Correction to “Exploring land surface temperature earthquake precursors: A focus on the Gujarat (India) earthquake of 2001”



This article corrects:

  1. Exploring land surface temperature earthquake precursors: A focus on the Gujarat (India) earthquake of 2001 Volume 38, Issue 15, Article first published online: 5 August 2011

[1] In the paper “Exploring land surface temperature earthquake precursors: A focus on the Gujarat (India) earthquake of 2001” by M. Blackett et al. (Geophysical Research Letters, 38, L15303, doi:10.1029/2011GL048282, 2011), there were several minor formatting errors in sections 4.3 and 4.4. The correct sections 4.3 and 4.4 are given here.

4.3. Method 2: Extended LST Differencing (Based on Multiple Years)

[2] Method 1 differences two years of data (a and b), so that the resulting measure (ΔLSTab) is as much influenced by b (the ‘baseline’ year) as by a (the year of interest). To mitigate this influence, we repeat the approach of Ouzounov and Freund [2004] using Region A data, but extend it by deriving a ‘climatological average’ LST to which 2001 could be independently compared. We then applied the same procedure to all six years, calculating the spatially averaged mean LST for each DOY (d) for each year (2001, 2002, …, 2006), Region A, giving LST_A(d)year. Then, for a given DOY, up to six values are available (2001, 2002, …, 2006); these were averaged to provide ‘climatological’ mean DOY values (equation image) which were subtracted from the respective daily (d) values (LST_A(d)year). This quantifies the LST difference between the date of interest and the corresponding six-year mean (ΔLSTA,year). We take ΔLSTA,year as our ‘anomaly’ measure, with up to 6 values for a given day of year.

4.4. Method 3: Robust Satellite Technique (RST)

[3] A detailed description of the RST as applied to remotely sensed LST data is given by Filizzola et al. [2004]. In summary, the technique functions by comparing the LST for a particular pixel and DOY (LSTr) to both the spatial mean of that particular scene (LST_A(d)year and LST_B(d)year, Regions A and B, respectively) and to the temporal mean (over multiple years considered) of LST for that particular pixel and DOY (equation imager). This is normalized by the standard deviation (again, over multiple years) of the LST values for that particular pixel and DOY (σ[LSTr]). The aim is to provide a method of isolating pixels whose LST signal appears thermally anomalous when compared with the longer-term local spatial average [Tramutoli, 2007].

[4] Application of the RST results in the derivation of an index value (RI) for each pixel (here given for Region A): RI = {(LSTrLST_A(d)year) − equation imager}/σ[LSTr], where RI represents the LST departure from the spatio-temporal historical ‘average’, weighted by its historical variability [Genzano et al., 2009]. This index value is derived from the more general Absolutely Local Index of Change of the Environment (ALICE) of Tramutoli [1998]. When applied to seismic monitoring it is often referred to as the Robust Estimator of TIR Anomalies (RETIRA) [Tramutoli et al., 2005; Aliano et al., 2008a]. For a specific year, the RETIRA index value (RI) for a given pixel and day can be interpreted as the number of standard deviations its LST (LSTr) is from that pixel's mean (equation imager) for that DOY, over all years considered, adjusted for each scene's spatial mean.

[5] Aliano et al. [2008a] suggested that the RST can, in some cases, be impacted by the presence of cloud-related data ‘gaps’. We explored and confirmed this using a set of LST simulations (see auxiliary material).1 Despite potential bias in the RETIRA index caused by cloud-cover variations or other data gaps, use of the RST has continued in seismic thermal precursor studies [e.g., Aliano et al., 2008b; Genzano et al., 2009; Pergola et al., 2010].

[6] We applied the RST to the six-year (2001–2006) MODIS LST dataset of Region A, so as to further examine the data at the scale used by Ouzounov and Freund [2004]. We then, as have other studies [e.g., Qiang et al., 1997; Choudhury et al., 2006; Genzano et al., 2007], applied the RST to a much larger region (Region B). In these applications the six-year mean and standard deviation for each MOD11A1 pixel and DOY (equation imager and σ[LSTr], respectively), were calculated using a 15-day moving window of LST data, centered on the DOY in question. This ensured that even during persistent cloud cover or other data gaps, a significant number of observations (up to 15 per DOY for each of the six years, or 90 values) contributed to calculating equation imager and σ[LSTr] for each pixel.

[7] Here, we take the number of pixels (NA and NB) exceeding a selected RI threshold in Regions A and B, respectively, as a measure of the degree to which the MOD11A1 data of a particular date contains LST ‘anomalies’; NA and NB are expressed as a percentage of the total number of useable land pixels within the scene (Percentage NA and Percentage NB, respectively). In previous studies, RI values have been classified as ‘anomalous’ using various thresholds, for example: >2.0, >2.5, >3.0 and >3.5 [Tramutoli et al., 2005]; ≥2.0 and ≥3.0 [Genzano et al., 2007]; and ≥2.0, ≥2.5 and ≥3.0 [Pergola et al., 2010]. We found that over all DOY considered (2001–2006), the percentage of ‘anomalous’ pixels for different thresholds was for Region A [B]: RI ≥ 2.0 (2.27% [1.44%]), RI ≥ 2.5 (0.62% [0.29%]), and RI ≥ 3.0 (0.13% [0.06%]). We use RI ≥ 2.5 to represent ‘anomalous’ pixels.