Search

Geoscience applications have been using sophisticated machine learning methods to model complex phenomena. These models are described as black boxes since it is unclear what relationships are learned. Models may exploit spurious associations that exist in the data. The lack of transparency may limit user’s trust, causing them to avoid high performance models since they cannot verify that it has learned realistic strategies. EXplainable Artificial Intelligence (XAI) is a developing research area for investigating how models make their decisions. However, XAI methods are sensitive to feature correlations. This makes XAI challenging for high-dimensional models whose input rasters may have extensive spatial-temporal autocorrelation. Since many geospatial applications rely on complex models for target performance, a recommendation is to combine raster elements into semantically meaningful feature groups. However, it is challenging to determine how best to combine raster elements. Here, we explore the explanation sensitivity to grouping scheme. Experiments are performed on FogNet, a complex deep learning model that uses 3D Convolutional Neural Networks (CNN) for coastal fog prediction. We demonstrate that explanations can be combined with domain knowledge to generate hypotheses about the model. Meteorological analysis of the XAI output reveal FogNet’s use of channels that capture relationships related to fog development, contributing to good overall model performance. However, analyses also reveal several deficiencies, including the reliance on channels and channel spatial patterns that correlate to the predominate fog type in the dataset, to make predictions of all fog types. Strategies to improve FogNet performance and trustworthiness are presented.

HighlightsLeaf nitrogen percentage in corn was estimated using various vegetation indices derived from UAVs.Eight machine learning methods were compared to find the most accurate model for nitrogen estimation.The most influential vegetation indices were determined for estimation of leaf nitrogen.Abstract. Nitrogen (N) is the most critical component of healthy plants. It has a significant impact on photosynthesis and plant reproduction. Physicochemical characteristics of plants such as leaf N content can be estimated spatially and temporally because of the latest developments in multispectral sensing technology and machine learning (ML) methods. The objective of this study was to use spectral data for leaf N estimation in corn to compare different ML models and find the best-fitted model. Moreover, the performance of vegetation indices (VIs) and spectral wavelengths were compared individually and collectively to determine if combinations of VIs substantially improved the results as compared to the original spectral data. This study was conducted at a Mississippi State University corn field that was divided into 16 plots with four different N treatments (0, 90, 180, and 270 kg ha-1). The bare soil pixels were removed from the multispectral images, and 26 VIs were calculated based on five spectral bands: blue, green, red, red-edge, and near-infrared (NIR). The 26 VIs and five spectral bands obtained from a red-edge multispectral sensor mounted on an unmanned aerial vehicle (UAV) were analyzed to develop ML models for leaf %N estimation of corn. The input variables used in these models had the most impact on chlorophyll and N content and high correlation with leaf N content. Eight ML algorithms (random forest, gradient boosting, support vector machine, multi-layer perceptron, ridge regression, lasso regression, and elastic net) were applied to three different categories of variables. The results show that gradient boosting and random forest were the best-fitted models to estimate leaf %N, with about an 80% coefficient of determination for the different categories of variables. Moreover, adding VIs to the spectral bands improved the results. The combination of SCCCI, NDRE, and red-edge had the largest coefficient of determination (R2) in comparison to the other categories of variables used to predict leaf %N content in corn. Keywords: Corn, Gradient boosting, Machine learning, Multispectral imagery, Nitrogen estimation, Random forest, UAV, Vegetation index.

