Globally, thermodynamics explains an increase in atmospheric water vapor with warming of around 7%/°C near to the surface. In contrast, global precipitation and evaporation are constrained by the Earth’s energy balance to increase at ∼2–3%/°C. However, this rate of increase is suppressed by rapid atmospheric adjustments in response to greenhouse gases and absorbing aerosols that directly alter the atmospheric energy budget. Rapid adjustments to forcings, cooling effects from scattering aerosol, and observational uncertainty can explain why observed global precipitation responses are currently difficult to detect but are expected to emerge and accelerate as warming increases and aerosol forcing diminishes. Precipitation increases with warming are expected to be smaller over land than ocean due to limitations on moisture convergence, exacerbated by feedbacks and affected by rapid adjustments. However, these temperature-dependent changes offset rapid atmospheric adjustments to radiative forcings which tend to increase precipitation over land relative to the oceans. These factors therefore drive complex changes in the regional water cycle in time and space, some examples of which will be discussed. Thermodynamic increases in atmospheric moisture fluxes amplify wet and dry events, driving an intensification of precipitation extremes. The rate of intensification can deviate from a simple thermodynamic response due to in‐storm and larger‐scale feedback processes, while changes in large‐scale dynamics and catchment characteristics further modulate the frequency of flooding in response to precipitation increases. Changes in atmospheric circulation in response to radiative forcing and evolving surface temperature patterns are capable of dominating water cycle changes in some regions. Moreover, the direct impact of human activities on the water cycle through water abstraction, irrigation, and land use change is already a significant component of regional water cycle change and is expected to further increase in importance as water demand grows with global population. This talk will summarize recent advances in understanding past and future large-scale responses in the water cycle.

Measurements of dust size usually obtain the optical or the projected area-equivalent diameters, whereas model calculations of dust impacts use the geometric and the aerodynamic diameters. As such, accurate conversions between the four types of diameters are critical. However, most current conversions assume dust is spherical, which is problematic as numerous studies show that dust is highly aspherical. Here, we obtain conversions between different diameter types that account for dust asphericity. Our conversions indicate that optical particle counters using optical diameter to determine dust size underestimate dust geometric diameter at coarse sizes. We further use the diameter conversions to obtain a consistent observational constraint of size distributions of emitted dust in terms of geometric and aerodynamic diameters. The resulting size distributions are coarser than accounted for by parameterizations used in climate models, which which underestimate the mass of emitted dust within 10≤D_geo≤20 μm by a factor of ~2 and do not account for dust emission with D_geo≥20 μm. This finding suggests that current models substantially underestimate coarse dust emission.

Snow and ice melt processes are a key in Earth’s energy-balance and hydrological cycle. Their quantification facilitates predictions of meltwater runoff as well as distribution and availability of fresh water. They control the balance of the Earth’s ice sheets and are acutely sensitive to climate change. These processes decrease the surface reflectance with unique spectral patterns due to the accumulation of liquid water and light absorbing particles (LAP), that require imaging spectroscopy to map and measure. Here we present a new method to retrieve snow grain size, liquid water fraction, and LAP mass mixing ratio from airborne and spaceborne imaging spectroscopy acquisitions. This methodology is based on a simultaneous retrieval of atmospheric and surface parameters using optimal estimation (OE), a retrieval technique which leverages prior knowledge and measurement noise in an inversion that also produces uncertainty estimates. We exploit statistical relationships between surface reflectance spectra and snow and ice properties to estimate their most probable quantities given the reflectance. To test this new algorithm we conducted a sensitivity analysis based on simulated top-of-atmosphere radiance spectra using the upcoming EnMAP orbital imaging spectroscopy mission, demonstrating an accurate estimation performance of snow and ice surface properties. A validation experiment using in-situ measurements of glacier algae mass mixing ratio and surface reflectance from the Greenland Ice Sheet gave uncertainties of ±16.4 μg/g_ice and less than 3%, respectively. Finally, we evaluated the retrieval capacity for all snow and ice properties with an AVIRIS-NG acquisition from the Greenland Ice Sheet demonstrating this approach’s potential and suitability for upcoming orbital imaging spectroscopy missions.

