Applying ML to local climate patterns

🌍 Climate scientists often face a trade-off: Global Climate Models (GCMs) are essential for long-term climate projections — but they operate at coarse spatial resolution, making them too crude for regional or local decision-making. To get fine-scale data, researchers use Regional Climate Models (RCMs). These add crucial spatial detail, but come at a very high computational cost, often requiring supercomputers to run for months. ➡️ A new paper introduces EnScale — a machine learning framework that offers an efficient and accurate alternative to running full RCM simulations. Instead of solving the complex physics from scratch, EnScale "learns" the relationship between GCMs and RCMs by training on existing paired datasets. It then generates high-resolution, realistic, and diverse regional climate fields directly from GCM inputs. What makes EnScale stand out? ✅ It uses a generative ML model trained with a statistically principled loss (energy score), enabling probabilistic outputs that reflect natural variability and uncertainty ✅ It is multivariate – it learns to generate temperature, precipitation, radiation, and wind jointly, preserving spatial and cross-variable coherence ✅ It is computationally lightweight – training and inference are up to 10–20× faster than state-of-the-art generative approaches ✅ It includes an extension (EnScale-t) for generating temporally consistent time series – a must for studying events like heatwaves or prolonged droughts This approach opens the door to faster, more flexible generation of regional climate scenarios, essential for risk assessment, infrastructure planning, and climate adaptation — especially where computational resources are limited. 📄 Read the full paper: EnScale: Temporally-consistent multivariate generative downscaling via proper scoring rules ---> https://lnkd.in/dQr5rmWU (code: https://lnkd.in/dQk_Jv8g) 👏 Congrats to the authors — a strong step forward for ML-based climate modeling! #climateAI #downscaling #generativeAI #machinelearning #climatescience #EnScale #RCM #GCM #ETHZurich #climatescenarios

Monsoon reliance: 60% of agriculture depends on it, but climate change is disrupting patterns. Transboundary water stress: AI can help manage Indus Water Treaty conflicts with India. High vulnerability: Ranked 5th in Global Climate Risk Index (floods/droughts cost ~$4B/year). Here are few topics where one can use AI/ ML with Hydrologic and Hydro meteorological analysis 1. AI-Enhanced Monsoon Forecasting for Pakistan Focus: Improve short-to-medium-term monsoon predictions (onset, intensity, spatial distribution) using LSTMs, Transformers, or hybrid AI models. Data: IMD, PMD (Pakistan Meteorological Department) records, satellite data (CPC, GPM), and oceanic indices (ENSO, IOD). Application: Reduce agricultural losses (e.g., Punjab’s rice/wheat belts) and flood risks in Sindh/Balochistan. 2. Flash Flood Early Warning with IoT & AI Challenge: Sudden floods in mountainous regions (KPK, Gilgit-Baltistan) due to erratic rainfall. Solution: Real-time sensor networks (river gauges, soil moisture) + CNN-RNN hybrids for flood modeling. Social media NLP (Urdu/Pashto text mining) to validate crowdsourced flood reports. 3. Drought Prediction for Indus Basin Agriculture Problem: Increasing droughts in southern Punjab & Sindh threaten food security. ML Approach: Satellite data (MODIS/SPEI) + clustering (k-means) to classify drought severity. Explainable AI (XAI) to pinpoint climate drivers (e.g., delayed monsoons, heatwaves). 4. Glacial Lake Outburst Flood (GLOF) Risk Assessment Focus: Monitor Himalayan/Karakoram glaciers (e.g., Shisper Glacier surges). Tools: U-Net segmentation on Sentinel-2/Landsat images to track glacial lakes. Bayesian Networks to predict GLOF triggers (temperature, snowfall). 5. AI for Reservoir & Irrigation Management Issue: Poor water allocation in Indus River System (Tarbela, Mangla dams). AI Solutions: Reinforcement Learning (RL) to optimize dam releases during monsoons. Federated Learning for cross-province data sharing (without raw data exchange). 6. Urban Flood Modeling for Islamabad, Peshawar, Karachi & Lahore Challenge: Poor drainage + intense rainfall = chronic urban flooding. Method: Graph Neural Networks (GNNs) to model drainage networks. Digital Twins simulating floods under SSP2 4.5/SSP5 8.5 scenarios. 7. Cyclone & Storm Surge Prediction for Coastal Pakistan Threat: Rising cyclones (e.g., 2022’s Cyclone Biparjoy) hitting Sindh coast. AI Tools: Physics-informed Neural Networks (PINNs) for surge modeling. Transformer models to predict cyclone tracks using ERA5 data. 8. Community-Driven Flood Alerts via AI + Citizen Science Idea: Mobile apps for farmers in flood-prone areas (e.g., Jacobabad) to report rainfall/floods. ML Integration: Train models on crowdsourced data + satellite inputs for hyper-local warnings.

