A dynamic water vapor correction method for the retrieval of land surface temperatures from the advanced very high resolution radiometer

Abstract

We present a method which permits retrievals of land surface temperatures (LSTs) from AVHRR (advanced very high resolution radiometer) radiances sensed through atmospheres which may contain a large and strongly varying water vapor content. This new method is an extension of the dynamic water vapor (DWV) algorithm which was designed to retrieve sea surface temperatures. The generalization to LST retrievals recognizes that in general, land emissivities are unknown, may be spectrally dependent, and are less than unity. Because the LST retrieval problem is inherently underconstrained (there are more unknowns than radiative transfer equations), some knowledge of surface emissivity is required in order to establish upper and lower bounds on surface temperature. We demonstrate our method by comparing DWV‐LST retrievals with point surface measurements made by a cluster of eight infrared thermometers (IRTs) deployed over a grasslands prairie site in eastern Kansas in July and August 1989; this deployment was part of the First International Satellite Land Surface Climatology Project (ISLSCP) Field Experiment (FIFE). We find that several of the AVHRR images supplied on FIFE CD‐ROM contain navigation errors of ∼30 km, consistent with a misidentification of the Tuttle Reservoir ground control point. After correcting the navigation, we identified the IRT pixels and computed the bias and rms errors for a DWV‐IRT comparison. For night passes we obtained agreement to.+0.39±1.11 K, while for day passes the comparison yielded +4.09±3.10 K. The large daytime bias is probably the result of the IRT readings not being representative of the ∼1 km2‐scale areas sensed by AVHRR (the IRT views vegetation; the AVHRR field of view includes warmer, less well vegetated surfaces). Our results show that while a fixed‐coefficient, global split‐window algorithm does not perform well in the relatively moist FIFE atmosphere, it is quite feasible to use the DWV‐LST to derive a local split‐window algorithm whose coefficient is tuned on a per‐pass basis.