Comparison of Vegetation Indices from Various Remote Sensing Sources
Head-to-head comparison of LSA SAF, Sentinel-2 and drone derived vegetation data
In recent years, severe weather regimes in European countries are becoming more and more frequent and intense. Areas located mostly in Southern Europe experienced extreme weather events during the summertime, such as droughts and heat waves. These events strongly affect vegetation and can even result in wildfires, turning the green, healthy areas into less fertile land. Reduced productivity of vegetation can also be a consequence of other factors such as pests and other natural hazards that affect the growth rate.
Vegetation monitoring can serve to protect vegetation and increase its productivity. In the frame of vegetation monitoring, vegetation indices have been proved to be beneficial. In general, these indices reflect the health of vegetation and can be calculated from various sources. Normalised Difference Vegetation Index (NDVI) and Fractional Vegetation Cover (FVC) are two of the most well-known vegetation indices.
Introducing vegetation indices and their use
NDVI is one of the most widespread vegetation indices that has been used for remote sensing surface analyses of vegetation cover for many years. NDVI correlates to the amount of chlorophyll in a plant and consequently its health. Its values range between -1 and 1, with values up to 0.2 usually indicating an area without vegetation cover, from 0.2 to 0.3 the emergence of vegetation or weak growth, and from 0.6 to 0.8 lush vegetation, including the tree canopies. To calculate NDVI, the amount of red and near-infrared light reflected by the observed object is required. Chlorophyll, which is important in the process of photosynthesis, absorbs the red part of the spectrum and reflects the nearby infrared part. Thus, the ratio between the two reflections provides information about the condition of the plant.
FVC determines the fraction of the surface covered by vegetation. Value 1 indicates that the area is completely covered with vegetation and in good condition, lower values are typical for areas with little vegetation. Values around 0 indicate bare soil. It can be used for monitoring of the vegetation state, similarly to NDVI. The nature of the index enables the precise monitoring of deforestation and desertification processes.
Figure 1 - Analysed vineyard from a drone perspective (left). Analysed vineyard (indicated with a point marker), shown for the Sentinel-2 NDVI product on 28 August 2020. Photo taken from EO Browser (https://apps.sentinel-hub.com/eo-browser/).
Figure 2 - FVC values on 28 August 2020 with a point marker indicating the analysed vineyard location.
Vegetation indices can therefore help monitoring the phenological phases of plants over a longer period of time, identifying agricultural tasks (such as mowing), and locating areas affected by drought or pests. We analysed NDVI and FVC over a vineyard area in Northeastern Slovenia (Fig. 1) in the year 2020. NDVI is derived from Sentinel-2 polar-orbit satellites, while FVC is calculated from the MSG (Meteosat Second Generation) geostationary satellite prepared in the frame of LSA SAF (LSA-421). The long FVC data set available since 2004 provides a climatology that can be used as reference for comparison of the state of the vegetation in 2020, allowing the detection of drought-related events. The spatial resolution of the product is 3 km at nadir. Fig. 2 displays FVC over Southeastern Europe.
On the other hand, Sentinel-2 is a relatively new mission, so the climatology of NDVI could not be calculated. As another source of independent NDVI measurements drone overflights at the observed area were performed. Drone measurements were supported by the Interreg Alpine Space project at the Alpine drought observatory. Drone overflights enabled the comparison and verification of the satellite derived measurements. The spatial resolution of the drone shots is approximately 10-15 cm while the spatial resolution of the Sentinel-2 product is 10 m.
The comparison of vegetation data
Sentinel-2 as well as MSG satellites provide regular and high-quality measurements. The satellite’s precise and consistent orbit time enables us to collect regular data. Another promising source of remote sensing derived data are drone overflights, which have a much finer resolution than satellite images. An additional advantage of drone footage is that recordings are also possible in cloudy conditions, as the drones can fly under the clouds. The main disadvantage of drone overflights, however, is that there are usually no continuous measurements available. Therefore, the multi-year data sets usually do not exist, which makes such data less useful for standalone statistical analyses.
Results of the comparison are shown in Fig. 3. The green curve shows the FVC values during 2020, with reference values shown in black. This comparison shows that FVC is mostly well above reference values in 2020 (top plot). This is consistent with ground observations, as there was no pronounced drought in 2020. The observations indicate that spring was the driest part of the year 2020, which is again consistent with the satellite measurements.
The bottom plot shows the comparison of drone and satellite derived NDVI calculations. Drone measurements are represented with the red curve and red dots depict the drone measurements that are temporally aligned with the satellite data. Satellite data (blue dots) are consistent with drone measurements and are in all cases within the standard deviation of drone measurements which suggests a high quality of the Sentinel-2 product.
Figure 3 - FVC over the 2020 vegetation season compared to reference values (top) and a comparison of NDVI calculated from Sentinel-2 (blue dots) and drone measurements (red line) over the 2020 vegetation season.
Conclusion and final remarks
Benefits of using remote sensing methods in agriculture and agronomy are growing from year to year, as the resolution of meteorological satellites is improving, and at the same time, such methods are becoming more affordable. By using remote sensing derived vegetation indices, alternative agricultural approaches can be put into practice, such as identification of drought or pest hit areas, selective fertilisation, selective pesticide spraying and optimised irrigation. In this way, optimal water management can be achieved, and the use of fertilisers and pesticides can be reduced, as vegetation indices can be used to control the condition of vegetation over larger areas and the use of resources is applied only over stress areas.
Drought detection requires a fine temporal resolution. Droughts can be detected by analysing the anomalies of a certain vegetation index, which requires a year-round data set for a no-drought year or a multi-year average of vegetation index. These indices are also affected by many other factors, which are not necessarily related to drought (limited vegetation growth due to the lack of nutrients in the soil, pests, ...) and this should be taken into account.
The combination of various vegetation indices has the potential to improve drought detection. LSA SAF data can provide daily updates and, together with the long-time reference, give a good insight into the status of vegetation. The limitation of this approach is that analysed areas need to be large and homogeneous enough due to the 3 km spatial resolution of satellite data. On the other hand, the Sentinel-2 measurements provide more spatial details because of its increased spatial resolution, which is crucial for finer scale vegetation analyses. Similarly, the drone recordings are of very high quality but are unfortunately available only for the 2020 season for the purpose of this study. Thus, there is no real reference point and it would be impossible to detect drought by using only drone observations.
Although there were no drought events in this study, the results look promising and could be used for future vegetation studies. The merging of various remote sources of data should result in improved tracking and detection of future drought events. While vegetation data from geostationary satellites gives a very good general overview of the vegetation conditions, the more detailed drone measurements would be extremely useful for looking at the conditions at the specific locations. The MTG’s (Meteosat Third Generation) improved spatial resolution is expected to increase the accuracy of such vegetation studies in the future.