Remote sensing data: surface radiometric temperature and vegetation indexes

Two satellite sensors with different spatial and temporal resolutions were used as a source of surface radiometric temperature (TRAD) values: MODIS (Moderate Resolution Imaging Spectroradiometer)

and Landsat 7ETM+ and 8OLI. The first sensor has daily coverage, with 250 meters and 1 km spatial resolution for the visible and the thermal bands respectively. The thermal product MYD11A1, which supplies TRAD with the atmospheric and emissivity effects corrected was used. For Las Majadas, 40 days

in 2008 and 2011 were analyzed and for Santa Clotilde site 65 days between 2012 and 2014. Eleven cloudless Landsat 7ETM + & 8OLI images (path 201 and row 33) coincident with the study period and the series of data from the Santa Clotilde ECT without gaps were also acquired and processed (DOY 124 for 2012 and 110, 182, 190, 198, 206, 214, 310, 318, 342, 350 for 2013). The images were already geo-referenced, with spatial resolutions of 30 m in the shortwave bands and 60/100 m in the thermal band, depending on whether the satellite was 7ETM+ or 8OLI. Atmospheric and surface emissivity effects were corrected by an atmospheric radiative transfer model MODTRAN4 (Berk et al., 1998). The lack of available atmospheric data required for an in-situ atmospheric characterization led us to use MODIS satellite-derived atmospheric profiles of air temperature and humidity (MOD07 product) which, according to Jimenez-Muñoz et al. (2010), provides an RMSE of 0.6 K in radiometric temperature estimates compared to locally measured profiles. The followed procedure is described in detail in Annex I.

The dates used in the analyses have been selected from the data series from both ECT, discarding days according to the following criteria: (a) periods with gaps due to instrument failure, (b) lacking thermal information in the ECT pixel of the image due to clouds, (c) unsuitable footprint. The selection was made in an attempt to capture the seasonal variability of dehesa; the dates are shown in Table 3.1. However, in Santa Clotilde the first data series collected by the ECT had several gaps, mainly due to the set-up and different tests performed on the tower and the instruments, and it was not always possible to capture an image every month. Nevertheless, the data collected was distributed as homogeneously as possible, taking into account these imitations.

Table 3.1: MODIS selected dates for Santa Clotilde and Las Majadas study sites.

Year Month Day Year Month Day Year Month Day Year Month Day

Santa Clotilde site Santa Clotilde site Santa Clotilde site Las Majadas site

2012 6 21 2013 7 31 2014 6 4 2008 8 3 2012 7 21 2013 8 1 2014 6 9 2008 8 13 2012 7 23 2013 8 2 2014 6 26 2008 8 27 2012 7 28 2013 8 3 2014 6 29 2008 9 7 2012 7 29 2013 8 13 2014 7 9 2008 11 1 2013 3 15 2013 8 25 2014 7 17 2008 12 2

2013 6 13 2013 8 27 Las Majadas site 2008 12 12

2013 6 26 2013 8 28 2008 1 4 2008 12 22 2013 7 5 2013 8 30 2008 1 20 2008 12 25 2013 7 6 2013 9 1 2008 1 30 2011 3 22 2013 7 7 2013 9 3 2008 3 18 2011 3 31 2013 7 8 2013 9 6 2008 3 21 2011 4 7 2013 7 12 2013 9 8 2008 3 23 2011 4 10 2013 7 13 2013 9 12 2008 4 6 2011 5 20 2013 7 15 2014 1 7 2008 4 27 2011 6 29 2013 7 16 2014 1 20 2008 4 29 2011 7 29 2013 7 17 2014 2 2 2008 5 3 2011 8 8 2013 7 20 2014 3 4 2008 5 17 2011 10 17 2013 7 21 2014 3 6 2008 6 6 2011 11 6 2013 7 22 2014 3 21 2008 6 12 2011 11 17 2013 7 24 2014 4 7 2008 6 25 2011 11 26 2013 7 26 2014 4 30 2008 7 6 2011 12 3 2013 7 29 2014 5 10 2008 7 16 2011 12 6 2013 7 30 2014 5 25 2008 7 31 2011 12 19

For the estimation of NDVI and the derivation of the leaf area index and the fractional cover, red and near infrared (NIR) bands were used, as well as blue band for the green fraction (Table 3.2), following the same procedure as described in Chapter 2, section 2.2.2.1.

