Analysis of drought variables derived from Earth-Observations in Kenya

The project was an internship with NASA SERVIR. It involved analysing the performance of 11 indicators of drought in Kenya derived from earth-observations.

Kenya has 47 counties which are classified based on their level of aridity, as shown in the Figure below. The counties mainly affected by droughts are arid and semi-arid (ASAL), making up 80% of the country.

Figure 1: Classification of the counties in Kenya into arid, semi-arid and non-ASAL

The pre-processing and analysis of the raster files of the variables over 20 years analysis was carried out using python. The final output was mapped using QGIS. The methodology used is shown below:

Figure 2:The workflow of the image processing carried out in the project

Detectability of the Variables

Historical data of the 11 variables were analysed on their ability to detect droughts were based on the 4 huge droughts reported in Kenya using python codes. Of the 11 variables used, severity, SPI12, ESI, NDVI, and TCI could have the capabilities to detect major droughts from the non-drought periods.

Figure 3: Image carousel of TCI, SPI12, severity, NDVI, and ESI from 2008 to 2019 for the selected ASAL counties (left) and non-ASAL counties(right)

Drought variability

The 5 variables (severity, SPI12, ESI, NDVI and TCI) were further analysed to determine the time lag of each variable in detecting drought. From this analysis, it is possible to determine which variable can be used to forecast drought before it is severe.

From the image carousel below, we can see that severity is the only variable that showed drought persistence even after the rains throughout the drought period.

Figure 4: The trends of TCI, severity, NDVI and ESI of counties most affected by the 04/2011-09/2011 drought(left) and 09/2017-02/2018 drought(right)

Comparison of the variables to the currently used variable

Finally, time series maps for the variables that could detect drought were compared with the currently used variable (Vegetation Condition Index(VCI)) from NDMA. The variables were compared based on the drought that happened between September 2017 to February 2018, as shown in the image below.

NDVI, SPI12, TCI and ESI had different spatially and temporally patterns compared to the VCI. The NDVI is expected to give the same results; however, that was not clearly observed.

The severity had a close temporal trend with VCI except during the rains. Historical data of the 11 variables were plotted and their ability to detect droughts were based on the 4 huge droughts reported in Kenya. Of the 11 variables used, severity, SPI12, ESI, NDVI, and TCI could have the capabilities to detect the major droughts from the non-drought periods.

Figure 5: Maps of severity, TCI, SPI12, ESI and NDVI (left) compared to the map of VCI (right) between 09/2019 to 02/2018

You can access some of the python scripts I used through: https://drive.google.com/drive/folders/1c61ryOLEzBt00m4AlMlAhR6Mz_d6-1nT?usp=share_link

Page updated

Google Sites

Report abuse