Our Challenge

AEDES is tackling the SMASH YOUR SDGs challenge under the LIVING IN OUR WORLD category. Specifically we are tackling Goal 3: GOOD HEALTH AND WELL BEING under the UN Sustainable Development Goals.

Dengue is now at epidemic levels in the Philippines with over 271,000 cases and more than 1,100 deaths as of August 2019. Data on dengue takes time to be manually gathered which hampers the health sector’s ability to deal with the threat. Dengue is spread between infected cases through the Aedes Aegypti mosquito and mosquitoes are known to breed in damp locations and stagnant water pools.

Our Solution

We propose an automated information portal that correlates dengue cases and deaths with real-time data from climate, google searches, and satellite maps, giving an advance indicator of when dengue will emerge and potential dengue hotspot locations. This portal is accessible publicly but is targeted towards public health and local government agencies to give them advanced notice of dengue outbreaks and help prioritize resources.

Logically, the behaviors captured by the datasets in our solution are:

1. First, precipitation and temperature climate creates mosquito-breeding environments
2. Mosquitoes spread and get infected by existing dengue cases, thereby spreading the disease
3. New infections cause alarm which drives internet searches for dengue
4. Dengue cases result in deaths
5. Cases and deaths are reported to public health officials

Therefore by detecting #1 and #3 we can address and hopefully disrupt the disease cycle before an epidemic spreads.

How It Works

The service relies on 3 data sets:

• Global Data: Satellite spectral data from Sentinel-2 Copernicus and Landsat 8
• Local Data: Climate data from DOST-PAGASA
• Google Data: Google search trends for 'dengue' and related terms

We propose to forecast dengue cases using a combination of:

• Climate Data: Average monthly temperature and monthly rainfall (precipitation) readings from local weather stations
• Google Data: Search index for 'dengue', 'dengue symptoms', 'dengue fever', and 'dengue medicine'
• Lagged values for Climate factors and Google search data

We propose to detect likely mosquito hotspots using Satellite spectral readings to calculate indices:

• FAPAR: Fraction of Absorbed Photosynthetically Active Radiation, reference here.
• NDVI: Normalized Difference Vegetation Index, reference here.
• NDWI: Normalized Difference Water Index, reference here.

For our initial prototype we decided to focus on four areas:

• National Capital Region (Quezon City as a focus)
• Eastern Visayas (Tacloban City as a focus)
• Western Visayas (Iloilo City as a focus)
• ARMM (Cotabato City as a focus)

We selected the above areas due to availability of local weather station in these locations. We decided on NCR as a default and added the three other cities due to the observed spike in Google Searches for dengue related terms indicating possible prevailing public panic in these areas.

NASA RESOURCES

We relied on satellite spectral bands from Landsat 8 (NASA) and Sentinel-2 Copernicus (ESA) missions to generate vegetation indices and water indices to derive stagnant water areas where mosquitos are found. Vegetation indices are calculated using FAPAR and NDVI calculations while water indices are derived from NDWI calculations.

We used the following spectral bands from the satellites as base data for the calculations:

• Sentinel - Green
• Sentinel - Red
• Sentinel - Near-Infrared (NIR)
• Landsat 8 - Green
• Landsat 8 - Red
• Landsat 8 - Near-Infrared (NIR)

Indices were calculated as follows:

• NDVI, FAPAR = (NIR - Red) / (NIR + Red)
• NDWI = (Green - NIR) / (Green + NIR)

Once calculated the readings were extracted as long-lat coordinates via QGIS.

dengue FORECAST MODELS

We derived the dengue forecasts using multivariate modeling on climate and google search index data and selected the top 3 fitting models for each location. The resulting models are as follows, along with correlation (R) and r-squared (R2) readings:

