Validation of Simulated MM5-Coupled PROMET Land Surface Temperatures Using MSG/SEVIRI Data
Putzenlechner, B.; Zabel, F.; Muerth, M.; Mauser, W.
LMU Munich, GERMANY
Permanent progress in land surface process modeling has improved both spatial and temporal resolution and process representation. In particular, coupling of atmospheric models with detailed land surface models has improved detailed understanding of land-surface-atmosphere feedbacks. However, high resolution model output requires adequate validation techniques, which leads to an increasing need for independent external information often of both high temporal and spatial resolution. A crucial variable in land surface mass exchange and energy balance is land surface temperature (LST), as it controls latent and sensible heat flux. LST can be derived from thermal remote sensing data, thereby offering the advantage of directly validating spatially distributed model outputs, providing that the satellite meets the spatial and temporal resolution demands.
Aim of this work was to examine whether high resolution coupled land-surface-atmosphere process simulation is capable of correctly simulating diurnal cycles of LST. We used time series of spatial LST distributions derived from MSG/SEVIRI for validation because of their high temporal and spatial resolution. We use SEVIRI time series of three cloudless days of different seasons (Fig.1) to be able to compare the influence of different plant development stages on simulated LST. LST was retrieved using a split window approach (Niclós et al., 2011) and compared with hourly modeled LST output of the physically based hydrology and plant growth model PROMET (Mauser & Bach, 2009). PROMET uses an iterative approach to close the energy and mass balance of the land surface, which results in a modeled equilibrium LST for each time step and pixel. In this study, PROMET was driven by downscaled meteorological data (Zabel et al., 2012), provided by the MM5 regional climate model which was driven by ERA-Interim reanalysis data for the year 2005. We chose Southern Bavaria/Germany as a study region, which is characterized by high variability of land use types. To consider the effect of different spatial resolution of MSG/SEVIRI data (app. 4x6 km) and the simulated PROMET data (1x1 km), the satellite data were resampled to the higher resolution of the model. A heterogeneity index (HI) was further developed, representing a measure for land use variability concerning differences of scales of data sources. During the validation we considered differences in LST as a function of HI, date, time of day and land use type.
The comparison of measured SEVIRI-LST and simulated PROMET-LST show good overall agreement for all three seasons. Our results also indicate that land use types, dates and time of day are in close relation to differences between modeled and retrieved LST, whereas the HI exerts a minor influence (Fig.2). Furthermore, the evaluation of diurnal cycles revealed that the model is capable of modeling LST for the whole study region with a coefficient of determination ranging up to 0.99, depending on date and respective land use type (Fig.3). Non-aggregated, pixel-based analysis indicated differences that temporarily exceeded the range of error of LST retrieval of ±1.5 K (Fig.4). We discuss different factors affecting the model performance that are in close relation to the model constellation such as meteorological bias as well as accuracy of land use type assignment and seasonal vegetation dynamics.
The study illustrates both the potential of current physically based models to describe dynamics of LST as a key variable in the energy budget as well as the validation potentials of SEVIRI data. It makes evident that area distributed evaluation of model outputs by using remote sensing data can meet the increased demand for adequate validation methods that arises with more explicit process representation, especially due to model coupling.
Figure 1: Example for MSG/SEVIRI derived LST on August 30, 2005 at 11 am and localization of the study area.
Figure 2: Difference of MSG/SEVIRI derived LST and LST modeled by PROMET in relation to HI for all dates and all pixels in the study area; HI of 0 means that all surrounding pixels within 3x3 km2 (representing spatial resolution of MSG/SEVIRI) are classified as the same land use types, whereas an HI of 1 indicates that all surrounding pixels have different land use classes than the reference pixel; gray area: range of error of LST retrieval (±1.5 K).
Figure 3: Evaluation of diurnal cycles for the whole study region for different dates (columns) and land use types (rows), here presented for agricultural land and grassland.
Figure 4: Mean difference of MSG/SEVIRI-LST and PROMET-LST on August 30, 2005.
Mauser, W., Bach, H. (2009): PROMET – Large scale distributed hydrological modeling to study the impact of climate change on the water flows of mountain watersheds, J. Hydrol., 376, pp. 362–377, doi:10.1016/j.jhydrol.2009.07.046.
Niclós, R., Galve, J. M., Valiente, J. A., Estrela, M. J., Coll, C. (2011): Accuracy assessment of land surface temperature retrievals from MSG2-SEVIRI data. Remote Sensing of Environment, 115, pp. 2126-2140, doi:10.1016/j.rse.2011.04.017.
Zabel, F., Mauser, W., Marke, T., Pfeiffer, A., Zängl, G., Wastl, C. (2012): Inter-comparison of two land-surface models applied at different scales and their feedbacks while coupled with a regional climate model. Hydrol. Earth Syst. Sci., 16, pp. 1017-1031, doi:10.5194/hess-16-101.