics-based models.

The use of drones for mosquito surveillance and control

In recent years, global health security has been threatened by the geographical expansion of vector-borne infectious diseases such as malaria, dengue, yellow fever, Zika and chikungunya. For a range of these vector-borne diseases, an increase in residual (exophagic) transmission together with ecological heterogeneity in everything from weather to local human migration and housing to mosquito species' behaviours presents many challenges to effective mosquito control. The novel use of drones (or uncrewed aerial vehicles) may play a major role in the success of mosquito surveillance and control programmes in the coming decades since the global landscape of mosquito-borne diseases and disease dynamics fluctuates frequently and there could be serious public health consequences if the issues of insecticide resistance and outdoor transmission are not adequately addressed. For controlling both aquatic and adult stages, for several years now remote sensing data have been used together with predictive modelling for risk, incidence and detection of transmission hot spots and landscape profiles in relation to mosquito-borne pathogens. The field of drone-based remote sensing is under continuous change due to new technology development, operation regulations and innovative applications. In this review we outline the opportunities and challenges for integrating drones into vector surveillance (i.e. identification of breeding sites or mapping micro-environmental composition) and control strategies (i.e. applying larval source management activities or deploying genetically modified agents) across the mosquito life-cycle. We present a five-step systematic environmental mapping strategy that we recommend be undertaken in locations where a drone is expected to be used, outline the key considerations for incorporating drone or other Earth Observation data into vector surveillance and provide two case studies of the advantages of using drones equipped with multispectral cameras. In conclusion, recent developments mean that drones can be effective for accurately conducting surveillance, assessing habitat suitability for larval and/or adult mosquitoes and implementing interventions. In addition, we briefly discuss the need to consider permissions, costs, safety/privacy perceptions and community acceptance for deploying drone activities.

Using Deep Learning and Very-High-Resolution Imagery to Map Smallholder Field Boundaries

The mapping of field boundaries can provide important information for increasing food production and security in agricultural systems across the globe. Remote sensing can provide a viable way to map field boundaries across large geographic extents, yet few studies have used satellite imagery to map boundaries in systems where field sizes are small, heterogeneous, and irregularly shaped. Here we used very-high-resolution WorldView-3 satellite imagery (0.5 m) and a mask region-based convolutional neural network (Mask R-CNN) to delineate smallholder field boundaries in Northeast India. We found that our models had overall moderate accuracy, with average precision values greater than 0.67 and F1 Scores greater than 0.72. We also found that our model performed equally well when applied to another site in India for which no data were used in the calibration step, suggesting that Mask R-CNN may be a generalizable way to map field boundaries at scale. Our results highlight the ability of Mask R-CNN and very-high-resolution imagery to accurately map field boundaries in smallholder systems.

RID—Roof Information Dataset for Computer Vision-Based Photovoltaic Potential Assessment

Computer vision has great potential to accelerate the global scale of photovoltaic potential analysis by extracting detailed roof information from high-resolution aerial images, but the lack of existing deep learning datasets is a major barrier. Therefore, we present the Roof Information Dataset for semantic segmentation of roof segments and roof superstructures. We assessed the label quality of initial roof superstructure annotations by conducting an annotation experiment and identified annotator agreements of 0.15–0.70 mean intersection over union, depending on the class. We discuss associated the implications on the training and evaluation of two convolutional neural networks and found that the quality of the prediction behaved similarly to the annotator agreement for most classes. The class photovoltaic module was predicted to be best with a class-specific mean intersection over union of 0.69. By providing the datasets in initial and reviewed versions, we promote a data-centric approach for the semantic segmentation of roof information. Finally, we conducted a photovoltaic potential analysis case study and demonstrated the high impact of roof superstructures as well as the viability of the computer vision approach to increase accuracy. While this paper’s primary use case was roof information extraction for photovoltaic potential analysis, its implications can be transferred to other computer vision applications in remote sensing and beyond.

