Optimizing Vegetation Monitoring over East Africa
The beginning of drought is characterized by lower amounts of rainfall. This rainfall shortage progresses to decreased soil moisture, increased land surface temperature, and disrupted growth of affected vegetation. Thus the monitoring of vegetation conditions allows us to successfully track a drought’s progression.
Numerous studies make use of remote sensing and land surface models to monitor changes in vegetation condition. Two particularly common variables studied are the Normalized Difference Vegetation Index (NDVI) and evapotranspiration (ET), estimates by models or remote sensing. NDVI is a vegetation index created using the red and near-infrared reflectance measurements recorded by earth observing satellite sensors. The Moderate Resolution Imaging Spectroradiometer (MODIS) satellite imagery, which is available at 250m and updated every 8 days, has allowed NDVI to be used in a number of studies investigating drought progression at larger scales. ET is the sum of water surface evaporation, soil moisture evaporation, and plant transpiration from Earth’s surface to the atmosphere. ET’s dependence on soil moisture and land cover makes it very useful in assessing crop biomass and crop yields and how both are affected by drought. While both variables have demonstrated utility in monitoring vegetation conditions in drought-affected areas, there has been little research comparing the effectiveness of NDVI and ET in a given location.
A recently published study sought to conduct this comparison, assessing the effectiveness of NDVI and ET in monitoring vegetation conditions across East Africa. The study, co-authored by NASA Harvest Partner Dr. Amy McNally alongside Dr. Shahriar Pervez, Dr. Kristi Arsenault, Dr. Michael Budde, and Dr. James Rowland, used MODIS-derived NDVI, as well as two differently calculated ET variables. The first form of ET was calculated using the relationship between water evaporation and change in remotely sensed surface temperature; while the second was derived using a land surface model (LSM) that accounts for both remotely sensed and ground-based data.
The comparison was conducted using a Triple Collocation analysis, a statistical method that quantifies random error by calculating the root mean square error (RMSE) for each of the variables under study. Quantifying how the random error of each model shifted over the study area allowed the team to determine the effectiveness of each model over different land cover types and areas with different percentages of vegetation cover.
The authors found that, given its reliance on surface evaporation and plant transpiration, ET becomes a more reliable indicator of vegetation condition as the amount of vegetation increases. They also found that NDVI and RS-derived ET struggled in areas with both very high and very low vegetation cover due to signal saturation and signal-to-noise- ratios respectively. The authors suggest that incorporation of ground-based data in its calculation makes LSM ET the best indicator in these low- and high-vegetation cover areas. All three variables were found to be good indicators in moderately vegetated areas.
Additionally, the study determined that both LSM and RS-derived ET is a better indicator of vegetation condition during extreme drought, while NDVI is the better indicator during periods of higher than normal rainfall.
While this study focused on East Africa, the authors note that it can be applied to any other drought or food insecure region in the world and used to optimize vegetation monitoring. The full study can be read here.