D&B stands for DALEC and BETHY, the two models upon which D&B are based.
D&B builds on the strengths of these two component models,
in that it combines the dynamic simulation of the carbon pools and canopy phenology of DALEC with
the dynamic simulation of water pools, and the canopy model of photosynthesis and energy balance of BETHY.
D&B is presented in Knorr et al. 2024.
D&B can indeed be considered as one generic model, we’ve chosen the name to indicate the heritage of the two model components
It is not better than any other state-of-the-art ecosystem model but it has been developed with focus on data assimilation. For this reason we avoid undue complexity that does not improve the realism of the simulation, which allows D&B to run on the spatial resolutions of the footprint of EO data and at which the outputs of interest. D&B further comes with a set of observation operators, allowing D&B to simulate analogues of a a diverse array of observational data streams. The code of D&B and its observation operators is complemented with derivative code, so-called tangent and adjoint code, which allows its efficient operation within the variational TCCAS.
The spatiotemporal resolution of D&B is flexible.
As an example, a 13 year experiment run at a single site which is covered by two Plant Functional Types (PFTs) takes about 0.5 sec in forward mode and about 3 sec for the adjoint (derivative) on a hardware platform equipped with Intel® CoreTM i9-7940X processor (19.25M Cache, up to 4.30 GHz), 128GB RAM, and NVMe SSD using a Linux operating system. As another example, when running a 13 year experiment for a region composed of 432 grid cells with up to 6 PFTs the D&B model takes ~300 sec and its adjoint (derivative) ~1700 sec on the same hardware described above. As a further example, an assimilation run at one site with 2 PFTs over XX years with 19 iterations and 70 model evaluations takes about 160 sec on a notebook with Intel® CoreTM i7-8550U processor (base frequency 1.8GHz, maximal turbo frequency 4GHz, and 8MB cache), 16GB RAM, and NVMe SSD.
D&B needs several static driving fields providing, e.g., soil properties and vegetation cover, and some meteorological dynamic driving fields, such as precipitation, temperature and radiation. For details refer to the user manual.
Yes, it provides both carbon fluxes and pools as well as a range of further variables related to the energy and water cycles, for details we refer to the user manual.
For the units we refer to the user manual.
Yes, as long as the spatial extent of a grid cell/footprint exceeds the height of the vegetation, the 2 flux scheme provides a good approximation of the radiative transfer within the canopy soil system.
Yes, but we have not yet run it globally.
Yes. The model is prognostic, i.e. it integrates the carbon, water and energy balance in time. For a prognostic future experiment D&B requires (similar as to a diagnostic experiment over past times) meteorological dynamic driving fields (e.g., from a climate model prediction) over the simulation period as input.
Yes, for example, it includes a two flux scheme of the radiative transfer in the optical domain. For details refer to the following article Knorr et al. 2024.
See user manual for instructions.
For the installation of the model on a given computing platform a basic environment of standard software needs to be available, see user manual for details:
The example make include file makes use of the nf-config utility that comes
with the NetCDF Fortran library installation. In case this library is installed
into a custom location on your system [NETCDF-FORTRAN-INSTALLDIR] you will
likely need to update the macro NFCONFIG in mk.compile to
NFCONFIG = [NETCDF-FORTRAN-INSTALLDIR]/bin/nf-config
Also, in this case you will most likely need to extend the linker options
by the flag "-Wl,-rpath=[NETCDF-FORTRAN-INSTALLDIR]/lib64".
The linker options are defined by the macro LDFLAGS in the same file.
To run D&B in forward mode at least the following list of input files must be accessible to the model, where the directory input must be in the same directory as the D&B executable:
The use of D&B does not necessarily require a HPC platform, whereby the memory requirement naturally depends on the spatio-temporal setup of the application.
The core of D&B is written in Fortran90. The selection of a not quite up-to-date version of Fortran eases the generation of derivative codes (which are required for the assimilation system) of the model with software tools of Automatic Differentiation. In addition, the compilation of the model code requires a preparatory step implemented in python, which generates a module defining the spatio-temporal dimensions in dependence of the forcing data in use. It is certainly possible to interface D&B with other programming languages, at least as a wrapper on the Operating system level.
Yes, we have tested it on a number of hardware architectures. So far D&B has only been run under Unix/Linux type operating systems.
We have not (yet) implemented a GUI.
Yes, it is available here under the AGPL license.
Yes.
Yes.