Weakly Supervised Spatial Deep Learning for Earth Image Segmentation Based on Imperfect Polyline Labels

In recent years, deep learning has achieved tremendous success in image segmentation for computer vision applications. The performance of these models heavily relies on the availability of large-scale high-quality training labels (e.g., PASCAL VOC 2012). Unfortunately, such large-scale high-quality training data are often unavailable in many real-world spatial or spatiotemporal problems in earth science and remote sensing (e.g., mapping the nationwide river streams for water resource management). Although extensive efforts have been made to reduce the reliance on labeled data (e.g., semi-supervised or unsupervised learning, few-shot learning), the complex nature of geographic data such as spatial heterogeneity still requires sufficient training labels when transferring a pre-trained model from one region to another. On the other hand, it is often much easier to collect lower-quality training labels with imperfect alignment with earth imagery pixels (e.g., through interpreting coarse imagery by non-expert volunteers). However, directly training a deep neural network on imperfect labels with geometric annotation errors could significantly impact model performance. Existing research that overcomes imperfect training labels either focuses on errors in label class semantics or characterizes label location errors at the pixel level. These methods do not fully incorporate the geometric properties of label location errors in the vector representation. To fill the gap, this article proposes a weakly supervised learning framework to simultaneously update deep learning model parameters and infer hidden true vector label locations. Specifically, we model label location errors in the vector representation to partially reserve geometric properties (e.g., spatial contiguity within line segments). Evaluations on real-world datasets in the National Hydrography Dataset (NHD) refinement application illustrate that the proposed framework outperforms baseline methods in classification accuracy.

ALWAES

For an online delivery platform, accurate physical locations of merchants are essential for delivery scheduling. It is challenging to maintain tens of thousands of merchant locations accurately because of potential errors introduced by merchants for profits (e.g., potential fraud). In practice, a platform periodically sends a dedicated crew to survey limited locations due to high workforce costs, leaving many potential location errors. In this paper, we design and implement ALWAES, a system that automatically identifies and corrects location errors based on fundamental tradeoffs of five measurement strategies from manual, physical, and virtual data collection infrastructures for online delivery platforms. ALWAES explores delivery data already collected by platform infrastructures to measure the travel time of couriers between merchants and verify all merchants' locations by cross-validation automatically. We explore tradeoffs between performance and cost of different measurement approaches. By comparing with the manually-collected ground truth, the experimental results show that ALWAES outperforms three other baselines by 32.2%, 41.8%, and 47.2%, respectively. More importantly, ALWAES saves 3,846 hours of the delivery time of 35,005 orders in a month and finds new erroneous locations that initially were not in the ground truth but are verified by our field study later, accounting for 3% of all merchants with erroneous locations.

Mobility-prediction and energy optimization for multi-channel multi-interface ad hoc networks in the presence of location errors

We present a mobility-prediction and energy optimization solution for multi-channel multi-interface (MCMI) ad hoc networks in the presence of location errors. This solution includes routing of the MCMI communication links that adapt to dynamic channel, traffic conditions, interference and mobility of nodes. We start first with implementing a novel cross-layer routing solution in order to share information between network and MAC layer, the benefit of this technique is to collect information about the channel quality and residual energy of the nodes and send them directly to the network layer. Next, we present a mobility-prediction model using Kalman filter to predict accurate locations and enhance routing performance, through estimating link duration and selecting reliable routes. The performance of proposed mechanism is measured using NS2.35 simulations with different scenarios and varying load in a network. Comparative analysis of simulation results shows better performance of our protocol (ME-MCMI AODV) in terms of reducing end-to-end delay, total dropped packets and increasing network lifetime and packet delivery ratio (PDR).

Estimation of earthquakes location errors distribution for LUMINEOS local seismic network in Poland.

