Satellite-Based Change Recognition and Tracking

Interpreting Change

ForWarn’s change products show NDVI departure from an eclectic mix of causes including climate stress, variation in the timing of seasons, biotic and abiotic disturbance agents, and successional recovery. At any given time, a complex of causes are often at work coincidentally in a landscape: wildfires often burn when vegetation is drought-stressed, logging and regrowth often occur in adjacent stands, and defoliation from storms or insects can occur in areas of active land cover change. Add to this the complication of year-to-year variability in seasonal transitions, and a landscape can harbor numerous anomalies, some of which have ready explanation while others, especially subtle differences, can remain uncertain.

Indicators of cause

Extent—The spatial scale of an anomaly is the best indicator of whether the cause is some seasonal climate factor or disturbance. In that seasonal phenology curves are not entirely consistent from one year to the next, current conditions can be more or less green at any given period compared to baseline years. Early or late springs, summer drought conditions and anomalous snowpack create regional patterns that manifest themselves locally, often according to topography. The challenge for the interpreter is to recognize local anomalies as part of a larger phenomenon (Figures 1, 2, 3).

Delayed spring in the Central Appalachians in 2013
Figure 1: The timing of spring in the Eastern US is highly variable from year to year. This change product compares NDVI for May 8, 2013 to the mean for the respective periods for all prior years since 2000. Note that it is only the high elevations of NC, TN, VA and WV that lag 10 to 12%. By this date, lower elevations have already greened up and are normal (blue), although many of these areas were similarly behind the mean earlier in the spring of 2013.
Areas hit hard by the 2011 drought in Texas
Figure 2: Drought often is sub-regional in scale as shown by this July 3, 2011 change product that compares NDVI to the period mean for all prior years since 2000. Most of the affected area is grassland, shrubland or irrigated cropland. Color differences within this massive anomaly relate to local differences in drought impacts related to the species that have changed. Portions of the regions were also affected by wildfires that also contribute to decline.
Areas where the 2007 frost was severe
Figure 3: Hard spring frosts are caused by regional weather patterns, so when severe, freezes create broad patterns. This remarkable frost event in April 5-9, 2007 followed weeks of abnormally warm temperatures and hit the mid elevation forests of Missouri, Arkansas and Tennessee hard. Similar patterns can result from early frosts in the fall. This change maps compares conditions soon after the event to those of the same period in 2006.

In the Western US, wildfires have grown in area in recent years so that they can affect moderate sized watersheds and landscapes. As the location of large disturbance events is generally well known from ancillary datasets, they are unlikely to be confused with the sub-regional scale of seasonal climate anomalies.

In general, the more local the anomaly, the longer the list of possible causes that should be considered. Regional climate factors can sometimes have local manifestations that with too close an inspection appear to be isolated. At local scales, logging, defoliation, fire or development are also among the possible explanations, so efficient assessments start by ruling out large-extent climate explanations.

Shape, edge and pattern of intensity—The structure of an anomaly within a landscape provides critical information about both the agent that caused it and the response of the vegetation. Responsiveness is often tied to vegetation type and topography, and these, along with other drivers such as insect population dynamics or fire weather contribute to pattern. Biotic and abiotic disturbances vary in the texture of their severity, as shown by the following examples:

