SOCFramework for soil organic carbon models with Luke weather database

Introduction

The purpose of this work with Luke weather database, Yasso07 and Yasso20 models is to propose an initial framework that could be expanded to include other soil organic carbon (SOC) models and perhaps other (European) weather database systems.

The implementation of the intial framework is done with mixed Python and R environment. Partially because limited time available and partially to utilize work done before in various projects.

Although the implementation consists of small pieces of python and R programs, and are naturally exposed to user via GitHub, one of the design goals was that no special programming skills are required by the user. Short introduction to R and how to call functions in R should be adequate.

NB: If you are only interested the Yasso family of SOC models see the official YASSOmodel repository for Yasso Fortran, R interface and graphical user interface implementations.

Installation of necessary software

First, the precursory software consists naturally of Python and R interpreters. This framework has been constructed with Python 3.9 and RSudio 2022.07.0/R version 4.1.2 but Python version above 3.6 and earlier or later R versions should work as well. This has not been tested though. Second, Fortan compiler is needed to compile Yasso models. The free Fortran compiler is gfortran. Third, git is needed to access GitHub.

In Luke this prerequisite software should be available from Software Center for Windows 10. Otherwise ask Luke support how to install missing programs on Windows 10.

The rest of the software can be download from GitHub. These include Yasso model implementations and LukeWeather consisting the implementation of simple queries to Luke weather database.

Once R, Python, Fortran and Git are installed take the step-by-step instructions how to build this initial framework into working order. The minor issue may be that you may need to spend some time in terminal command line of the operating system.

Download Yasso models, LukeWeather and SOCFramework

Yasso07 is located in YASSO07 GitHub repository. You will notice Code menu where you can download the zip file. To use git write the following command in terminal:

git clone https://github.com/YASSOmodel/YASSO07

Yasso20 is located in YASSO20. In the same way download zip file or use git:

git clone https://github.com/YASSOmodel/Yasso20

To download LukeWeather with git:

git clone https://github.com/LukeEcomod/LukeWeather.git

LukeWeather GitHub repository is private. Ask for membership (Samuli Launiainen, Luke) in LukeEcomod GitHub group to gain access.

Finally, download SOCFramework:

git clone https://github.com/LukeEcomod/SOCFramework

R and Python interoperability

Install in R/RStudio the reticulate R package for Python interoperability.

Compile Yasso models

Compile Yasso07 and Yasso20 models as shared libraries. Their Fortran implementations are in y07_subroutine.f90 and yassofortran.f90 respectively located in their project model directories. First, both Fortran implementations must be edited:

• Replace every REAL with DOUBLE PRECISION (both Yasso07 and Yasso20 models).
• Replace type cast to REAL with DBLE (i.e. double precision) in Yasso20 model.
• Yasso20 model is inside Fortran 90 module declaration. Remove the module declaration.

The Yasso Fortran implementations will be called via .Fortran foreign function interface in R. .Fortranis admittedly old and meant for Fortran 77. Using this interface double precision floating point variables must be used both in R and in Fortran. .Fortran does not support Fortran 90/95 module declarations either but the usage is simple and straightfoward. With more modern .C or .Call function interfaces one ends up writing R function calling C function calling C function calling Fortran function.

Ask for help if in doubt how to do these edits. To compile Yasso07 type in terminal:

gfortran -shared -fPIC -O2 -o  y07_subroutine.so y07_subroutine.f90 

To compile Yasso20:

gfortran -shared -fPIC -O2 -o   yassofortran.so yassofortran.f90

Copy y07_subroutine.so and yassofortran.so to SOCFramework.

Install LukeWeather

First, to create Python virtual environment type in terminal¹:

python -m venv venv/lukeweather

See LukeWeather and the README_setup file for details to install LukeWeather to lukeweather virtual environment. In short activate first the virtual enviroment. On Linux or Mac type:

source venv/lukeweather/bin/activate
(lukeweather)

On Windows 10 type (PowerShell):

venv/lukeweather/Scripts/activate
(lukeweather)

Note the lukeweather prompt appearing. Then install setuptools and wheel packages:

(lukeweather) pip install setuptool wheel

In LukeWeather directory you should see setup.py file. Install LukeWeather as wheel package into your lukeweather virtual environment (uninstall previous installation):

(lukeweather) python setup.py  bdist_wheel 
(lukeweather) pip uninstall LukeWeather
(luekweather) pip install  dist/LukeWeather-1.0-py3-none-any.whl

Test the installation:

(lukeweather) python
>>>import LukeWeather.fmidata
>>>import LukeWeather.checkfmidata
>>>quit()

You should not get any errors. Type deactivate to quit virtual environment. If you are interested in implementation details LukeWeather is documented for Doxygen. Generate the final document by running doxygen with Doxyfile input that parses and formats Python file contents.

Run the Yasso models with SOCFramework

Start R/RStudio and change directory to SOCFramework.

You should have the same Python version for R/RStudio and terminal command line. If you have only one Python installation this should be the case. Otherwise errors may turn up using reticulate package. If this happens the workaround is to from the terminal (PowerShell) command line first activate the lukeweather virtual environment and in that virtual envronment from the command line start R/RStudio.

