> For the complete documentation index, see [llms.txt](https://docs.alphageo.ai/llms.txt). Markdown versions of documentation pages are available by appending `.md` to page URLs; this page is available as [Markdown](https://docs.alphageo.ai/methodology/methodology-docs/migration-monitor-methodology.md).

# Migration Monitor - Methodology

### Introduction

Climate-driven relocation is rarely a linear response to physical hazards. The decision to stay or leave a flood-prone area is shaped by inundation severity, but equally by whether a household can afford to move, whether local jobs anchor them in place, and whether housing markets have already priced in the risk. Wealthier residents may exit early while lower-income households remain precisely because flood-exposed land is the only affordable option left, making climate migration as much an economic story as an environmental one.  

To disentangle these forces, this model estimates how much of future population change is driven specifically by flooding versus ordinary demographic trends separating the baseline trajectory (what population would do under normal demographic conditions) from the climate consequence (the additional population loss or gain attributable specifically to flood risk). The model provides a direct empirical measure of where flood-driven displacement is already embedded in demographic trends and where it is accelerating, turning climate risk into a concrete, quantifiable signal for population change. 

### Data Sources

<table data-header-hidden data-header-sticky><thead><tr><th width="262.20001220703125"></th><th></th><th></th></tr></thead><tbody><tr><td><strong>Feature</strong> </td><td><strong>Source</strong> </td><td><strong>Time Period</strong> </td></tr><tr><td>Population Count </td><td>US Census (Decennial) </td><td>2000, 2010, 2020 </td></tr><tr><td>Future Population </td><td>NASA SEDAC / Hauer (2019) </td><td>2020–2100 </td></tr><tr><td>Population Density </td><td>US Decennial Census Gazetteer </td><td>2010 </td></tr><tr><td><p>Total Jobs  </p><p>(job density, job growth rate) </p></td><td>Census LODES / LEHD </td><td>2004–2013 </td></tr><tr><td><p>Socioeconomic data  </p><p>(median income, home values, employment rates) </p></td><td>American Community Survey (ACS) </td><td>2018 </td></tr><tr><td>Flood Inundation </td><td>AlphaGeo </td><td>2025 and 2050  </td></tr></tbody></table>

**Future population projections** are drawn from the NASA SEDAC dataset produced by Hauer (2019), which provides county-level SSP projections in five-year increments from 2020 to 2100 for all five scenarios. These county totals are disaggregated to the block level using each block's proportional share of its county's 2010 population, preserving fine-grained spatial variation within counties.&#x20;

### Building the Statistical Model&#x20;

The model uses Ordinary Least Squares (OLS) regression to learn the relationship between local characteristics and observed population change. The outcome variable of regression is the annualized percentage change in census block population observed between 2000 and 2020, used to train the model on how flood exposure and local economic conditions jointly shaped past demographic change. Beyond flood exposure, the model controls for a range of structural factors: job density captures the concentration of employment and its pull on residents; job growth rate (the compound annual growth rate of jobs from 2004 to 2013) proxies for pre-period economic momentum; population density reflects how built-out an area already is; median income and median home value describe the economic character of the area; and employment rate measures the share of the labour force actively working. Together these controls isolate the flood signal from broader economic and demographic drivers that independently shape where people choose to live.&#x20;

Two model specifications are compared using stepwise AIC (Akaike Information Criterion) selection: a required model containing only flood exposure interactions, and a full model incorporating all predictors including socioeconomic controls, geographic coordinates, and squared (quadratic) terms to capture nonlinear effects. The better-fitting model judged by AIC is selected separately for each US state, allowing flood impacts to vary across regional contexts.&#x20;

The central identification strategy mimics a treatment-control experiment. Census blocks with meaningful flood inundation are the treated group; similar unflooded blocks serve as controls. Propensity score matching is applied before modelling to ensure treated and control blocks are comparable on observable characteristics, reducing the risk that flood effects are confounded with pre-existing differences in income or density. Interaction terms between flood inundation and treatment status let the model estimate how the intensity of flooding shapes demographic outcomes. Coefficients that are not statistically significant are set to zero before forecasting, preventing overfitting to noise.&#x20;

### Output&#x20;

All three outputs are expressed as a percentage rate over the 30-year window from approximately 2020 to 2050.&#x20;

**Baseline Future Change:** What would population growth look like if flooding did not exist? We calculate this as the annualized percentage change between the SSP2 current and future populations over a 30-year horizon to 2050. SSP2\_CUR and SSP2\_FUT are each the average of their two nearest SSP2 snapshots (2020+2025 and 2050+2055 respectively).

