Moment Distance Metric Assessing the Moment Distance Metric as an Indicator of Spring Green-up Using MODIS NBAR Data and Lilac/Honeysuckle Bloom Dates Eric Ariel L. Salas1, Mark D. Schwartz2, and Geoffrey M. Henebry1 1Geographic Information Science Center of Excellence (GIScCE), South Dakota State University, Brookings, SD 57007-3510 2Geography, University of Wisconsin-Milwaukee, Milwaukee, WI 53201-0413 Overview Approach Results [1] Results [2] We present a new metric called the Moment Distance (MD) that characterizes the shape of the reflectance curve in simple geometric operations. We compare 70 lilac and honeysuckle phenophase sequences from 2008 with land surface phenologies described by NDVI and MD time series based on MODIS NBAR 0.05 degree data. The MD metric takes advantage of the seven bands available in the NBAR product and at each composite calculates the moment distances among the bands. Comparison between the NDVI and the MD reveals that MD can provide crisper identification of the timing of phenophase sequences. The disparity in spatial scale between the 0.05 degree MODIS data and the point observations is a source of potential error. However, if clonal plants are to be useful in sentinel networks, then the inference domain of the observations must extend beyond the immediate environmental neighborhood of the ground observations. Results also suggest that the onset of senescence may be clearly delimited due to the inclusion of both the visible and shortwave infrared bands in the MD calculation; however, we do not have ground observations of senescence at this point. We will be pursuing this analysis further with finer spatial resolution datasets. Monitoring the timing of blooming phases of lilac and honeysuckle may provide insight into the species ecosystem responses to seasonal change. NDVI (Tucker, 1979 ) offers a way to look at the land surface phenologies using MODIS NBAR 0.05 degree data. In addition, we introduced the Moment Distance to help provide a more receptive metric to detecting the timing of the phenophases. How will the MD compare against the NDVI in identifying the timing of phenophase sequences? How sensitive is the MD in catching the timing (even in the absence or presence of important MODIS bands in the calculation)? These questions lead to our goals: (1) To compare lilac and honeysuckle phenophase sequences against the 2008 land surface phenologies described by NDVI and MD time series (2) To look at the changes of the MD curve vis-à-vis the phenophase sequences by varying the number of MODIS bands. A. MD vs. NDVI for Phenophase Timing Detection NDVI shows increasing values starting from around the 100th day and then peaks around the 200th day. MD illustrates flatness during the early days before it starts to spike around the 140th day. We divided the samples based on the location of the phenophase and the timing of the curve peaks of the MD. First group with End Bloom near or at the foot of the peak (left figure) and second group with End Bloom at the slope of the peak (right figure). *Best 6-band combination for MODIS-MD based on the MD average within the range considered; **‘Very low’ to ‘No relationship’ was evident for most of the range of days considered. For the first group, once the EDB is reached, the curve begins to peak. The average EDB for the samples of the First Group is 148 days. MD values during the blooming period play between the 40-60 range. The average EDB for samples in the Second Group is 161 days with curve flattening at the phenophase sequences less evident. Results also show that at low NDVI, MD is observed to flatten out. MD has a lesser average correlation against the NDVI in the range 97 to 153 days, which covers the phenophase sequence (left figure). The disparity is caused by the presence or absence of the flattening. MD vs. Ave. NDVI gives a high r2 considering all days (right figure). These results use 7 MODIS NBAR 0.05 bands. *Best 5-band combination for MODIS-MD based on the MD average within the range considered; **Not advisable to omit the Red and NIR. The graphs below show variations of the MD slope within the phenophase sequence (97 to 153 days) when a MODIS band was deleted one at a time. Both figures exhibit importance of the Red band in vegetation studies as negative slopes were obtained. Moment Distance Metric Moment Distance (MD) Explained The MD is a matrix of distances computed from two selected reference wavelengths to every wavelength within the specified wavelength range. The distances involved are the diagonal lengths from the abscissa of a reference wavelength to a point on the reflectance curve. Datasets MODIS NBAR 0.05 Degree Data The 2008 MODIS NBAR images are available every 8 days. The products are provided in global fields at a 0.05 degree resolution in a geographic latitude-longitude projection. MODIS NBAR has 7 bands – covering from around 400 nm to 2000 nm. B. Lilac and Honeysuckle Samples We utilized 70 lilac and honeysuckle phenophase sequences collected in 2008 (see figure below). Five phenophase sequences consist of First Leaf (FL), Leaf 95% (95%), First Bloom (FTB), Full Bloom (FLB), and End Bloom (EDB) – described in number of days. The Lilac and Honeysuckle datasets are part of the “Plant Phenology Programs” of the developing USA-National Phenology Network (www.usanpn.org) and are available online for free through the USA National Climatic Data Center. Refer to Schwartz and Caprio (2003) “North American First Leaf and First Bloom Lilac Phenology Data” for additional information.  Reference Conclusions End Bloom usually falls near the foot of the MD peak. Results have shown that MD has the capability of the NDVI in detecting Lilac/Honeysuckle peak due to greenness. MD, however, is able to associate ‘flatness’ to phenophase sequence, thus adding more vital information. The 7 MODIS NBAR bands have shown to be sufficient in finding vegetation peak seasons. Red and NIR should never be omitted in the MD analysis, while band 2130 nm may not be essential for the early part of the phenophase. MD Distance Reflectance (%) B. MD Sensitivity with Varying MODIS Bands One of the advantages of the MD is its capability to supplement the loss or absence of an important band and still come up with satisfactory relationships. We tested its curve shape detection potential through deletion of bands. The left figure below shows the NDVI and MD curves after the elimination of 2130 nm band (using only 6 MODIS bands). The right figure uses 5 MODIS bands for the MD, deleting Red and 2130 nm. 1 2 3 4 5 6 7 MD Index A References MD Index B Schwartz, M.D. and J.M. Caprio. 2003. North American First Leaf and First Bloom Lilac Phenology Data, IGBP PAGES/World Data Center for Paleoclimatology Data Contribution Series # 2003-078. NOAA/NGDC Paleoclimatology Program, Boulder CO, USA. Tucker, C.J. Red and photographic infrared linear combinations for monitoring vegetation. 1979. Remote Sensing of Environment, 8, 127-150. MD Index C We varied the number of MODIS bands to check the sensitivity of the MD approach to phenophase timing. Contacts Eric Salas: eric.salas@sdstate.edu [http://globalmonitoring.sdstate.edu] Geoffrey Henebry: geoffrey.henebry@sdstate.edu [http://globalmonitoring.sdstate.edu] Mark Schwartz: mds@uwm.edu [https://pantherfile.uwm.edu/mds/www/] Equation of the MD: where: x1 = first reference wavelength point x2 = second reference wavelength point