Limits...
Global data for ecology and epidemiology: a novel algorithm for temporal Fourier processing MODIS data.

Scharlemann JP, Benz D, Hay SI, Purse BV, Tatem AJ, Wint GR, Rogers DJ - PLoS ONE (2008)

Bottom Line: MODIS data are composited over 8- or 16-day time intervals that pose unique problems for temporal Fourier analysis.Applying standard techniques to MODIS data can introduce errors of up to 30% in the estimation of the amplitudes and phases of the Fourier harmonics.The finer spatial and temporal resolution, combined with the greater geolocational and spectral accuracy of the MODIS instruments, compared with previous multi-temporal data sets, mean that these data may be used with greater confidence in species' distribution modelling.

View Article: PubMed Central - PubMed

Affiliation: Spatial Ecology and Epidemiology Group, Department of Zoology, University of Oxford, Oxford, United Kingdom.

ABSTRACT

Background: Remotely-sensed environmental data from earth-orbiting satellites are increasingly used to model the distribution and abundance of both plant and animal species, especially those of economic or conservation importance. Time series of data from the MODerate-resolution Imaging Spectroradiometer (MODIS) sensors on-board NASA's Terra and Aqua satellites offer the potential to capture environmental thermal and vegetation seasonality, through temporal Fourier analysis, more accurately than was previously possible using the NOAA Advanced Very High Resolution Radiometer (AVHRR) sensor data. MODIS data are composited over 8- or 16-day time intervals that pose unique problems for temporal Fourier analysis. Applying standard techniques to MODIS data can introduce errors of up to 30% in the estimation of the amplitudes and phases of the Fourier harmonics.

Methodology/principal findings: We present a novel spline-based algorithm that overcomes the processing problems of composited MODIS data. The algorithm is tested on artificial data generated using randomly selected values of both amplitudes and phases, and provides an accurate estimate of the input variables under all conditions. The algorithm was then applied to produce layers that capture the seasonality in MODIS data for the period from 2001 to 2005.

Conclusions/significance: Global temporal Fourier processed images of 1 km MODIS data for Middle Infrared Reflectance, day- and night-time Land Surface Temperature (LST), Normalised Difference Vegetation Index (NDVI), and Enhanced Vegetation Index (EVI) are presented for ecological and epidemiological applications. The finer spatial and temporal resolution, combined with the greater geolocational and spectral accuracy of the MODIS instruments, compared with previous multi-temporal data sets, mean that these data may be used with greater confidence in species' distribution modelling.

Show MeSH

Related in: MedlinePlus

Examples of temporal Fourier processed artificially generated MODIS data for two years using (a) a standard TFA algorithm and (b) a standard TFA algorithm applied to spline-interpolated data.The signal (black line) is a daily time series artificially generated by summing annual, bi-annual and tri-annual cycles of known, randomly chosen amplitudes and phases. The ‘satellite sample’ (blue vertical lines) samples this signal at the MODIS 16-day interval and on the MODIS mid-sample date, which gives unequal intervals spanning each year end (upper tick marks on x-axis). The Fourier fit (red vertical lines) is the fit to the satellite signal that ignores this beginning/end of year anomaly and thus assumes a constant interval throughout, corresponding to the 23 images per year of the satellite sample (lower tick marks on x-axis). In (b) the daily spline fit (yellow line) is the cubic spline fit to these irregular satellite sample data. The Fourier fit (red vertical lines) is the TFA fit to the spline fit resampled every 5 days (lower tick marks on x-axis). Notice that there are no end-of-year anomalies here, resulting in a more accurate estimate of the harmonics used to generate the signal. The end of year anomaly is also present in MODIS data that run for only one year and hence affects TFA outputs in the same way, but is more clearly demonstrated visually in multi-year data.
© Copyright Policy
Related In: Results  -  Collection


