Precipitation flag (precipitating or not; stratiform or convective) is a key parameter for us to make betterretrieval of precipitation characteristics as well as to understand the cloud-precipitation physicalprocesses. The Global Precipitation Measurement (GPM) Core Observatory’s Microwave Imager (GMI)and Dual-Frequency Precipitation Radar (DPR) together provide ample information on globalprecipitation characteristics. As an active sensor in particular, DPR provides an accurate precipitationflag assignment, while passive sensors like GMI were traditionally believed not to be able to tell apartprecipitation types. Using collocated precipitation flag assignment from DPR as the “truth”, this project employs machinelearning models to train and test the predictability and accuracy of using passive GMI-only observationstogether with ancillary atmosphere information from reanalysis. Precipitation types are classified intothe following classes: convective, stratiform, convective-stratiform mixed, no precipitation, and otherprecipitation. Sub-sampling with different probabilities is employed to construct a balanced trainingdataset. A variety of classification algorithms are tested, including Support Vector Machines, NaiveBayes, Random Forests, Gradient Boosting, and Neural Networks (Multilayer Perceptron Network), andtheir results are evaluated and compared. The trained model has ~ 85% of prediction accuracy for everytype of precipitation. High-frequency channels (166 GHz and 183 GHz channels) and 166 GHzpolarization difference are found among the most important factors that contribute to the modelperformance, which shed light on future instrument channel selection.

Lunar mare basalts are the products of their corresponding parent magma compositions, sourced from the lunar upper mantle. The lunar mantle has been repeatedly modeled through numeric simulations to reflect lunar magma ocean (LMO) crystallization, resulting in an early-stage anorthositic crust and immediately underlying, late-stage, KREEP-rich and ilmenite-rich layer. This negatively buoyant layer is expected to have induced mixing with the underlying mantle, potentially to the core-mantle boundary. The lunar mare basalts, in this context, reflect mantle sources that are variably mixed between pristine mantle compositions and the dense ilmenite-rich layer. In order to constrain the geometry of lunar mantle heterogeneity, we simultaneously examined multiple mare basalt characteristics to extract significant multivariate patterns that might lend insight into the nature of this mixing-induced heterogeneity. Using two fundamental machine learning approaches and a newly compiled database of Apollo basalt characteristics (ApolloBasaltDB), we conducted a preliminary investigation, holding the assumptions that 1) mare basalts are assumed to retain the majority of their original characteristics at the time of extrusion: texture, isotopic age, major element composition, mineral mode, and geographic occurrence; 2) negligible basalt alteration occurred due to the lack of an atmosphere; and 3) impact gardening did not have significant bearing on final geographic location of basalt samples based on our nearside spatial partitions. The results of cluster and principal component analyses over changing spatial basalt groupings suggest that lunar nearside changes in major element concentrations and mineral modes vary spatially. Al2O3 concentrations increase in diversity within the Procellarum KREEP Terrane (PKT) compared to older regions immediately exterior to the eastern PKT, while a general nearside trend appears to suggest that ilmenite (TiO2) diversity comes at the expense of plagioclase (Al2O3) diversity. Cluster analysis suggests PKT perimeter rifting may have tapped into increasingly Ti-rich sources as rifting proceeded SE to NW. By establishing such trends over varying spatial scales through multivariate processing, the changing strengths of these surface correlative patterns may indicate the changing conditions (including temporal) of the immediately underlying mantle at the time of extrusion.

