Analyzing Spatiotemporal Climate Event Patterns

🛰️ New paper out 🌍 How can we detect #landslides reliably when #satellite imagery only captures them as brief moments in time? Landslides are not static events — they develop from stable terrain, rupture abruptly, and then transition through complex recovery patterns. Yet most #EarthObservation models only ever see a single snapshot, because existing datasets rarely preserve the temporal sequence behind each event. To address this challenge, my PhD student Paul Höhn (Technical University of Munich & German Aerospace Center (DLR)) curated #Sen12Landslides, a global, multi-temporal dataset designed to capture landslides as evolving processes rather than isolated shapes. It includes around 75,000 annotated events across 15 diverse regions, paired with time series of Sentinel-1 SAR, Sentinel-2 optical imagery, and Copernicus DEM. Each sample is structured into a standardized sequence with pre-, event-, and post-event timestamps. When we trained spatio-temporal deep learning models on this data, the impact of temporal context became clear: F1 scores exceeded 0.83, refined annotations improved performance by 4–7%, and models demonstrated strong generalization to regions not included in training. These results highlight how essential temporal information is for understanding landslides at scale. We hope Sen12Landslides supports advances in global landslide detection, multi-sensor fusion, spatio-temporal anomaly detection, and Earth observation foundation models. 👉 Dataset: https://lnkd.in/d6r94br7 👉 Code: https://lnkd.in/d89m4qgc 👉 Paper: https://lnkd.in/dinvcVzT This work is made possible through valuable inputs from our domain expert Robert Behling (GFZ Helmholtz Centre for Geosciences) and from our lab alumni Konrad Heidler (who is currently with Hula Earth).

Continuing on the #Urbanization and #DroughtRisk theme, one of the important questions comes up is - How should we do this analysis, when many cities globally may not have the data? We address this need by developing and adopting a data-fusion framework , which is now published as 🏭 "Huang, Shuzhe, et al. "Urbanization-induced spatial and temporal patterns of local drought revealed by high-resolution fused remotely sensed datasets." #RemoteSensingofEnvironment 313 (2024): 114378." 📌 Our findings revealed that urbanization led to more intense peak drought intensity and average drought severity. In addition, urban drought fields showed lower effective radius, indicating more concentrated drought towards urban regions. 📍 From the text ....." we initially proposed a two-step fusion framework, integrating both surface (i.e., gridded data)-surface and point (i.e., in-situ data)-surface fusion. The framework was applied to generate daily precipitation and average/maximum/minimum air temperature at a 1 km resolution through the integration of high-resolution remotely sensed datasets ... 📍 "comparison of our fused data with CPC, ERA5-Land, CMFD, CHIRPS, IMERG, and TMPA products confirmed its capability in capturing local-scale meteorological dynamics by improving spatial resolution from 0.1°-0.25° to 1 km. Utilizing these high-resolution datasets, we quantified urbanization's impacts on local drought across 52 major cities ..." 📍 "We found that urbanization significantly magnified extreme Standardized Precipitation Evapotranspiration Index (SPEI) and drought severity in 69.2% and 61.5% of these cities, respectively. The effects of urbanization on extreme SPEI were amplified by the increase of urbanization rates, with a slope of −0.24 (p < 0.05). To further examine the spatial patterns of urbanization-induced local drought, we proposed a drought spatial field identification method.." #Droughts #IPCC #CityClimate Jackson School of Geosciences at The University of Texas at Austin Cockrell School of Engineering, The University of Texas at Austin University of Texas Center for Space Research #UTcityClimateCoLab

