\begin{table}%t1 \caption{\label{table:polaris}SPIRE observations of the Polaris flare.} \par %\centerline {\small \begin{tabular}{cccccccc} \hline\hline $\lambda$ & Pixel size & {\it FWHM} & $G$ & $S_0$ & Noise & $\gamma$ & $P_{0}$ \\ ($\mu$m) & (arcsec) & (arcsec) & & (MJy sr$^{-1}$) & (MJy sr$^{-1}$) & & (Jy$^2$ sr$^{-1}$) \\ \hline &&&&&&& \\[-3mm] 250 & 6 & 18.1 & $3.3\pm0.5$ (0.01) & $20.1\pm3.1$ (0.1) & 1.26 & $-2.65\pm0.10$ & $5\pm 2\times 10^3$\\ 350 & 10 & 25.2 & $1.7\pm0.3$ (0.008) & $10.1\pm1.6$ (0.05) & 0.55 & $-2.69\pm 0.13$ & 4$\pm 2 \times 10^3$\\ 500 & 14 & 36.9 & $0.7\pm0.1$ (0.003) & $4.3\pm0.7$ (0.02) & 0.34 & $-2.62\pm 0.17$ & $1\pm 1 \times 10^3$\\ \hline \end{tabular}} \tablefoot{Columns~4 and~5: gain and offset coefficients of the SPIRE-IRIS~100~$\mu$m correlations. The uncertainty on $G$ represents the rms of the ratio $S(\lambda)/S(100~\mu{\rm m})$ once the offset $S_0$ is removed from the SPIRE data. Similarly the uncertainty on $S_0$ is the rms of $S(\lambda) - G\times S(100~\mu{\rm m})$. These two uncertainties are correlated, but they give more realistic estimates compared to the statistical ones obtained with the linear regression fit (given in brackets). Column~6: noise level estimated directly on the power spectrum for $k>0.75k_{\rm max}$ ($3.75