A gusty wind by typhoons is one of the major natural hazards and has been the most threatening windstorm in urban districts. In recent year some major cities in Japan have experienced extreme winds during typhoon landfalls. For example, Typhoon Jebi (2018) caused extreme wind gusts in Osaka and Kyoto, while Typhoon Faxai (2019) and Hagibis (2019) produced high winds in Tokyo and neighboring cities. Urban roughness obstacles exert significant influences on the magnitude of wind gustiness. With the growing urbanization globally, the quantification of turbulent winds in densely built, urban districts is important to the assessment and prediction of risks of wind damages and the understanding of the underlying physical mechanisms. Influences of densely built urban environments on the occurrence of wind gusts in urban districts during the typhoon landfalls are studied by merging mesoscale meteorological and building-resolving large-eddy simulations (LES), which allows an explicit representation of the complicated building structures while retaining the strong mesoscale perturbations from the typhoon. The actual building data of Osaka, Kyoto, and Tokyo are used in the building-resolving LES computational domains. With the successful reproduction of the track and intensity of the typhoon in meteorological simulations, the simulated winds at the simulated boundary-layer top are used to quantify the wind gusts in the urban district. The maximum wind gust in the analysis area of Osaka during the landfall of Typhoon Jebi is found to exceed 60 m/s, which is comparable to the wind speed at the height of about 300 m. Such wind gusts are generated by the instantaneous downward momentum transfer in areas, where buildings of great height variability are clustered. The instantaneous wind gusts are found to be the strongest for moderate building packing density. The results suggest that the risks of wind damages are mostly likely to be maximized in urban districts of high building-height variability and moderate packing density.

To support the development of the Global Forecast System (GFS) physics suite and identify opportunities for improving the model physics in the UFS, the Developmental Testbed Center (DTC) conducted an array of analyses for evaluating the operational GFSv15 forecasts and the experimental forecasts using the GFSv16beta physics suite distributed with the UFS Medium-Range Weather Application v1.0 public release. Five-day GFSv16beta forecasts for one boreal winter season were generated using the operational GFS analyses as initial conditions. The evaluation metrics included tools from Model Evaluation Tools (MET) and in-house process-oriented diagnostics. The evaluations focused on the perpetuating GFS forecast errors pertaining to the planetary boundary layer (PBL), land-surface, cumulus, radiation, and cloud processes. The runs using GFSv16beta outperformed the operational GFSv15 with respect to the root-mean-square errors of large-scale environmental variables and the anomaly correlation coefficient for 500 hPa geopotential height. Nevertheless, larger biases associated with key physical processes were identified in the GFSv16beta forecasts. For example, the global precipitation forecast skill degrades and a dry bias remains in the tropics, suggesting a persistent problem in the cumulus scheme. The near-surface and boundary-layer cold biases are larger over most continents and polar regions, which is partly related to the systematic negative temperature errors in the GFS analysis. The overestimated near-surface wind speed particularly at night in the northeastern U.S. implies that the surface drag may be underrepresented. Excessive short-wave radiation reaching the ground in the high-latitudes of the summer hemisphere appears to be related to low cloud liquid and ice water path. These and other results will be described in this presentation.

The Global Warming Potential (GWP) is a widely used single-number metric used to compare the climate change effects of various greenhouse gases. Although GWP has an established role in international climate agreements, it is unphysical, unintuitive, arbitrary, ignores the time dependence of emission sources, and is in some cases misleading. The same doubts have been expressed by the convening lead author of the relevant chapter in the First Assessment Report of the Intergovernmental Panel on Climate Change, in which GWP was introduced. GWP and the related CO2-equivalent methodology have no place in describing the effects of climate change mitigation strategies beyond a 20 year horizon. An important element of sound policy making is the quantitative analysis of policy interventions. This work argues for the broader use of temperature change trajectories in educating policymakers and the public about greenhouse gas control. Modeling tools include general circulation models, related reduced complexity models, and simpler models of global mean temperature change. Examples are presented illustrating the misleading nature of GWP and the CO2-equivalent methodology. These include the greenhouse gas impacts of (1) single year and multiyear emissions of methane, (2) natural gas and coal use in the electric power sector, (3) pipeline and LNG transport of natural gas, (4) natural gas flaring, and (5) livestock herd reduction. Responding to climate change is likely to require global investments measured in trillions of dollars. Clear and accurate communication among and between scientists, policy makers. and the general public is essential. GWP is a crude, inaccurate tool that and the general public is essential. GWP is a crude, inaccurate tool that played a constructive role in the formulation of international agreements. Using better modeling tools can help make climate policy discussions more scientifically rigorous while demystifying the criteria upon which policy choices are made.