The two demonstrations for Yasso07 and Yasso20 are called yasso07.r and yasso20.r respectively.

NB: In the beginning of both files the line:

use_virtualenv('~/venv/lukeweather')

activates lukeweather python virtual environment. The argument string for path must be edited to point to the installation location.

There also are two sample files awenh.csv and yassoinit.csv for litter infall and initial values. They follow Excel input for Yasso server in Luke.

Weather

To intialize litter stock variables in R/RStudio source both R files. To retrieve weather data for Yasso07 in R/RSudio type:

> source('yasso07.r') 
> y07weather <- yasso07.weather(E,N,'user_name','password','2016-01-01','2022-12-31')

You will need read permission for weather database. Contact Arto Aalto at Luke. E and N contain East and North coordinates for Pudasjärvi that can be found in Luke weather database. To be precise: by default the closest point for E and N is found in the weather database assuming Euclidian space. This is a bit slower than defining E and N as exact grid point by setting the last argument exact_location=TRUE (see yasso07.r for details). You should see y07weather to contain the following data frame:

> y07weather

AnnualMeanTemperature	AnnualPrecipitation	AnnualAmplitude	Year
5.055495	303.4	11.99055	2016
2.216164	556.8	12.14222	2017
3.051233	429.9	17.06365	2018
2.173699	554.8	14.35161	2019
4.001366	696.7	11.68103	2020
1.950685	587.5	16.50726	2021
3.177839	559.8	13.42097	2022

For Yasso20 weather data type:

> source('yasso20.r')
> y20weather <- yasso20.weather(E,N,'user_name','password','2016-01-01','2022-12-31')

Identically to yasso07.weather the last argument exact_location=TRUE assumes the given coordinate point is some grid point in the weather database. You should see the following data frame in y20weather:

> y20weather

MonthlyMeanTemperature	AnnualPrecipitation	Year
[-7.896774193548388, -8.739285714285716, -4.12...	556.8	2017
[-8.087096774193547, -13.782142857142857, -9.6...	429.9	2018
[-14.14516129032258, -8.164285714285713, -5.17...	554.8	2019
[-4.419354838709677, -6.462068965517242, -3.58...	696.7	2020
[-11.312903225806453, -14.450000000000001, -4....	587.5	2021
[-9.790322580645162, -8.082142857142859, -3.79...	559.8	2022

Yasso20 uses the monthly temperature means instead annual temperature means with temperature amplitude in Yasso07. Because year 2016 did not have full 12 months of data it was dropped from the results.

Soil organic carbon simulation

To run Yasso07 soil carbon model with data for this demonstration type:

> source('yasso07.r')
> y07soc <- yasso07.soc(yasso07.skandinavian,y07weather,litter.start.stock,df.litter)

The result for SOC dynamics is in y07soc data frame:

> y07soc

Year	A	W	E	N	H	LitterSize
Start	8.235963	0.8931869	0.929875	9.127769	44.40272	0
2016	7.046699	0.7649162	0.8055626	8.412239	44.40604	0
2017	6.107139	0.6417022	0.7020420	7.573297	44.40797	0
2018	5.523495	0.5845530	0.6137222	6.849109	44.40844	0
2019	4.840161	0.4932341	0.5297692	6.156678	44.40765	0
2020	4.284145	0.4471506	0.4989971	5.373577	44.40532	0
2021	4.142418	0.4258699	0.4270894	4.947869	44.40256	0
2022	3.504887	0.3695343	0.3722988	4.487991	44.39891	0

 								   	   											   |

To run Yasso20 carbon model with the same initial stock and annual litter input:

> source('yasso20.r')
> y20soc <- yasso20.soc(yasso20.parameters,y20weather,litter.start.stock,df.litter)

The result for SOC dynamics is in y20soc data frame:

> y20soc

Year	A	W	E	N	H	LitterSize
Start	8.235963	0.8931869	0.9298725	9.127769	44.40272	0
2017	7.261067	0.7582581	0.8375732	8.973893	44.36322	0
2018	6.404966	0.6709506	0.7329615	8.837831	44.31272	0
2019	5.738415	0.5894140	0.6450267	8.533247	44.26371	0
2020	5.088575	0.5281767	0.5983642	8.132536	44.21187	0
2021	4.870214	0.5015199	0.5091259	7.962147	44.15661	0
2022	4.277270	0.4437070	0.4416660	7.678748	44.09608	0

For implementation details see comments in yasso7.r and yasso20.r and also LukeWeather GitHub repository.

Futher work

For this framework grid_day table in Luke weather database is used. It is updated daily giving contemporary daily weather data from 2016 to present-day (given day or two delays with database updates). The table grid10_day should be straightforward to include within a day or two. It is static and not updated giving daily weather data from 1961 to 2018.

Include other Soil organic carbon models. Yours truly is not so familiar with them so it is difficult to say how long that would take.

Include other (European) weather databases. Again, yours truly is not familiar with them so it is difficult to say how long that would take.

Footnotes

1. Yours truly has in home directory venv subdirectory under which all python virtual environments are created. ↩