An Overview of the Applications of Earth Observation Satellite Data: Impacts and Future Trends

As satellite observation technology develops and the number of Earth observation (EO) satellites increases, satellite observations have become essential to developments in the understanding of the Earth and its environment. However, the current impacts to the remote sensing community of different EO satellite data and possible future trends of EO satellite data applications have not been systematically examined. In this paper, we review the impacts of and future trends in the use of EO satellite data based on an analysis of data from 15 EO satellites whose data are widely used. Articles that reference EO satellite missions included in the Web of Science core collection for 2020 were analyzed using scientometric analysis and meta-analysis. We found the following: (1) the number of publications and citations referencing EO satellites is increasing exponentially; however, the number of articles referencing AVHRR, SPOT, and TerraSAR is tending to decrease; (2) papers related to EO satellites are concentrated in a small number of journals: 43.79% of the articles that were reviewed were published in only 13 journals; and (3) remote sensing impact factor (RSIF), a new impact index, was constructed to measure the impacts of EO satellites and to predict future trends in applications of their data. Landsat, Sentinel, MODIS, Gaofen, and WorldView were found to be the most significant current EO satellite missions and MODIS data to have the widest range of applications. Over the next five years (2021–2025), it is expected that Sentinel will become the satellite mission with the greatest influence.

High Resolution, Annual Maps of Field Boundaries for Smallholder-Dominated Croplands at National Scales

Mapping the characteristics of Africa’s smallholder-dominated croplands, including the sizes and numbers of fields, can provide critical insights into food security and a range of other socioeconomic and environmental concerns. However, accurately mapping these systems is difficult because there is 1) a spatial and temporal mismatch between satellite sensors and smallholder fields, and 2) a lack of high-quality labels needed to train and assess machine learning classifiers. We developed an approach designed to address these two problems, and used it to map Ghana’s croplands. To overcome the spatio-temporal mismatch, we converted daily, high resolution imagery into two cloud-free composites (the primary growing season and subsequent dry season) covering the 2018 agricultural year, providing a seasonal contrast that helps to improve classification accuracy. To address the problem of label availability, we created a platform that rigorously assesses and minimizes label error, and used it to iteratively train a Random Forests classifier with active learning, which identifies the most informative training sample based on prediction uncertainty. Minimizing label errors improved model F1 scores by up to 25%. Active learning increased F1 scores by an average of 9.1% between first and last training iterations, and 2.3% more than models trained with randomly selected labels. We used the resulting 3.7 m map of cropland probabilities within a segmentation algorithm to delineate crop field boundaries. Using an independent map reference sample (n = 1,207), we found that the cropland probability and field boundary maps had respective overall accuracies of 88 and 86.7%, user’s accuracies for the cropland class of 61.2 and 78.9%, and producer’s accuracies of 67.3 and 58.2%. An unbiased area estimate calculated from the map reference sample indicates that cropland covers 17.1% (15.4–18.9%) of Ghana. Using the most accurate validation labels to correct for biases in the segmented field boundaries map, we estimated that the average size and total number of field in Ghana are 1.73 ha and 1,662,281, respectively. Our results demonstrate an adaptable and transferable approach for developing annual, country-scale maps of crop field boundaries, with several features that effectively mitigate the errors inherent in remote sensing of smallholder-dominated agriculture.

Application of Artificial Neural Networks for Mangrove Mapping Using Multi-Temporal and Multi-Source Remote Sensing Imagery