Flood science has historically been trapped between two extremes: hydrodynamic models that are highly accurate but computationally expensive, or global models that are too coarse (>1 km) to capture critical local vulnerabilities. Bridging this divide requires a fundamental shift from physics-based deduction to data-driven induction, a challenge that has defined my research over the last four years. This week, I am very happy to share that I have formalized this solution by submitting my Ph.D. thesis at Hong Kong Baptist University: "Towards GeoAI-based Data-driven Flood Management Solutions: A Synergistic Machine Learning and Earth Observation Framework" As illustrated, the thesis establishes a scalable GeoAI framework built on three synergistic pillars: 1. High-Dimensional Earth Observation (The Data) Leveraging multi-temporal global data streams (Landsat, Sentinel) to transition the field from data scarcity to data abundance. 2. Planetary-Scale Geo-Computation (The Platform) Utilizing cloud clusters (Google Earth Engine) and HPC (Shaheen-III) to democratize processing power, enabling the analysis of petabyte-scale geospatial data without traditional hardware constraints. 3. Machine Learning Analytics (The Engine) We systematically benchmarked 14 ML architectures to resolve the "accuracy-efficiency" trade-off, establishing a robust modeling engine. This framework was first operationalized across Pakistan's diverse landscapes to reveal that 95 million people reside in high-risk zones, before being scaled globally to produce the first harmonized 30 m flood susceptibility baseline. The Output: Global Flood Susceptibility Map (GFSM v1) By applying a climate modeling scheme (across 192 climate zones), we produced the first globally harmonized, 30 m resolution flood susceptibility baseline derived entirely from open-access data. This research addresses the "data equity deficit" in the Global South, where 89% of flood-exposed populations reside, often without high-resolution risk data. Next Steps: I will be releasing the open-source code, the GFSM v1 dataset, and the GEE web applications in the coming weeks. If you are interested in the work, feel free to drop a message to dicsuss further possibilities! For more info, feel free to check my updated portfolio: www.waleedgeo.com #geoai #earthengine #floodrisk #remotesensing #hkbu #datascience #gfsm #flood

Infrastructure in #northern regions is increasingly threatened by climate change, mainly due to #permafrost thaw. This has direct implications for community #safety, #economic development, and long-term #resilience of built assets across Arctic and sub-Arctic regions. A key challenge in integrating climate change into geotechnical and #infrastructure #design in northern regions is the representation of the surface energy balance (SEB). In our newly published paper in Cold Regions Science and Technology, titled “Ground surface boundary condition methods for analysis of climate-driven permafrost thaw: A comparative study and long-term projections for Nunavik, Canada,” led by my former PhD student Dr. Ali Gheysari, we present a data-driven, #machinelearning-based approach to represent ground surface thermal forcing in permafrost simulations using #ERA5-Land climate reanalysis data through 2100. We compare this ML-based method with SEB heat-flux approaches and traditional n-factors to evaluate how different ground surface boundary condition methods influence projections of climate-driven permafrost thaw. Using #Nunavik as a case study, we provide a comparative assessment of commonly used modeling approaches and show that surface forcing choices can significantly alter long-term #thaw predictions. These differences directly affect #risk assessments and #engineering design decisions for infrastructure systems in cold regions. Our objective was to identify the most effective approach for predicting ground surface temperatures to support climate-resilient design of northern infrastructure. Results indicate that the ML-based method outperforms both SEB heat-flux and n-factor approaches, with substantially lower prediction errors. The feasibility of long-term thermal analysis using ML-predicted ground surface temperatures is demonstrated through a permafrost case study in #Salluit, where active layer thickness and talik development are projected under moderate and extreme climate scenarios by the end of the 21st century. We also discuss the applicability and limitations of surface boundary condition methodologies, including the limited suitability of n-factors for long-term analysis and the sensitivity of SEB heat-flux methods to input data and thermal imbalance. The findings highlight the importance of selecting appropriate boundary condition methodologies to improve the reliability of geotechnical analyses in cold regions. Link to the paper: https://lnkd.in/gAHZ4VkG

