This GitHub Repository covers the 'Rice Damage Model' of the Impact Based Forecasting project of 510. It has been executed in collaboration with the German and Philippines Red Cross, the Food and Agriculture Organization, the Philippines Department of Agriculture and Philippine Rice Research Institure. The main aim was to develop a model that can be used to predict damages to rice fields, on municipality level, before a typhoon makes landfall. To this extent, three sets of models have been implementend; a binary classification, a multiclass classifcication and a regression.
The dependent variable used in the model is the percentage of standing rice area damaged.
The features used cover a set of exposure and vulnerability indicators (municipality specific) and a set of hazard indicators (typhoon and municipality specific).
Exporsure and vulnerability indicators:
- Area (km^2)
- Latitude
- Longitude
- Perimeter (m)
- Coast length (m)
- Coast binary
1 if the municipality is by the coast, 0 if it is not - Mean elevation (m)
- Mean ruggedness (m)
- Ruggedness stdv (m)
- Mean slope (%)
- Slope stdv (%)
- Coast - Perimeter ratio
- Poverty percentage
Hazard indicators:
- Maximum 6 hour rainfall (mm/h)
This reflects the maximum rainfall intensity on a 6 hour time interval. The time period covered is 72 hours before the typhoon makes landfall. Thus, this variables shows the maximum 6 hour rainfall intensity in mm/h in the 72 prior to the typhoon making landfall. - Maximum 24 hour rainfall (mm/h)
Same definition as the maximum 6 hour rainfall, but now for 24 hours. Thus, this variables shows the maximum 24 hour rainfall intensity in mm/h in the 72 prior to the typhoon making landfall. - Minimum track distance (km)
- Maximum wind speed (m/s, 1 minute average)
The binary classification models predicts whether the damages is above or below 30%
The multiclass classification has three classes:
- 0 - 30%
- 30% - 80%
- > 80%
The regression model predicts on a continuous scale.
This repo contains the data and code used in the projects including documentation to explain the steps taken. The two main files are:
-
The data preparation notebook
This notebook contains a step-by-step process of obtaining and processing all the data, resulting in the final input data set used in the model. -
The model results notebook
This notebook contains a step-by-step process of training and testing a set of models, to find the optimal ones, on the input data obtained.
The folder contain all supporting files, brief documentation and code comments.