Using Network Science to Optimize Disaster Relief in the Philippines.

Fork this project in Github here

Play around with our interactive web apphere

Its been a while, everyone!

I’m really excited to share about this work because it took years in the making, almost 9 years to be exact!

During this past decade, a series of supertyphoons – Yolanda (Haiyan), Odette (Rai) Pablo (Bopha), and Sendong (Washi) to name a few– have damaged entire regions and carved trauma into our generation’s collective memory.

As young scientists, we felt that it is duty to give back to the country by contributing a physicists’ perspective on disaster science.

To accomplish this goal, Dr. Reinabelle Reyes created and led the DOST ReliefOpsPH project back in 2014 to assess and identify improvements in country-wide relief operations through simulations, and later on, formed Project GeoNetworks with Dr. May Lim and I around 2017 to analyze the transport network of the Visayas region in the Philippines using complex network theory.

We were doing the work on our spare time, more like a fun hobby than a formal project. Admittedly, we got caught in each of our own career shifts and hit a couple of dead ends, so it took us about 7 years to finish a publication last 2023 in the Philippine Journal of Science.

Better late than never, right?

In this blog, I want to share with you the accompanying app to our 2023 paper. The app is a tool that provincial level planners can use to simulate relief delivery travel times from the source hub at the provincial capital, going to each of the component LGU towns, and in addition, compare and assess the effect of adding a new hub to a component LGU town of their choice.

We invite you to test out our app below!

What the app does#

The relief delivery times computed by the tool are the estimated travel times to each component LGU from the food hub nearest to them (either the original capital hub or the new hub). We did this by representing the land and sea transport network of Visayas as a mathematical object known as a graph, where road intersections and ports represent nodes, whereas road segments and sea routes constitute the edges. This method allowed us to programatically trace routes from any two points in the area.

The outputs are (1) simulated routes maps and (2) delivery travel time savings when the +1 new hub scenario is compared to the baseline single hub scenario.

Shown below is an example of an output simulation for Bohol province. The left map shows the baseline scenario, with Tagbilaran, the province capital, as the only relief hub. With a new food hub in the town of Ubay at the west end of the island, it will take only 1.2 hrs to reach all LGUs in the province, which is a 52.0% reduction from the baseline (2.5 hrs).

We do have important assumptions to disclose. In this idealized simulation:

• Deliveries to all LGUs start simultaneously from the food hubs.
• Deliveries do not incorporate stopovers in towns they would pass by, but only on its destination.
• All land and sea routes are undamaged and considered fully passable.

These assumptions will not hold in real-life scenarios, where disruption is to be expected from already non-ideal conditions. Still, the maps generated by the simulation can serve to illustrate the potential impact of activating new food hubs to more efficient relief distribution for nearby areas and may be used as a starting point for more detailed projections.

Datasets used#

We constructed the Visayas land and sea transport network using open data. For the road networks, we used available OpenStreetMap (OSM) data from June 2024 and cleaned it up so only vehicle-passable roads are kept. Sea routes connecting the islands of Visayas were identified from three sources in 2018: ferry routes in (1) OSM (2) Google Maps, and (3) the Visayas General Logistics Planning Map created by Logistics Cluster of the United Nations World Food Programme (UN-WFP) from long-term engagements with LGUs in the aftermath of Typhoon Haiyan hitting the region in November 2013.

We also obtained the coordinates of the local government municipal/city hall (or the market/plaza if latter is not available), to serve as the town center for the simulations. It is assumed that all disaster relief efforts start in these town centers and aid is received on this location.

Key Results from the paper#

Our paper included more insights from the Visayas-wide baseline scenario, where we considered regional level operations:

• We have estimated that most Visayas towns can be reached within 4 hours from a regional food hub. Town centers in northern Negros and southeastern Panay in Region 6 were reached in about 5 hours, whereas the rest of the region is reached in 6–8 hours. Remarkably, travel times to Samar Island in Region 8 are longer (~ 7 hours) than to Leyte Island (~ 5 hours) despite being in the same region and having similar geodesic distance from their regional food hub.

• Assuming an undamaged network state and normal land and sea traffic conditions, one-half and three-quarters of the whole Visayas population can theoretically be reached within 3.5 and 5.5 hours, respectively.

• However, it would take about 13 hours to reach the remaining one-quarter of the population to finally complete the delivery to all Visayas town centers. Travel to these few farther towns involves long coastal highways traversing sea routes.

We created the app with the intention to further localize the applicability of our insights, and to help LGUs in planning new food hubs at the provincial level.

Next Steps#

We have barely scratched the surface of the possibilities of the model transport network that we made. Some next steps that we have discussed are the following.

• Incorporating damage scenarios in the app. Given our model, we can estimate how long of a delivery delay is expected given a certain % of road nodes being damaged. This development is in the works!
• Consider road hazard exposures and vulnerabilities. Currently, our simulations assume a homogenous random probability of damage to all road nodes, but we all know that is not the case in real life. For example, some roads are in landslide prone areas, while others are either of lower grade or poorly maintained. These can be included in the damage algorithm as another parameter.
• New region? We wrote the code so it would be easy for any programmer to create and simulate any region in the Philippines, or in the world. Would anyone of you be interested in modelling your home region?

I hope this was useful! Thank you so much and see you in the next blog!