Spatial validation of large‐scale land surface models against monthly land surface temperature patterns using innovative performance metrics

Abstract

Land surface models (LSMs) are a key tool to enhance process understanding and to provide predictions of the terrestrial hydrosphere and its atmospheric coupling. Distributed LSMs predict hydrological states and fluxes, such as land surface temperature (LST) or actual evapotranspiration (aET), at each grid cell. LST observations are widely available through satellite remote sensing platforms that enable comprehensive spatial validations of LSMs. In spite of the great availability of LST data, most validation studies rely on simple cell to cell comparisons and thus do not regard true spatial pattern information. The core novelty of this study is the development and application of two innovative spatial performance metrics, namely, empirical orthogonal function (EOF) and connectivity analyses, to validate predicted LST patterns by three LSMs (Mosaic, Noah, Variable Infiltration Capacity (VIC)) over the contiguous United States. The LST validation data set is derived from global High‐Resolution Infrared Radiometric Sounder retrievals for a 30 year period. The metrics are bias insensitive, which is an important feature in order to truly validate spatial patterns. The EOF analysis evaluates the spatial variability and pattern seasonality and attests better performance to VIC in the warm months and to Mosaic and Noah in the cold months. Further, more than 75% of the LST variability can be captured by a single pattern that is strongly correlated to air temperature. The connectivity analysis assesses the homogeneity and smoothness of patterns. The LSMs are most reliable at predicting cold LST patterns in the warm months and vice versa. Lastly, the coupling between aET and LST is investigated at flux tower sites and compared against LSMs to explain the identified LST shortcomings.

Calibration of a parsimonious distributed ecohydrological daily model in a data-scarce basin by exclusively using the spatio-temporal variation of NDVI

“…Discharge represents an integrated catchment response, and hence provides only limited insight on the lumped behavior of a catchment (Stisen et al, 2011;Koch et al, 2016a;Michaud and Sorooshian, 1994;Reed et al, 2004;Smith et al, 2013). In that sense, Conradt et al (2013) provided several examples for large simulation errors within the model domain and they mentioned, among others, the outcomes given by Feyen et al (2008), Merz et al (2009) and Smith et al (2012).…”

On the importance of multiple-component evaluation of spatial patterns for optimization of earth system models – A case study using mHM v5.6 at catchment scale

“…The connectivity analysis is applied individually on cells that exceed a given threshold and those that fall below, which is referred to as low and high 25 phase, respectively. Following Koch et al (2016b), the root-mean-square-error between the connectivity at all percentiles of the observed ( ( ) ) and the simulated ( ( ) ) pattern denotes a tangible pattern similarity metric and can be calculated as: The average RMSE score of the low and the high phase is employed as the pattern similarity score for the connectivity analysis and is referred to as connectivity throughout the manuscript.…”

“…Outside the hydrogeology community, connectivity analyses have also been conducted to describe spatial patterns of soil moisture (Grayson et al, 2002;Western et al, 2001) or land-surface temperature (Koch et al, 2016b). Following the classification of Renard and Allard (2013), the connectivity analysis of a continuous variable is conducted via three steps: (1) a series of threshold percentiles decomposes the domain 15 into a series of binary maps, (2) the binary maps undergo a cluster analysis that identifies spatially connected clusters and (3) the transition from many disconnected clusters to a single connected clusters can be quantified by principles of percolation theory (Hovadik and Larue, 2007).…”

“…In this study, we rigorously test SPAEF and compare it with two additional spatial performance metrics; namely fractions skill score (Roberts and Lean, 2008) and connectivity analysis (Koch et al, 2016b). All three metrics are applied in a spatial pattern oriented calibration of a catchment model using the multiscale Hydrologic Model (mHM: Samaniego et al, 2010).…”

“…At regional scale, Mendiguren et al (2017) applied a spatial pattern oriented model evaluation based on remote sensing estimates of evapotranspiration to diagnose shortcomings of the national hydrological model of Denmark. At large scale, Koch et al (2016b) utilized land surface temperature retrievals to evaluate large scale land surface models across the continental U.S..…”