getmorefigures.php?uid=PMC2171368&req=5

pone-0001408-g003: Examples of temporal Fourier processed artificially generated MODIS data for two years using (a) a standard TFA algorithm and (b) a standard TFA algorithm applied to spline-interpolated data.The signal (black line) is a daily time series artificially generated by summing annual, bi-annual and tri-annual cycles of known, randomly chosen amplitudes and phases. The ‘satellite sample’ (blue vertical lines) samples this signal at the MODIS 16-day interval and on the MODIS mid-sample date, which gives unequal intervals spanning each year end (upper tick marks on x-axis). The Fourier fit (red vertical lines) is the fit to the satellite signal that ignores this beginning/end of year anomaly and thus assumes a constant interval throughout, corresponding to the 23 images per year of the satellite sample (lower tick marks on x-axis). In (b) the daily spline fit (yellow line) is the cubic spline fit to these irregular satellite sample data. The Fourier fit (red vertical lines) is the TFA fit to the spline fit resampled every 5 days (lower tick marks on x-axis). Notice that there are no end-of-year anomalies here, resulting in a more accurate estimate of the harmonics used to generate the signal. The end of year anomaly is also present in MODIS data that run for only one year and hence affects TFA outputs in the same way, but is more clearly demonstrated visually in multi-year data.

Mentions: An example time series of the artificial MODIS data subjected to standard TFA is shown in Figure 3a. The slippage between the observed and fitted values at the end of the first year is visible. TFA has assumed an equal interval throughout the 2-year time period of these artificial data, and predicts the signal at these intervals throughout. This assumption affected the estimated values of amplitudes and phases, so that the fitted values (in red) did not capture the signal satisfactorily (obvious at the very first fitted value, but noticeable throughout). This would also occur if only one year's worth of MODIS-type data were analysed by TFA since the method assumes the time series continues, as measured, forever. Application of the new TFA algorithm to the same time series is shown in Figure 3b. The 5-day spline interpolated time series, itself an accurate representation of the input time series, was described very accurately by TFA, and provides more accurate estimates of the input values of mean, amplitudes and phases.


Global data for ecology and epidemiology: a novel algorithm for temporal Fourier processing MODIS data.

Scharlemann JP, Benz D, Hay SI, Purse BV, Tatem AJ, Wint GR, Rogers DJ - PLoS ONE (2008)

Examples of temporal Fourier processed artificially generated MODIS data for two years using (a) a standard TFA algorithm and (b) a standard TFA algorithm applied to spline-interpolated data.The signal (black line) is a daily time series artificially generated by summing annual, bi-annual and tri-annual cycles of known, randomly chosen amplitudes and phases. The ‘satellite sample’ (blue vertical lines) samples this signal at the MODIS 16-day interval and on the MODIS mid-sample date, which gives unequal intervals spanning each year end (upper tick marks on x-axis). The Fourier fit (red vertical lines) is the fit to the satellite signal that ignores this beginning/end of year anomaly and thus assumes a constant interval throughout, corresponding to the 23 images per year of the satellite sample (lower tick marks on x-axis). In (b) the daily spline fit (yellow line) is the cubic spline fit to these irregular satellite sample data. The Fourier fit (red vertical lines) is the TFA fit to the spline fit resampled every 5 days (lower tick marks on x-axis). Notice that there are no end-of-year anomalies here, resulting in a more accurate estimate of the harmonics used to generate the signal. The end of year anomaly is also present in MODIS data that run for only one year and hence affects TFA outputs in the same way, but is more clearly demonstrated visually in multi-year data.
© Copyright Policy
Related In: Results  -  Collection

