Malaria in Papua New Guinea
A Risk Map for allocating health care facilities
1. Why Climate change is amplifying Malaria...
[PNG is] Rich in resources, poor in health
PNG is one of the biologically most diverse countries worldwide - however, the pictorial landscapes are also proving to be excellent breeding spots for mosquitos.
Due to climate change, mountainous regions experience higher temperatures, creating new breeding environments for the insect. Consequently, breakouts are shifting to previously unaffected places that are most often difficult to access (see Picture 1).
Picture 1: Flight delivery of mosquito nets to Lapalama, a remote mountain region (MAF 2023)
The combination of an insufficient hospital network and a fragile health system created a dangerous cocktail, ultimately resulting in a devastating malaria crisis.
Reflected in a report by the PNG Institute of Medical Research , the necessity for an up-to-date risk Map becomes clear. In the last decade, Malaria in PNG has experienced a comeback: From 2014 to 2019, the number of increased by ≈12x (50k to 600k). Among amplifiers such as a lack of doctors or insufficient infrastructure, the changing environmental paradigms for vector-borne diseases are moving more and more into the centre of research...
1.1. Why PNG needs a Malaria Risk Map
Illustration 1: Socio-technical malaria risk factors for PNG
It is important to note that malaria intensification only becomes a problem for humanity if there is no adequate response mechanism. Hence, the most pressing challenge to overcome PNG's malaria crisis is the strategic allocation of its scarce human, technological and infrastructural resources.
Environmental factors exacerbating malaria outbreaks can not be regarded as isolated. For this reason, it is important to also understand the underlying socio-economic amplifiers that fracture PNG's health system. The parameters showcased in Illustration 1 mutually influence each other - therefore, an effective risk map must consider and respond to all three spheres simultaneously.
1.2. Research question
All of the above demonstrates the necessity for a tool to:
- define public health strategies tailored to its geographical needs,
- promote effective pathways for its limited capacities.
For these reasons, the following research question has been formulated that will be analysed and visualised throughout this story map:
"How can GIS be utilised to integrate environmental factors into a malaria risk map for PNG, which can guide the strategic allocation of hospitals?"
2. Methodology
2.1. Area of study & socio-economic background
Papua New Guinea is located in the southwestern Pacific Ocean and encompasses the eastern half of New Guinea as well as its offshore islands. Its geographical landscape is mainly characterized by its diverse topography that includes mountains, dense rainforest and vast coastlines.
The country has a population of over 9 million people, with a demographic composition that is largely rural and dependent on subsistence agriculture. This uneven distribution of people across its rugged terrain poses significant challenges to hospital accessibility - which is one of the decisive factors that are currently hampering the development of a resilient hospital network.
As already indicated in the previous section, creating a risk map is a complex web of socio-economic and environmental factors. To gather all the necessary data, a long journey of trial and error through an obscene amount of sometimes very rusty websites began...
2.2. Materials & programs
- ArcMap software for spatial data analytics (CEU & University of the Aegean)
- Remote sensing for extraction of necessary imagery information (Google Earth Engine)
- Digital elevation model
- Main tools: Reclassification, Weightening overlay, Euclidian Distance, Multiple Ring Buffers
2.3. Framework
Before diving into the data-collection adventure, a brief explanation follows of WHY these environmental parameters were chosen.
After some extensive research on the implementation of GIS for vector-borne diseases, I came across one case study about Malaria in Mozambique that set the framework for my project. In this paper, the authors summarised the parameter values into three categories (low, medium, and high risk). Furthermore, it contained the weighting scale for each of the parameters. The reason why the latter is so crucial is because the different factors do not have the same risk impact on the prevalence of malaria. For example, temperature therefore bears a higher risk potential for Malaria prevalence than slope.
Illustration 2: Process from concept to practical data analysis
The reclassification and weighting are necessary to create the final risk map. In the next section, I will showcase the technical "behind-the-scenes" along the three steps (Illustration 2).