PH Aggregated

• Model 1 (R 0.75, R2 0.56): Search (Dengue Medicine) Lag 1, Month 11, Search(Dengue Fever) Lag 1, Month 2, Average Temperature Lag 3, Month 5
• Model 2 (R 0.73, R2 0.53): Search (Dengue Medicine) Lag 1, Month 11, Month 2, Average Temperature Lag 3
• Model 3 (R 0.75. R2 0.57): Search (Dengue Medicine) Lag 1, Month 11, Search (Dengue Fever) Lag 1, Month 3, Month 2, Average Temperature Lag 3

NCR

• Model 1 (R 0.84, R2 0.7): Average Rainfall Lag 3, Month 9, Average Temperature Lag 3, Month 6
• Model 2 (R 0.84, R2 0.71): Average Rainfall Lag 3, Month 9, Month 4, Average Temperature Lag 3, Month 7, Month 6
• Model 3 (R 0.83, R2 0.69): Month 6, Month 9, Average Rainfall Lag 3, Month 7

Eastern Visayas

• Model 1 (R 0.75, R2 0.57): Month 7, Average Temperature Lag 2, Search (Dengue) Lag 1, Month 6, Month 9, Month 10
• Model 2 (R 0.76, R2 0.58): Month 7, Average Temperature Lag 2, Search (Dengue) Lag 1, Month 6, Month 9, Month 11, Month 10
• Model 3 (R 0.73, R2 0.53): Search (Dengue) Lag 1, Month 6, Month 10, Month 7

Western Visayas

• Model 1 (R 0.9, R2 0.81): Month 10, Month 4, Search (Dengue) Lag 1, Search (Dengue Fever) Lag 1, Search (Dengue Symptoms) Lag 3, Month 2, Average Rainfall Lag 2, Average Rainfall Lag 3, Search (Dengue Symptoms) Lag 1, Month 9, Search (Dengue) Lag 2, Month 8, Search (Dengue) Lag 3, Average Temperature Lag 2, Average Temperature Lag 3, Search (Dengue Symptoms) Lag 2, Search (Dengue Fever) Lag 3
• Model 2 (R 0.89, R2 0.79): Month 10, Month 4, Search (Dengue) Lag 1, Search (Dengue Fever) Lag 1, Search (Dengue Symptoms) Lag 3, Average Rainfall Lag 2, Average Rainfall Lag 3, Search (Dengue Symptoms) Lag 1, Month 9, Search (Dengue) Lag 2, Month 8, Search (Dengue) Lag 3, Average Temperature Lag 2, Average Temperature Lag 3, Search (Dengue Symptoms) Lag 2, Search (Dengue Fever) Lag 3
• Model 3 (R 0.9, R2 0.81): Month 10, Month 5, Month 4, Search (Dengue) Lag 1, Search (Dengue Fever) Lag 1, Search (Dengue Symptoms) Lag 3, Month 2, Average Rainfall Lag 2, Average Rainfall Lag 3, Search (Dengue Symptoms) Lag 1, Month 9, Search (Dengue) Lag 2, Month 8, Search (Dengue) Lag 3, Average Temperature Lag 2, Average Temperature Lag 3, Search (Dengue Symptoms) Lag 2, Search (Dengue Fever) Lag 3

ARMM

• Model 1 (R 0.87, R2 0.76): Month 3, Month 5, Month 8, Month 9, Month 12, Search (Dengue) Lag 1, Month 1, Average Temperature Lag 1, Search (Dengue Symptoms) Lag 1, Average Rainfall Lag 1, Average Rainfall Lag 3, Search (Dengue) Lag 3, Search (Dengue) Lag 2, Search (Dengue Symptoms) Lag 2, Average Temperature Lag 3
• Model 2 (R 0.86, R2 0.74): Month 3, Month 5, Month 8, Month 9, Search (Dengue) Lag 1, Month 1, Average Temperature Lag 1, Search (Dengue Symptoms) Lag 1, Average Rainfall Lag 1, Average Rainfall Lag 3, Search (Dengue) Lag 3, Search (Dengue) Lag 2, Search (Dengue Symptoms) Lag 2, Average Temperature Lag 3
• Model 3 (R 0.81, R2 0.66): Month 8, Month 12, Month 9, Search (Dengue) Lag 1, Average Temperature Lag 1, Search (Dengue Symptoms) Lag 1, Search (Dengue) Lag 3, Search (Dengue Symptoms) Lag 2, Average Temperature Lag 3

