A document by Danish Tantary. Click on the document to view its contents.

In recent years, Machine Learning (ML) techniques have gained the attention of the hydrological community for their better predictive skills. Specifically, ML models are widely applied for streamflow predictions. However, limited interpretability in the ML models indicates space for improvement. Leveraging domain knowledge from conceptual models can aid in overcoming interpretability issues in ML models. Here, we have developed the Physics Informed Machine Learning (PIML) model at daily timestep, which accounts for memory in the hydrological processes and provides an interpretable model structure. We demonstrated three model cases, including lumped model and semi-distributed model structures with and without reservoir. We evaluate the first two model structures on three catchments in India, and the applicability of the third model structure is shown on the two United States catchments. Also, we compared the result of the PIML model with the conceptual model (SIMHYD), which is used as the parent model to derive contextual cues. Our results show that the PIML model outperforms simple ML model in target variable (streamflow) prediction and SIMHYD model in predicting target variable and intermediate variables (for example, evapotranspiration, reservoir storage) while being mindful of physical constraints. The water balance and runoff coefficient analysis reveals that the PIML model provides physically consistent outputs. The PIML modeling approach can make a conceptual model more modular such that it can be applied irrespective of the region for which it is developed. The successful application of PIML in different climatic as well as geographical regions shows its generalizability.

Uncertainty quantification and characterization in changing climate scenarios can have a direct impact on the efforts to mitigate and adapt. Chaotic and non-linear nature of atmospheric processes results in high sensitivity to initial conditions resulting in considerable variability. Multiple model ensembles of Earth System Models are often used to visualize the role of parametric uncertainties in mean and extreme attributes of precipitation trends in various time horizons. However, studies quantifying the role of internal variability in controlling extreme precipitation statistics in decadal and interdecadal scales are limited. In this study, we use a thirty one-member ensemble of Community Earth System Model Large ensemble project and thirty-one ensembles from Coupled Model Intercomparison Project 5 (CMIP5) to quantify the relative contribution of uncertainty due to internal variability in the depth and volatility of Indian Summer Monsoon Rainfall extremes of different durations and frequencies. We find that in the short-term and long-term, the role of internal variability in extreme precipitation indices is comparable to the uncertainty arising from structural differences in the model captured through multiple model ensembles. Further, we show that combining outputs from multiple initial condition runs generated to span the range of internal climate variability can help us reduce uncertainty in infrastructure design relevant Depth Duration and Frequency (DDF) curves.

Disasters triggered by extreme precipitation events i.e., landslides, debris flows, and floods cause devastating damages to lives, infrastructure, and the economy. Under a warming climate, precipitation extremes and the occurrence of debris flows are further expected to intensify. Driven by extreme runoff, the triggering of debris flows can be simultaneous. Their concurrent occurrence multiplies complexity in decision-making during emergencies. Despite advancements in geotechnics and network science, a systematic framework to analyze the impact of debris flows on road networks is lacking. While network science-based approaches work on large-scale, geotechnics-based damage assessments are done solely on a site-to-site basis. Here we develop an integrated approach to analyze the impacts of simultaneous debris flows on road networks. The approach includes a novel infinite slope-based one-dimensional numerical model that simulates runoff-induced erosion and a network science-based mathematical model for road failures. This study covers multiple catastrophic events of debris flows that occurred in different geological and climate settings i.e., post-earthquake, post-volcanic, and post-wildfire environments. We perform spatio-temporal simulations of initiation and runout of debris flows and calculate the damage caused on individual road segments. We validate the model results using metadata. Our results show even remote local disturbances caused by successive debris flows upstream may lead to complete cascading disruption of the network downstream. Our unified strategy opens avenues to understand the resilience of critical infrastructure networks against catastrophic debris flows.

The warming climate intensifies the frequency and intensity of extreme precipitation events, leading to increases in precipitation-induced disasters. Precipitation-induced disasters such as flooding, landslide, and debris flow possess the potential risk of damage to socio-economic activity. The losses due to concurrent hazards in a region not only depend on the intensity and frequency but also socio-economic condition, topography, and exposure to the affected region. Recent advancements in risk mapping have shown approaches to measure the vulnerability to disaster but not accounting for concurrent hazards can lead to underestimation of risk. Here we propose the framework to assess the risk of concurrent precipitation-induced disasters while incorporating socio-economic, topographic, and land use information. In Kerala, India, the Periyar river basin is selected as a testbed for analysis considering 2018 extreme precipitation events. We perform 2D hydrodynamic flood inundation modeling to analyze the spread of the flood with the Spatio-temporal simulations of shallow landslides and debris flows using infinite slope-based stability and erosion models to identify the exposure of disaster. We evaluate socio-economic vulnerability and topographic vulnerability using disparate techniques from census demographic data and digital elevation model data respectively and exposure using land use information. The risk mapping is performed at the taluka (sub-district) level in the Periyar basin. Our results show better land-use planning considering multi-hazard vulnerability assessments reduces the exposed risk and would be beneficial for risk mitigation measures in high-risk areas