How to assess drought in data scarcity areas? A case study in Kruger National Park (South Africa)

Mediterranean savanna ecosystems exhibit a hydrological regime with marked wet and dry seasons. This variability makes droughts a recurrent hazard that impacts water supply, food security, wildlife, and economy. Consequently, a thorough drought definition and monitoring are essential to foresee drought impact, to support decision-making processes, and to design mitigation and adaptation strategies. However, drought definition is not easy, specifically, in data-scarce areas, where ground observations are sparse or unavailable. In these cases, satellite remote sensing and geospatial data are a valuable alternative to in situ information.  

This study evaluates the usefulness of satellite-based indicators to characterize drought dynamics in these data scarce environments. Special emphasis was put in two aspects: (i) exploring lagged relationships and cascading effects among different drought types and (ii) the capacity of different remote-sensing-based indexes to capture agricultural drought in several land cover of these environments. The Kruger National Park in South Africa (KNP) over the period 2000 – 2023 was selected as pilot case area due to the characteristic recurrence of droughts.  

Therefore, meteorological, agricultural, and hydrological droughts were computed over four different land cover classes: savanna, forest, grassland, and cropland. Each of these areas were identified using ESACCI Land Cover (1992 – 2015). Meteorological drought was defined through the Standardized Precipitation Index (SPI) computed using the ERA5-Land precipitation data (25km). Agricultural drought was analyzed using three different methods, with different levels of complexity. First, the 16-day MOD13Q1 NDVI product. Second, the daily Evaporative Stress Index (ESI), defined as the ratio of actual to reference evapotranspiration (ET), as a proxy for ecosystem water stress. Actual ET was estimated using a Two Source Energy Balance (TSEB) model driven by MOD11A1 Land Surface Temperature (LST). Third, a MODIS-based Composite Drought Index (CDI) derived from air temperature, precipitation, and NDVI was also considered. Finally, hydrological drought was assessed through the Standardized Streamflow Index (SSI) derived from Global Flood Awareness System (GloFAS) v4 river discharge data.  Drought events were identified using standard thresholds for SPI, SSI, and CDI, while ESI and NDVI thresholds were defined by land cover and month to account for phenology. 

Regarding the connection between different droughts, the preliminary results show for all the classes analyzed that only the more severe meteorological droughts, that is those occurring during 2003-04 and 2015-16, have a direct impact on agricultural drought. The effect on hydrological drought is lumped in comparison. When comparing the different agricultural drought methods, we found that NDVI is the index more sensitive to changes, particularly in non-forested areas which are more dependent on precipitation, while ESI is better representing abrupt fluctuation in forests and savannas. On the contrary, CDI poses a more homogenous value, what makes it overestimate the presence of droughts. 

These initial results show the potential of coupling different spatial data sources, geospatial and remote-sensing-based to define different droughts and their relations in data scare regions. In addition, they allow providing some initial recommendations about their different responses depending on the land cover analyzed. 

Acknowledgments: This work is part of the grant RYC2022-035320-I, funded by MCIN/AEI/10.13039/501100011033 and FSE+.