SCIENTIFIC MACHINE LEARNING · WATER & CLIMATE

Mohammad A. Farmani

Ph.D. researcher building differentiable physical models, graph neural networks, and geospatial data pipelines for environmental prediction.

Full-time opportunities across the United States · Expected Ph.D. graduation: May 2027
Tucson, Arizona · farmani@arizona.edu

Contact me ↗GitHub ↗LinkedIn ↗

Interested in ML engineering, scientific ML research, geospatial data engineering, and physical climate risk modeling.

EXPERIENCE

Research with an engineering foundation

Graduate Research Assistant

August 2022–Present

University of Arizona · Hydrology and Atmospheric Sciences

  • Developed a memory-efficient differentiable Noah-MP framework combining PyTorch, native C++ forward kernels, and custom backward/adjoint operators for regional parameter learning without storing full long-window computation graphs.
  • Reimplemented and validated major Noah-MP physics against the reference model at double precision, comparing 116 model variables at hourly timesteps.
  • Modularized vegetation, soil, and snow physics to support interchangeable physical and learned process representations, with training workflows for snow water equivalent, evapotranspiration, runoff, and streamflow.
  • Developed RiverGraphNet, a graph-attention routing model connecting gridded runoff to directed river networks; evaluated learned routing against RAPID under common runoff forcing.
  • Built pipelines for multi-terabyte climate and geospatial datasets using xarray, Dask, SQL, DuckDB, PostGIS, Zarr, Parquet, and GeoParquet.
  • Implemented reproducible HPC experiments with SLURM, GitHub Actions, automated testing, and linting; studied the influence of meteorological forcing uncertainty on hydrologic behavior.
RiverGraphNet research ↗

Researcher

October 2018–January 2022

Tarbiat Modares University · Climate and flood modeling

  • Developed machine-learning climate downscaling workflows for high-resolution precipitation and generated intensity-duration-frequency curves for future extreme-rainfall analysis.
  • Modeled urban flood dynamics with SWMM and HEC-RAS, combining GIS/DEM-based watershed characterization, rainfall disaggregation, and flood-risk analysis.

SELECTED TECHNICAL WORK

Models, pipelines, and usable software

Differentiable Noah-MP

Built a differentiable interface around a land-surface simulator for gradient-based physical parameter learning. Combines process equations, native C++ computation, and PyTorch integration.

Focus: Differentiable simulation · Scientific computing · HPC

Research experience ↑

RiverGraphNet

Developed graph-attention routing using river topology, channel attributes, and temporal runoff history to study streamflow propagation.

Focus: Graph neural networks · Geospatial modeling · Hydrology

Research overview ↗ · Code ↗ · Dashboard ↗

Streamflow prediction

Built LSTM and Transformer workflows with held-out basin evaluation, a FastAPI service for completed runs, and a Streamlit dashboard. Includes automated tests and CI.

Focus: PyTorch · Model evaluation · Research software

Project ↗ · Code ↗

Hydrologic analysis tools

Developed Python tools for soil moisture memory and baseflow analysis, making research methods available as reusable software.

Focus: Time series · Python packages · Environmental analysis

Explore the tools ↗

Coffee suitability & forest-loss analysis

YAML-driven geospatial workflow combining NDVI, rainfall, soil moisture, and forest-loss layers. Harmonizes inputs to a common monthly grid and applies configurable suitability thresholds. Includes a clearly labeled synthetic demo.

Python · Remote sensing · Raster processing

Explore the repository ↗

Remote data jobs: SQL analysis

PostgreSQL analysis of job-posting data to compare skill demand and reported salaries, with reproducible Python visualizations. Results describe the source dataset rather than the current job market.

SQL · PostgreSQL · Data visualization

Explore the repository ↗
Additional data science and coursework projects
  1. Bayesian Inference of 3D Line Parameters (Metropolis–Hastings MCMC)
    Implemented custom MCMC sampler to infer 3D line endpoints from noisy 2D camera images.
    Used Gaussian priors and multi-view likelihood to achieve accurate MAP reconstruction.
  2. Fairness in Insurance Pricing & Vaccine Misinformation Ethics (Data Ethics)
    Analyzed independence, separation, and sufficiency criteria in ML fairness.
    Conducted ethical analysis on criminalization of vaccine misinformation using real-world scenarios.
  3. Tucson Car Collision Dashboard (Data Visualization – Quarto, R)
    Built interactive dashboard visualizing traffic accidents (2018–2023).
    Used heatmaps, time series, and violation-based statistics to support road safety planning.
  4. Crash Severity Prediction (Data Mining – ML)
    Trained and evaluated Random Forest, Decision Tree, KNN models on Massachusetts crash data.
    Achieved ~78% accuracy; identified environmental and demographic risk factors.
  5. Global Temperature Trends (Time Series + ML)
    Analyzed NASA GISS data (1880–2023) using Mann–Kendall, regression, STL decomposition.
    Applied ML classifiers (XGBoost, RF) to detect and forecast warming anomalies.
  6. Neural Networks & Codabench NLP Competition (Deep Learning)
    Built feedforward NN from scratch (NumPy); Developed Keras architectures (deep vs wide MLP, ReLU vs tanh).
    Implemented CNN and RNN models for MNIST and YouTube spam datasets; Built multi-label NLP classifier using BERT.
  7. Streamflow Prediction Dashboard (LSTM + Transformers)
    Built a modular ML workflow for streamflow prediction using sequence models across CAMELS-style catchments.
    Developed CLI-driven training/evaluation pipelines and deployed an interactive Streamlit dashboard for exploring model results.
  8. Smart Water Distribution and Optimization System (SWDOS) (SQL/DB)
    Designed conceptual–logical–physical database models integrating IoT sensor data.
    Implemented advanced SQL queries: JOINs, subqueries, CASE, aggregation. Built schema for real-time water monitoring and decision support.

TECHNICAL SKILLS

Tools behind the work

Machine learning & simulation
PyTorch, differentiable simulation, automatic differentiation, custom backward operators, graph neural networks, LSTMs, Transformers, scikit-learn, TensorFlow, XGBoost.
Programming & scientific computing
Python, C++, Fortran, SQL, NumPy, SciPy, pandas, Linux, Bash.
Geospatial & climate data
xarray, Dask, GeoPandas, rasterio, GDAL, PostgreSQL/PostGIS, DuckDB, NetCDF, GRIB2, Zarr, HDF5, Parquet, GeoParquet.
Engineering & delivery
SLURM, HPC workflows, Git/GitHub, GitHub Actions, automated testing, reproducible environments, FastAPI, Streamlit.
Water & climate modeling
Noah-MP, RAPID, NGen hydrofabric, SWMM, HEC-RAS, climate downscaling, river routing, land-surface processes, flood modeling.

EDUCATION

Hydrology meets data science

SELECTED PUBLICATIONS

Research outputs

More publications ↗

Awards & Honors

2025 – Arizona Hydrological Society (AHS) Academic Scholarship
2025 – 2nd Place, CAP Award for Water Research
2025 – Shlomo and Yael Neuman Hydrology Scholarship
2025 – Water Prediction Innovators Summer Institute
2025 – CIROH HydroLearn Fellowship
2024 – 2nd Place, Groundwater Student Contest, AGU24
2024 – John W. & Margaret Harshbarger Fellowship
2023 – Roots for Resilience Fellowship, Data Science Institute
2023 – Best Oral Presentation, El Dia Conference
2023–25 – Hydrologist Scholarship, US Army Corps of Engineers – ERDC CHL