Areas affected by tornadoes in 2011
Figure 4: Wind and ice storms often cause impacts with continuous gradients from the hardest hit core areas to a barely recognizable edge. Such impact gradients are even detectable across the narrow path of a tornado, and can correlate with tree uprooting in the hardest hit area, branch and leaf damage farther from the epicenter, to just ephemeral leaf stripping at the edge. The historic spring tornado outbreak in the Southeast shows wind severity gradients both across and along several tornadoes. Note in particular the massive and destructive storm that cut through forests between Tuscaloosa and Birmingham, Alabama. (June 1, 2011; 1 Year Standard Baseline).
Patchiness within the 2012 Whitewater Baldy Fire
Figure 5: While large wildfires often have sharp edges for suppression reasons, the area within a fire perimeter can exhibit strong patchiness caused by local differences in fire behavior and effects, and even patterns of responsiveness across differing canopy structures and compositions. Understory fires beneath dense conifer canopy will show up less strongly than will fire in adjacent woodlands. This phenomenon is suggested by New Mexico’s 2012 Whitewater Baldy Fire which caused more percent change in NDVI on the drier southwest slope of Mogollon Mountain (lower left of the burn perimeter shown by the black line) than on more densely treed northeast slopes. NDVI change here reflects the pre-burned state of the vegetation which was partially influenced by recent fires. A large part of the yellow and orange anomalies outside the fire perimeter are drought effects to open woodland vegetation that has a drought-sensitive grass or shrub component. Due to vegetational responsiveness, their change from drought is greater than that of nearby lightly-burned and denser conifer forests. (Sep. 28, 2012; 1 Year Standard Baseline).
Gypsy Moth defoliation in PA and NY
Figure 6: Insect defoliations such from the Gypsy Moth prefer host plants that conform to certain topographic positions. This often creates structure in the anomaly that is superimposed unto the broader scale bullseye core and periphery structure of an outbreak that is related to population dynamics and other factors. This Gypsy Moth outbreak during the early summer of 2013 shows conditions for July 11, 2013 compared to those of same period of 2012 (1 Year Standard Baseline).

Change behavior—The strongest asset of ForWarn is that because of its use of MODIS, it provides a continuous record of departure that allow analysts to explore the onset, developmental speed, spread and duration of change. This tool reveals changes that are short lived, or ephemeral, that could otherwise go unnoticed, such as leaf stripping events caused by insect defoliation, wind, ice, or low severity spring fire that can all be followed by leaf recovery.
When change is more severe, the onset speed or rate of change can be particularly revealing to interpret cause. For example, logging in an area can progress over the course of several weeks to months, while fire or storm effects are often immediate.

Individual insect defoliators often have a distinctive phenology of their own, so subtle differences in seasonal timing can reveal which of several possible candidates is present. Wildfire or prescribed fire seasons are also different from one region to the next, being more typically spring-associated in the east and a phenomenon of summer and fall in the West. Local or specialized knowledge is often critical for making sense of when causes are likely to be likely.
As a near-real-time emerging disturbance detection tool, ForWarn’s Early Detect product is particularly useful as it can detect change within a few days, clouds permitting. Persistence of the disturbance can also be a useful tool for characterizing disturbance behavior, as ephemeral disturbances, such as defoliation from hail in Spring, operate on much shorter time scales than more substantive disturbances that can cause tree mortality.

Expectations by cause

ForWarn’s viewer includes a broad suite of ancillary maps and data to help users diagnose the cause of observed changes using the indicators described above.

Early/late spring/fall—The onset and progression of spring is highly variable from year to year with a range of 3 weeks in much of the East. Fall is similarly variable, though often from different weather causes. During a late year in mountainous regions, anomalies show up elevationally, as there can be no anomaly from seasonal timing before or after the greenup transition period occurs. Anomalies often pass upward with elevation or northward latitudinally as the local spring wave passes according to this transitional sensitivity. Early fall onset is often triggered by a regional cold snap that affects broad areas more uniformly.

EXTENT: Regional to sub-regional
PATTERN: Consistent with topographic expectations
BEHAVIOR: Gradual to rapid anomalies in spring or fall

Drought--Drought is a sub-regional to regional phenomenon, but it can cause a finer scale patterned response due to the responsiveness of the vegetation present. In particular, open woodlands with a grass component are far more responsive to drought than is closed canopy conifer forest. Drought stress is persistent across periods, but can be alleviated after a single rain event. Tree or shrub mortality from drought can be subtle and impossible to recognize in near-real-time, and is particularly difficult to recognize when weather-sensitive grass is present and drought is persistent for more than a year; but detailed retrospective analyses can document this decline.