To understand the influence of the Sun on Earth's system, long and accurate measurements of solar irradiance are a prerequisite. The available direct measurements of solar irradiance since 1978 are clearly not sufficient for this purpose. This stimulated development of models used to reconstruct past solar irradiance variations from alternative observations. The main driver of the irradiance variations on time scales of days to millennia is the evolution of the solar surface magnetic field in form of dark sunspots and bright faculae and network. Therefore, models require input data describing the contributions of these various magnetic regions on the Sun at earlier times. Unfortunately, records that can be used to describe the facular and network contributions are barely longer than the direct irradiance measurements. Thus, irradiance reconstructions to earlier periods have to rely on sunspot data alone. Data that have hardly been used for solar irradiance reconstructions until now are full-disc solar observations in the Ca II K line. Such data exist since 1892 from various observatories and include all the needed information describing faculae and the network. However, they are plagued by a bunch of various problems and artefacts, and recovering the non-linear response of the photographic material to the radiation is non-trivial since the required information is usually missing, too. We have developed a method to process Ca II K observations from various sources and demonstrated the higher accuracy achieved by our method compared to other techniques presented in the literature. Here we use the carefully reduced Ca II K observations from multiple archives to reconstruct solar irradiance variations.

Post-seismic deformation following large earthquakes offers insights into the rheology of the lithosphere and upper asthenosphere. The Mojave, in southern California, is one of the best studied regions on Earth, yet key questions, about fault slip rates and rheological heterogeneity, remain unanswered. Unprecedented geodetic coverage of the 2019 Ridgecrest earthquakes provides an opportunity to test whether rheological models developed for the Mojave, from the Landers, Hector Mine and El Major Cucapah earthquakes, are applicable north of the Garlock fault, and to place bounds on the effects of local rheological heterogeneties associated with the Coso volcanic field. This volcanic field, which is located to the NW of the Mw7.1 rupture trace, is a region of high heat flow and geothermal activity. The locally high temperatures in the Coso volcanic field are likely to be associated with low viscosities compared to the surrounding regions, and high pore pressures due to the hydrothermal activity. The aftershock sequence associated with the Ridgecrest earthquakes shows a notable absence of large magnitude earthquakes in this region. We use variational Bayesian independent component analysis to isolate postseismic deformation in GPS time series around the earthquakes. We present models of the possible poroelastic, afterslip and viscoelastic response driven by coseismic stress changes in the July 2019 Ridgecrest earthquakes and investigate the possible effect of the Coso volcanic field. By modelling a series of different afterslip geometries, and viscoelastic rheologies we identify features of the GPS- and InSAR-derived surface deformation which are diagnostic of different post-seismic mechanisms and rheological heterogeneities.

Metal compounds abundant in Early Earth are thought to play an important role in the origins of life. Certain iron-sulfur minerals, for example, are proposed to have served as primitive metalloenzyme cofactors for their ability to catalyze organic synthesis processes and facilitate electron transfer reactions. An inherent difficulty with studying the catalytic potential of many metal compounds is the wide range of data and parameters to consider when searching for individual minerals and ligands of interest. Detecting mineral-ligand pairs that are structurally analogous enables more relevant selections of data to study, since structural affinity is a key indicator of comparable catalytic function. However, current structure-oriented approaches tend to be subjective and localized, and do not quantify observations or compare them with other potential targets. Here, we present a mathematical approach that compares structural similarities between various minerals and ligands using molecular similarity metrics. We use an iterative substructure search in the crystal lattice, paired with benchmark structural similarity methods. This structural comparison may be considered as a first stage in a more advanced analysis tool that will include a range of chemical and physical factors when computing mineral-ligand similarity. This approach will seek relationships between the mineral and enzyme worlds, with applications to the origins of life, ecology, catalysis and astrobiology.