3. Data Analysis
3.1. Breaking down the problem
Each of these factors plays a critical role in the lifecycle and distribution of malaria vectors and parasites. In the context of my research question, these environmental factors are crucial for understanding and managing malaria risk.
average temperature (Tmean), precipitation (PP), altitude (Alt), slope (SLP), distance to water body (DTWB), distance to road (DTR), Normalized Difference Index (NDVI), land use and land cover (LULC), malaria prevalence (Mal prev) and population density (pop dens)
Temperature: Malaria transmission is optimal between 28 °C and 32 °C, as these temperatures allow the malaria parasite to develop efficiently within its mosquito vector. Below 18 °C and above 40 °C, the parasite's development is hindered. In Papua New Guinea (PNG), varying altitudes affect local temperatures, thus influencing malaria transmission patterns, with highland areas experiencing lower risk due to cooler temperatures.
Precipitation: Rainfall influences mosquito breeding sites; moderate increases in precipitation can enhance breeding opportunities, while excessive rainfall can eliminate them. PNG's tropical climate, characterized by frequent and heavy rainfall, can both facilitate and disrupt mosquito breeding.
Altitude and Topography: Altitude affects temperature and, consequently, malaria transmission. PNG's mountainous regions may have reduced malaria transmission due to cooler temperatures at higher elevations. Topography, particularly in smaller regions, significantly impacts malaria risk by influencing microclimatic conditions.
Slope and Land Cover: Flat areas tend to accumulate water, increasing malaria risk by creating more breeding sites. Conversely, land cover types like forests and swamps can either increase or decrease the incidence of malaria, depending on the local ecosystem's characteristics. In PNG, the diverse landscape, including farmlands, pastures, and forests, affects the distribution and density of mosquito populations.
Distance from Roads and Water Bodies: Proximity to roads and water bodies can significantly influence malaria incidence. Closer access to roads and water bodies often correlates with higher malaria risk due to the increased presence of breeding sites and human activity facilitating vector-human contact.
3.2. Data Collection
Data | Source | Evaluation of data processing: |
|---|---|---|
Temperature | Easy | |
Precipitation | Easy | |
Altitude/Slope | Difficult | |
LULC | Intermediate | |
DTWB | Intermediate | |
DTR | Intermediate | |
Pop. density | Easy | |
Malaria Prev | Intermediate | |
NDVI | google earth engine | Difficult |
Data Sources
What was striking to me, was that there were nearly no useful data sets provided by the National Papua New Guinea Environment . Most of the data provided was the data was neither complete nor up-to-date. Therefore, I mostly relied on global data sources that had to be masked accordingly.
For simplicity purposes, I will not explain the data collection behind each set of data - for some of the datasets, the process was quite straightforward. Others, however, especially the ones that required high resolutions were not so easy to fetch.
To showcase the processes behind of one of the more tricky parameters, I will explain the data collection of the Normalized Difference Vegetation Index (NDVI). As I was not able to find a suitable NDVI dataset for PNG, I had to get creative as I was reaching a dead end. After some research, the only way to scrape this data for my purposes was to use Google Earth Engine.
Luckily, I already have some previous coding experience and could get familiar with the program quite easily. Here is a brief step-by-step tutorial on how I managed to download the data.
3.2.1. NDVI data collection
- Open Google Earth Engine and log in to your Google Account and create a project
- Enter the code in the blank script editor
- Run the Code...
- After running the code, it will initiate the export process of NDVI Data to my Google Drive
- Once the status reads "Completed", I downloaded the NDVI data from Google Drive ("NDVI_PNG_2022")
import ee
# Initialize the Earth Engine API
ee.Initialize()
# Define the region of interest (ROI) using a GeoJSON polygon for Papua New Guinea
roi_geometry = ee.Geometry.Polygon([
[[141.0000, -7.0000], [155.0000, -7.0000], [155.0000, -12.0000], [141.0000, -12.0000], [141.0000, -7.0000]]
])
# Load the Landsat 8 Surface Reflectance data for NDVI calculation
landsat = ee.ImageCollection("LANDSAT/LC08/C01/T1_SR").filterBounds(roi_geometry)
# Define a function to calculate NDVI
def calculate_ndvi(image):
ndvi = image.normalizedDifference(['B5', 'B4']).rename('NDVI')
return image.addBands(ndvi)
# Map the function over the Landsat collection
landsat_ndvi = landsat.map(calculate_ndvi)
# Select the NDVI band and filter by date
ndvi_collection = landsat_ndvi.select('NDVI').filterDate('2022-01-01', '2022-12-31')
# Export the NDVI data to Google Drive
export_task = ee.batch.Export.image.toDrive({
'imageCollection': ndvi_collection,
'description': 'NDVI_PNG_2022',
'region': roi_geometry,
'scale': 30,
'maxPixels': 1e13
})
# Start the export task
export_task.start()
# Check the status of the export task
print('Exporting NDVI data, task ID:', export_task.id)
3.3. Performing geospatial analysis
1. Extraction by Mask
- This step allows you to extract specific areas based on another data layer (the "mask")
- Navigate to "Arc Toolbox" > "Spatial Analyst Tools" > "Extraction" > "Extract by Mask."
