Remote sensing, or Earth Observation (EO), is increasingly used to understand Earth system dynamics and create continuous and categorical maps of biophysical properties and land cover, especially based on recent advances in machine learning (ML). ML models typically require large, spatially explicit training datasets to make accurate predictions. Training data (TD) are typically generated by digitizing polygons on high spatial-resolution imagery, by collecting in situ data, or by using pre-existing datasets. TD are often assumed to accurately represent the truth, but in practice almost always have error, stemming from (1) sample design, and (2) sample collection errors. The latter is particularly relevant for image-interpreted TD, an increasingly commonly used method due to its practicality and the increasing training sample size requirements of modern ML algorithms. TD errors can cause substantial errors in the maps created using ML algorithms, which may impact map use and interpretation. Despite these potential errors and their real-world consequences for map-based decisions, TD error is often not accounted for or reported in EO research. Here we review the current practices for collecting and handling TD. We identify the sources of TD error, and illustrate their impacts using several case studies representing different EO applications (infrastructure mapping, global surface flux estimates, and agricultural monitoring), and provide guidelines for minimizing and accounting for TD errors. To harmonize terminology, we distinguish TD from three other classes of data that should be used to create and assess ML models: training reference data, used to assess the quality of TD during data generation; validation data, used to iteratively improve models; and map reference data, used only for final accuracy assessment. We focus primarily on TD, but our advice is generally applicable to all four classes, and we ground our review in established best practices for map accuracy assessment literature. EO researchers should start by determining the tolerable levels of map error and appropriate error metrics. Next, TD error should be minimized during sample design by choosing a representative spatio-temporal collection strategy, by using spatially and temporally relevant imagery and ancillary data sources during TD creation, and by selecting a set of legend definitions supported by the data. Furthermore, TD error can be minimized during the collection of individual samples by using consensus-based collection strategies, by directly comparing interpreted training observations against expert-generated training reference data to derive TD error metrics, and by providing image interpreters with thorough application-specific training. We strongly advise that TD error is incorporated in model outputs, either directly in bias and variance estimates or, at a minimum, by documenting the sources and implications of error. TD should be fully documented and made available via an open TD repository, allowing others to replicate and assess its use. To guide researchers in this process, we propose three tiers of TD error accounting standards. Finally, we advise researchers to clearly communicate the magnitude and impacts of TD error on map outputs, with specific consideration given to the likely map audience.

The shear volumes of data generated from earth observation and remote sensing technologies continue to make major impact; leaping key geospatial applications into the dual data and compute-intensive era. As a consequence, this rapid advancement poses new computational and data processing challenges. We implement a novel remote sensing data flow (RESFlow) for advancing machine learning to compute with massive amounts of remotely sensed imagery. The core contribution is partitioning massive amounts of data into homogeneous distributions for fitting simple models. RESFlow takes advantage of Apache Spark and the availability of modern computing hardware to harness the acceleration of deep learning inference on expansive remote sensing imagery. The framework incorporates a strategy to optimize resource utilization across multiple executors assigned to a single worker. We showcase its deployment in both computationally and data-intensive workloads for pixel-level labeling tasks. The pipeline invokes deep learning inference at three stages; during deep feature extraction, deep metric mapping, and deep semantic segmentation. The tasks impose compute-intensive and GPU resource sharing challenges motivating for a parallelized pipeline for all execution steps. To address the problem of hardware resource contention, our containerized workflow further incorporates a novel GPU checkout routine and the ticketing system across multiple workers. The workflow is demonstrated with NVIDIA DGX accelerated platforms and offers appreciable compute speed-ups for deep learning inference on pixel labeling workloads; processing 21 028 TB of imagery data and delivering output maps at area rate of 5.245 sq.km/s, amounting to 453 168 sq.km/day-reducing a 28 day workload to 21 h.