Proper assessment of the economic risk from hazards is an important prerequisite toward enhancing resilience and is often overlooked or underestimated in importance. This research describes a method of assessing risk due to extreme cold temperature, hail, lightning, and tornado in 2050, utilizing projections from well-respected model output, with Louisiana as a case study. Our approach improves upon previous hazard risk assessments by considering the magnitude of the exposed population in weighing the property loss. This makes the approach her preferable over previous risk assessments. Furthermore, the present research uses current model projections to estimate changes in future conditions of the hazard presence. Finally, our use of downscaled data to the census-block circumvents the complications of examining hazards at the county level, particularly in cases for which population is unevenly distributed in the county. Results suggest that extreme cold temperature and tornado are by far the costliest of the four hazards in terms of property loss, although tornado loss is inherently difficult to project due to the unpredictable nature of individual tornado paths. Both extreme temperatures and hail are projected to decrease in loss as temperatures warm, especially in the New Orleans area, where population may decrease. The lightning hazard, while small and likely underestimated due to assignment of lightning damage to the phenomena in which it is embedded, is projected to increase, both on an absolute and per capita basis. Our results can assist environmental planners in protecting life and property, while also promoting hazard resilience and environmental, economic, and social sustainability.

Erma Yulihastin 1,2, Tri Wahyu Hadi1, Nining Sari Ningsih3, Muhammad Ridho Syahputra1 1Atmospheric Sciences Research Group, Faculty of Earth Sciences and Technology, Institut Teknologi Bandung, Bandung, 40132, Indonesia 2Center of Atmospheric Sciences and Technology, National Institute of Aeronautics and Space, Bandung, 40175, Indonesia 3Oceanography Research Group, Faculty of Earth Sciences and Technology, Institut Teknologi Bandung, Bandung, 40132, Indonesia The diurnal cycles of precipitation over the northern coast of West Java have been studied using the Tropical Rainfall Measuring Mission (TRMM) Real Time Multi-satellite Precipitation Analyses (MPA-RT) products with records spanning from 2000 to 2016, with emphasis on the occurrences of early morning precipitation peaks. Diurnal precipitation over the study area during November to March is basically characterised by precipitation peaks that occur in the afternoon to evening time (15:00–21:00 LT) but secondary peaks in night to morning time (01:00–07:00 LT) are also pronounced in January and February. Harmonic analysis method was then applied on data of January and February to objectively determine the diurnal phase and classify the timing of precipitation for each day into three categories, i.e. afternoon-to-evening precipitation (AEP), early morning precipitation (EMP), and late morning precipitation (LMP) with peaks that occur in the time windows of 13:00–24:00 LT, 01:00–04:00 LT, and 05:00–12:00 LT, respectively. In terms of frequency of occurrence, AEP, EMP, and LMP constitute 55 %, 26.1 %, and 18.9 % of total samples of precipitation events. In spite of the smallest percentage, EMP events are characterised by seaward (as well as landward) propagation, flat phase distribution, and large mean amplitudes. The propagating characteristics of EMP are more prominent, with indications of stronger connectivity between precipitation systems over land and ocean, when data are composited by taking the 99th percentile values in each grid to represent extreme precipitation events. The flat phase distribution of EMP events suggests that the timing of coastal precipitation is not necessarily locked to the phase of land/sea-breezes, thus, allowing precipitation to occur more randomly.