- Define the "Input Raster" as one of your environmental factors (e.g., temperature, precipitation, DEM) and the "Extract by Mask Data" as another data layer (e.g., high malaria prevalence areas)
- This will create a new raster containing only the values from the input raster that fall within the areas defined by the mask data.
2. Converting Vector to Raster
- Navigate to "Arc Toolbox" > "Conversion Tools" > "To Raster" > "Feature to Raster"
- Define the input features as your LULC data and choose appropriate settings for cell size and output format. This will create a raster layer representing land cover types
3. Multiple Ring Buffers
- Navigate to "ArcToolbox" > "Analysis Tools" > "Buffer"
- Define the "Input Features" as your high malaria prevalence data layer
- In the "Buffer Distance(s)" section, specify the desired distances for your multi-buffer rings. You can enter multiple distances separated by semicolons (e.g., "5; 10; 15" km)
- Choose an appropriate "Dissolve Option" (e.g., "Dissolve All") if you want a single feature with multiple rings
- Define an output name and location for the new feature class containing the multi-buffer zones
- Run the tool. This will create a new vector layer representing the multi-buffer zones around the high malaria prevalence areas
4. Distance Analysis
a) Distance to Water Bodies (DTWB) • To calculate the distance from each location to the nearest water body, navigate to "Arc Toolbox" > "Spatial Analyst Tools" > ”Distance." • Define the "Input Raster" as your raster representing water bodies and choose an appropriate measurement method (Euclidean distance) for calculating the distance • This will create a new raster layer showing the "Distance to Water Bodies" (DTWB) for each location
b) Distance to roads • Repeat the process above using the "Proximity" tool • Define the "Input Raster" as your roads network data and calculate the distance to roads, creating a new "Distance to Roads" (DTR) raster layer
5. Reclassifying Factors
- Navigate to "Arc Toolbox" > "Spatial Analyst tools" > "Reclassify." This allows you to categorize each environmental factor data layer (temperature, precipitation, DEM, etc.) into risk classes (low, medium, high) based on their influence on malaria transmission.
- Select each environmental factor data layer one by one and define suitable/provided risk categories with appropriate ranges of values for each class
6. Weighted Overaly analysis
- The final step involves combining all the reclassified data layers into a single integrated malaria risk map
- Navigate to "Arc Toolbox" > "Spatial Analyst tools" > "Overlay" > "Weighted Overlay."
- Define each reclassified environmental factor layer (temperature, precipitation, DEM, etc.), DTWB, DTR, and potentially LULC as input rasters
- Assign weights to each factor based on its relative influence on malaria risk. Research and expert knowledge will guide you in determining these weights
- The weighted overlay tool will then combine these layers considering the assigned weights, resulting in the final integrated malaria risk map.
4. Results & Analysis
The provided malaria risk map of Papua New Guinea (PNG) illustrates the spatial distribution of malaria risk in relation to healthcare facilities across the country. The risk levels are categorized into three classes: low (green), medium (yellow), and high (red). Additionally, buffers have been created around hospitals, represented by coloured circles, with varying distances delineated by different colours: purple for 5 km, blue for 6-10 km, and pink for 11-15 km.
- Spatial Distribution of Risk:
- High-risk areas (red) are predominantly closer to the coast in the south, and the islands in the north and appear to be concentrated in certain remote regions in the south.
- Low to medium-risk areas (green to yellow) tend to be more inland and along the mountain chain. This indicates a possible interaction between higher regions that are colder and therefore have lower malaria prevalence, due to less hospitable conditions for malaria vectors or better access to healthcare facilities.
- Healthcare Access:
- The buffers around hospitals indicate proximity to healthcare, which is crucial for effective malaria treatment and management - there is a high concentration in the centre of PNG due to the higher population density.
- In some regions, especially in the south-east and big parts of the west, there are vast "Hospital deserts" - being in a medium-risk area becomes much more of a threatening situation as the hospitals are far out of the buffer zones.
