Methodology documentation
This page presents climate risk assessments at the national and district levels. It helps users understand the key components and the level of risk across the country.
The methodologies for compiling maps showing risk, impact and vulnerability for three hazards (floods, soil loss/degradation and deforestation) affecting the eight target states are described below.
Data Collection and Preparation
The first step involved acquiring spatial datasets relevant to flood risk, soil loss, and deforestation. The datasets procured include :
Topography: A Digital Elevation Model (DEM) of Malawi from SRTM.
Hydrology: A nationwide dataset of major rivers in Malawi obtained from ESRI. Gridded annual precipitation data for 1981 to 2010 were based on the CHIRPS dataset (Funk et al, 2015).
Flow accumulation, slope, drainage density and Topographic Wetness Index (TWI) maps created using the DEM data: These steps were performed using the PCRaster plugin in QGIS. The datasets were necessary to determine relative flood risk in different parts of the country.
Soils: Vectorized UN FAO datasets delineating soil types, used to estimate erosion susceptibility and infiltration capacity parameters.
Land Use and Land Cover (LULC) data: ESACCI 300m land cover classification maps for 1992 to 2020 from Copernicus.
Population: A 2020 population density raster from HDX.
All datasets were reprojected to UTM 36S and a 300m x 300m horizontal resolution to improve rendering speed and ensure consistency in spatial analysis.
Topography: A Digital Elevation Model (DEM) of Malawi from SRTM.
Hydrology: A nationwide dataset of major rivers in Malawi obtained from ESRI. Gridded annual precipitation data for 1981 to 2010 were based on the CHIRPS dataset (Funk et al, 2015).
Flow accumulation, slope, drainage density and Topographic Wetness Index (TWI) maps created using the DEM data: These steps were performed using the PCRaster plugin in QGIS. The datasets were necessary to determine relative flood risk in different parts of the country.
Soils: Vectorized UN FAO datasets delineating soil types, used to estimate erosion susceptibility and infiltration capacity parameters.
Land Use and Land Cover (LULC) data: ESACCI 300m land cover classification maps for 1992 to 2020 from Copernicus.
Population: A 2020 population density raster from HDX.
All datasets were reprojected to UTM 36S and a 300m x 300m horizontal resolution to improve rendering speed and ensure consistency in spatial analysis.
Flood Risk, Impact, and Vulnerability Mapping
Flood Risk: The methodology used is an adaptation of the approach developed by Negese et al. (2022), whereby flood prone areas are mapped using a combination of multi-criteria decision-making (MCDM) and analytical hierarchy process (AHP) applied to the GIS raster datasets described above.
Raster datasets for eight selected flood-controlling factors were compiled from different online sources and using a variety of open-source processing tools available in QGIS 3.26. 300m annual rainfall from 1981 to 2010 are based on CHIRPS data. Digital elevation model (DEM) were downloaded from NASA’s Shuttle Radar Topography Mission (SRTM) were. Slope, flow accumulation, drainage density and TWI raster maps were then calculated in QGIS using the DEM dataset as input. A vector dataset of major rivers in Malawi obtained from ESRI was then used to create a distance-to-rivers raster using the SAGA proximity raster tool. Finally, 300m resolution ESA CCI land cover maps for 1991 to 2022 were downloaded from Copernicus.
Each of the eight flood-controlling factors was reclassified on a scale of 1 to 5, with 1 representing very low flood susceptibility and 5 representing very high susceptibility. The AHP method (Saaty, 1987) was then applied to assign relative importance to each factor. The first step in AHP involved constructing a pairwise comparison matrix, where each factor is compared against every other factor to determine its relative importance in causing floods. For simplicity, we classified the relative importance of the eight factors according to Negese et al. (2022). After constructing the pairwise comparison matrix, the values were normalized by dividing each value in the matrix by the column sum. The final weights were calculated by averaging the normalized values of each factor. The final weights represent the relative importance of each factor in determining flood susceptibility.
To create the maps of the eight selected target states showing relative flood risk in each Traditional Authority (TA), the number of pixels in each of the five flood risk categories in each TA was divided by the total land area of the TA to obtain five flood risk percentages. The overall distribution of TA-level flood risk percentages nationwide was then evaluated to delineate TA-wide flood risk classifications ranging from 1 to 5. The initial flood risk category percentage thresholds are as follows:
Raster datasets for eight selected flood-controlling factors were compiled from different online sources and using a variety of open-source processing tools available in QGIS 3.26. 300m annual rainfall from 1981 to 2010 are based on CHIRPS data. Digital elevation model (DEM) were downloaded from NASA’s Shuttle Radar Topography Mission (SRTM) were. Slope, flow accumulation, drainage density and TWI raster maps were then calculated in QGIS using the DEM dataset as input. A vector dataset of major rivers in Malawi obtained from ESRI was then used to create a distance-to-rivers raster using the SAGA proximity raster tool. Finally, 300m resolution ESA CCI land cover maps for 1991 to 2022 were downloaded from Copernicus.