<p align="center"><em><mark style="color:$info;">B_CHG_FUT = [ (SSP2_FUT - SSP2_CUR) / SSP2_CUR] x (100 / 30)</mark></em> </p>

<p align="center"><sub><mark style="color:$info;">SSP2_CUR = average of SSP2 projections for 2020 and 2025 | SSP2_FUT = average of 2050 and 2055</mark></sub></p>

**Projected Population:** This is our climate-adjusted population estimate. It is built from the current SSP2 baseline and adds all the modelled effects — flood exposure, economic conditions, geography, and everything else our statistical model learned.&#x20;

<p align="center"><em><mark style="color:$info;">P_FUT = SSP2_CUR + (SSP2_CUR x 30/10000 x Σ modeled effects)</mark></em> </p>

**Projected Future Change:** Annualized percentage change in population using the climate-adjusted projection.

<p align="center"><em><mark style="color:$info;">P_FUT_CHG = [(P_FUT − SSP2_CUR) / SSP2_CUR] × (100 / 30)</mark></em></p>

**Climate Consequence:** This is the key output. The extra population change attributable specifically to flood risk. It is the gap between the climate-adjusted projection and the baseline. It answers how much of the change is due to flooding.&#x20;

<p align="center"><em><mark style="color:$info;">CLIM_CONSQ = P_FUT_CHG - B_CHG_FUT</mark></em> </p>

<p align="center"><sub><mark style="color:$info;">Negative value = flood risk is pushing people out faster than natural trends. Positive = flood risk areas still growing.</mark></sub></p>

### Result Interpretation&#x20;

A negative CLIM\_CONSQ means that flood risk is accelerating population loss beyond what demographic trends alone would predict, people are leaving or not arriving, specifically because of flooding. A value near zero indicates that flood risk has little additional demographic effect in that location, and natural trends dominate. A positive value means a flood-exposed area is still growing despite risk, which may reflect high desirability (waterfront access, strong job markets) or an undervaluation of long-term hazard. Together, these outcomes map the full spectrum of how flooding reshapes where people live. These results give planners, investors, and policymakers a clear and granular picture of where flood risk is driving demographic change, thus supporting smarter decisions on where to build, invest, and intervene.&#x20;

### References&#x20;

1. Hauer, M. E., & Center for International Earth Science Information Network (CIESIN), Columbia University. (2020). Georeferenced U.S. county-level population projections, total and by sex, race and age, based on the SSPs, 2020–2100 (Version 1.00). NASA Socioeconomic Data and Applications Center (SEDAC). <https://doi.org/10.7927/DV72-S254>&#x20;
2. U.S. Census Bureau. (2020). Decennial Census \[Block-level population counts, P001001]. <https://www.census.gov/programs-surveys/decennial-census.html>&#x20;
3. U.S. Census Bureau. (2018). American Community Survey 5-year estimates \[Tables B19013, B25077, B23025]. <https://www.census.gov/programs-surveys/acs>&#x20;
4. U.S. Census Bureau. (2013). LEHD Origin-Destination Employment Statistics (LODES8) \[Workplace Area Characteristics, C000]. <https://lehd.ces.census.gov/data/lodes/LODES8>&#x20;
5. U.S. Census Bureau. (2010). Census Gazetteer Files — Tracts \[Land area, ALAND10]. <https://www.census.gov/geographies/reference-files/time-series/geo/gazetteer-files.html>&#x20;
6. AlphaGeo. (2025). Flood inundation layers \[H3 resolution 8, 100-year return period, 2025 & 2050 scenarios]. AlphaGeo proprietary dataset.&#x20;


---

# Agent Instructions
This documentation is published with GitBook. GitBook is the documentation platform designed so that both humans and AI agents can read, navigate, and reason over technical content effectively. Learn more at gitbook.com.

## Querying This Documentation
If you need additional information that is not directly available in this page, you can query the documentation dynamically by asking a question.

Perform an HTTP GET request on the current page URL with the `ask` query parameter:

```
GET https://docs.alphageo.ai/methodology/methodology-docs/migration-monitor-methodology.md?ask=<question>
```

The question should be specific, self-contained, and written in natural language.
The response will contain a direct answer to the question and relevant excerpts and sources from the documentation.

Use this mechanism when the answer is not explicitly present in the current page, you need clarification or additional context, or you want to retrieve related documentation sections.