Pore connectivity, a topological characteristic of pore structure, is oftentimes more important than the geometrical aspects in controlling fluid flow and mass transport in porous natural rocks as well as their associated utilities in energy and environmental stewardship. A different extent of pore connectivity can be reflected in the proportion of isolated pore space not connected to the surface of natural rocks. This work presents the multi-approach and multi-scale laboratory studies to investigating the proportion of isolated pore space of, and its resultant anomalous fluid flow and radionuclide movement in, generic geological barrier materials (clay sediment, crystalline rock, salt rock, shale, tuff). The samples include clay sediments of Wakkanai formation at Horonobe underground research center in Hokkaido of Japan, Opalinus clay of Mt. Terri Underground Research Laboratory as well as granodiorite from the Grimsel Test Site in Switzerland, salt rock from Waste Isolation Pilot Plant in New Mexico, various shales (Barnett, Eagle Ford and Wolfcamp from Texas), and welded tuff in Yucca Mountain in Nevada. Working with sample sizes from <75 μm to several centimeters, the experimental approaches include the independent quantification of both (1) surface-accessible pore space with various probing fluids (e.g., helium in expansion, water in vacuum saturation and nuclear magnetic resonance, mercury in intrusion porosimetry, nitrogen in gas physisorption, and Wood’s metal in high-pressure impregnation and micron-scale tracer mapping using laser ablation-ICP-MS); and (2) total (both connected and isolated) porosity by small angle X-ray scattering. In summary, our evolving complementary approaches provide a rich toolbox for tackling the pore structure characteristics in geological barrier materials, and associated fluid flow & radionuclide transport, implicated in their long-term performance in natural and engineered systems of a nuclear waste repository.

The clear-sky and all-sky global mean energy flow system of the Earth as the solution of four radiative transfer constraint equations were presented in Zagoni (CERES STM 2020). These equations have their theoretical origin in the concept of radiative equilibrium for stratified atmospheres (Schwarzschild 1906, Eq. 11; Milne 1930, Eq. 93-95; Goody 1964, Chamberlain 1978, Eq. 1.2.29-1.2.30; Eq. 2.115; Goody and Yung 1989, Eq. 2.146 and Eq. 9.5; Houghton 2002, Eq. 2.13; Andrews 2010, Eq. 3.50-3.51; Pierrehumbert 2010, Eq. 4.44-4.45, etc.). We showed that each of the equations is justified by 19 years of CERES data within ±3 Wm-2 and the all-sky equations by the IPCC-AR5 (2013) Fig. 2.11 global energy budget estimate within ±2 Wm-2. The individual flux components, both for clear-sky and all-sky are valid within ± 1 Wm-2 at the TOA and within ± 4 Wm-2 at the surface. A fifth equation was introduced in Zagoni (EGU 2020) showing the place of a non-observable flux component (atmospheric window radiation) in the theoretical system; and a sixth equation allowed to extend the arithmetic structure to total solar irradiance. The set of these equations has a solution for LWCRE = 1 as unit flux, and the whole system of the global mean atmospheric energy flows can be described by small integers, in their own units, separately for the clear-sky, the cloudy sky and (as their weighted sum) for the all-sky case. OLR(all) = 9, OLR(clear) = 10, ULW = 15, G(all) = 6, G(clear) = 5 units = 133.40 Wm-2 with TSI = 51 units = 1360.68 Wm-2. Here we introduce the cloudy versions of the examined classic radiative transfer constraint equations and present the all-sky global mean energy budget as the weighted sum of the clear-sky and cloudy-sky energy flow system. “Clouds” are regarded as a single IR-opaque layer, represented by an effective cloud area fraction. Essential characteristics of the climate (like the clear-cloudy energy exchange; the surface, TOA, in-atmosphere and net CREs; and the greenhouse effect) are explained as the cooperation of the clear-sky and cloudy-sky geometric arrangements expressed in the transfer equations. This way, the theoretical description of the Earth’s global mean energy budget is complete.