🌍 #Publication_Alert! 🌿🌧 Excited to share my latest research as a #corresponding_author in the #Climate_Services (Q1) Journal! Our study, "#Predicting_precipitation and #NDVI utilization of the multi-level linear mixed-effects model and the #CA_Markov_simulation_model," reconstructs the spatiotemporal evolution of precipitation and NDVI in the #Loukkos_watershed, providing scenarios for their recent and future trends. 📊📉 🔬 #Key_Highlights: ✅ Analyzed precipitation and NDVI time series data (1999–2019) to assess trends and associations. ✅ Predicted NDVI changes for 2029 and 2040 using the CA-Markov model. ✅ Simulated future precipitation (2019–2040) using the multi-level linear mixed-effects model (LME) with data from ten meteorological stations. ✅ Findings provide critical insights into vegetation-climate dynamics, essential for sustainable water and land management in changing climatic conditions. This research enhances our understanding of environmental transformations, supporting data-driven decision-making for climate resilience and ecosystem conservation. 🌱🌏 Read more: 🔗 DOI: https://lnkd.in/eNpF2FBz #ClimateChange #RemoteSensing #Precipitation #NDVI #SustainableDevelopment #MachineLearning #CA_Markov #EnvironmentalScience #Research #ClimateResilience

Top Drivers of Urban Flooding in Denver, Colorado Our recent work connecting spatial rain and geographic data to municipal reports of urban flooding shows us new insights on drivers of urban flooding. Rainfall is not enough to predict a street flood occurrence, information about the geography is important, as well. In our study we used service request reports from the City and County of Denver Department of Transportation and Infrastructure Wastewater Management Division. These service reports are written by any concerned or affected individual who has encountered street flooding spanning 2000-2019 in Denver. In addition, we collected spatial rain data courtesy of the Mile High Flood District Rain Gauge Network and looked at geographic variables from the US Census. We connected service reports to rain events based on time of occurrence and location. We could connect service requests to rain gauges via Thiessen polygons. Our first attempt of analysis was to find a threshold for rainfall to predict street flooding. But with no strong predictors, we chose to incorporate spatial variables in a linear regression. Our linear regression showed that population density and 5-minute max intensity are both nearly as strong predictors of pluvial flood reports. This information can be used to improve city flash flood warning systems which currently rely on rainfall depth over a ten-minute time span. Perhaps different alert systems can utilize a 5-minute max intensity for areas of a specific population density. Our analysis also found that census tracts with higher social vulnerability see higher rates of street flooding. We used the Social Vulnerability Index from the Centers of Disease Control and Prevention (CDC). We suggest further work studying better ways to track urban flooding that are more even and regular. Biases and errors may be present by only looking at human-created reports. Urban flooding is predicted to become a serious problem for American cities with aging infrastructure and more intense storms being predicted in the future. For more information, see the full article: DeSousa, Stacie, Aditi S. Bhaskar, Christa Kelleher, and Ben Livneh. 2024. “Understanding Spatiotemporal Patterns and Drivers of Urban Flooding Using Municipal Reports.” Hydrological Processes 38(12): e70028. doi:10.1002/hyp.70028/ https://lnkd.in/grEK4EUU

🌍 Proud to Share Our Latest Research Publication! I am delighted to share that our research paper titled: “Spatiotemporal Assessment of Extreme Rainfall Events in the Wainganga River Basin Using CMIP6 Climate Models” has been published in Natural Hazards (2026). This research provides a detailed spatiotemporal investigation of extreme rainfall dynamics in the Wainganga River Basin — one of Central India’s ecologically and economically vital river systems, and also one of most vulnerable hotspots for extreme events. 🔎 Major Findings of the Study: 📈 Extreme rainfall events are projected to increase significantly in the future, especially under the high-emission SSP5 scenario. 🌧️ Short-duration, high-intensity rainfall (1-day extremes) shows strong amplification, indicating a rise in flash-flood-type events. 📊 Total annual precipitation (PRCPTOT) is projected to increase by ~20–35% in the latter half of the century. ⚡ Indices such as R10, R20, R30, RX1, RX5, RX7, R95P, and R99P indicate increasing frequency and intensity of heavy and very heavy rainfall days. 🌊 The eastern ridge of the basin emerges as a future flood hotspot, with intensified extreme precipitation. 🌵 The western ridge shows early-century drought dominance, followed by increasing wet spells toward the end of the century. 📉 Consecutive Dry Days (CDD) initially increase (near future), while Consecutive Wet Days (CWD) rise significantly toward the end of the century. 🔬 Mann–Kendall trend analysis confirms statistically significant increasing trends in most extreme rainfall indices under SSP5. Overall, the study clearly indicates a shift toward more intense, concentrated, and extreme rainfall patterns, increasing the risks of floods, hydrological instability, and ecosystem stress in the basin. 🔍 Methodological Strength Multi-Model Ensemble (MMEA) using six CMIP6 GCMs 14 ETCCDI & ET-SCI rainfall extreme indices Pixel-level (97 grids) basin-wide assessment I sincerely express my deep gratitude to Dr. Manish Nema for his continuous support, cooperation, and invaluable guidance throughout this journey. His mentorship has been instrumental in shaping this research. I gratefully acknowledge my research scholar, Abhishek Raj Gupta’s, hard work and commitment, which greatly enhanced the quality of this research. This work reinforces the urgent need for climate-resilient water resource planning and adaptive strategies in vulnerable river basins. #ClimateChange #ExtremeEvents #CMIP6 #Hydrology #WaterResources #ClimateScience #Sustainability #ResearchPublication #NaturalHazards #WaingangaRiver