Mangroves, as unique coastal wetlands with numerous benefits, are endangered mainly due to the coupled effects of anthropogenic activities and climate change. Therefore, acquiring reliable and up-to-date information about these ecosystems is vital for their conservation and sustainable blue carbon development. In this regard, the joint use of remote sensing data and machine learning algorithms can assist in producing accurate mangrove ecosystem maps. This study investigated the potential of artificial neural networks (ANNs) with different topologies and specifications for mangrove classification in Iran. To this end, multi-temporal synthetic aperture radar (SAR) and multi-spectral remote sensing data from Sentinel-1 and Sentinel-2 were processed in the Google Earth Engine (GEE) cloud computing platform. Afterward, the ANN topologies and specifications considering the number of layers and neurons, learning algorithm, type of activation function, and learning rate were examined for mangrove ecosystem mapping. The results indicated that an ANN model with four hidden layers, 36 neurons in each layer, adaptive moment estimation (Adam) learning algorithm, rectified linear unit (Relu) activation function, and the learning rate of 0.001 produced the most accurate mangrove ecosystem map (F-score = 0.97). Further analysis revealed that although ANN models were subjected to accuracy decline when a limited number of training samples were used, they still resulted in satisfactory results. Additionally, it was observed that ANN models had a high resistance when training samples included wrong labels, and only the ANN model with the Adam learning algorithm produced an accurate mangrove ecosystem map when no data standardization was performed. Moreover, further investigations showed the higher potential of multi-temporal and multi-source remote sensing data compared to single-source and mono-temporal (e.g., single season) for accurate mangrove ecosystem mapping. Overall, the high potential of the proposed method, along with utilizing open-access satellite images and big-geo data processing platforms (i.e., GEE, Google Colab, and scikit-learn), made the proposed approach efficient and applicable over other study areas for all interested users.

Using satellite imagery to understand and promote sustainable development

Accurate and comprehensive measurements of a range of sustainable development outcomes are fundamental inputs into both research and policy. We synthesize the growing literature that uses satellite imagery to understand these outcomes, with a focus on approaches that combine imagery with machine learning. We quantify the paucity of ground data on key human-related outcomes and the growing abundance and improving resolution (spatial, temporal, and spectral) of satellite imagery. We then review recent machine learning approaches to model-building in the context of scarce and noisy training data, highlighting how this noise often leads to incorrect assessment of model performance. We quantify recent model performance across multiple sustainable development domains, discuss research and policy applications, explore constraints to future progress, and highlight research directions for the field.

Identifying Spatiotemporal Patterns in Land Use and Cover Samples from Satellite Image Time Series

The use of satellite image time series analysis and machine learning methods brings new opportunities and challenges for land use and cover changes (LUCC) mapping over large areas. One of these challenges is the need for samples that properly represent the high variability of land used and cover classes over large areas to train supervised machine learning methods and to produce accurate LUCC maps. This paper addresses this challenge and presents a method to identify spatiotemporal patterns in land use and cover samples to infer subclasses through the phenological and spectral information provided by satellite image time series. The proposed method uses self-organizing maps (SOMs) to reduce the data dimensionality creating primary clusters. From these primary clusters, it uses hierarchical clustering to create subclusters that recognize intra-class variability intrinsic to different regions and periods, mainly in large areas and multiple years. To show how the method works, we use MODIS image time series associated to samples of cropland and pasture classes over the Cerrado biome in Brazil. The results prove that the proposed method is suitable for identifying spatiotemporal patterns in land use and cover samples that can be used to infer subclasses, mainly for crop-types.

Using convolutional neural network for predicting cyanobacteria concentrations in river water

Machine learning modeling techniques have emerged as a potential means for predicting algal blooms. In this study, synthetic spatio-temporal water quality data for a river section were generated with a 3D water quality model and used to investigate the capability of a convolutional neural network (CNN) for predicting harmful cyanobacterial blooms. The CNN model displayed a reasonable capacity for short-term predictions of cyanobacteria (Microcystis) biomass. In the nowcasting of Microcystis, the CNN performance had a Nash-Sutcliffe Efficiency (NSE) of 0.87. An increase in the forecast lead time resulted in a decrease in the prediction accuracy, reducing the NSE from 0.87 to 0.58. As the spatial observation density increased from 20% to 100% of the input image grids, the CNN prediction NSE had improved from 0.70 to 0.84. Adding noise to the data resulted in accuracy deterioration, but even at the noise amplitude of 10%, the accuracy was acceptable for some applications, with NSE = 0.76. Visualization of the CNN results characterized its performance variations across the studied river reach. Overall, this study successfully demonstrated the capability of the CNN model for cyanobacterial bloom prediction using high temporal frequency images.