There are growing evidences that the warming marginal seas impact the torrential rainfall events in many parts of the world. The sea surface temperature (SST) in the East Asian marginal seas (EAMS) exhibits rapid rise during the last decades compared to the global mean, but its impact is still a matter of debate. Here we assess the impact of the recent warming trend of the EAMSs on the torrential rainfall event that caused devastating flooding and landslides in Kyushu Island located in the western part of Japan in July 2017, using a high-resolution cloud permitting model. The warming trend of EAMS with atmospheric warming since 1980s increases the simulated 12-hr precipitation to 6.8%, corresponding to approximately 10% increase per 1 K of SST rise, consistent with the previous data analysis study. Slight increase in heat and moisture supplies due to the warming EAMS has impacted a conditionally unstable condition of air masses flowing into an area of the precipitation, leading to more vigorous convective systems. Moisture budget analysis indicates that the changes in the amount of precipitation is not responsible for the availability of precipitable water but the intensification of the convective system. Since the change in precipitation amount exhibits high sensitivity to the SST trends, use of multiple SST datasets is desirable to provide a more reliable estimate and its uncertainty.

While Artistic techniques may help scientists visualize data and communicate results to their colleagues, I believe the Arts can play a more significant role in communicating scientific information to the general public. Modes of presentation of information determine the impact of reception. Data on a page or screen is received cerebrally. Psychologists posit that concepts received emotionally can have a more lasting, deeper impact on an audience. That is what the arts (film, dance, drama, painting) can achieve. My paper will demonstrate through a particular project that concept. It will involve the presentation of a recent 56 foot long piece I’ve completed called “THE WORLD MAP SERIES” : https://www.dianeburko.com/new-page-1m which speaks to the impact of Climate Change on glaciers and coral reefs. This 10-part series represents my bearing witness to melting ice and coral bleaching, as well as exchanges with glaciologists and marine biologists over the past few years. It will also serve as an example of the artistic process.

Soil moisture is an essential measurement to manage water and improve crop production. However, agricultural research in the Yaqui Valley (in northwestern Mexico) with extensive wheat fields (Triticum sp.) have focused on other monitoring schemes (e.g. remote sensing) with less attention to soil moisture. Most of this cultivated soil contains up to ~ 50% clay, which results in changes to soil properties from wet to dry conditions and challenges in the implementation of in-situ measurements of soil moisture. For this research, we selected a 1-ha wheat field in the Yaqui Valley representative of a typical flood irrigation system. We measured meteorological variables (ClimaVUE™50), and soil moisture for the winter crop-cycle from December 2019 to Abril 2020. Volumetric water content (VWC) was recorded from 5 to 50 cm using two TDR (SoilVUE™10), one located in the bottom of the furrow under bare conditions, and the other on the top under the vegetated condition for further integration and comparison. A Cosmic Ray Neutron Sensor (CRNS) was located alongside the meteorological sensor. The universal calibration equation was used to estimate VWC based on neutron counts. The comparison from the CRNS and the integrated TDR (5 to 50 cm) resulted in an RMSE of 0.02 m3m-3 and an r2 = 0.73. While both technologies respond to water inputs, the CRNS is a more reliable measurement during the dry-down periods when the high-clay soil cracks to the extent of 40 cm where soil is exposed to air. During this driest period, recorded VWC at 50 cm was, on average, 0.25 m3 m-3, while measurements with the CRNS was on average, 0.16 m3 m-3. Interestingly, both sensors peaked at 0.56 m3 m-3 during the flood irrigation event.