<p>Estimation of hypocenter location errors  is not a simple task. These errors are influenced by many factors. The most important are: the quality of velocity model, the configuration of stations in the observation network and the noise level recorded at stations. While the network configuration affects the error distribution in a deterministic manner, the noise level is largely random. It means that the uncertainties cannot be determined in a deterministic way and only statistical approach can be used. There are several methods for estimating location errors for particular seismic network. Some techniques use synthetic seismograms to calculate the detection range related to each station. However, this approach requires very precise knowledge of the geological model, which is not always possible. Instead, in this work we present a different approach, which uses only phase data for events included in the catalog. In this method, the detection range for each station is estimated using the detection probability (Schorlemmer & Woessner, 2008) used for both P- and S- waves first arrivals. The usefulness of this approach is discussed assuming the shape of  LUMINEOS seismic network which operates in the Legnica-Głogów Copper District (LGCD), Poland. In the LGCD region seismic activity is related to three deep underground copper mines. Every year thousand of seismic events with magnitudes up to M4.0 are registered here. Some of them are followed by tragic mining collapses and are widely felt by local residents.</p>

A Guide to Pre-Processing High-Throughput Animal Tracking Data

Modern, high-throughput animal tracking studies collect increasingly large volumes of data at very fine temporal scales. At these scales, location error can exceed the animal step size, confounding inferences from tracking data. Cleaning the data to exclude positions with large location errors prior to analyses is one of the main ways movement ecologists deal with location errors. Cleaning data to reduce location error before making biological inferences is widely recommended, and ecologists routinely consider cleaned data to be the ground-truth. Nonetheless, uniform guidance on this crucial step is scarce. Cleaning high-throughput data must strike a balance between rejecting location errors without discarding valid animal movements. Additionally, users of high-throughput systems face challenges resulting from the high volume of data itself, since processing large data volumes is computationally intensive and difficult without a common set of efficient tools. Furthermore, many methods that cluster movement tracks for ecological inference are based on statistical phenomena, and may not be intuitive to understand in terms of the tracked animal biology. In this article we introduce a pipeline to pre-process high-throughput animal tracking data in order to prepare it for subsequent analysis. We demonstrate this pipeline on simulated movement data to which we have randomly added location errors. We further suggest how large volumes of cleaned data may be synthesized into biologically meaningful residence patches. We then use calibration data to show how the pipeline improves its quality, and to verify that the residence patch synthesis accurately captures animal space-use. Finally, turning to real tracking data from Egyptian fruit bats (Rousettus aegyptiacus), we demonstrate the pre-processing pipeline and residence patch method in a fully worked out example. To help with fast implementations of our pipeline, and to help standardise methods, we developed the R package atlastools, which we introduce here. Our pre-processing pipeline and atlastools can be used with any high-throughput animal movement data in which the high data volume combined with knowledge of the tracked individuals biology can be used to reduce location errors. The use of common pre-processing steps that are simple yet robust promotes standardised methods in the field of movement ecology and better inferences from data.

Mitigation of Geostationary Lightning Mapper Geolocation Errors

The geolocation of lightning flashes observed by spaceborne optical sensors depends upon a priori assumptions of the cloud-top height (or, more generally, the height of the radiant emitter) as observed by the satellite. Lightning observations from the Geostationary Lightning Mappers (GLMs) on Geostationary Operational Environmental Satellite 16 (GOES-16) and GOES-17 were originally geolocated by assuming that the global cloud-top height can be modeled as an ellipsoidal surface with an altitude of 16 km at the equator and sloping down to 6 km at the poles. This method produced parallax errors of 20–30 km or more near the limb, where GLM can detect side-cloud illumination or below-cloud lightning channels at lower altitudes than assumed by the ellipsoid. Based on analysis of GLM location accuracy using a suite of alternate lightning ellipsoids, a lower ellipsoid (14 km at the equator, 6 km at the poles) was implemented in October and December 2018 for GLM-16 and GLM-17, respectively. While the lower ellipsoid slightly improves overall GLM location accuracy, parallax-related errors remain, particularly near the limb. This study describes the identification of optimized assumed emitter heights, defined as those that produce the closest agreement with the ground-based reference networks. Derived using the first year of observations from GOES-East position, the optimal emitter height varies geographically and seasonally in a manner consistent with known meteorological regimes. Application of the optimal emitter height approximately doubles the fraction of area near the limb for which peak location errors are less than half a GLM pixel.