LanDAT: An Introduction

Vegetation is one of the most familiar features of a landscape, easily appreciated by the human eye. Changes in vegetation over time are less apparent, but these changes are crucially important for understanding the sustainability of landscapes.

Landscapes usually harbor a mix of different kinds of vegetation and land uses. As both human activities and natural processes alter landscapes through time, the benefits and conservation values provided by the land’s inherent productivity and variety also change.

While change is inevitable, in resilient landscapes change does not necessarily imply the loss of those core functions we value the most. Resilience can be thought of as the capacity of a landscape to maintain its core identity and functions even while change is ongoing. 

People expect certain benefits from the lands they inhabit or manage. Land management is often about enhancing and sustaining these benefits, which are sometimes described as ecosystem services. But this is a moving target—managers and planners must make decisions in the context of change driven by climate variability, natural disturbances, and other processes. Understanding landscapes through the lens of resilience can therefore be a valuable tool when sustainability is the goal.​

The map below illustrates how drivers of landscape change interact spatially with ecosystem services. In this case, the clean water provided by forests is impacted by land use intensity, and the impact varies from place to place. Housing Density (left) is a good indicator of land use intensity, which impacts a range of forest-based ecosystem services. People depend on forests for drinking water (right) to varying degrees in different watersheds.

In most landscapes, multiple stressors affect a wide variety of ecosystem services, and their cumulative impacts are not always clear. Understanding landscape resilience can help us to assess sustainability in an integrative way rather than piecemeal. After all, land management never affects just one resource. To achieve this, we need tools for measuring landscape change and assessing what that change implies about resilience.

LanDAT is a new tool for assessing how landscapes change, based on satellite-observed changes in vegetation. These observations connect to landscape resilience through a series of insights about vegetation and ecosystem behavior.

Seasonal changes in vegetation, known as phenology, tell us a great deal about the types of vegetation present, the influence of human activities, and how things like climate and topography shape vegetation types. Using land surface phenology (discussed below), LanDAT data products and maps help us understand landscape capacities and potential changes in ecosystem services when vegetation changes.

The data underlying LanDAT are time series of an index of vegetation greenness, the Normalized Difference Vegetation Index (NDVI). From these data, vegetation change over time and across landscapes can be characterized in many ways. NDVI data are gathered daily across the globe by two satellite-based sensors (MODIS). The data are filtered and smoothed to generate an eight-day, cloud-free composite time series that spans the period of MODIS record from 2000 until present for most of North America.

In an NDVI time series like the one plotted below, change over time at one pixel from 2000 to present is read from left to right. The pin in the map shows where this particular pixel is located in West Virginia, and the background map displays the mean growing season greenness (NDVI) for every pixel.

NDVI plots differ depending on the type of vegetation present and anything that may impact the vegetation over time. The plots below show changes in example pixels from 2000 to 2018.

The plot at left shows strong disturbance to relatively evergreen vegetation caused by a tornado, followed by rapid recovery. The plot at right shows impact from the invasive Hemlock Woolly Adelgid, which has caused widespread mortality in evergreen Hemlock trees in the Appalachian region. As affected trees declined in this pixel over several years, NDVI also declined—especially in winter months when the evergreen NDVI component prevails. The pixel is more deciduous in recent years due to the loss of evergreen trees.

Phenology is the seasonal timing of recurring biological events. Examples are spring leaf-out and the loss of leaves in fall. Broad changes across whole landscapes are referred to as land surface phenology. Land surface phenology reflects seasonal events as well as land use changes and irregular events like those shown above. This is useful because different ecosystems have different phenologies, and changes in phenology are indicators of ecosystem change.

To measure various aspects of land surface phenology, we examine the NDVI data in circular plots illustrating the annual cycle. The same data are shown in this plot as in the map viewer above. Features such as the strength of seasonality and the length of the growing season can be quickly calculated from these plots in a standard way across the map. The red line in the center of the plot points toward the middle of the growing season, and its length indicates seasonality. A shorter line would indicate more evergreen vegetation.

Some important aspects of land surface phenology are mapped below. All eleven phenology variables mapped by LanDAT are available in the Map Viewer.