In previous work (Hairston et al., GRL doi 10.1029/2018GL077381, 2018) we showed the topside F-layer (~850 km) electron temperatures measured by two DMSP spacecraft as they flew through the Moon’s shadow during the 21 August 2017 eclipse exhibited a series of non-uniform, banded decreases rather than a broad and smooth temperature decrease. We found that making a “mask” of the shadow of the Moon eclipsing the existing active regions on the sun’s surface created a pattern on the ionosphere showing where the gradient of the EUV from the active regions was greatest. The complex pattern of these areas from the mask at the F-peak altitude at 300 km corresponded to the areas in the topside F-layer where the DMSP observed the bands of cooled electrons. We have expanded this work to examine about a dozen other eclipses including the most recent 21 June 2020 eclipse. We repeatedly observed the same banded pattern in the electron temperatures in almost all the DMSP eclipse passes, thus demonstrating this is a repeatable phenomenon. Since the DMSP series of spacecraft form a constellation of four operational satellites with the same plasma instrument package making multiple measurements of the shadow at different local times, and sometimes within 10-15 minutes of each other, we can use these observations to map the shape and evolution of these cooling band patterns as the eclipse’s shadow passes over the Earth’s ionosphere. Here we will present our first detailed analysis of the two eclipses that occurred on 20-21 May 2012 and 2 July 2019. Both these eclipses have passes through the duskside by two spacecraft within a few minutes of each other, thus allowing us to examine the evolution of the pattern. We are using these events to determine the empirical patterns seen in the electron temperature decreases during eclipses and to explore the mechanism causing the cooling of the plasma and how it is transported from the F-peak region to the topside ionosphere.

The drivers of tree growth are one critical question in forest ecology and conservation. However, the measurement of tree growth is a difficult task that requires novel methods to improve accuracy and broaden the understanding of the effect of climate on tree metabolism and carbon accumulation. In this context, isotopes variation along woody tissues is a strong tool that provides new information about tree metabolism, growth rate, and the effect of climatic variation on these processes at high temporal resolution. Here, we obtained woody samples of two tree species two individuals per species (n = 4) from the Biogeographic Choco; Region in Colombia, one of the most humid regions of the planet without dry periods (mean annual temperature 25.9C and rainfall over 7200 mm). We measured 18O and 13C on these samples across some rings in each one to obtain intra-annual variation. Using these data, we assessed if isotopes variation in wood is correlated with climatic variation, explicitly precipitation regimen indicators employing Pearson correlation and linear mixed effect models. We found that both isotopes are correlated negatively with ring width. We also found that 18O is high negative correlated with precipitation indicators, rather than 13C. Our results suggest that isotopes variation are surrogates of tree growth in humid and non-seasonal forests. Besides, the 18O accumulation, which is strongly related to rainfall during the less rainy month (February: 370 mm on average), could be a better indicator of the effect of precipitation on the woody tissue rate change. However, 13C is more related to tissue formation processes. In conclusion, we found evidence of intra-annual variation in isotopes and tree growth in one hyper-humid forest challenging the effect of the dry season of tree growth and potentially suggesting the water excess as an additional limiting factor controlling growth rhythms in tropical trees.

Reliable, high-resolution sea ice forecasts may help contribute to safety of Indigenous marine mammal hunting and community travel in the Arctic. At the same time, sea ice forecasts could also be useful predictors of seasonal harvest success, including potential harvest shortfalls with impacts at the community level. However, large scale measures of sea ice concentration and ice extent are not sufficient indicators for predicting marine mammal harvest success in Bering Strait communities. Weekly sea ice and weather forecasts for Alaskan communities in the Bering Strait were analyzed from the Sea Ice for Walrus Outlook (SIWO) to identify sea ice and weather thresholds that greatly affected Indigenous marine mammal hunting activity and access to marine mammals during the spring harvest. The forecast of the timing for ice-free waters around a coastal community generally aligned with local observations. However some hunters continued to hunt by traveling longer distances from shore to reach marine mammals. This emphasizes the importance of maintaining coastal and marine forecasts even when coastal areas around communities appear ice-free. It was also apparent that while wind forecasts generally predicted the local wind conditions that were observed, the forecasts sometimes omitted important changes in wind direction, speed or ice movement, that occasionally resulted in ocean and ice conditions that required local search and rescue efforts or left hunters “stuck in the ice” overnight. The use of seasonal sea ice forecasts from models have not yet been explored with community partners as a potential tool to plan for upcoming subsistence hunting seasons, although the current use of SIWO is recognized as a tool to record hunter observations in “our words and descriptions” for current events and for future use and analysis. An evaluation of the Sea Ice for Walrus Outlook is being conducted, and the results of this new study will be useful to support future development of forecast information, and avenues for sharing of timely community observations.

