Community-Led Flood Forecasting - Thailand 2011

Innovation in Crisis: Hyperlocal Flood Predictions That Saved Homes.

Community-driven forecasting platform using self-learning models to answer one question - "Will my barrier hold?"

Bangkok Flood Prediction 2011 is an emergency response system that provided real-time flood forecasts when official models failed during Thailand's worst natural disaster. Using ARIMA time series models and spatial interpolation, it predicted water levels 5 days in advance across Bangkok's neighborhoods—empowering 13 million residents to protect their homes when government forecasts were unavailable.

All predictions were shared freely via WooWooWuu.de, filling a critical information void during Thailand's $45B flood crisis.

Predictive water level compared with the three types of floow barriers in the area

Project Overview

This project was developed in October-November 2011 during the great Thailand floods to provide real-time flood predictions for residents and authorities in the Bangkok Metropolitan Area (BMA). The predictions helped individuals and communities assess flood risk, determine whether flood protection barriers would be sufficient, and make critical decisions about protecting their homes and evacuating.

The Problem

During the 2011 floods, the Bangkok area faced a critical information gap:

• Insufficient monitoring infrastructure: Lack of monitoring stations and water level measurement points across the metropolitan area
• No suitable models: After the flood inundated greater Bangkok, existing hydrological models became ineffective for urban flood prediction
• Limited situational awareness: People in flood-affected areas could not determine if water levels were rising or falling
• No advance forecasting: The government could not forecast and announce the flood situation in advance due to the complexity of flood movement in urban areas

Animation of water level forecasting in the Bangkok area showing a brief rise followed by the decrease after a week

The Solution

This project addressed these challenges through a data-driven forecasting approach:

• Self-driving predictions: Used existing water level data to forecast water levels 5 days ahead
• Spatial interpolation: Created flood maps showing water levels across individual neighborhoods
• Actionable information: Enabled residents to make informed decisions about barrier reinforcement or relocating valuables
• Real-time dissemination: All predictions were released publicly through WooWooWuu.de website and WooWooWuu Facebook page, created specifically for flood relief

Logo of the website where the flood infomration and forecastes were puhlished

What People Needed to Know

Residents in flooded areas had simple but critical questions:

• Do I need to prepare more sandbag barriers?
• Should I relocate valuable items to higher floors?
• Will the water level rise or fall in the next few days?
• Will flood walls at 2.5m, 2.8m, or 3.0m MSL be sufficient for my area?

This project empowered individuals to mitigate their own risk with hyperlocal flood predictions.

Animation showing predicted water levels across Bangkok area, highlighting the risng level in the southern zones

---

Historical Context

The 2011 Thailand floods were among the worst flooding disasters in the country's history:

• Affected over 13 million people
• Caused estimated damages of $45.7 billion USD
• Lasted from July to December 2011
• Required large-scale evacuations and emergency responses

This project represents a real-time scientific response to a major natural disaster, developed under emergency conditions to fill a critical information void.

Water situation charts showing the water levels (blue bars) compared with the flood wall height (green bars)

---

Project Structure

Flood2011/
├── README.md                      # This file
├── docs/
│   ├── CLAUDE.md                  # Technical guide for developers
│   ├── Logo/                      # Project logos and graphics
│   └── reports/                   # Excel reports and official statements
│       └── Statements/            # Official documentation
├── data/
│   ├── raw/                       # Original CSV time series data
│   │   ├── seawl24.csv            # 24-hour sea water level observations
│   │   ├── seaPrd.csv             # Sea level predictions
│   │   └── wloct.csv              # October water level measurements
│   ├── gis/                       # GIS data organized by type
│   │   ├── base_layers/           # Bangkok district boundaries
│   │   ├── bma_output/            # BMA output shapefiles with predictions
│   │   ├── arcgis_projects/       # ArcGIS .mxd map documents
│   │   └── rasters/               # Raster imagery and KML files
│   └── processed/                 # Exported/processed data
│       └── Export/                # Date-specific shapefiles and exports
├── scripts/
│   ├── arima/                     # ARIMA time series prediction models
│   │   ├── arimaBKK.R             # Initial prediction (Oct 21-30, 2011)
│   │   ├── arimaBKK-2.R           # Updated prediction (Nov 8-30, 2011)
│   │   ├── testAR.R               # Model testing and selection
│   │   └── *.csv                  # Historical climate data (GCM, ocean indices)
│   └── spatial/                   # Spatial interpolation scripts
│       ├── interpolateV1.R        # IDW interpolation (initial)
│       ├── interpolateV2.R        # IDW interpolation (improved)
│       └── readSHP.R              # Shapefile utilities
├── output/
│   ├── predictions/               # ARIMA model prediction outputs
│   │   ├── AR_BKK10Nov/           # Nov 10, 2011 predictions
│   │   ├── AR_BKK10Nov2/          # Updated Nov predictions
│   │   └── AR_BKK25oct/           # Oct 25, 2011 predictions
│   ├── maps/                      # Forecast maps and visualizations
│   │   ├── forecasting/           # Daily status maps and animations
│   │   │   ├── BKK*.png           # Daily flood status maps
│   │   │   └── *.gif              # Animated flood progression
│   │   └── IDW*.pdf               # Spatial interpolation outputs
│   └── charts/                    # Water level time series charts
│       ├── WL Charts/             # Water level forecast charts
│       └── WL-ts/                 # Time series visualizations
└── archives/                      # Archived RAR files