📊🤖 𝗦𝗼𝗶𝗹 𝗠𝗼𝗶𝘀𝘁𝘂𝗿𝗲 𝗔𝗻𝗼𝗺𝗮𝗹𝘆 & 𝗠𝗮𝗰𝗵𝗶𝗻𝗲 𝗟𝗲𝗮𝗿𝗻𝗶𝗻𝗴 𝗙𝗼𝗿𝗲𝗰𝗮𝘀𝘁 — 𝗦𝗼𝗺𝗮𝗹𝗶𝗹𝗮𝗻𝗱 (𝗟𝗮𝘁𝗲𝘀𝘁 𝗠𝗼𝗻𝘁𝗵) I conducted a 𝐌𝐚𝐜𝐡𝐢𝐧𝐞 𝐋𝐞𝐚𝐫𝐧𝐢𝐧𝐠 (𝐌𝐋)–based forecasting analysis to capture and predict 𝐬𝐡𝐨𝐫𝐭-𝐭𝐞𝐫𝐦 𝐬𝐨𝐢𝐥 𝐦𝐨𝐢𝐬𝐭𝐮𝐫𝐞 𝐯𝐚𝐫𝐢𝐚𝐛𝐢𝐥𝐢𝐭𝐲 across 𝐒𝐨𝐦𝐚𝐥𝐢𝐥𝐚𝐧𝐝, enhancing drought early-warning capability. Combined with anomaly mapping, this analysis helps identify 𝐝𝐫𝐲𝐧𝐞𝐬𝐬 𝐡𝐨𝐭𝐬𝐩𝐨𝐭𝐬, 𝐦𝐨𝐧𝐢𝐭𝐨𝐫 𝐡𝐲𝐝𝐫𝐨𝐥𝐨𝐠𝐢𝐜𝐚𝐥 𝐬𝐭𝐫𝐞𝐬𝐬, and 𝐬𝐮𝐩𝐩𝐨𝐫𝐭 𝐞𝐚𝐫𝐥𝐲 𝐰𝐚𝐫𝐧𝐢𝐧𝐠 𝐬𝐲𝐬𝐭𝐞𝐦𝐬. 🔍 𝗔𝗽𝗽𝗿𝗼𝗮𝗰𝗵 • 𝐃𝐚𝐭𝐚𝐬𝐞𝐭: ERA5-Land Soil Moisture (ECMWF) • 𝐁𝐚𝐬𝐞𝐥𝐢𝐧𝐞: 2000–2025 monthly climatology • 𝐌𝐋 𝐅𝐨𝐫𝐞𝐜𝐚𝐬𝐭𝐢𝐧𝐠: 6-month Machine Learning time-series model (Prophet) • 𝐏𝐫𝐨𝐜𝐞𝐬𝐬𝐢𝐧𝐠: Moisture anomaly computation, clipping, spatial integration • 𝐏𝐥𝐚𝐭𝐟𝐨𝐫𝐦𝐬: 𝐆𝐨𝐨𝐠𝐥𝐞 𝐄𝐚𝐫𝐭𝐡 𝐄𝐧𝐠𝐢𝐧𝐞 (data extraction) & 𝐑𝐒𝐭𝐮𝐝𝐢𝐨 (ML modeling + visualization) 🌦️ 𝗞𝗲𝘆 𝗜𝗻𝘀𝗶𝗴𝗵𝘁 The anomaly map reveals 𝐬𝐩𝐚𝐭𝐢𝐚𝐥 𝐝𝐞𝐯𝐢𝐚𝐭𝐢𝐨𝐧𝐬 𝐢𝐧 𝐬𝐨𝐢𝐥 𝐦𝐨𝐢𝐬𝐭𝐮𝐫𝐞, while the ML model provides a 𝐩𝐫𝐞𝐝𝐢𝐜𝐭𝐢𝐯𝐞 𝐞𝐚𝐫𝐥𝐲-𝐰𝐚𝐫𝐧𝐢𝐧𝐠 𝐬𝐢𝐠𝐧𝐚𝐥 for potential moisture deficits. Machine Learning enhances traditional drought monitoring by capturing non-linear temporal patterns that conventional methods often miss. This is essential for improving agricultural planning, range management, and climate resilience in Somaliland. 🛰️ 𝐓𝐨𝐨𝐥𝐬 𝐔𝐬𝐞𝐝 Machine Learning • Time-Series Forecasting • GEE • RStudio • Remote Sensing #Somaliland_Climate_Insights

🌍 Turning Satellite Data into Climate Action What if we could predict floods and map carbon stocks before disaster strikes or forests are lost? That’s exactly what GeoAI makes possible. In my recent work, I built a spatial ML workflow that: • Identifies flood-prone areas using multi-source satellite imagery and terrain data • Estimates carbon stock across forested landscapes • Integrates climate variables for predictive insights Tools & Techniques: Python, Google Earth Engine, ArcGIS, QGIS The result? Data-driven maps that support disaster planning, climate adaptation, and environmental decision-making — not just mapping what has happened, but modeling what could happen next. 💡 Why this matters: From local authorities to NGOs, actionable geospatial intelligence can save lives, protect forests, and guide climate policy. 🔬 I’m curious: How is your organization leveraging satellite data or GeoAI for environmental resilience? #GeoAI #SpatialML #GIS #FloodRisk #CarbonMapping #ClimateTech #EnvironmentalAnalytics #RemoteSensing

