Benchmarks for the Urban-PLUMBER model evaluation project Phase 1 (AU-Preston)

AU-Preston benchmarks for the Urban-PLUMBER project
These timeseries data are associated with the study:

Evaluation of 30 urban land surface models in the Urban-PLUMBER project: Phase 1 results
doi: https://doi.org/10.1002/qj.4589
These are empirical models of various complexity used as benchmarks to evaluate land surface models in the project. The benchmarking process follows the methods in the PLUMBER project (Best et al., 2015).
Benchmarks
Physically based

Manabe_1T: A simple 'slab and bucket' model (Fig. 1a) based on physical principles (i.e. conservation of energy, mass and momentum). The impervious (built) fraction is simulated using a one-tile slab scheme (Best, 2005). For the pervious fraction a simple representation allows precipitation to fill a store which overflows when full, and otherwise freely evaporate (Manabe, 1969). At each timestep, the impervious and pervious tile outputs are calculated and aggregated with a weighted mean.
Empirical (out-of-sample)

REG1-SWdown: Linear regression with one variable (SWdown, e.g. Fig. 1b) is used separately to predict SWup, LWup, Qh, Qle, and Qtau. At night predicted values are constant where SWdown = 0.
REG2-SWdown-Tair: Two-variable (SWdown and Tair) linear regression (e.g. Fig. 5c) provides some information at night and more generally for variables closely dependent on temperature (e.g. Lwup and Qh).
KM3-SWdown-Tair-RH: Following PLUMBER's conceptual arguments, three predictor variables (SWdown, Tair and RH data) are split into three (low, medium and high) to create 3^3=27 groups, for which independent regressions are trained. K-means clustering is used to determine the training data groups (e.g. Fig. 1d). To use this piecewise regression benchmark, at each time step the input data's proximity to one of the 27 cluster centroids is determined to select the regression to apply. This benchmark equates to PLUMBER's EMP3KM27 (Best et al., 2015), which based on common metrics outperformed all their participating land surface models when predicting sensible and latent heat fluxes across 20 sites.
Empirical (in-sample)

KM3-IS-SWdown-Tair-RH: This follows the KM3-SWdown-Tair-RH method, but trained with in-sample data only (i.e. AU-Preston). This should outperform an out-of-sample model because of the reuse of the forcing data, but performance is expected to degrade if applied to dissimilar conditions (i.e. another site).
KM4-IS-SWdown-Tair-RH-Wind: The k-means approach is applied again, but with an additional variable (wind speed), which increases the clusters to 81 (3^4) given the above rationale. Wind speed provides information that helps to predict turbulent heat and momentum fluxes.
Authors
Emperical benchmarks were developed by Mathew Lipson: https://orcid.org/0000-0001-5322-1796
The physical benchmark was developed by Martin Best: https://orcid.org/0000-0003-4468-876X
Co-authors for the associated manuscript are:
Mathew Lipson, Sue Grimmond, Martin Best, Gab Abramowitz, Andrew Coutts, Nigel Tapper, Jong-Jin Baik, Meiring Beyers, Lewis Blunn, Souhail Boussetta, Elie Bou-Zeid, Martin G. De Kauwe, Cécile de Munck, Matthias Demuzere, Simone Fatichi, Krzysztof Fortuniak, Beom-Soon Han, Maggie Hendry, Yukihiro Kikegawa, Hiroaki Kondo, Doo-Il Lee, Sang-Hyun Lee, Aude Lemonsu, Tiago Machado, Gabriele Manoli, Alberto Martilli, Valéry Masson, Joe McNorton, Naika Meili, David Meyer, Kerry A. Nice, Keith W. Oleson, Seung-Bu Park, Michael Roth, Robert Schoetter, Andres Simon, Gert-Jan Steeneveld, Ting Sun, Yuya Takane, Marcus Thatcher, Aristofanis Tsiringakis, Mikhail Varentsov, Chenghao Wang, Zhi-Hua Wang
References
Observational data
Lipson, M., Grimmond, S., Best, M., Chow, W., Christen, A., Chrysoulakis, N., Coutts, A., Crawford, B., Earl, S., Evans, J., Fortuniak, K., Heusinkveld, B. G., Hong, J.-W., Hong, J., Järvi, L., Jo, S., Kim, Y.-H., Kotthaus, S., Lee, K., Masson, V., McFadden, J. P., Michels, O., Pawlak, W., Roth, M., Sugawara, H., Tapper, N., Velasco, E., and Ward, H. C.: Data for "Harmonized gap-filled dataset from 20 urban flux tower sites" for the Urban-PLUMBER project, https://doi.org/10.5281/zenodo.7104984, 2022.
Benchmark references
Best, M. J., Abramowitz, G., Johnson, H. R., Pitman, A. J., Balsamo, G., Boone, A., Cuntz, M., Decharme, B., Dirmeyer, P. A., Dong, J., Ek, M., Guo, Z., Haverd, V., Hurk, B. J. J. van den, Nearing, G. S., Pak, B., Peters-Lidard, C., Santanello, J. A., Stevens, L., and Vuichard, N.: The Plumbing of Land Surface Models: Benchmarking Model Performance, Journal of Hydrometeorology, 16, 1425–1442, https://doi.org/10.1175/JHM-D-14-0158.1, 2015.
Best, M. J.: Representing urban areas within operational numerical weather prediction models, Boundary-Layer Meteorol, 114, 91–109, https://doi.org/10.1007/s10546-004-4834-5, 2005.
Manabe, S.: CLIMATE AND THE OCEAN CIRCULATION: I. THE ATMOSPHERIC CIRCULATION AND THE HYDROLOGY OF THE EARTH'S SURFACE, Monthly Weather Review, 97, 739–774, https://doi.org/10.1175/1520-0493(1969)097<0739:CATOC>2.3.CO;2, 1969.