Using satellite imagery to track the pulse of vegetation recovery

The success of vegetation restoration efforts requires careful monitoring over many years to ensure that minimum thresholds are met with regards to vegetation cover, species makeup, function, and the health of the vegetation ecosystem. With the help of Nayani Ilangakoon (CU EarthLab) and Steve Perkins (Chevron) I have been exploring how we can use satellite imagery and weather data to monitor the success of vegetation restoration measures at McKinley Mine, a closed open pit coal mine, in north-western New Mexico (see map and figure below).

For this work I used historical satellite imagery from the Landsat 8 system going back to 2013 to explore how NDVI (Normalized Difference Vegetation Index), a common measure of the health or 'greenness' of vegetation that can be derived from the Landsat 8 data, changes over time for different areas of the mine, which have been revegetated at different times and using different approaches. I used as the basis for my analysis a database of 'seeding units' that provides information on the location/extent, age, and type of vegetation restoration that has been performed at McKinley Mine (see map below).

I built a workflow to automate the extraction of the average NDVI values of the different seeding units from each of the Landsat 8 images; in total after removing satellite images with cloud and snow there were a total of 70 images from 2013 to 2021. The extracted 'time series' of NDVI values (see examples below) for the different seeding units stimulated a variety of questions that I hoped to answer:

1. How does NDVI vary with time for the different seeding units?
2. How does NDVI vary over the short and long-terms?
3. How does the type of seeding impact NDVI?
4. What is the impact of rainfall on the NDVI trends for the seeding units?

Investigating the short and long term trends in NDVI

When we look at the plot above we can see for each year that there is typically a peak in NDVI in the spring timeframe. This corresponds to the 'green-up' of vegetation, particularly annual and perennial grasses and flowering plants, that occurs in the spring followed by drying-out and browning (senescence) of the vegetation into the summer and fall. When you compare the maximum NVDI value against total precipitation (from gridMET) for each year there does seem to be a loose relationship in that the NDVI values are higher in wetter years. This is not unexpected but the relationship is not as strong as I would expect and perhaps points to other factors being important such as temperature and whether precipitation is in the form of snow or rain.

To better discern the long-term trends in NDVI I used an approach called a 'moving window' to smooth out the strong month to month variation. This technique worked by calculating an average NDVI across a two year period and shifting this period from month to month across the time series. The plot below shows an example of applying this algorithm to the time series for a single seeding unit.

I investigated the significance of the age and type of vegetation restoration measure using various summary metrics that I calculated from the smoothed NDVI time series data. In particular I calculated an 'NDVI anomaly' that represented the difference in average NDVI between year 0 (when the seeding unit was restored) and subsequent years. The plot below shows for each seeding unit how the NDVI anomaly changes with time and also what type of restoration was performed.

In general the plot below shows that post-restoration a majority of the seeding units have a positive and persistent NDVI anomaly consistent with the growth of new vegetation. That said there are seeding units where we see a reduction in the NDVI anomaly with time. Also, when we look at the type of restoration measure the plot does seem to indicate that seeding units that have been 're-seeded' do appear to have NDVI anomalies that are consistently trending down with the age. This is particularly interesting as these areas may indicate ongoing issues; re-seeding occurs with areas where restoration end-points are not being met and the revegetation has to start from scratch.

I also wanted to investigate whether there was any relationship between NDVI anomaly and annual precipitation for the seeding units. The plot below suggests that there isn't a relationship as we don't see higher NDVI anomalies in wetter years.

Identifying areas of concern

Using a simple threshold of the NDVI anomaly we can identify and map seeding units that may be of concern with regards to meeting vegetation restoration end points. In the figures below I have plotted maps of seeding units that have negative NDVI anomaly values (red) after three and five years respectively; seeding units that have positive values are indicated in green. What is particularly interesting is that the area towards the East of the map seems to show growing areas of concern between the two years. These may be areas that require follow-up with regards to the success of vegetation restoration.