Phenological metrics can be derived from satellite data in several ways. Some researchers use complex mathematical models. Others employ threshold-based approaches that use either relative or pre-defined (global) reference values at which vegetative activity is assumed to begin. For example, seasonal midpoint NDVI (SMN) is a threshold-based approach that uses relative reference values to derive metrics.
This approach delineates SOS, for instance, using the midpoint between minimum and maximum NDVI values, thus tying the threshold to the seasonal amplitude of individual pixels and, in turn, the dynamic characteristics of each pixel. By contrast, in a pre-defined threshold approach, if the threshold value in a study using NDVI data is set at 0.099, SOS is the point at which NDVI values reach this point at the beginning of the growing season. This method can be effective for deriving the start of season in localized areas with relatively uniform land cover. But difficulties arise when using it to determine SOS over large areas with varying soil background characteristics or land cover types.
At USGS/EROS, we calculate phenological metrics from time-series NDVI data using a curve derivative method, which employs a backward-looking or delayed moving average (DMA). In essence, DMA values are predicted values based on previous observations along a time-series NDVI curve. In this approach, smoothed NDVI data values are compared to a moving average of the previous n observations to identify departures from an established trend. The trend change is defined as the point where the smoothed NDVI values become larger than those predicted by the DMA. This departure point is labeled as the start of the growing season (SOS). The end of the growing season is calculated in a similar manner, with the moving average run in reverse. Once these two parameters are defined, additional metrics are readily derived.
Below are other science projects associated with this project.
Data Smoothing - Reducing the "Noise" in NDVI
Validation
Challenges in Deriving Phenological Metrics
Deriving Phenological Metrics from NDVI
Data Smoothing - Reducing the "Noise" in NDVI
- Overview
Phenological metrics can be derived from satellite data in several ways. Some researchers use complex mathematical models. Others employ threshold-based approaches that use either relative or pre-defined (global) reference values at which vegetative activity is assumed to begin. For example, seasonal midpoint NDVI (SMN) is a threshold-based approach that uses relative reference values to derive metrics.
This approach delineates SOS, for instance, using the midpoint between minimum and maximum NDVI values, thus tying the threshold to the seasonal amplitude of individual pixels and, in turn, the dynamic characteristics of each pixel. By contrast, in a pre-defined threshold approach, if the threshold value in a study using NDVI data is set at 0.099, SOS is the point at which NDVI values reach this point at the beginning of the growing season. This method can be effective for deriving the start of season in localized areas with relatively uniform land cover. But difficulties arise when using it to determine SOS over large areas with varying soil background characteristics or land cover types.
At USGS/EROS, we calculate phenological metrics from time-series NDVI data using a curve derivative method, which employs a backward-looking or delayed moving average (DMA). In essence, DMA values are predicted values based on previous observations along a time-series NDVI curve. In this approach, smoothed NDVI data values are compared to a moving average of the previous n observations to identify departures from an established trend. The trend change is defined as the point where the smoothed NDVI values become larger than those predicted by the DMA. This departure point is labeled as the start of the growing season (SOS). The end of the growing season is calculated in a similar manner, with the moving average run in reverse. Once these two parameters are defined, additional metrics are readily derived.
- Science
Below are other science projects associated with this project.
Data Smoothing - Reducing the "Noise" in NDVI
The reflected light waves that satellite sensors detect coming from vegetation on the Earth's surface can be altered or blocked by a variety of phenomena, including aerosols and clouds in the atmosphere as well as changing illumination patterns and the angle at which the satellite views the ground at any given time. These phenomena introduce "noise" into raw satellite data. To address this problem...Validation
Documenting the effectiveness of the phenology metrics is difficult given the current scarcity of ground verification data at the appropriate scale (i.e., vegetation-canopy phenology). However, current efforts by the National Phenology Network to build and organize ground observation databases will go far toward addressing these shortcomings. In the meantime, two examples of coincident ground...Challenges in Deriving Phenological Metrics
Regardless of the method used, it is difficult to create algorithms sufficiently robust to derive phenological metrics from certain types of real time-series NDVI curves (as opposed to modeled or simulated data). For example, in desert shrublands (see below), time series NDVI shows little seasonal amplitude. Ideally, algorithms should be able to identify these regions and assign them a value such...Deriving Phenological Metrics from NDVI
Plotting time-series NDVI data produces a temporal curve that summarizes the various stages that green vegetation undergoes during a complete growing season. Such curves can be analyzed to extract key phenological variables, or metrics, about a particular season, such as the start of the growing season (SOS), peak of the season (POS), and end of the season (EOS). These characteristics may not...Data Smoothing - Reducing the "Noise" in NDVI
The reflected light waves that satellite sensors detect coming from vegetation on the Earth's surface can be altered or blocked by a variety of phenomena, including aerosols and clouds in the atmosphere as well as changing illumination patterns and the angle at which the satellite views the ground at any given time. These phenomena introduce "noise" into raw satellite data. To address this problem...