This study explored whether the global temperature had any role in the spread and vulnerability to COVID-19 and how that knowledge can be used to arrest the fast spreading disease. It highlighted that for transmitting the virus, global temperature played an important role and a moderately cool environment was the most favourable state. Whereas, the risk from the virus was reduced significantly for warm places and countries. Based on the temperature of March and April, various degree of vulnerability was identified and countries were specified. The maximum reported case, as well as death, was noted when the temperature was in the range of around 275°K(2°C) to 290°K(17°C). Countries like the USA, UK, Italy and Spain belonged to this category. The vulnerability was moderate when the temperature was less than around 275°K(2°C) e.g., Russia, parts of Canada and a few Scandinavian countries. For temperature 300°K(27°C) and above, a significantly lesser degree of vulnerability was noted. Countries from South Asian Association for Regional Cooperation, South East Asia, the African continent and Australia fell in that category. This work discussed that based on the variability of temperature, countries can switch from one vulnerability state to another. That influence of temperature on the virus and results of previous clinical trials with similar viruses provided a useful insight that regulating the level of temperature can offer remarkable results to arrest and stop the outbreak. Based on that knowledge, some urgent and simple solutions are proposed, which are practically without side effects and very cost-effective too.

Gold mining has played a significant role in Ghana’s economy for centuries. Regulation of this industry has varied over time and while large-scale mining is prevalent in the country, prevalence of artisanal mining, or Galamsey has escalated throughout Ghana in recent years. These mines are not only harmful to human health due to the use of Mercury in the amalgamation process, but also leave a significant footprint on terrestrial ecosystems, degrading and destroying forested ecosystems in the region. This study used machine learning and Google Earth Engine to quantify the footprint of artisanal gold mines in Ghana and understand how conversion of forested regions to mining has changed from 2002-2019. We used Landsat imagery and a random forest classification to classify areas of anomalous NDVI loss during this time period and used WorldView image collections to assess the accuracy of the model. We then used a 3-year moving average to calculate the year of maximum derivative NDVI values. We used this calculation to identify the year of conversion to mining. Within the study area of Southwestern Ghana, our analysis showed that approximately 35,000 ha of vegetation were converted to mining. The majority of this mining occurred between 2014 and 2017. Additionally, around 700 ha ha of mining occurred within protected areas defined by the World Database on Protected Areas. Often, artisanal mining appears to be co-located with rivers such as the Orin and Ankobra Rivers, demonstrating the potential of these mines to affect access to clean drinking water. Through the process of gold extraction, these mines leave a distinct footprint with a series of ponds following these major rivers. However, while the footprints of these ponds are spatially distinct, our model does not distinguish between active and inactive ponds if no remediation actions are taken following inactivity. Future research should work towards distinguishing between active and inactive mining sites to better understand current levels of mining activity in Ghana.