From interpreting data to scenario modeling of subduction events, numerical modeling has been crucial for studying tsunami generation by earthquakes. Seafloor instruments in the source region feature complex signals containing a superposition of seismic, ocean acoustic, and tsunami waves. Rigorous modeling is required to interpret these data and use them for tsunami early warning. However, previous studies utilize separate earthquake and tsunami models, with one-way coupling between them and approximations that might limit the applicability of the modeling technique. In this study, we compare four earthquake-tsunami modeling techniques, highlighting assumptions that affect the results, and discuss which techniques are appropriate for various applications. Most techniques couple a 3D Earth model with a 2D depth-averaged shallow water tsunami model. Assuming the ocean is incompressible and that tsunami propagation is negligible over the earthquake duration leads to technique (1), which equates earthquake seafloor uplift to initial tsunami sea surface height. For longer duration earthquakes, it is appropriate to follow technique (2), which uses time-dependent earthquake seafloor velocity as a time-dependent forcing in the tsunami mass balance. Neither technique captures ocean acoustic waves, motivating newer techniques that capture the seismic and ocean acoustic response as well as tsunamis. Saito et al. (2019) propose technique (3), which solves the 3D elastic and acoustic equations to model the earthquake rupture, seismic wavefield, and response of a compressible ocean without gravity. Then, sea surface height is used as a forcing term in a tsunami simulation. A superposition of the earthquake and tsunami solutions provides the complete wavefield, with one-way coupling. The complete wavefield is also captured in technique (4), which utilizes a fully-coupled solid Earth and ocean model with gravity (Lotto & Dunham, 2015). This technique, recently incorporated into the 3D code SeisSol, simultaneously solves earthquake rupture, seismic waves, and ocean response (including gravity). Furthermore, we show how technique (3) follows from (4) subject to well-justified approximations.

In this study, we revisit the role of curvature in modifying frontal stability. We first consider the statement “fq < 0 implies potential for instability”, where f is the Coriolis parameter and q is the Ertel potential vorticity (PV). This is true for any inviscid baroclinic flow. It is also evident in the transition of a governing equation for circulation within a front from elliptic to hyperbolic form as the discriminant changes sign. However, for curved fronts, an additional scale factor enters the discriminant owing to conservation of absolute angular momentum, L, leading to Solberg’s (1936) generalization of the Rayleigh criterion. In non-dimensional form, this expression also generalizes the classical instability criterion of Hoskins (1974) by accounting for centrifugal forces: modification of the front’s vertical shear and stratification owing to curvature tilts the absolute vorticity vector away from its thermal wind state and, in an effort to conserve the product of non-dimensional PV (q’) and absolute angular momentum (L’), this alters Rossby and Richardson numbers permitted for stable flow. The criterion, Φ’=L’q’ < 0, is then investigated in non-dimensional parameter space representative of low-Richardson-number vortices. An interesting outcome is that, for Richardson numbers near one, anticyclonic flows increase in q’, while cyclonic flows decrease in q’. Though stabilization is muted for anticyclones (owing to multiplication by L’), the de-stabilization of cyclones is robust, and may help to explain an observed asymmetry in the distribution of submesoscale coherent vortices in the global ocean.

The National Water Center (NWC), part of the National Weather Service’s (NWS) Office of Water Prediction (OWP), seeks to provide forecast flood inundation maps (FIM) as services at the national scale along 2.7+ million river reaches. The National Water Model (NWM) forecasts stream discharges along these reaches at varying time horizons. Building on the methodology of previous versions, OWP FIM 3.0 seeks to map NWM stream discharges to inundation extents with a modified version of the Height Above Nearest Drainage method for mapping stages. FIM 3.0 produces required datasets for generation of FIMs, which are relative elevations, reach-level catchments, rating curves and crosswalk tables, to convert the NWM discharge forecasts to inundation extents as a post-processing step. Computational tests estimate that producing these FIM required datasets for the continental U.S. will be around 600 total CPU hours. Only publicly available input datasets from the NWM or NHDPlusHR are required. The latest methods leveraged within FIM 3.0 include the use of consistent reach lengths and deriving FIM outputs at lower stream densities (downstream of River Forecast Center forecast points) to mitigate catchment boundary limitations. FIM 3.0 is primarily parallelized across Hydrologic Unit Codes (HUC) 4, 6, or 8 as processing units which can be selected by the user at run time. FIM capability improvements were measured by comparisons to engineering scale maps using contingency statistics. Future versions will support NWM geospatial dataset production, usage of Lidar elevations, and more advanced hydrological methods. FIM 3.0 is highly configurable, modular, and portable due to its software architecture and containerization. All software dependencies are open-source including GDAL/OGR, TauDEM, and RichDEM. The source code and Docker container are expected to be made available publicly for community review and development.

