My Tran

Causal Inference in Environmental Policy

Satellite Embeddings • Causal Inference • Policy Evaluation

From space to counterfactuals: I bring satellite data into causal policy analysis to reveal the physical world behind policy outcomes.

Explore My Research →

Most environmental policy studies stop at

"Did it work there?"

I ask a different question:

"What physical and social conditions made it work, and where else do those conditions exist?"

⚠️

The Problem

Policies fail when deployed in the wrong environments.

📊

What Policy Sees

GDP, yields, prices, enrollment, emissions.

🛰️

What Matters

Soil moisture, vegetation stress, heat, water flow.

My research uses three tools:

1

Satellites

Direct observation of land, climate, infrastructure, and human activity

2

Causal Inference

Isolate what would have happened without the policy

3

Bayesian Decision Theory

Choose policies under uncertainty across space

My Methodological Pipeline

Not just "Did it work?" but "Where in the physical world will it work again?"

🛰️ Measure
🔍 Infer
🗺️ Transfer
📊 Stakeholders
⚖️ Decide
🛰️

Measure

What is happening on the ground?

Satellite remote sensing, computer vision, and geospatial data processing extract ground truth from high-dimensional imagery: damage maps, spectral signatures, temporal dynamics across complex landscapes.

Tools: Sentinel-2, SMAP, BRIGHT, TensorFlow, PyTorch, Vision Transformers, Google Earth Engine

Computer Vision & Remote Sensing

Feature extraction from satellite data to detect changes and classify land-use transitions. Foundation models (PrithVi, CLIP) create learned representations for ecological similarity.

View Project →
🔍

Infer

What changed because of the intervention?

Causal inference via synthetic control, embedding-based matching, and pre-treatment balance diagnostics isolates treatment effects when randomization is impossible and confounding is latent and spatial.

Methods: Synthetic Control, PrithVi V2 embeddings, MODIS FIRMS, Spatial Statistics

Prescribed Burn Evaluation via Satellite-Based Synthetic Control

Embedding-informed donor selection improves causal inference. Embedding-matched controls reduce pre-treatment error by 49%, revealing that low-intensity prescribed fire reduces wildfire risk by 4.4%.

View Project →
🗺️

Transfer

Where else should this policy be applied?

Policy transportability assessment combines biophysical suitability mapping, spatial heterogeneity analysis, and institutional feasibility to reveal where evidence-based interventions can credibly be scaled.

Framework: Water balance modeling, spatial statistics, institutional analysis

Spatial Policy Transportability: AWD in Japan

Vietnam's water-saving agricultural policy evaluated for Japan transfer. Water balance modeling reveals AWD is feasible in only 18% of Japan's paddies. Spatial fragmentation raises extension costs 3.8×.

View Project →
📊

Stakeholders

How do different stakeholders respond to policy?

Economic modeling of policy incidence, firm-level distributional effects, and general equilibrium dynamics reveals unintended consequences and identifies which stakeholders bear the true cost.

Tools: Firm-level econometrics, general equilibrium modeling, trade policy analysis

Trade Policy Incidence: Vietnam–China HRC Tariffs

Vietnam's 23–28% anti-dumping tariffs on Chinese steel. Firm-level analysis reveals inverted incidence: mills profit, manufacturers lose. 35% exemptions create institutional capture.

View Project →
⚖️

Decide

Which policy should be chosen under uncertainty?

Bayesian model comparison, posterior predictive checks, and PSIS-LOO validation select among competing policy mechanisms when data are limited and outliers violate normality assumptions.

Framework: PyMC, PSIS-LOO, Bayesian inference, heavy-tailed distributions

Bayesian Model Comparison: Argentina Fiscal Policy

Which model best predicts Argentina's government wage spending? Student-T cubic model wins decisively—200 million times more likely. Heavy tails capture outliers; cubic captures non-linear thresholds.

View Project →