National Aeronautics and Space Administration

Validation at stage 2 has been achieved for the MODIS Collection 6 (C6) Land Surface Temperature product (MOD11).

The MODIS MOD11 (Terra) and MYD11 (Aqua) C6 products have been validated at Stage 2 via a series of field campaigns conducted in 2000-2007, and over more locations and time periods through radiance-based validation studies. Accuracy is better than 1K (0.5K in most cases) at all sites except arid regions, similar to the previous C5 LST product.

For the daily 1 km products, slightly higher errors may occur at large view angles and in semi-arid regions. Higher errors in some arid regions are apparent for these products generated by the split-window algorithm, due to effects of heavy dust aerosols and greater uncertainties in classification-based surface emissivities. For the 6km (MOD11B1) product generated by the day/night algorithm, these errors are also evident, due to the strong influence of incorporating the split-window algorithm. The error in land surface temperatures (LSTs) contaminated by clouds and heavy aerosols can be very large (4-11K or even larger). It is very difficult for the MODIS cloud-mask to discriminate all pixels affected by clouds and heavy aerosols from clear-sky pixels, particularly near cloud edges and/or with sub-pixel clouds. The LSTs that are severely contaminated by clouds and heavy aerosols are removed in Level-3 MODIS LST products using empirical constraints on temporal variations in clear-sky LSTs.

Three refinements were implemented in the C6 production code for daily LST:

• refined the original viewing angle dependent split-window algorithm by adding two sets of coefficients for retrieval of both the daytime and nighttime LSTs for bare soil pixels in the hot and warm bare soil zone (HAWBSZ), within latitude range from −38° to 49.5°
• adjusted the emissivity difference in bands 31 and 32 for bare soil pixels in the HAWBSZ
• incorporated the new refined split-window algorithm into the day/night algorithm for retrieving surface emissivity and temperature values at 6 km grids in the HAWBSZ

The algorithm was tested with C6 Aqua and Terra MODIS input data over a period of one year or longer for selected tiles. The resulting C6 level-2 LST products were evaluated at five bare soil sites in north Africa, where radiosonde observation data for atmospheric temperature and water vapor profiles are available. Validation tests were performed using 12 R-based validation data sets with the following results: the mean LST error was within ±0.6 K in 10 validation data sets, 0.8 K in one data set (for daytime LSTs at SVU site) and 1.9 K in another set (for daytime LSTs at Farafra site). The standard deviation of the C6 LST validation errors were less than 0.5 K in all 12 validation data sets, however the daytime LST validation error at the Farafra site was 0.7 K. Statistical and viewing angle dependence analyses of the retrieved surface emissivity values also indicate that the C6 level-3 LST products are much better than the previous collections of LST products.