Glacial acceleration is the largest source of uncertainty in sea-level-rise assessment, according to the Intergovernmental Panel on Climate Change. Of the different types of glacial acceleration, surging is the least understood. In this paper, we demonstrate how a combination of automated algorithms dedicated to analysis of two entirely different observation types - satellite altimetry from NASA’s ICESat-2 and satellite imagery from Planet SkySat - can aid in advancing glaciology, utilizing state-of-the art remote sensing /Earth observation technology. NASA’s Ice, Cloud and land Elevation Satellite ICESat-2, launched 15~September~2018, carries the first space-borne multi-beam micro-pulse photon counting laser altimeter system, the Advanced Topographic Laser Altimeter System (ATLAS). ATLAS observations are collected in three pairs of weak and strong beams with 0.7m nominal along-track spacing (under clear-sky conditions). The recording of the observations as a photon-point cloud requires a dedicated algorithm for identification of signal photons and determination of surface heights. As a solution, we developed the density-dimension algorithm for ice surfaces, the DDA-ice. ATLAS data analyzed with the DDA-ice allow determination of heights over heavily crevassed ice surfaces, which are characteristics of accelerating glaciers. The study presented here builds on a special multi-component data set, obtained through synoptic observations of an Arctic glacier system during surge (Negribreen, Svalbard): Airborne altimeter and image data collected during our ICESat-2 validation campaign, and SkySat image data from a special acquisition collected as part of NASA’s Commercial Smallsat Data Acquisitions Pilot program. These are complemented by WorldView (Maxar) and ESA Sentinel-1 data. With a spatial resolution of 0.7-0.86m, SkySat data and WorldView lend themselves to automated classification of crevasse types. Altogether, we obtain a characterization in 3 dimensions that allows discrimination of ice-surface types from surging glaciers (Negribreen) and continuously fast-moving and accelerating glaciers (Jakobshavn Isbrae) based on morphological characteristics.

Detecting and locating microearthquakes from continuous waveform records is the fundamental step in microseismic processing. Dense local networks and arrays have introduced the possibility to detect large numbers of far weaker events, but when viewed on seismic records from individual stations their waveforms are often obscured by noise. Furthermore, areas of interest for microseismic monitoring often feature extremely high event rates, highlighting the limitations of traditional techniques based on phase picking and association. In order to maximise the new insights gained, we require fully automated techniques which can exploit modern recordings to produce highly complete earthquake catalogues containing few artefacts. QuakeMigrate is a new modular, open-source python package providing a framework to efficiently, automatically and robustly detect and locate microseismicity. The user inputs continuous seismic data, a velocity model or pre-calculated look-up table and list of station locations. Instead of reducing the raw waveforms to discrete time picks, they are transformed (by amplitude, frequency and/or polarisation analysis) to continuous functions representing the probability of a particular phase arrival through time. These ‘onset functions’ from stations across the network are then migrated according to a travel-time look-up table and stacked to perform a grid-search for coherent sources of energy in the subsurface. This enables detection of earthquakes at close to or below the signal-to-noise ratio at individual stations, and implicitly associates phase arrivals even at very small inter-event times. We demonstrate the flexibility and power of this approach with examples of basal icequakes detected at the Rutford Ice Stream, Antarctica, dike- and caldera-collapse induced seismicity at Bárðarbunga central volcano, Iceland, and the aftershock sequence from a M5 earthquake at Mt. Kinabalu, North Borneo. The modular nature of the workflow and wide range of automatic plotting options makes parameter choice straightforward, and robust event location uncertainty statistics facilitate filtering to produce a robust catalogue. QuakeMigrate also outputs phase picks and local magnitude estimates, with an architecture designed to promote further community-driven extension in future.