Floods, especially large floods, are one of the most important factors affecting river morphology and ecosystems. Due to the effects of floods on landforms, the analysis of the geomorphological consequences of floods is important. In this study, the effects of an extreme flood that happened on April 1, 2019, in the Khorramabad River Basin located in western Iran are investigated. This research aims to investigate the effects of floods on channel changes as well as estimating the amount of flood-related erosion and sedimentation of the Khorramabad River. To do so, Sentinel-2 satellite images, Google Earth images, geological and topographic maps, and field investigations have been used in this research. First, the river channel was extracted from cloud-free multispectral Sentinel-2 images taken before (March 22, 2019) and after (April 21, 2019) the flood. Then, the amounts of erosion, sedimentation, and also unchanged areas were calculated using the Union tool in ArcGIS. Furthermore, using Google Earth images, cross-sectional profiles were prepared in two sections of the river, and changes in the depth and width of the riverbed were studied. The results show that the flood has significantly affected the width and depth of the channel, and the most prominent impacts of the flood were shifts of the riverbed and severe bank erosion in the meandering areas. According to the field surveys, floodwater occupied floodplains around the river channel and destroyed the adjacent agricultural lands and gardens. The analysis of river channel maps (before and after the flood) indicates that the greatest changes in the river channel (46.4% of the total changes) occurred as riverbed widening in the downstream and midstream areas due to riverbank erosion as well as river channel shifts. The changes caused by sedimentation account for 18.6% of the total changes, and the remaining 35% of the river channel area did not change at all.

Deep learning has already been proved as a powerful state-of-the-art technique for many image understanding tasks in computer vision and other applications including remote sensing (RS) image analysis. Unmanned aircraft systems (UASs) offer a viable and economical alternative to a conventional sensor and platform for acquiring high spatial and high temporal resolution data with high operational flexibility. Coastal wetlands are among some of the most challenging and complex ecosystems for land cover prediction and mapping tasks because land cover targets often show high intra-class and low inter-class variances. In recent years, several deep convolutional neural network (CNN) architectures have been proposed for pixel-wise image labeling, commonly called semantic image segmentation. In this paper, some of the more recent deep CNN architectures proposed for semantic image segmentation are reviewed, and each model’s training efficiency and classification performance are evaluated by training it on a limited labeled image set. Training samples are provided using the hyper-spatial resolution UAS imagery over a wetland area and the required ground truth images are prepared by manual image labeling. Experimental results demonstrate that deep CNNs have a great potential for accurate land cover prediction task using UAS hyper-spatial resolution images. Some simple deep learning architectures perform comparable or even better than complex and very deep architectures with remarkably fewer training epochs. This performance is especially valuable when limited training samples are available, which is a common case in most RS applications.

Streamflow forecasting is critical for real-time water resources management and flood early warning. In this study, we introduce a novel attention-based Long-Short Term Memory (LSTM) cell deep learning (DL) model for postprocessing streamflow simulations which will herein be referred to as Self-activated and Internal Attention LSTM, or SAINA-LSTM. In this model, we incorporate an improved self-attention mechanism into the inner structure of the LSTM cell to increase focus on the more important time points, thereby enhancing the information flow of the cell. Performance of the SAINA-LSTM is then compared against that of the current National Weather Service’s operational streamflow forecast ensemble postprocessor (EnsPost), a recently-developed multiscale alternative (MS-EnsPost), a robust machine learning algorithm (Gradient Boosting), and two other deep learning algorithms (LSTM and Gated Recurrent Unit (GRU)). Forecast performance in four basins in different climatological regimes of the United States are compared. Several deterministic evaluation metrics are examined for one to seven-day-ahead predictions of daily flows. SAINA-LSTM reduce the biases in simulations of low, medium, and high daily flows. The other two deep learning models also outperform EnsPost and MS-EnsPost in most cases. The results highlight the relatively poor performance of EnsPost, particularly in low flow conditions. Generally, the novel SAINA-LSTM model outperforms other models in low, medium, and high ranges of flow and for 1- to 7-day ahead forecasts in all three highly nonlinear and non-snow-driven study basins. In snow-driven basins, due to low nonlinearity, the three deep learning models are relatively comparable and significantly improved over statistical models. The results of the comparative evaluation demonstrate the capability of SAINA-LSTM to reduce the RMSE of daily-predicted flow by up to 20 percent compared to MS-EnsPost. The promising result shown here is a motivation for extending this research to also cover ensemble forecasting using the novel model developed in this work.

