Lab 04: Resilience of the Chaco forests to drought disturbance
Introduction
In this assignment we will try to find an answer to the question Do the remaining woodlands in the Chaco lose their ability to recover after disturbance? In particular, we will analyze vegetation greenness after drought events. The Chaco – as you probably know by now – is a major deforestation frontier region in the world. Over 30% of the woodlands have been lost since the 1980s, primarily for expansion of row crop agriculture and cattle ranching. In assignment #2 we have focused on the implications of agricultural expansion for carbon fluxes in the region using a carbon bookkeeping approach. In this assignment we focus on the areas that remain unaffected so far. One key element of this assignment is that we return to \(\textbf{R}\) in order to run the analysis. We have prepared a number of materials for you, including some of the code that you will need to successfully read and manipulate the data. Yet, you will have revisit some of the techniques and materials from module M1 if you feel insecure, for example on how to do filtering operations and/or selections. The data you will be dealing with are in parts quite large, so (a) don’t be surprised that some operations may take a moment to calculate, and (b) think about efficiency when processing the data. We will give some hints on how to address (b) along the way, but we also want you to come up with your own ideas on how to improve the efficiency of the process. More generally, this lab is based heavily on the paper by Gazol et al.1 – you can find the paper in moodle. Make sure that you carefully read it before and while doing the analysis.
Go to moodle, download the .Rmd-file and open it in \(\textbf{R}\). Get familiar with the structure of the script and carefully read the descriptions. You will see that we have prepared some code, for example on how to read in the datasets successfully so that you can start working with the data in form of a tibble. You will also find in moodle the input data for this assignment which is a multi-band raster file that we will use in this lab. We ask you in this lab again to submit a .html file as well as some quiz questions (multiple choice as well as calculated numbers).
Exercise I:
Go into the .Rmd-file and follow the instructions on how to read the raster file into R. Update the path according to the location of where you have stored your data. Executing the code chunk will take the raster file and store it as a variable which you now can conveniently manipulate. Explore the dataset! It contains (a) an annual time series 2002-2021 of the 90th percentile of MODIS Aqua+Terra EVI data (hereafter: EVI-data, band names: y2002, y2003,…) (b) a layer that indicates the precipitation anomaly for the year 2013 compared to the 1981-2017 average, (c) the distance of a pixel to frontier areas (Baumann et al. 2022)2, and (d) the year of the onset of the frontier (onset_yr). Load the data into \(\textbf{R}\) using the provided code chunk and examine the dataset in \(\textbf{R}\). Have a look as well at the other bands and the properties of the image. Once you are done with the examination it is time to prepare the data for our analysis. As you remember, we will focus on areas that have not experienced agricultural expansion yet. Differently put, we want to mask those areas that are outside of the agricultural frontier. This means, we need to have a look at the layer onset_yr which contains information when a pixel became a frontier between the period 1985-2020. All other values in that layer do not represent frontier areas. Using this information, remove all areas from the dataset that are frontier areas. After masking, plot the same layer again to control your result.
Exercise II:
Now that we know the dataset a bit, it is time to manipulate the data for our analysis. As you can read in the Gazol et al. (2018) paper and the paper by Lloret et al. (2011)3, for our analysis we need to define a pre-drought period to which we compare the drought and the post-drought years. The drought event that we are looking at was in 2013, so naturally we would define all years prior to that year as the reference period. However, in the Chaco the Chaco experienced a less severe drought in 2004/2005 whose effect we don’t want to consider in our analysis. Hence, we define the pre-drought period as 2006-2012. Calculate the mean EVI for the period 2006-2012. Once this is done, we want to examine the difference in the EVI values from the drought year to the reference period. Using the knowledge you gained from the code we provided for the previous operation, please calculate the difference between the pre-drought EVI and the EVI of the year 2013. We call this layer ‘drought severity’ hereafter in the document.