Whistler-mode waves have often been proposed as a plausible mechanism for pitch angle scattering and energization of electron populations in the solar wind. Theoretical work suggested that whistler waves with wave vectors parallel to the interplanetary magnetic field must counter-propagate (sunward) to the electrons for resonant interactions to occur. However, recent studies reveal the existence of obliquely propagating, high amplitude, and coherent waves consistent with the whistler-mode. Initial results from a particle tracing simulation demonstrated that these waves were able to scatter and energize electrons. That simulation was limited and did not examine a broad range of electron distributions. We have adapted the original particle tracing code for the solar wind with wave parameters observed by the STEREO satellites and to model core, halo and strahl electrons. Simulations are run to record the response of a wide initial phase space volume with uniform waves and wave packets. Using a Hamiltonian analysis, resonant responses at different harmonics of the cyclotron frequency are included in the simulation. A numerical integration scheme that combines the Hamiltonian analysis and the relativistic 3d particle tracing deployed on a high performance cluster enables accurate mapping and large-scale statistical studies of phase space responses. Observations of electron distributions from WIND at 1 AU are used for normalization. This enables extrapolations for core, halo, and strahl electrons evolution with the numerical Green’s function method. Results provide evidence for pitch angle broadening of the strahl and energization of core and halo electrons. This model can also provide results that are applicable to a number of different wave-particle interactions in the heliosphere for comparison to in-situ measurements.

To manage potential risks due to H2 leaks into the near-surface geosphere from H2 underground storages (e.g. salt caverns, aquifer), reliable monitoring methods along with a precise knowledge of the geochemical environmental impacts are necessary. Thus, the evolution of some prominent parameters in soil and aquifers can be determined: gas concentrations, redox potential, ionic balance and trace elements. As part of the ROSTOCK’H project, Ineris simulated H2 leakage by injection of dissolved H2 into a shallow aquifer (~20 m deep) in an experimental site within the Paris basin. This experiment aimed to testing advanced monitoring techniques and studying hydrogeochemical impacts at shallow depths. The aquifer water has calcium-bicarbonate facies and a neutral pH. Eight piezometers were aligned over 80 m according to the aquifer main flow (west-east). Hydrogeochemical monitoring devices were set up. One of the piezometers was equipped with a completion connected to a Raman probe and a specific Mid-IR cell for continuous measurement of aqueous gases. At the experiment outset, 5 m3 of water were extracted from the aquifer to be saturated with H2 under atmospheric conditions, before being reinjected through the injection well. About 100 LSTP of dissolved H2 (concentration of 1,8 mg/L) was injected in the aquifer. The H2 injection was preceded by the injection of underground water containing tracers (He(aq), uranine and LiCl) in order to warn the H2 plume arrival in the piezometers located downstream of the injection well. The concentrations of aqueous gases (He, H2, N2, O2, CO2, H2S and CH4) were measured in a control piezometer (20 m upstream) and in six piezometers up to 60 m downstream. Thus, the maximum H2 contents were detected up to 20 m downstream of the injection well: 0.6 mg/L at 5 m, 0.17 mg/L at 7 m then 1.8 µg/L of H2 at 10 and 20 m during the first week. Following the H2(aq) addition, the aquifer physico-chemistry has been modified: low increase in pH, high decrease in redox potential and O2(aq). These results confirm the feasibility of detecting and monitoring H2 in shallow aquifers in very low concentration conditions and highlight the potential impacts. This is of first importance for establishing the surveillance and security aspects related to with H2 storage.