EXTENT: Regional to sub-regional
PATTERN: Strongest on exposed sites with drought-sensitive vegetation
BEHAVIOR: Very gradual onset; persistent for several periods

Insects, pests and diseases--Insects and diseases have wide ranging effects related to the phenology of the host species and agents involved. Damaging insects are often revealed by the timing of the outbreak and the topographic position or vegetation type affected.

EXTENT: Regional to sub-regional
PATTERN: Conforms to host distribution; often irregular or measles-like
BEHAVIOR: Often a gradual onset; ephemeral to persistent, depending on the season and severity

Wind or hail–Wind storms often produce a complex array of impacts that can confound NDVI detections. Leaf stripping is most common, and if the storm occurs in spring, trees can regrow the leaves lost or other vegetation on the site can compensate and fill the void. The specific type of storm changes expectations, as a hurricane or derecho can affect a broad area while tornadoes, straight-line winds or local downbursts are often localized.

EXTENT: Local, landscape, to sub-regional, depending on type
PATTERN: Conforms to storm behavior; tornadoes are linear, hurricanes are diffuse
BEHAVIOR: Sudden onset; ephemeral to persistent, depending on the season and severity

Wildland fire--The responsiveness of MODIS NDVI to fire varies considerably according to the fire’s impact on understory and overstory foliage. In many systems, loss of foliage need not be an indicator of mortality, as grass and many shrubs resprout. Relative mortality of conifer cover is indicated by change in the non-snow winter (evergreen) NDVI, while the deciduous fraction is indicated by the increase over winter’s value. These fractions often change pre- and post-fire.

EXTENT: Local to landscape
PATTERN: Variable depending on patch severity and prior vegetation responsiveness
BEHAVIOR: Sudden onset, but often with progressive landscape behavior; usually persistent

Logging—Clearcut logging patches are usually nearby other cutting unit, particularly in industrial timber lands. Even-aged conifer plantations have rows of trees, suggesting commercial timber land use. Selection logging or restoration treatments that use thinning open the canopy which results in less change than seen with clearcut logging. Either can increase the importance of drought sensitive understory, particularly when conifers are being removed. Large patches often emerge over the course of several weeks to months as groups of trees are successively removed.

PATTERN: Patches are of equal size to phenomenon observed on aerial photos
BEHAVIOR: Gradual development; lasting impacts

Development-- Urban development tends to be associated with new or existing roads and near previously developed areas; but ridgeline wind turbines and gas wells can cause change in more remote areas. NDVI drops can appear progressively, with tree removal followed weeks to months later by bare soil excavation. After development is complete, NDVI eventually achieves a reduced stable state given the mix of vegetation, structures and infrastructure on the site.

PATTERN: Blocky to linear
BEHAVIOR: Gradual onset; permanent impacts

Succession--ForWarn’s ability to track vegetation recovery is among its greatest strengths, and this shows up in ForWarn’s change imagery as two shades of blue when conditions are compared with post-disturbance years. It is common to see transitions in both overall NDVI and the phenological profile which suggests changes in vegetation type, such as succession from grass, to shrub, to more conifer dominated stands.

EXTENT: Local to landscape, consistent with the prior disturbance
PATTERN: Consistent with disturbance or topographic controls
BEHAVIOR: Gradual with a multi-year progression

Clouds (false positive)--ForWarn’s 24-day data-collection window effectively removes most clouds and cloud shadows, however some 24 day periods lack even a single day without clouds and this artificially reduces NDVI values. The Viewer includes two different True Color cloud assessment products that show natural color at 1km resolution for the 1 year standard and 1 year Early Detect products. Look for white or grey on the Viewer’s True Color maps.

PATTERN: Small patches; randomly occurring or in cloud-prone areas
BEHAVIOR: Usually ephemeral

For an additional discussion of available ancillary datasets and links, look under Overview/Datasets for Assessment Support.