Satellite imagery often exhibits large spatial extent areas that encompass object classes with considerable variability. This often limits large-scale model generalization with machine learning algorithms. Notably, acquisition conditions, including dates, sensor position, lighting condition, and sensor types, often translate into class distribution shifts introducing complex nonlinear factors and hamper the potential impact of machine learning classifiers. This paper investigates the challenge of exploiting satellite images using convolutional neural networks (CNN) for settlement classification where the class distribution shifts are significant. We present a large-scale human settlement mapping workflow based-off multiple modules to adapt a pretrained CNN to address the negative impact of distribution shift on classification performance. To extend a locally trained classifier onto large spatial extents areas we introduce several submodules: First, a human-in-the-loop element for relabeling of misclassified target domain samples to generate representative examples for model adaptation; second, an efficient hashing module to minimize redundancy and noisy samples from the mass-selected examples; and third, a novel relevance ranking module to minimize the dominance of source example on the target domain. The workflow presents a novel and practical approach to achieve large-scale domain adaptation with binary classifiers that are based-off CNN features. Experimental evaluations are conducted on areas of interest that encompass various image characteristics, including multisensors, multitemporal, and multiangular conditions. Domain adaptation is assessed on source–target pairs through the transfer loss and transfer ratio metrics to illustrate the utility of the workflow.

Remote sensing in combination with deep learning has become instrumental for efficiently and accurately classifying land-use and land-cover across large geographic areas. These technologies have also been successful in characterizing urban environments in terms of their structural units, structure types, or morphological regions. In these approaches, an urban area is partitioned into regions that exhibit homogeneous physical characteristics. However, existing approaches are typically limited to a single city, use inconsistent typologies, and lack scalability and generalization capacity. In this study, we propose an urban structural units categorization scheme and demonstrate its utility by applying it to 13 cities. Inspired by the lack of scalability and generalization capacity in urban structural units mapping, we extend the reach of deep learning and conduct a set of classification experiments in all 13 cities. These experiments offer insights into the strengths and limitations of deep neural networks for classifying urban structural units over diverse geographic regions and on heterogeneous collections of satellite imagery. The efficacy of the proposed deep learning approach is compared to a baseline method of multiscale image features and support vector machines. Our validation on five cities shows that better performance is achieved with deep neural networks. Additionally, we evaluate the impact of input size, model depth, and spatial pyramid pooling to assess the generalization capacity of deep neural networks.

As satellite imagery collections continue to grow at an astonishing rate, so is the demand for automated and scalable object detection and segmentation. Scaling computational activities demand models that generalize well across various challenges that can hamper progress, including diverse imaging and geographic conditions, sampling bias in training data, manual ground truth collection, tooling for model reuse and accountability assessment, and poor model training strategies. A great deal of progress has been made on these challenges. We contribute to the improvement through further development of ReSFlow, a workflow that breaks the problem of model generalization into a collection of specialized exploitations. ReSFlow partitions imagery collections into homogeneous buckets equipped with exploitation models trained to perform well under each bucket.s specific context. Essentially, ReSFlow aims for generalization through stratification. Therefore, within a bucket, exploitation is a homogeneous process that mitigates heterogeneity challenges, including the number of training data and data biases that can occur over varied conditions. Furthermore, custom model architectures and rich training strategies effective for within-bucket conditions can be developed. Meanwhile, across buckets, performance metrics support systematic views of the workflow leading to optimal data labeling allocations and indications that further specialization is warranted. Herein, we discuss the formation of models during the framework's “Offline Initialization” stage. Lastly, we exploit the inherent parallelism due to bucketing to introduce model reuse and demonstrate efficacy by reducing an 89-day manual data labeling cost to zero-days in a new area of interest.

Understanding how people occupy open spaces is important for research in support of population modeling, policy, national security, emergency response, and sustainability. For the past decade, there has been an increase in research toward capturing and reporting population dynamics and patterns of life at the building level and in some open public spaces such as cemeteries and parks. This is done through observation models developed from local sociocultural information acquired at various spatiotemporal scales to inform night, day, and episodic population occupancy estimates (people/1000 sq ft). Sociocultural information for cemeteries and parks is scarcely available and often collected manually. The process is not only marred by inconsistencies but is laborious and time consuming. In this study, we leverage convolutional neural networks (CNNs) and satellite imagery to derive grave site counts as proxy variables to support scalable and accurate sociocultural data required in a population observation model. Through a hybrid workflow (weak localization plus regression model), we characterize a large scale automation process to counting of grave sites. We evaluate and demonstrate the efficacy of proposed workflow using out-of-data set large satellite imagery and establish its broader impact on cemetery observation models.