Seismic hazard due to fluid invasion in hydraulic fracturing, wastewater disposal, and enhanced geothermal systems have become a concern for industry and nearby residents. Some of the challenges associated with the fluid-induced seismic hazard are the estimation of the spatial effects of these industry operations as well as the presence or absence of aftershock triggering. In some cases (e.g. Geysers, California, Hoadley gas field of Alberta), aftershocks triggering do occur, while in other cases (e.g. Soultz‐sous‐Forêts, France), this is not the case. First, to address the spatial effects, using several previously published high-resolution well-log data, we first show that there is a tendency that porosity within the basement resembles fractional Gaussian noise (fGn), while above the basement it resembles fractional Brownian motion (fBm). Based on this observation, we introduce a novel conceptual model of the fluid-induced seismicity in disordered porous media by integrating the notion of fluid diffusion and invasion percolation with spatially correlated permeability and porosity. We find that our model does not only capture the observed variations in frequency-magnitude distribution of seismic events but it also exhibits a much slower decay in seismic activity at large distances for fBm compared to fGn. Second, to address the presence of aftershock triggering, we also introduce nonlinear viscoelastic effects in our model to augment the failure mechanics. This allows us to test whether the presence or absence of aftershocks is coupled to the validity of a time scale separation between fluid dynamics and nonlinear viscoelastic response, for example. Our findings can be directly incorporated in the seismic hazard assessment related to fluid injections.

Urban expansion is a global trend primarily determined by overpopulation, particularly in developing countries like India. The pattern and boundary of urban expansion could be observed and modelled on a Spatio-temporal dimension in the Asan watershed, situated in western Doon Valley, Uttarakhand, India. The specific objectives of the present study are: 1) Monitoring Land Use Land Cover (LULC) change detection and its impact on the hydrological regime of the watershed, 2) Generation of urban growth pattern map to identify the urban expansion along with the directions and distance from the watershed centroid. To achieve the objectives, the four LULC maps of different years are generated through Landsat and LISS-IV dataset from 1980 to 2016. The eight classified classes of LULC includes: build-up (urban and rural), cropland, fallow land, plantation, evergreen needle leaf forest, deciduous broadleaf forest, mixed forest, and water bodies. The LULC change analysis concludes that the built-up area continuously expanding at an alarming rate, nearly four folds in the last 36 years. The major areal portion of cropland and plantation are converted into build-up at a faster rate after 1995 whereas the areal extent of water bodies are also reduced. Based on monitoring of urban growth, the wind rose tool revealed that the southwest direction is the dominant direction of the urban expansion till 1980 and 1980-1995 which contribute 48% and 47.61% respectively of the total urban area of the watershed. However, in the period 1995-2008 and 2008-2016, the eastern direction is predominant, which contribute 36.44% and 35.27% respectively from the total urban areal expansion. This reflects a situation where the eastern direction is expected to dominate in the next years. Consequences of urban expansion create pressure on the into the watershed which leads to degradation of natural resources, forest environment and hydrological elements of the watershed.

The subseasonal Madden-Julian oscillation (MJO) is among the most important modes of tropical variability on the planet and is a dominant driver of subseasonal-to-seasonal (S2S) prediction globally. The past decade has seen substantial advances in MJO prediction using dynamical forecast models, which now routinely outperform traditional statistical MJO forecasts (e.g. multiple linear regression models). At the same time, an increasing body of literature has demonstrated that machine-learning methods represent a new frontier in Earth science, opening the door to more advanced statistical forecast models of the MJO. In this study, we explore whether state-of-the-art machine learning methods can be used to make real-time MJO forecasts that outperform traditional statistical models and do comparably well to dynamical models. In particular, we utilize neural networks trained on observational tropical fields to attempt to make skillful forecasts of MJO convection out to several weeks lead time. Through contrasting the machine-learning models’ behavior with simpler statistical models and dynamical forecast models, we explore the advantages and disadvantages of statistical versus dynamical forecasts. A novel aspect of our analysis is the use of cutting-edge techniques to allow us to visualize how our neural network models makes their predictions. These techniques, such as layer-wise relevance propagation, can lead to new insights into regions of MJO predictability, allowing us to better interpret sources of MJO prediction skill within the machine-learning model. We further diagnose whether our machine-learning models contain well-known aspects of MJO prediction found in dynamical models, such as an increase in prediction skill during boreal winter or during certain phases of the stratospheric quasi-biennial oscillation.