The overall character of a pixel can be described using a combination of phenology variables. To understand patterns across the larger landscape in terms of this overall character, we reduce the collection of variables to a few fundamental ones that capture nearly all the important variation. This is done through a statistical method called factor analysis. We then group pixels into 500 classes—phenoclasses—based on their similarity to one another in these basic phenological dimensions. For this, we use a statistical clustering procedure.

Pixels may change classes from year to year depending on land use, ecological disturbance, and plant growth—anything that affects the vegetation strongly. As seen in the example plot at left, the progression of NDVI through the year can be different in different years. This drives change in phenological variables and in phenoclasses.

The maps below show the phenoclass of every pixel for two different years to illustrate how phenology tracks landscape change. These maps are color composites, with a color assigned to each of three variables that determine the phenoclasses. The legend at left describes the combination of these variables. 

Interact with these layers and see data plots for this location in the  Map Viewer .

Beyond understanding annual changes in small areas—single pixels—we can also quantify broader, long-term landscape dynamics using phenology. We can think of a landscape as a collection of adjacent pixels and characterize their collective change over time.

Landscapes behave in a variety of ways, ranging from simple to complex, highly organized to chaotic. The mix of phenoclasses within a landscape and how they change from year to year provides a way to define and measure landscape behavior. More specifically, we quantify landscape dynamics through time by applying information theory to our measures of land surface phenology.

Information Theory, which generally deals with the coding of information, may seem like an unlikely fit for describing landscapes. Actually, information theory is used by ecologists to describe the complexity and dynamics of ecological systems. In LanDAT, it helps us understand how signals (phenological changes) reveal landscape patterns and processes.  

For example, Shannon Entropy measures the overall diversity phenoclasses that have been observed over the years within landscapes—a way to assess the complexity of landscapes and their behavior. When many different phenoclasses are observed, the value is higher.

Mutual Information measures the predictability of phenoclasses from one year to the next. It asks, how much does knowing the phenoclasses in one year tell you about those observed the next year? Predictability can arise from progressive change that is regular or organized, or from long-term stability.

Ascendency combines Mutual Information and overall productivity (NDVI), so that more predictable and productive landscapes have higher values. The term comes from the observation that undisturbed ecosystems tend to become more predictable and productive (ascendant) over time.

Conditional Entropy is the part of landscape complexity which is unpredictable or disordered—essentially the complement of Mutual Information. It can reflect unpredictable disturbances, novel states and pathways of change, and other kinds of seemingly chaotic complexity. 

Many landscapes display relatively stable dynamics in which yearly changes occur at the pixel level, but the overall collection of phenoclasses in the landscape remains largely unchanged over the long term—a dynamic equilibrium. Other landscapes exhibit more directional change: the composition of phenoclasses has shifted over the long term. These ‘landscape trajectories’ suggest the kinds of change we might expect to continue to see, perhaps until some new dynamic equilibrium is reached.

LanDAT tools help to distinguish the direction of change and project longer-term conditions—assuming that the dynamics revealed through the period of observation (2000-present) remain in play. These trajectories are estimated at the landscape level using phenoclasses, and they can be used to estimate long-term trajectories in individual phenology variables.

The maps below show estimated long-term increases and decreases for two example LanDAT metrics. Such insights reflect the longer-term resilience or durability of a landscape’s defining vegetative features, which are closely bound to the ecological characteristics and functions that underpin a wide range of ecosystem services. The pins on the map illustrate some of the long-term change that is evident.

For example, Pacific Northwest forests have been strongly impacted by wildfire and by tree mortality from insects and disease in recent decades. Depending on the timing of such events, landscapes may be in a long-term phase of recovery (yellow areas) or decline (blue to purple areas), both driven by large-scale disturbance. Whereas fire can cause rapid and obvious disturbance followed by gradual recovery, insect outbreaks can cause more gradual declines.

Here, increase suggests an overall shift of the growing season to later in the calendar year. Long-term shifts of the growing season can be driven by earlier or later spring growth, fall decline, or both. Changes can also be due to subtle shifts in the timing of productivity within the growing season.