Once you have done this, we want to take a look at the data to see whether 2013 was really an unusual year, i.e., whether the EVI values of that year are substantially smaller than from previous and subsequent years. For simplicity, we want to do this at the aggregate level. This means that we calculate the average over all non-frontier pixels. Doing these aggregate summaries is easier to do in a tibble compared to a raster dataset. Thus, in a first step we need to convert our raster dataset into a tibble. We have prepared some code chunks that do that for you. Execute the code and examine the new tibble. Once you understand the data structure, summarize the data so that you get one EVI value for each year that represents the mean EVI value of that year for all non-frontier pixels in the Dry Chaco. Important: the data are for storage reasons provided as 16bit Unsigned Integer values. The data range is 0-10000, but EVI is defined with a range 0-1. Before you answer the questions below, make sure you scale the data back to the normal EVI value range.
Next, we want to see whether the areas (i.e., pixels) of drought severity and the precipitation anomaly correspond to each other. The basis for this analysis will again be the tibble you created in the previous step – yet, for visualization purposes, we want you to make this comparison based on a random sample of 10,000 points. Choose an appropriate visualization for this comparison – be creative here: add a trend line or a local regression. Combine the plot with the previous plot (i.e., the EVI time series)
Exercise III:
Now that we have gotten an overview of the data, we want to assess the resilience of the remaining forests in the Chaco to the drought. To do that we will calculate a variety of different indicators for what can be interpreted as resilience of forest systems to drought event. We will adapt the concept of the paper by Gazol et al. (2018) and modify it in a way that we generate a spatial representation of the resilience indicators the authors propose. You will heavily rely on the data you prepared for exercises I and II, so if you don’t feel comfortable with the data processing, return to the code you’ve written so far and revisit the parts where you felt it was most difficult. In the Gazol et al. paper a number of different resilience indices are being presented. Below is a summary of these, but make sure you have a detailed look into the paper, and specifically to figure 5 where the authors present the results. Be aware that their work is to large extents non-spatial in the sense that they do not look at spatial distributions like we do:
The resistance index (Rt) quantifies the difference between EVI during the drought year and the preceding non-drought period (i.e., the capacity of trees to buffer the drought stress and continue growing during drought), whereas the recovery index (Rc) accounts for the growth reaction following the drought year (i.e., the difference EVI between the drought year and the following 3 years). The resilience index (Rs) quantifies the difference in EVI before and after the drought year (i.e., the capacity of trees to recover EVI values similar to those observed before the drought). Finally, the relative-resilience index (rRs) is the difference between Rs and Rt (Lloret et al., 2011).
Using the knowledge you gained in terms of manipulating your data in R, calculate the four indices in R and visualize them. When you calculate them, make sure you define the pre-drought period in the same way we did in exercise I. Plot the four maps inside a code chunk. In addition, answer the following question:
In this last bit, we want to create a combination of the indices Rt and Rc. Specifically, we want to assess how long it takes for the forests after the drought to recover to the pre-drought state. Depending on the status of the forest, the speed of recovery can substantially vary. As a result, one can interpret this speed of recovery as an indicator of ecosystem resilience. Your task now is to assess whether this phenomenon is reflected in our time series of EVI-data. Specifically, we ask how many years it took for EVI values to get back within 95% of the pre-disturbance level (i.e., the averaged values for the period 2006-2012)? We will do this in two ways: (1) in a non-spatial, aggregated way; (2) in a spatially explicit way.
Use the provided code-chunks to run these two analyses. Start with the non-spatial analysis and use the tibble/data.frame for this purpose. The general idea for this process is to record for each observation the first year when its EVI value returned back to pre-drought conditions (i.e., within 95%). This will require some thinking on the order of the processing, so that you do not overwrite important information. Once you have established a workflow for the non-spatial analysis, you can repeat the same processing technique for the spatial data.
Last, but not least, we want to better understand why some forest areas appear to be less resilient than others. One way to do that is to compare the outcome of this analysis to a recent satellite image with a high enough resolution that allows you to interpret and infer on potential processes that are/were happening on the ground. We have provided you with a line of code that allows you to convert the layer you have created before into a GeoTIFF file.
Gazol, A. et al. (2018). Forest resilience to drought varies across biomes. Global Change Biology, 24, 2143-2158↩︎
Baumann, M. et al. (2022). Frontier metrics for a process-based understanding of deforestation dynamics. Environmental Research Letters, 17, 095010↩︎
Lloret, F., et al. (2011). Components of tree resilience: effects of successive low-growth episodes in old ponderosa pine forests. Oikos 120, 1909-1920↩︎