The reduction of visibility adversely affects land, marine, and air transportation. Thus, the ability to skillfully predict fog would provide utility. We predict fog visibility categories below 1600 m, 3200 m and 6400 m by post-processing numerical weather prediction model output and satellite-based sea surface temperature (SST) using a 3D-Convolutional Neural Network (3D-CNN). The target is an airport located on a barrier island adjacent to a major US port; measured visibility from this airport serves as a proxy for fog that develops over the port. The features chosen to calibrate and test the model originate from the North American Mesoscale Forecast System, with values of each feature organized on a 32 × 32 horizontal grid; the SSTs were obtained from the NASA Multiscale Ultra Resolution dataset. The input to the model is organized as a high dimensional cube containing 288 to 384 layers of 2D horizontal fields of meteorological variables (predictor maps). In this 3D-CNN (hereafter, FogNet), two parallel branches of feature extraction have been designed, one for spatially auto-correlated features (spatial-wise dense block and attention module), and the other for correlation between input variables (variable-wise dense block and attention mechanism.) To extract features representing processes occurring at different scales, a 3D multiscale dilated convolution is used. Data from 2009 to 2017 (2018 to 2020) are used to calibrate (test) the model. FogNet performance results for 6, 12 − and 24 − h lead times are compared to results from the High-Resolution Ensemble Forecast (HREF) system. FogNet outperformed HREF using 8 standard evaluation metrics.

In urban planning and transportation management, the centrality characteristics of urban streets are vital measures to consider. Centrality can help in understanding the structural properties of dense traffic networks that affect both human life and activity in cities. Many cities classify urban streets to provide stakeholders with a group of street guidelines for possible new rehabilitation such as sidewalks, curbs, and setbacks. Transportation research always considers street networks as a connection between different urban areas. The street functionality classification defines the role of each element of the urban street network (USN). Some potential factors such as land use mix, accessible service, design goal, and administrators&rsquo, policies can affect the movement pattern of urban travelers. In this study, nine centrality measures are used to classify the urban roads in four cities evaluating the structural importance of street segments. In our work, a Stacked Denoising Autoencoder (SDAE) predicts a street&rsquo, s functionality, then logistic regression is used as a classifier. Our proposed classifier can differentiate between four different classes adopted from the U.S. Department of Transportation (USDT): principal arterial road, minor arterial road, collector road, and local road. The SDAE-based model showed that regular grid configurations with repeated patterns are more influential in forming the functionality of road networks compared to those with less regularity in their spatial structure.

The deep convolutional neural network (DCNN) has recently been applied to the highly challenging and ill-posed problem of single image super-resolution (SISR), which aims to predict high-resolution (HR) images from their corresponding low-resolution (LR) images. In many remote sensing (RS) applications, spatial resolution of the aerial or satellite imagery has a great impact on the accuracy and reliability of information extracted from the images. In this study, the potential of a DCNN-based SISR model, called enhanced super-resolution generative adversarial network (ESRGAN), to predict the spatial information degraded or lost in a hyper-spatial resolution unmanned aircraft system (UAS) RGB image set is investigated. ESRGAN model is trained over a limited number of original HR (50 out of 450 total images) and virtually-generated LR UAS images by downsampling the original HR images using a bicubic kernel with a factor × 4 . Quantitative and qualitative assessments of super-resolved images using standard image quality measures (IQMs) confirm that the DCNN-based SISR approach can be successfully applied on LR UAS imagery for spatial resolution enhancement. The performance of DCNN-based SISR approach for the UAS image set closely approximates performances reported on standard SISR image sets with mean peak signal-to-noise ratio (PSNR) and structural similarity (SSIM) index values of around 28 dB and 0.85 dB, respectively. Furthermore, by exploiting the rigorous Structure-from-Motion (SfM) photogrammetry procedure, an accurate task-based IQM for evaluating the quality of the super-resolved images is carried out. Results verify that the interior and exterior imaging geometry, which are extremely important for extracting highly accurate spatial information from UAS imagery in photogrammetric applications, can be accurately retrieved from a super-resolved image set. The number of corresponding keypoints and dense points generated from the SfM photogrammetry process are about 6 and 17 times more than those extracted from the corresponding LR image set, respectively.