From Past Patterns to Future Warnings: My Climate ML Project Over the last few weeks, I built a Machine Learning–powered early warning system for monsoon temperature anomalies turning decades of climate data into actionable insights. Here’s what I did: Pulled historical seasonal temperature data from a MySQL database Engineered climate-specific features: rolling trends, seasonal contrasts, lag variables Used Isolation Forest to detect past unusual years no labels required Leveraged Prophet to forecast the next 10 years Flagged future anomaly risks to help plan for agriculture, water management & disaster readiness Visualized past vs. future anomalies for crystal-clear decision-making Why it matters: In India, a single unusual monsoon can disrupt food supply chains, impact millions of livelihoods, and cost billions in losses. This project transforms raw climate records into predictive intelligence that ministries, researchers, and communities can act on. Next step: Integrating rainfall & atmospheric pressure to build a multi-factor climate risk model and deploying it as a real-time API for decision-makers. If you work in climate tech, agriculture, disaster management, or AI for social impact, I’d love to connect and exchange ideas. #MachineLearning #ClimateChange #AI #DataScience #Forecasting #Prophet #IsolationForest #AgricultureTech #SustainableDevelopment #EarlyWarningSystem #AI #MachineLearning #DeepLearning #DataScience #AIResearch #NeuralNetworks #AIAutomation #BigData #AIInnovation #AIEthics #AIRevolution

I’m excited to share that I co-authored a new research article published in Discover Geoscience(Springer Nature): Machine learning-based flood inundation mapping and LULC classification in Ahoada West, Rivers State, Nigeria using satellite imagery This work explores how integrating Sentinel-1 SAR, Sentinel-2 imagery, geospatial analysis, and machine learning can improve flood mapping and land use/land cover (LULC) classification in data-scarce environments like the Niger Delta. 🔍 Key highlights: Multi-temporal analysis (2018–2023) of flood dynamics Application of ML models including Random Forest, SVM, KNN, XGBoost, and Decision Tree Random Forest achieved the best performance (91.3% accuracy, Kappa = 0.89) Significant increase in flood extent in 2022, with ~22.5% of the study area highly inundated Rapid expansion of built-up areas (+12.6%), largely at the expense of wetlands and vegetation 🌍 Impact: The study demonstrates how combining remote sensing and machine learning can support flood risk assessment, environmental monitoring, and sustainable land-use planning, especially in vulnerable regions. I’m proud to have contributed to this work alongside an amazing team of researchers. This is another step toward leveraging data science and geospatial intelligence for real-world environmental challenges. 📄 You can read the full article here: https://lnkd.in/eB35aAn9 #MachineLearning #RemoteSensing #GIS #FloodMapping #ClimateChange #Geospatial #DataScience #EnvironmentalScience #Nigeria #Research

𝐍𝐞𝐰 𝐑𝐞𝐬𝐞𝐚𝐫𝐜𝐡 𝐂𝐨𝐦𝐩𝐚𝐫𝐢𝐧𝐠 𝐃𝐞𝐞𝐩 𝐋𝐞𝐚𝐫𝐧𝐢𝐧𝐠 𝐀𝐩𝐩𝐫𝐨𝐚𝐜𝐡𝐞𝐬 𝐟𝐨𝐫 𝐅𝐥𝐨𝐨𝐝 𝐅𝐨𝐫𝐞𝐜𝐚𝐬𝐭𝐢𝐧𝐠 Flood forecasting presents unique ML challenges: multi-modal data fusion (meteorological, geographical, and soil variables), high-resolution spatial modeling, and capturing complex temporal dynamics. While foundation models promise transfer learning benefits, they can struggle with domain adaptation from global patterns to local contexts. Eric Wanjau and Samuel Maina explored data-driven approaches to flood extent forecasting in Rwanda, a region particularly vulnerable to flooding due to its mountainous terrain and increasingly frequent heavy rainfall. They compared three approaches: - A standard U-Net architecture - A ClimaX variant trained from scratch - Fine-tuned ClimaX model ClimaX is a transformer-based weather and climate foundation model. They found that a ClimaX variant trained from scratch with a linear projection decoder outperformed both the U-Net baseline and the fine-tuned ClimaX models. Perhaps most interesting was that pre-training on coarse global climate data didn't transfer effectively to the high-resolution local forecasting task in Rwanda. This suggests that foundation models might need region-specific pre-training at appropriate resolutions to bridge the gap between global patterns and local flood dynamics. https://lnkd.in/evmPuRje #ComputerVision #MachineLearning #FloodForecasting #FoundationModels