ForWarn

Satellite-Based Change Recognition and Tracking

Near-Real-Time Products

The near-real-time monitoring need

Our response to disturbances such as wildfire, pests, wind or ice storms is often limited by knowing where the disturbance occurred. Existing disturbance tracking programs were not designed to track all forest or other vegetation types at high frequency, and this can contribute to delays in awareness, and if warranted, response. With its 8-day product frequency, ForWarn can recognize ephemeral biologically important disturbances that in some cases may go entirely unnoticed.

The systematic use of broad-scale remote sensing is different than past approaches to remote sensing that were designed to assess isolated disturbance events. For example, when and where large wildfires occur, burn severity assessments help land managers decide where to stabilize soil and mitigate suppression activities. Similarly, when hail or tornadoes strike populated areas or croplands, emergency efforts map the area affected to efficiently channel resources where they are needed most. In contrast, ForWarn tracks the entire landscape, all the time. In this sense, continuous monitoring from remote sensing is akin to the plot-based Forest Inventory and Analysis (FIA) program, although ForWarn is at 5.4 hectare resolution and updated weekly.

Historical baselines

ForWarn’s near-real-time products compare current NDVI to seasonally-appropriate NDVI values derived from it historical NDVI database. Historical baselines have been devised to capture the diverse needs of assessors and the different ways that change manifests itself across disturbances and vegetation types. Rather than presume that a single date or condition in the past is normal, ForWarn’s suite of baselines provide different contexts or perspectives on the current condition, and in many instances, more than one baseline gives the assessor the necessary insights to help diagnose the cause of the change and its implications.

NDVI values can reflect both disturbances and climate variation. Some baselines are more relevant than others for understanding which of these two causes is at work. The 3 year and All year maximum baselines provide the highest greenness observed for that period, and so approximate the concept of potential vegetation in a disturbance-free and climatically optimal year (although only within the 2000 to prior year time frame). Areas that were disturbed in 2001 (or thereafter) will have a lower NDVI than they did in 2000 until they recover. Therefore, baselines that use the max are a useful way to identify areas of disturbance.

However, maximum baselines are less useful for detecting disturbance in climate-sensitive vegetation types where the maximum observed NDVI reflects extreme productivity rather than a sense of background normal. For example, during an average year, grasslands may appear disturbed relative to the All-Year Maximum NDVI. To know if an area is experiencing drought stress or an early or late spring or fall, the All-Year Mean for the period is better suited. In many respects, the All-Year Mean is analogous to our everyday understanding of “normals” or “averages” used by meteorologists.

Figure 1: The implications of using different baselines for comparing 2012’s NDVI is suggested by the variable land surface phenology of the deciduous forests of Great Smoky Mountains National Park. 2012 experienced an unusually early spring and fall, but a slightly-below-normal summer. When subsequent springs are compared to 2012, they will show up as anomalous, unless they occur earlier yet. In contrast, by any historical baseline except for the 1 prior year, 2012’s fall NDVI was lower than expected.

Current NDVI values are calculated in two ways. All but the Early Detect product use the maximum NDVI value of the 24-day moving window. (That is, ForWarn uses the best cloud-free, and greenest observation obtained). While needed to ensure a clean, cloud free value, the approach often makes it difficult to detect disturbances during the growing season until the entire 24 day window has passed. Even with continuous clear skies, a wildfire on day 2 of the 24 day window leaves day 1 green values that will provide the maximum NDVI for the period. As this is the pre-fire condition, detection is delayed.

To achieve a more near-real-time emphasis on recency, Early Detect uses an adaptive length window so that disturbances might be captured after one day or in a similar length of time as needed by the standard baseline within the 24 day period if clouds persist. This approach is not without its shortcomings, as recent observations can provide marginal NDVI due to somewhat poor atmospheric quality or thin clouds. Acceptance of these NDVI values can cause apparent anomalies that are not vegetational.

Product short name

Calculation of current NDVI value

Calculation of baseline NDVI  value

Recommended use

Early Detect 1 Year

Adaptive window length, as image quality allows, to emphasize recency

The maximum value of the same period for the prior year

For near real time detection of disturbances as they occur or immediately after they occur when there is an urgent need for tracking progression for immediate response. The 1 year context is useful when the prior year was climatically similar

Standard 1 Year

The maximum of the 24 day window

The maximum value of the same period for the prior year

For near real time detection of disturbances when there is a less urgent need for immediate response or when the need for confidence is greater. This product helps isolate annual change in areas that see a lot of change, such as industrially logged areas.

Standard 3 Year

The maximum of the 24 day window

The maximum value of the same period for the prior 3 years

For near real time detection of disturbances that have occurred over the medium term (i.e., last three years).

Standard 5 Year

The maximum of the 24 day window

The maximum value of the same period for the prior 5 years

For near real time detection of disturbances that have occurred over the medium term (i.e., last five years).

Standard All Year Max

The maximum of the 24 day window

The maximum value of the same period for all prior years since 2000.

For detection of disturbances that have occurred anytime since 2000. It is particularly useful for capturing cumulative effects of disturbances and recovery across landscapes, but it can be hard to interpret in vegetation types that are sensitive to extreme variability in productivity from climate, such as forests with a grass component.

Standard All Year Mean

The maximum of the 24 day window

The maximum value of the same period for all prior years since 2000.

For isolating the effects of disturbance from climate in climate-sensitive vegetation such as areas of drought or when disturbances occur during spring or fall as the mean can provide a measure of central tendency or “climate normal” in areas that haven’t experienced much disturbance.

Phenotype Max

The maximum of the 24 day window

The maximum of annual phenotypes since 2000 for the same period

For understanding disturbance or climate impacts compared to the normal behavior of previously similar phenologically-defined types.

Figure 2: Baselines tell different stories, particularly in landscapes with a lot of change. In northeastern California for Sep. 21, 2014, the Early Detect product shows larger burn areas in the fire just southeast of the “CA” than the Standard 1 year baseline, but it also shows more decline in the dryer eastern half of the image and along the foggy coast. Progressing from the 1-Year to the 3-Year the All-Year shows the cumulative decline from past fire events in addition to the 2014 drought impacts on sensitive vegetation in the east. This drought effect is lessened in the All-Year Mean, as it provides more average baseline values rather than the year of greatest productivity and that helps separate disturbances from climate impacts. Average long-term values make it hard to make sense of cumulative fire effects involving disturbance or recovery. The phenotype maximum baseline mutes the extreme values of the All-Year Maximum.