Using the Vegetation Condition Index combined with time-series analysis to monitor the health of the Maltese vegetation in the context of sustainable tourism development

Earth observation technologies can help tourism organizations to meet sustainable tourism development guidelines and management practices set by the World Tourism Organization, especially regarding the environmental dimension. In this context, the Malta Tourism Authority (MTA) is looking for an easy and reliable tool to assess vegetation health in order to monitor the impact of tourism on its environment. 

Based on this partnership, we chose to enhance the Vegetation Condition Index (VCI) using Sentinel-2 to assess the vegetation health of Malta from January 2017 to October 2022. This method consists of comparing the current NDVI values to the range of values observed in previous years. The VCI allows to determine where the observed value is situated between the extreme values (minimum and maximum). Lower and higher values are used as proxy to indicate the critical and optimal vegetation state conditions, respectively. 

We used the Seasonal and Trend decomposition using Loess to decompose the VCI time-series from three distinct vegetation types, namely cropland, grassland, and trees. This method uses locally fitted regression models to decompose a time series into trend, seasonal, and remainder components. Regarding vegetation health assessment of Malta, we noticed a period of drought in 2021 which was the result of a strong anomaly in October 2020. During this period, trees were the most affected type of vegetation. However, no correlation was found between tourists' inbound and vegetation health.

We based our validation on the fact that meteorological conditions are the main factors for vegetation health variations. Thus, we used total precipitation, and surface temperature variables from the ERA5 climate reanalysis database (ECMWF) as proxy for ground-truth data. We found that precipitation was “Granger causing” (statistical hypothesis test for determining whether one time series is useful in forecasting another) VCI and that it was cross-correlated (using Spearman correlation method) with VCI at 0.80, whereas temperature was negatively correlated with VCI at -0.91 meaning that our hypothesis was correct. 

Ultimately, we combined the produced information into a dashboard in order to display the information for the end-user. This visualization combined three distinct dimensions of vegetation health, namely the temporal dimension which displays long-term time-series, the spatial dimension which displays VCI maps with vegetation highlight layers that help for spatial contextualization, and the trend dimension which combines trends of VCI and the influencing factors.