Show All Figures
getmorefigures.php?uid=PMC2171368&req=5

pone-0001408-g003: Examples of temporal Fourier processed artificially generated MODIS data for two years using (a) a standard TFA algorithm and (b) a standard TFA algorithm applied to spline-interpolated data.The signal (black line) is a daily time series artificially generated by summing annual, bi-annual and tri-annual cycles of known, randomly chosen amplitudes and phases. The ‘satellite sample’ (blue vertical lines) samples this signal at the MODIS 16-day interval and on the MODIS mid-sample date, which gives unequal intervals spanning each year end (upper tick marks on x-axis). The Fourier fit (red vertical lines) is the fit to the satellite signal that ignores this beginning/end of year anomaly and thus assumes a constant interval throughout, corresponding to the 23 images per year of the satellite sample (lower tick marks on x-axis). In (b) the daily spline fit (yellow line) is the cubic spline fit to these irregular satellite sample data. The Fourier fit (red vertical lines) is the TFA fit to the spline fit resampled every 5 days (lower tick marks on x-axis). Notice that there are no end-of-year anomalies here, resulting in a more accurate estimate of the harmonics used to generate the signal. The end of year anomaly is also present in MODIS data that run for only one year and hence affects TFA outputs in the same way, but is more clearly demonstrated visually in multi-year data.
Mentions: An example time series of the artificial MODIS data subjected to standard TFA is shown in Figure 3a. The slippage between the observed and fitted values at the end of the first year is visible. TFA has assumed an equal interval throughout the 2-year time period of these artificial data, and predicts the signal at these intervals throughout. This assumption affected the estimated values of amplitudes and phases, so that the fitted values (in red) did not capture the signal satisfactorily (obvious at the very first fitted value, but noticeable throughout). This would also occur if only one year's worth of MODIS-type data were analysed by TFA since the method assumes the time series continues, as measured, forever. Application of the new TFA algorithm to the same time series is shown in Figure 3b. The 5-day spline interpolated time series, itself an accurate representation of the input time series, was described very accurately by TFA, and provides more accurate estimates of the input values of mean, amplitudes and phases.

Bottom Line: MODIS data are composited over 8- or 16-day time intervals that pose unique problems for temporal Fourier analysis.Applying standard techniques to MODIS data can introduce errors of up to 30% in the estimation of the amplitudes and phases of the Fourier harmonics.The finer spatial and temporal resolution, combined with the greater geolocational and spectral accuracy of the MODIS instruments, compared with previous multi-temporal data sets, mean that these data may be used with greater confidence in species' distribution modelling.

View Article: PubMed Central - PubMed

Affiliation: Spatial Ecology and Epidemiology Group, Department of Zoology, University of Oxford, Oxford, United Kingdom.

ABSTRACT

Background: Remotely-sensed environmental data from earth-orbiting satellites are increasingly used to model the distribution and abundance of both plant and animal species, especially those of economic or conservation importance. Time series of data from the MODerate-resolution Imaging Spectroradiometer (MODIS) sensors on-board NASA's Terra and Aqua satellites offer the potential to capture environmental thermal and vegetation seasonality, through temporal Fourier analysis, more accurately than was previously possible using the NOAA Advanced Very High Resolution Radiometer (AVHRR) sensor data. MODIS data are composited over 8- or 16-day time intervals that pose unique problems for temporal Fourier analysis. Applying standard techniques to MODIS data can introduce errors of up to 30% in the estimation of the amplitudes and phases of the Fourier harmonics.

Methodology/principal findings: We present a novel spline-based algorithm that overcomes the processing problems of composited MODIS data. The algorithm is tested on artificial data generated using randomly selected values of both amplitudes and phases, and provides an accurate estimate of the input variables under all conditions. The algorithm was then applied to produce layers that capture the seasonality in MODIS data for the period from 2001 to 2005.

Conclusions/significance: Global temporal Fourier processed images of 1 km MODIS data for Middle Infrared Reflectance, day- and night-time Land Surface Temperature (LST), Normalised Difference Vegetation Index (NDVI), and Enhanced Vegetation Index (EVI) are presented for ecological and epidemiological applications. The finer spatial and temporal resolution, combined with the greater geolocational and spectral accuracy of the MODIS instruments, compared with previous multi-temporal data sets, mean that these data may be used with greater confidence in species' distribution modelling.

Show MeSH
Related in: MedlinePlus