Each of the eight flood-controlling factors was reclassified on a scale of 1 to 5, with 1 representing very low flood susceptibility and 5 representing very high susceptibility. The AHP method (Saaty, 1987) was then applied to assign relative importance to each factor. The first step in AHP involved constructing a pairwise comparison matrix, where each factor is compared against every other factor to determine its relative importance in causing floods. For simplicity, we classified the relative importance of the eight factors according to Negese et al. (2022). After constructing the pairwise comparison matrix, the values were normalized by dividing each value in the matrix by the column sum. The final weights were calculated by averaging the normalized values of each factor. The final weights represent the relative importance of each factor in determining flood susceptibility.
To create the maps of the eight selected target states showing relative flood risk in each Traditional Authority (TA), the number of pixels in each of the five flood risk categories in each TA was divided by the total land area of the TA to obtain five flood risk percentages. The overall distribution of TA-level flood risk percentages nationwide was then evaluated to delineate TA-wide flood risk classifications ranging from 1 to 5. The initial flood risk category percentage thresholds are as follows:
| Criterion | TA-wide flood risk |
|---|---|
| >= 10% of pixels are “Very high” flood risk | Very high |
| < 10% of pixels are “Very high” + >= 20% of pixels are “High” + < 65% of pixels are “Moderate” + < 50% of pixels are “Low” | High |
| < 10% of pixels are “Very high” + < 20% of pixels are “High” + >= 65% of pixels are “Moderate” + < 50% of pixels are “Low” | Moderate |
| < 10% of pixels are “Very high” + < 20% of pixels are “High” + < 65% of pixels are “Moderate” + >= 50% of pixels are “Low” | Low |
| All others | Very low |
Flood Impact: At this initial stage and without more sophisticated spatial datasets showing critical infrastructure (roads, schools, hospitals, etc.), the flood impact map is based solely on population density. A 2020 population density dataset was obtained from Humanitarian Data Exchange (HDX). Impact scores were then assigned following a cursory assessment of the distribution of TA-level population densities nationwide. The selected thresholds were:
| Population (per km2) | Impact |
|---|---|
| >= 1450 | Very high |
| >= 350 and < 1450 | High |
| >= 150 and < 350 | Moderate |
| >= 75 and < 150 | Low |
| < 75 | Very low |
Flood Vulnerability: Vulnerability maps represent a simple combination of the appropriate risk and impact scores for each district. They are calculated as:
| Risk + Impact | Vulnerability |
|---|---|
| 9, 10 | Very high |
| 7, 8 | High |
| 5, 6 | Moderate |
| 3, 4 | Low |
| 1, 2 | Very low |
Soil Loss Risk, Impact and Vulnerability Mapping
Soil Loss Risk: The Revised Universal Soil Loss Equation (RUSLE) was used to estimate countrywide erosion. The formula is: E = R * K * C * SL * P, where E is soil erosion, R is the climatic erosivity index, K is the soil’s resistance to erosion, C represents the effect of vegetation cover, SL is the slope length/topography factor, and P reflects erosion control measures. In QGIS, the K factor (soil’s resistivity to erosion) is calculated by assigning specific K values for different soil types, such as Cambisol, Luvisol, and Lixisol. These values were added to a soil map GIS layer downloaded from the UN FAO using regular expressions to classify soil types.