Discussion
In light of the high prevalence of malaria coupled with limited healthcare infrastructure, it is advisable to prioritize the allocation of hospitals and healthcare resources in the western and southern regions of Papua New Guinea. These areas are particularly burdened by malaria, yet they lack sufficient medical facilities to adequately address the disease. By concentrating resources in these regions, PNG can better serve populations most affected by malaria while improving access to essential healthcare services.
Illsutration 3: Incidence Rate Malaria, 2020
Despite some deviations, my results go hand in hand with the actual incidence rate of Malaria in PNG. The incidence rate is defined as the "the number of newly diagnosed cases per 1000 population during a defined year" ( Malaria Atlas) . This shows how decisive natural components (and dangerous!) are for the disease outbreak.
Essentially, the battle against Malaria has become a race against the clock, as climate change is intensifying day by day. Therefore, I suggest the following three recommendations to improve the allocation of health care facilities in PNG:
- Hospital Allocation: Prioritize the establishment and enhancement of hospitals and healthcare facilities in the western and southern regions of Papua New Guinea, where malaria prevalence is high but healthcare infrastructure is lacking.
- Capacity Building: Invest in training and capacity-building programs for healthcare professionals in these regions to ensure they are equipped with the necessary skills and knowledge to diagnose, treat, and manage malaria cases effectively.
- Community Engagement: Engage local communities through awareness campaigns and health education programs to promote preventive practices such as the use of bed nets, seeking prompt treatment for fever, and participating in malaria control efforts.
More Maps...
Risk Map vs. Hospital Buffers
Environmental Risk Factors
Socio-technical Risk Factors
5. Limitations
For strategic planning, it is important to reflect the current situation but also consider future changes in environmental factors due to climate change, land use changes, and population growth.
However, I cannot create a Future Risk Map because of the inherent challenges of predicting complex phenomena. For malaria in Papua New Guinea (PNG), accurately modelling future risk requires considering factors like climate change, environmental alterations, and the potential emergence of drug-resistant strains. These variables are difficult to predict with certainty, and existing models for future malaria risk rely on assumptions that might not capture all the nuances of PNG's specific situation.
Another hurdle that I could not overcome was creating a map for the Slope and Elevation Data as the file was too large. This caused which caused the system to reboot itself many times causing me to lose all of my process over and over. Nevertheless, the slope and elevation for the whole process of the Risk map and applied in the weighted analysis.
One last detail that could not be improved is the missing data for the small island in the western part of PNG, just above the Solomon Island label - here, ArcMap was not generating values for this small part because some parameters had incomplete data.
6. Personal Reflection
I was alternating between frustration and excitement, once there was data available that was tailored for PNG. Luckily, at some point, I overcame this valley and began to approach errors differently (from anger to acceptance).
What was certainly not the most practical solution was to work simultaneously on my personal device (Mac Book), and the stationary PCs at CEU and on lucky days of good internet connection, I even used the virtual machine. Switching between those three possibilities workflow interrupted by uploading data to my USB or one-drive and reuploading it. This was for instance the case when I transferred the NDVI data from my Mac to stationary PCs.
Eventually, I gave up on the virtual machine and conducted all my work in the computer lab and my USB drive, which accelerated my working speed significantly.
Finding a balance between curiosity ("I want to find the best data") and being satisfied with the data was certainly not an easy task as I wanted to stick to my initial framework and incorporate all the data. Getting "lost" in data research, instead of moving forward is definitely a lesson that I will incorporate for future projects.
Thanks :)
Sources
Ferrao, J. L., Niquisse, S., Mendes, J. M., & Painho, M. (2018). Mapping and modelling malaria risk areas using climate, socio-demographic and clinical variables in Chimoio, Mozambique. International journal of environmental research and public health, 15(4), 795.
World Economic Forum. ‘Papua New Guinea Is Battling the World’s Oldest Pandemic: Malaria.’, 13 September 2021. https://www.weforum.org/agenda/2021/09/papua-new-guinea-malaria/ .
Zhao, X., Thanapongtharm, W., Lawawirojwong, S., Wei, C., Tang, Y., Zhou, Y., ... & Kaewkungwal, J. (2020). Malaria risk map using spatial multi-criteria decision analysis along Yunnan border during the pre-elimination period. The American Journal of Tropical Medicine and Hygiene, 103(2), 793.