---

Methodology

1. Time Series Prediction (ARIMA/AR Models)

Location: scripts/arima/

The project uses AutoRegressive Integrated Moving Average (ARIMA) models to forecast water levels:

• AR (AutoRegressive) Models: Orders 1-2 tested for simple predictions
• ARIMA Models: Various parameter combinations (p,d,q) tested with stepwise and non-stepwise selection
• ARIMAx (External Regressors): Used sea level and ocean climate indices as external predictors for improved accuracy

Key Scripts:

• arimaBKK.R: Initial prediction model using 25 days of water level data from September 2011
• arimaBKK-2.R: Updated model using 18 days of data for November predictions
• testAR.R: Comprehensive AR/ARIMA model testing with multiple GCM scenarios and ocean indices

Data Sources:

• Water Level Observations: Chao Phraya River at Royal Thai Navy Headquarters
• Sea Level Data: Tidal predictions and observations (24-hour cycles)
• Ocean Indices: ENSO, SST, and other climate indicators (1971-2009)
• Climate Models: ECHO-G GCM projections (A1B, A2, B1 scenarios) for robustness testing

Model Performance:

• Nash-Sutcliffe Efficiency (NSE): Measures predictive power (-∞ to 1, where 1 = perfect)
• Root Mean Square Error (RMSE): Measures prediction accuracy in meters
• Forecast Horizon: 5 days ahead with confidence intervals
• Validation: 50% calibration, 50% verification split-sample approach

Best Performing Model:

auto.arima(water_level, xreg=sea_level, d=1, stepwise=FALSE)

This ARIMAx model with sea level as external regressor provided the most accurate 5-day forecasts.

2. Spatial Interpolation

Location: scripts/spatial/

Inverse Distance Weighting (IDW) interpolation creates continuous flood risk maps from point observations:

• Grid Resolution: 1000m × 1000m
• Extension: 5000m buffer around observation points
• Input: Water level measurements from multiple stations across Bangkok
• Output: Raster surfaces showing predicted water levels across BMA
• Method: gstat::krige() function in R

Key Scripts:

• interpolateV1.R: Initial IDW implementation
• interpolateV2.R: Improved version with better grid generation (recommended)
• readSHP.R: Utilities for reading Bangkok administrative boundaries

Process:

1. Collect point measurements from distributed stations
2. Generate uniform grid across BMA
3. Apply IDW algorithm to interpolate between points
4. Overlay Bangkok district boundaries
5. Generate PDF maps showing flood levels by neighborhood

3. Flood Risk Assessment

Critical Thresholds:

• 2.5m MSL: General flood protection wall height
• 2.8m MSL: Top-up wall height
• 3.0m MSL: Extra top-up wall height

The predictions compared forecasted water levels against these thresholds to assess flood risk for different districts, helping residents determine if their barriers would hold.

4. Visualization and Dissemination

Location: output/maps/forecasting/

Daily flood status maps were generated showing:

• Current water levels
• Predicted water levels (5-day forecast)
• Areas at risk of inundation
• Flow direction and magnitude
• Comparison against flood wall heights

Map Versions:

• V2-V7: Progressive improvements in accuracy and detail throughout October-November
• Animations: GIF files showing flood progression over time
• Charts: Time series plots with prediction intervals and confidence bounds

Map showing the rate of water level changes in Bangkok, green indicates decreasing level, red is the increasing one. The size of each circle represents the magnitude of change

Public Release: All predictions and maps were published on:

• Website: WooWooWuu.de
• Social Media: WooWooWuu Facebook page

These platforms were created specifically for flood relief and provided free access to predictions for all residents.