Raster data for the K factor were created by rasterizing the soils vector layer. Similarly, for the C factor, values were based on land cover classifications. The slope-length factor (SL) is calculated from the slope length and m-exponent derived using QGIS’s raster calculator, accounting for slope percentages and angles, and applying formulas based on slope gradients. Finally, after calculating the necessary factors (K, C, and SL), total soil loss is computed by applying the RUSLE equation in the QGIS raster calculator, using the newly created input layers/fields for each factor. The output is a raster file representing the annual soil loss in tons per hectare, which helps identify areas susceptible to erosion. The nationwide pixel-specific soil loss values are then redistributed into five bins representing different levels of erosion risk. As was done for the flood risk maps, these values are then used to compute TA-level percentages belonging to each erosion risk category. The overall distribution of TA-level erosion risk percentages nationwide was then evaluated to delineate TA-wide erosion risk classifications according to the following criteria:
Raster data for the K factor were created by rasterizing the soils vector layer. Similarly, for the C factor, values were based on land cover classifications. The slope-length factor (SL) is calculated from the slope length and m-exponent derived using QGIS’s raster calculator, accounting for slope percentages and angles, and applying formulas based on slope gradients. Finally, after calculating the necessary factors (K, C, and SL), total soil loss is computed by applying the RUSLE equation in the QGIS raster calculator, using the newly created input layers/fields for each factor. The output is a raster file representing the annual soil loss in tons per hectare, which helps identify areas susceptible to erosion. The nationwide pixel-specific soil loss values are then redistributed into five bins representing different levels of erosion risk. As was done for the flood risk maps, these values are then used to compute TA-level percentages belonging to each erosion risk category. The overall distribution of TA-level erosion risk percentages nationwide was then evaluated to delineate TA-wide erosion risk classifications according to the following criteria:
| Criterion | TA-wide flood risk |
|---|---|
| < 95% of pixels are either “Very low” or “Low” + most other pixels are “Very high” | Very high |
| < 95% of pixels are either “Very low” or “Low” + most other pixels are “High” | High |
| < 95% of pixels are either “Very low” or “Low” + most other pixels are “Moderate” | Moderate |
| < 95% of pixels are “Very low” but >= 95% of pixels are either “Very low” or “Low” | Low |
| >= 95% of pixels are “Very low” flood risk | Very low |
Soil Loss Impact: The primary input dataset used to assess soil loss impacts was the 2020 CCI land cover map. The land cover types in the land cover map were first classified into six broad categories: forest, crop, shrub, wetland, urban and other (including water bodies). Because soil loss and erosion may have particularly detrimental economic impacts on farming and agriculture, we calculated the TA-wide percentage of pixels comprising “crop” land cover types in each TA. Crops include CCI categories “cropland rainfed,” “cropland rainfed herbaceous cover,” “cropland rainfed tree or shrub cover,” “cropland irrigated,” “mosaic cropland.”
The nationwide distribution of TA-level cropland percentages nationwide was then evaluated and impact classifications from 1 to 5 were assigned using the following criteria:
The nationwide distribution of TA-level cropland percentages nationwide was then evaluated and impact classifications from 1 to 5 were assigned using the following criteria:
| Cropland pixels | Vulnerability |
|---|---|
| >= 80 % | Very high |
| >= 60% and < 80% | High |
| >= 40% and < 60% | Moderate |
| >= 20% and < 40% | Low |
| < 20% | Very low |
Soil Loss Vulnerability: Vulnerability scores were calculated using the same method described above for floods.
Deforestation Risk, Impact and Vulnerability Mapping
Deforestation Risk: A national deforestation map layer was created by overlaying the 300m CCI land cover maps for the years 1992 and 2020. Pixels identified as having “forest” type land cover in each year were first extracted into binary rasters, with 1 used to represent forested pixels and 0 for pixels with other land cover types. A difference raster was then calculated to assess deforestation from 1992 to 2020. Unlike the flood and soil loss risk maps, the deforestation values were not binned prior to creating the risk maps. Instead, the distribution of deforested pixels per TA was evaluated to choose delineations for the five risk classes. The thresholds chosen were:
| Deforested pixels | Vulnerability |
|---|---|
| >= 2000 | Very high |
| >= 1500 and < 2000 | High |
| >= 1000 and < 1500 | Moderate |
| >= 500 and < 1000 | Low |
| < 500 | Very low |
This more simplistic methodology notably does not account for the size of each TA or for deforestation as a percentage of the TA’s area, which may not provide a holistic view of deforestation in Malawi. The smallest TAs tend to be mostly urban, however, and the choice of 2000 pixels as the minimum threshold for the “Very high” risk category allows us to adequately capture the deforestation observed in many mid-size TAs such as Chapananga (Chikwawa), Kapelula (Kasungu) and Kalembo (Balaka).
Deforestation Impact: At this initial stage, we are lacking important spatial data that could be used to better evaluate the environmental impacts of deforestation including locations of known biodiversity hotspots, protected areas or socioeconomic indicators that may help to evaluate population dependence on forest resources. For the time being, the population density raster described above in 2 was used to assess deforestation impact, rendering the impact maps for deforestation identical to those for flood impacts. This step is to be improved following the use of additional proxy data.
Deforestation Vulnerability: Socioeconomic factors, such as dependence on forest resources, legal protection status, and governance frameworks, were analyzed to determine vulnerability levels.
Deforestation Impact: At this initial stage, we are lacking important spatial data that could be used to better evaluate the environmental impacts of deforestation including locations of known biodiversity hotspots, protected areas or socioeconomic indicators that may help to evaluate population dependence on forest resources. For the time being, the population density raster described above in 2 was used to assess deforestation impact, rendering the impact maps for deforestation identical to those for flood impacts. This step is to be improved following the use of additional proxy data.
Deforestation Vulnerability: Socioeconomic factors, such as dependence on forest resources, legal protection status, and governance frameworks, were analyzed to determine vulnerability levels.