Researchers and end users using climate data face a challenge when they analyze the data they need. Data volumes are increasing very rapidly, and the ability to download all needed data is often no longer possible. Also, it can be complex to install, configure and use some advanced analysis tools on such large datasets. This is especially true when they are stored in a federated architecture like the ESGF. An example of a complex analysis tool used in climate research and adaptation studies is a tool to follow storm tracks. In the context of climate change, it is important to know if storm tracks will change in the future, in both their frequency and intensity. Storms can cause significant societal impacts, hence it is important to assess future patterns. Having access to this type of complex analysis tool is very useful, and integrating them with front-ends like the IS-ENES climate4impact (C4I) would enable the use of those tools by a larger number of researchers and end users. Integrating this type of complex tool is not an easy task. It requires significant development effort, especially if one of the objectives is also to adhere to FAIR principles. The DARE Platform enables research developers to faster develop the implementations of scientific workflows more rapidly. This work presents how such a complex analysis tool has been implemented to be easily integrated with the C4I platform. The DARE Platform also provides easy access to e-infrastructure services like EUDAT B2DROP, to store intermediate or final results and powerful provenance-powered tools to help researchers manage their work and data. This project has received funding from the European Union’s Horizon 2020 research and innovation programme under grant agreements N°824084 and N°777413.

Current machine learning based shear wave velocity (Vs) inversion using surface wave dispersion measurements utilizes synthetic dispersion curves calculated from existing 3-D velocity models as training datasets. It is shown in the previous studies that the performances of the resulting networks are dependent on the diversity of the training data. We present an improved semi-supervised deep learning algorithm-based method that incorporates both observed and synthetic surface wave dispersion curves in the network training process. The algorithm is termed Wasserstein cycle-consistent generative adversarial networks (Wcycle-GAN), which combines the architecture of cycle-consistent GAN with Wasserstein loss metrics in optimization. Different from conventional supervised deep learning approaches, the GAN architecture also extracts structural information from the observed surface wave dispersion data in the training process that may improve generalization of the resulting network. The cycle-consistent loss addresses soft constraints on the trained neural networks to be reversible and thus reduces the variance of the trained networks. The Wasserstein metric provides weaker topology for convergence and improves spatial continuity of the predicted shear velocity (Vs) models. We demonstrate these improvements by applying the Wcycle-GAN to 4066 fundamental mode Rayleigh wave phase and group dispersion curves obtained in Southern California (SC). In general, the 3-D Vs model predicted by the best training Wcycle-GAN is consistent with previous surface wave tomography studies of SC in the overlapping area, but with smaller data misfit, yields better spatial smoothing, and provides improved images of structures near faults and in the top 5 km. Our results indicate that the proposed Wcycle-GAN algorithm has strong training stability and generalization abilities.

This paper presents some pre-earthquake and coseismic water-level and stream-flow changes observed in Japan and Taiwan. The results suggest that: (1) Hydrological precursors do occur; (2) they can be observed at a relatively few sensitive sites; (3) these “sensitive” sites consistently show coseismic changes; (4) the mechanisms of these precursors can be understood, if crustal heterogeneity and pre-earthquake slow-slip events are included in their mechanism consideration. In Japan, the monitored well was sensitive, because it tapped a permeable aquifer connected to a nearby fault, which was under a hydraulic –pressure gradient caused by pumping activity in an underground gallery on the other side of the fault. Also, the fault was in a near-critical condition, such that leakage could be caused by a small crustal disturbance, such as seismic shaking, in the case of coseismic changes, or a stress increment, in the case of precursory changes. In Taiwan, both the sensitive well and the stream gauge were located on the hanging wall of the seismogenic fault of the magnitude-7.6 Chi-Chi earthquake in 1999. The hanging wall probably bulged before the earthquake, causing opening up of fractures along some secondary faults, and allowing stored groundwater to flow down to the monitored stream and caused the observed pre-earthquake stream-flow increase. The continued fracture-opening process toward greater depth then caused down-flow of water into greater depth of the crust and triggered the occurrence of slow-slip events, which propagated down-dip to where these faults met the seismogenic fault and caused slow-slip events to propagate up-dip the seismogenic fault toward the hypocenter, triggering the earthquake and the observed water-level precursor at a well located near the tip of the wall.