Table 3.2: MODIS and Landsat-7 ETM+ and -8 OLI wavelengths intervals for Blue, Red, NIR and TIR bands

MODIS Landsat 7TM Landsat 8TM

Blue B3 (0.459-0.479 μm) B1 (0.441-0.514 μm) B1 (0.452-0.512 μm) Red B1 (0.620 – 0.670 μm) B3 (0.631-0.692 μm) B4 (0.636-0.673 μm) NIR B2 (0.841 – 0.876 μm) B4 (0.772-0.898 μm) B5 (0.851-0.879 μm) TIR B31 (10.78 – 11.28 μm) B32 (11.77 – 12.27 μm) B6 (10.31 – 12.36 μm) B10 (10.60 – 11.19 μm) B11 (11.50 – 12.51 μm)

When the MODIS satellite was used, the NDVI for the first period (2012 and part of 2013) over Santa Clotilde was derived from the separate red and NIR bands daily reflectance data. For the

following periods and for Las Majadas, after testing, the behavior of the MODIS product MOD13Q1 with 250 m of resolution and fifteen days of frequency was regarded as accurate enough for our requirements (Figure 3.2). This product select an NDVI representative of the 15- day period, as the average of the two days with maximum NDVI and higher-quality information. For that reason, MOD13Q1 data are always higher than daily derived NDVI (Fig. 3.2). However, the similarity of the pairs of data, with RMSD between the estimated NDVI and that provided by MODIS equal to 0.03, with a relative error of 6% and the continuous nature of changes in NDVI, led to the decision to use the MOD13Q1 product directly, facilitating the process.

Figure 3.2: Comparison between NDVI derived from reflectance MODIS product and NDVI from MOD13Q1 product.

Landsat-7 ETM+ suffered a technical problem on 31st _{May 2003, related to the scan-line corrector,}

since when it has been operating without this instrument functioning properly. The sensor images the surface in a “zig-zag” pattern, resulting in some areas not being scanned. These areas are approximately 22% of a Landsat-7 scene (Storey et al., 2005), with the effect being greater on the east and west sides, with no missing values over the central line. In this case, Santa Clotilde is on the east part of the image, with losses due to this effect that could be visible on the results. No gap-filling techniques have been used.

MODIS images for 2014 over the Iberian Peninsula, listed in Table 3.1 were selected, to apply TSEB over the Andalusian dehesa.

3.2.2 Derivation of the oak LAI and total ecosystem LAI from remote-sensing data

In order to isolate the effect of the tree layer from the understorey component and to study its variability in the course of the year, the local LAI of the trees was derived from MODIS data for both locations, taking into account that the LAI index thus derived integrated the clumping effect. This was done for the period when the herbaceous layer was dry, assuming that the reflectance registered by the sensor corresponded only to the oaks. The estimated oak LAI results from Santa Clotilde were compared with field measurements of LAI in order to study the accuracy of the estimation. Seven days were analyzed following this procedure (21st _{June, 23}rd _{July and 23}rd _August

2013, 4th _{June, 17}th _{June, 30}th _{June and 17}th _{July, 2014). During the period with an active grass}

layer, with local LAI field measurements of both oaks and grass, an “average ecosystem LAI” weighted by the surface occupied by each component was derived and then compared with the MODIS-estimated index. Dates available for the analysis were 15th _{and 21}st _{May, 8}th _{April, 20}th

September, 17th _{and 31}st _{October and 14}th _{of November in 2013, and in 2014, 25}th _{April and 19}th

May.