🌧️⚠️ New Research Published! Proud to share our latest work in the Journal of Earth System Science (Springer). 📘 Full Article: https://lnkd.in/dpZykFCA 📄 Title: Analysis of flash flood incidents triggered by cloudbursts and heavy downpours over Jammu and Kashmir, India: Spatiotemporal characteristics and implications 🌍 What Our Study Reveals ✔️ 68 extreme rainfall-triggered flash flood incidents analysed across J&K (2011–2022) ✔️ Strong monsoon dominance, with most cloudburst events between May–August ✔️ Distinct elevation control, with peak occurrence between 3,100–4,100 m ✔️ High-susceptibility districts: Ganderbal, Pulwama, Kishtwar, Poonch & Doda ✔️ Basin-scale hotspots identified for targeted mitigation and preparedness ✔️ Provides a first-of-its-kind spatiotemporal dataset for Himalayan flash-flood research 🔎 Why This Matters Flash floods triggered by cloudbursts are emerging as one of the most destructive hydro-meteorological hazards in the Himalayas. Our findings offer: • Better understanding of risk zones • Inputs for early warning and forecasting systems • Scientific support for climate-resilient planning • A baseline for future hydrological and hazard modelling Grateful to my mentors Dr Harish Bahuguna Sir, co-authors and colleagues for their support throughout this work. Excited to continue contributing to Himalayan geohazard research and sustainable mountain development.

Precipitation Annual Change Detection using Google Earth Engine & Python I’ve been working on detecting annual precipitation changes using the differencing technique with Google Earth Engine and Python. This workflow uses NASA GPM IMERG monthly precipitation data (2015–2025) to analyze spatial and temporal rainfall variability. 🔍 What this workflow covers: Accessing GPM IMERG data through Google Earth Engine Defining a region of interest and clipping data to national boundaries Converting Earth Engine ImageCollections to Xarray using xee Aggregating monthly precipitation into annual totals Detecting year-to-year precipitation change using differencing Visualizing spatial patterns and mean annual change 📌 Why it matters: Understanding precipitation trends and changes is essential for climate studies, water resource management, drought assessment, and climate-resilient planning. This type of analysis is especially useful for regional climate monitoring and decision support. A great resource for learning advanced Earth Engine and Python integration. #GoogleEarthEngine #ClimateAnalysis #Precipitation #RemoteSensing #GIS #Python #Xarray #ClimateChange #EnvironmentalMonitoring