Collaboration with citizen-initiative partners during the crisis period

---

Technical Requirements

R Packages Required

# Time series analysis
library(timeSeries)
library(TSA)
library(forecast)
library(Kendall)
library(Rwave)
library(wmtsa)

# Spatial analysis
library(sp)
library(maptools)
library(gstat)
library(rgdal)

# General utilities
library(plotrix)
library(car)
library(DAAG)
library(lattice)

Software

• R: Statistical computing and graphics (version 2.x or higher)
• ArcGIS: Spatial mapping and visualization (for .mxd files)
• Excel: Data preprocessing and visualization

---

Running the Models

ARIMA Predictions

cd scripts/arima
Rscript arimaBKK.R      # October predictions
Rscript arimaBKK-2.R    # November predictions

Output: PDF charts in output/predictions/AR_BKK*/ARp-charts.pdf

Spatial Interpolation

cd scripts/spatial
Rscript interpolateV2.R

Output: Multi-page PDF in output/maps/ with interpolated flood surfaces

---

Key Data Files

Time Series Data

File	Description	Location
seawl24.csv	24-hour tidal cycle water level observations (Oct 11-24)	data/raw/
seaPrd.csv	Sea level predictions for future dates	data/raw/
wloct.csv	October water level measurements at Navy HQ	data/raw/
wlsep.csv	September water level data (Days 21-30)	scripts/arima/
wlsep2.csv	September water level data (Days 8-30)	scripts/arima/

GIS Data

File	Description	Location
bma_output.*	BMA shapefile with water level attributes	data/gis/bma_output/
BKK*.mxd	ArcGIS map documents for various dates	data/gis/arcgis_projects/

Reports

File	Description	Location
SeaLevel Navy Headquarter v2.xls	Comprehensive sea level observations	docs/reports/
Status WL*.xls	Water level status reports (versions 1-10)	docs/reports/

---

Results and Impact

Applications

The predictions enabled:

1. Individual risk mitigation: Residents could prepare barriers or evacuate based on 5-day forecasts
2. Community planning: Neighborhoods coordinated barrier construction and evacuation efforts
3. Infrastructure protection: Businesses and institutions protected assets based on predicted water levels
4. Public awareness: Daily maps showed which areas were at risk, reducing uncertainty

Public Dissemination

• Website: WooWooWuu.de hosted all daily predictions and maps
• Facebook: WooWooWuu page provided updates and answered questions from affected residents
• Free Access: All information provided at no cost during the emergency

This grassroots approach filled the information gap when official forecasts were unavailable, empowering individuals to protect themselves and their communities.

---

Data Sources

1. Royal Thai Navy Headquarters: Water level observations from Chao Phraya River
2. Thai Meteorological Department: Historical climate data
3. ECHO-G Global Climate Model: Future climate projections (1971-2100) - used for model robustness testing
4. Ocean Climate Indices: ENSO, SST, and other indicators (1971-2009) - used as external regressors

---

Important Notes

• All water levels are in meters above Mean Sea Level (MSL)
• Predictions were made daily during October-November 2011
• The project combined statistical models (ARIMA) with spatial interpolation (IDW) for comprehensive coverage
• File structure reflects iterative development during the emergency
• Some data files remain in scripts/arima/ as they are script-specific
• Coordinate system: UTM Zone 47N (projected coordinates)

---

Author

Project developed in 2011 by Werapol Bejranonda for Bangkok flood prediction and risk assessment.

Public Platforms:

• Website: WooWooWuu.de (created for flood relief)
• Facebook: WooWooWuu (flood information and community support)

---

License

Historical research project. Data sources should be properly attributed if reused.

---

References

• 2011 Thailand floods: Wikipedia
• ARIMA modeling: Box, G.E.P., & Jenkins, G.M. (1976). Time Series Analysis: Forecasting and Control
• Spatial interpolation: Li, J., & Heap, A.D. (2008). A review of spatial interpolation methods for environmental scientists

---

Acknowledgments

This project was developed under emergency conditions to serve the people in Thailand during one of Thailand's worst natural disasters. Special thanks to all who contributed data, provided feedback, and helped disseminate the predictions to those in need.