Our Impact

We visualized the resulting models on a web interface for easy navigation as well as the dengue hotspots on an interactive map that can zoom down to street level for public health sector targeting. Check the prototype here.

Through this solution we are addressing 2 key challenges for public health and local government officials:

• Get ahead of the lagged delay of dengue reporting by using real-time information (e.g. climate, searches) to infer if dengue cases and deaths are about to spike.
• Precisely anticipate areas that may be affected by dengue to prioritize health aid, supplies, and proactive fumigation to prevent the outbreaks.

Based on latest (August 2019) figures, dengue deaths are averaging 138 lives a month with 34,000 new cases emerging very month.

Every single day we reduce the lag of the response time we can save 5 lives and prevent 1,130 new cases.

Future Plans

We intend to finish the application and populate climate data, search data, cases, deaths, and satellite views for all regions in the Philippines. We hope to get funding for a comprehensive nationwide public health campaign to educate local government and health sector offices on the use of data to prevent dengue.

Beyond the Philippines, the application is also relevant to other countries that are suffering from dengue as well as other mosquito-borne diseases such as Zika and Chikungunya (same mosquito as dengue: Aedes Aegypti) and Malaria (Anopheles Mosquito).

We are committed to bringing the solution out for public use and bring it to the radar of the public health sector and we hope we can get NASA's support to help stem the existing dengue crisis.

Built With

We used QGIS for extraction of Landsat 8 and Sentinel-2 data. For our web application a vanilla web stack (HTML, CSS, Javascript, PHP). Charts were prepared using Chart.js. Mapping applications used ping OpenStreetMap on Mapbox API (free-tier). Statistical modeling was performed using Python on Jupyter notebooks with SciKit-Learn and StatsModels libraries.

Related Literature

The key research and prior work that provide the foundation for this solution:

• In Caro, et. al. (2016), "Forecasting and Data Visualization of Dengue spread in the Philippine Visayas Island group", researchers proposed a method to predict dengue cases in the Visayas regions through the use of an Artificial Neural Network (ANN) which predicted dengue cases (R 0.8) through the use of climate data (temperature and rainfall) and a lagged number of previous dengue cases.
• In Yang, et. al. (2016), "Advances in using Internet searches to track dengue", researchers used Google searches for 'dengue' and related keywords to predict dengue cases in Mexico, Brazil, Thailand, Singapore and Taiwan.
• In Chua, Tan, et. al. (2018), "Project Still Water", researchers proposed a methodology to detect dengue hotspots using FAPAR (Vegetation) and NDWI (Water) areas.
• In Stancalie, et. al. (2014), "Potential of using satellite based vegetation indices and biophysical variables for the assessment of the water footprint of crops", researchers showed the usefulness of vegetation indices and biophysical variables (NDWI, NDVI, FAPAR) for water footprint assessment.
• In Chan, Johnson (2012), "The Incubation Periods of Dengue Viruses", researchers establish the incubation periods for dengue. 95% of EIPs are between 5 and 33 days at 25C, and 2 and 15 days at30C, with means of 15 and 6.5 days, respectively which suggests that climate has an effect on the incubation of the disease.

Try it out

• Check the prototype here.
• Resources are also found at our Github page.
• Our original presentation is here.

View the walkthrough here:

Tags

#dengue, #aedes, #remotesensing, #python, #javascript, #html, #css, #chartjs, #landsat, #sentinel, #ndvi, #ndwi, #fapar, #sdg3