Dynamic Habitat Indices (DHIs)
The DHIs are calculated globally based on vegetation productivity estimates derived from Terra and Aqua MODIS Collection 5 satellite data. The DHIs are designed for biodiversity assessments and to describe habitats of different species.
Three individual indices comprise the DHIs (Fig. 1):
- DHI cum – cumulative DHI, i.e., the area under the phenological curve of a year
- DHI min – minimum DHI, i.e., the minimum value of the phenological curve of a year
- DHI var – seasonality DHI, i.e., the coefficient of variation of the phenological curve of a year
Fig. 1: Calculation of the three DHIs using productivity (p) at different time periods (t) over the course of a year. Cumulative DHI (DHI Cum) where productivity values are summed up for all the time periods over a year, Minimum DHI (DHI Min) where the minimum productivity value within a year is extracted, and Variation DHI (DHI Var) indicating the seasonality of the productivity by calculating the coefficient of variation using the standard deviation (σ) and the mean (μ) over a year. The number of time periods within a year is 46 for 8-day MODIS products and 23 for the 16-day products.
The DHIs are provided both as yearly product for each year for 2003 to 2014, and a composite product averaging the data over the time period of 2003 to 2014. Furthermore, the DHIs are derived from five different MODIS input data sets (Tab. 1). All DHIs have a spatial resolution of 1 km and the temporal resolution of the input data sets is either 8-day or 16-day.
Tab. 1: For the calculation of DHIs we used MODIS vegetation productivity products with a spatial resolution of 1 km and a temporal resolution or 8 or 16 days.
|Product||Index||Name||Platform||Temporal resolution||Spatial resolution|
|NDVI16||Normalized difference vegetation index||MOD13A2||Terra||16-day||1000 m|
|EVI16||Enhanced vegetation index||MOD13A2||Terra||16-day||1000 m|
|FPAR8||Fraction of absorbed photosynthetically active radiation||MOD15A2||Terra/Aqua||8-day||1000 m|
|LAI8||Leaf area index||MOD15A2||Terra/Aqua||8-day||1000 m|
|GPP8||Gross primary production||MOD17A2||Terra||8-day||1000 m|
Downloading and mosaicking of data
We downloaded the yearly MODIS HDF files, and converted them to GeoTIFFs. Dependent on the temporal resolution of the data product this resulted in 46 data sets (one for every 8 days for FPAR, LAI and GPP) or 23 data sets (one for every 16 days in the case of NDVI and EVI) for a given year. For each of these time steps in each year, we mosaicked all tiles together to get a global coverage per time step.
Quality assessment rules
To reduce noise due to clouds or haze, we extracted the associated quality assessment (QA) metadata to identify those pixels that had to be excluded from the analysis. Two different rules had to be applied due to the different quality assessment data provided for the FPAR, LAI and GPP data on one hand, and the NDVI and EVI data on the other hand. For FPAR/LAI/GPP, we applied a threshold for good pixels with a QA < 83 and for the NDVI/EVI data we only used pixels classified as ‘land’ or ‘ocean coastlines and lake shorelines’ (Explicit QA rule: QA<5’411 or (QA>=18’433 & QA<=21’798) or (QA>=34’817 & QA<=38’378) or (QA>=51’201 & QA<=54’574)).
Handling of fill values
For the FPAR and LAI data, we applied an additional rule for the fill values, which we set to zero or no data dependent on their land cover type. Fill values assigned to perennial snow and ice areas (fill value 252) as well as rock, tundra or desert (fill value 253) were set to zero, because these areas are ecologically meaningful in the calculation of the DHI, and vegetation cover is very low. In contrast, we assigned fill values reflecting urban/built-up and populated areas (fill value 250), permanent wetlands/inundated marshlands (fill value 251), as well as perennial salt and inland fresh water (fill value 254) to no data. This mask of no data areas was then also applied to the NDVI, EVI and GPP data sets to make the five DHIs data sets comparable.
Missing data in northern latitudes
In all the data sets we corrected for missing data at the start and the end of the season in the northern latitudes. Because the baseline for these areas is unknown we set these time steps at the beginning and end of the season to zero, if there were records of vegetation productivity during mid-season.
To eliminate noise from the raw data representing a yearly phenology curve, we applied a two-step filtering procedure. The steps included an iterative median and a Savitzky-Golay filter (see Fig. 2). The iterative median filter is based on Chen et al. (2004) and allows to reconstruct time-series. The main feature of the iterative median filter is to eliminate noise that reflects depressed raw values. The main difference of our procedure to the one of Chen et al. (2004) is that we use a median filter in the first step followed by the Savitzky-Golay.
Fig. 2: Two examples of the two filters within the smoothing process based on MODIS FPAR data. The first filter is an iterative median filter and second one is a Savitzky-Golay filter which is applied after the first filter.
The DHIs, especially minimum productivity, are relatively sensitive to occasional changes in data values that are typical for MODIS data, as well as climate events, such as droughts that can affect data for a given year. For some biodiversity analyses, such as analyses of species richness based on range maps, this interannual variation is not relevant. This is why we provide a single composite phenology curve from all MODIS data from 2003 to 2014 in addition to our DHIs for single years. The composite phenology curve represents the median value for each of the 12 observations that were available for each of the 23 or 46 time steps dependent on the MODIS input data. In addition, we only calculated the median if there were at least three valid pixels among the 12 possible years. We then used this composite penology curve to calculate the DHIs.
All the DHIs data that has been completed to date is available for downloading. Each link leads to a compressed file that contains a single composite RGB image of the three DHI components globally in GeoTIFF format: DHI cum (band 1), DHI min (band 2), DHI var (band 3). The data covers the whole globe and is in WGS84 (EPSG:4326) projection. If you have trouble uncompressing the data, please install 7-zip. Typical file size is around 700-800 Mb. The newest product is version 5.
|2003||download||in progress||in progress||in progress||download|
|2004||download||in progress||in progress||in progress||download|
|2005||download||in progress||in progress||in progress||download|
|2006||download||in progress||in progress||in progress||download|
|2007||download||in progress||in progress||in progress||download|
|2008||download||in progress||in progress||in progress||download|
|2009||download||in progress||in progress||in progress||download|
|2010||download||in progress||in progress||in progress||download|
|2011||download||in progress||in progress||in progress||download|
|2012||download||in progress||in progress||in progress||download|
|2013||download||in progress||in progress||in progress||download|
|2014||download||in progress||in progress||in progress||download|
|2015||download||in progress||in progress||in progress||download|
Codes for data processing
- Preparation of MODIS raw data (Python, GDAL)
- Smoothing of yearly phenology curves (Python, GRASS)
- Unpacking of MODIS QA (ODS tables)
Chen, J., P. Jönsson, M. Tamura, Z. Gu, B. Matsushita, and L. Eklundh. 2004. A simple method for reconstructing a high-quality NDVI time-series data set based on the Savitzky–Golay filter. Remote Sensing of Environment 91:332-344.