Page 19 - Transitioning Turfgrass
P. 19

th
                                                                           6  ETS Field Days



























          Figure 2 RGB image derived from Specim IQ hyperspectral data (a) and classification raster obtained with the
          hyperspectral analysis (b) of dataset collected in the plot with Achillea millefolium, Trifolium repens, and Festuca rubra.



          features (Sakowska et al., 2019). The dark pixels   turned  out  to  be  promising  in  the  classification
          that typically are due to the canopy shadows and   of plant species in turfgrasses. However, further
          can negatively affect the classification accuracy   analyses  are  needed  to  define  the  appropriate
          were excluded from the datasets. We considered   data collection method and to improve the classi-
          as dark pixels the pixels with a mean reflectance   fier performance.
          value lower than 0,01.
          We report here the results of hyperspectral analy-
          sis performed on two images (Fig. 1 and Fig. 2) as   References
          an example. Figure 1 (page 14) reports the RGB
          image and the raster resulting from the analy-  Aneece I.P., Epstein H., Lerdau M., 2017. Correlating species
          sis of the dataset collected in plot with T. repens   and spectral diversities using hyperspectral remote sensing in
                                                  early-successional fields. Ecology Evolution 7: 3475–3488.
          and F. rubra. We found that 12,1% were dark pix-
          el, while F. rubra reached 76,4% and T. repens   Bailey  K.L.,  Falk  S.,  Derby  J.A.,  Melzer  M.,  Boland  G.J.,
          11,5%. From the analysis of the dataset collect-  2013. The effect of fertilizers on the efficacy of the bioher-
                                                  bicide, Phoma macrostoma, to control dandelions in turf-
          ed in the plot with A. millefolium, T. repens, and   grass. Biological control. 65(1): 147-151.
          F. rubra (Fig. 2) we found that 7,6% were dark   Brede A.D., Duich J.H., 1984a. Initial mowing of Kentucky
          pixels (canopy shadows), while F. rubra reached   bluegrass-perennial ryegrass seed- ling turf mixtures. Agron-
          56,2%, T. repens 4,3%, and A. millefolium 31,9%.   omy Journal 76: 711–714.
          Comparing RGB images (Fig. 1a and 2a) and ras-  Brede A.D., Duich J.H., 1984b. Establishment characteristics
          ters (Fig. 1b and 2b), it is possible to observe that   of Kentucky bluegrass—Perennial ryegrass turf mixtures as
          the hyperspectral system can be used to classify   affected by seeding rate and ratio.  Agronomy Journal 76:
          plant species in a close mowing sward. However,   875–879.
          with this preliminary analysis, we found that dark   Cho M.A., Debba P., Mathieu R., Naidoo L., van Aardt J.,
          pixels represent a great percentage of the whole   Asner G.P., 2010. Improving discrimination of savanna tree
          image and this could strongly affect the accuracy   species through a multiple-endmember spectral angle map-
                                                  per approach: Canopy-level analysis.  Geoscience and Re-
          in the estimation of species cover. The technique   mote Sensing 48(11): 4133-4142.

                                                                                      15
   14   15   16   17   18   19   20   21   22   23   24