📜 Publication Alert 📜 Pleased to share our recent publication "Deconstructing the spatiotemporal characteristics of extreme precipitation events from multiple data products during Indian summer monsoon", published in the Journal of Hydrology: Regional Studies, Elsevier. I would like to appreciate the efforts of Mr. Sandipan Paul and guidance of Prof. Ramesh Teegavarapu in shaping this work. The manuscript can be accessed at: https://lnkd.in/dBmtnrXK 📒 This study analyzes the spatiotemporal characteristics of EPEs across the Indian subcontinent during the monsoon season, critical for the region’s water resources and agriculture. Using observational (IMD, APHRODITE), reanalysis (IMDAA, GLDAS, ERA5-Land), satellite (CHIRPS, PERSIANN-CDR), and hybrid (MSWEP) datasets, we assess their ability to reproduce EPE intensity, detectability, timing, trends, and statistical properties. Key outcomes: 🌍 The study reveals that EPE intensity and frequency are highest along India’s western coast and northeast, moderate in central regions, and lowest in arid western and peninsular areas. 🫧 Wet-to-wet, dry-to-dry, and wet-to-dry transitions follow similar regional patterns. 🛰️ Satellite datasets generally underestimate, while reanalysis datasets overestimate EPE intensities, introducing wet and dry biases in moderate-intensity event frequencies, respectively.  ⛈️ Results identify MSWEP as the most reliable alternative to IMD in data-scarce regions, providing valuable insights for hydrologic studies, climate impact assessments, disaster risk management and enhancing socio-economic resilience. Elsevier India

Very happy to share a new publication led by Santiago Duarte, marking an important milestone in his final steps of the doctoral research at IHE Delft Institute for Water Education and Delft University of Technology , The paper introduces a Spatiotemporal Non-Linear Dynamics Assessment (SNLDA) framework to evaluate how well ERA5-Land precipitation represents rainfall in a tropical basin, going beyond conventional metrics by explicitly accounting for spatiotemporal rainfall structures and nonlinear dynamical behavior. Using the Magdalena River Basin (Colombia) as a case study, the results reveal systematic limitations of ERA5-Land in capturing event structure, extremes, and nonlinear dynamics across different climatic regions. Congratulations to Santiago Duarte on this achievement. 🔗 https://lnkd.in/eAShGYQi #PhDResearch #Hydrology #MachineLearning #WaterResources

Under CMIP6 SSP3‑7.0, years with a flood hazard index exceeding 0.20 — rare in the near term — become increasingly frequent by mid‑ to end‑century. CCART captures both the spatial pattern and the temporal intensification signal. Today marks a major milestone for CCART — a fully open‑source, free‑to‑use, hazard‑ and country‑agnostic climate‑risk framework. This update focuses on the India Flood Module, built entirely for scientific transparency, public understanding, and long‑term climate‑risk planning. We now have a clean, fully reproducible pipeline that converts daily CMIP6 rainfall into annual rainfall‑extreme indicators for India, conditioned on CHIRPS observations to maintain physical realism. The land‑masking and susceptibility component uses the Flood Susceptibility Index (FSI), derived from INDOFLOODS catchment characteristics and gauge metadata. At this stage, susceptibility is computed only in regions where INDOFLOODS provides empirical information; extension into proxy basins using HydroBASINS will come in a future release. A big thank you to the INDOFLOODS team for making high‑quality observational data publicly available — it forms the backbone of the susceptibility layer. For future rainfall extremes (2027–2100), we use CMIP6 SSP3‑7.0, which provides a more plausible socio‑economic trajectory for India. Tail‑risk exploration using SSP5‑8.5 can be run through the same workflow. This pipeline includes: ·      Rolling 2‑day rainfall from CMIP6 ·      CHIRPS‑based 95th percentile thresholds (1995–2024) ·      Land‑only masking using FSI ·      Annual maximum 2‑day rainfall ·      Annual frequency of extreme 2‑day rainfall events This is not a full flood‑hazard model — it is the susceptibility + rainfall‑extreme foundation on which a hydrological module can be built. If you’re a hydrologist interested in collaborating on the next stage, I’d love to connect. CCART is an open‑source, country‑agnostic climate‑risk framework designed for directional risk estimation (e.g., TCFD) and long‑term planning. The methodology and code are fully open source and free to use. Documentation and scripts will be uploaded to GitHub in the coming days.