TUTORIAL-07: GLUE Analysis Laura Dobor, Péter Ittzés, Dóra Ittzés, Ferenc Horváth & Zoltán Barcza Training WS for Ecosystem Modelling studies Budapest, of May, 2014
BOX WROTE THAT „Essentially, all models are wrong, but some are useful"
MODULE X: Understanding basics of GLUE Analysis 1) Purpose and main features of GLUE = Generalized Likelihood Uncertainty Estimation AIM: to estimate parameter values NEEDS: based on observation data & Monte Carlo Experiment METHOD: compare the obs to different model results (based on different parameter sets) Find out which parameter set gives the best results to our observations? How to measure the goodness of the outputs? misfit calculation (root mean square error) likelihood calculation Greater LH value refers to better parameter set
2) Randomized parameters, output settings (MCE) + + Observational dataset >>> GLUE options GLUE needs careful preparation! A) Monte Carlo Experiment: -- randomize the parameters which you want to calibrate -- ask for the outputs wherefor you have observation b) Observation data: -- match your observation to one of the MuSo outputs (look up for the variable short name and code at the output variables) -- prepare your file in csv format and upload it to the database Run GLUE workflow at the portal!
MODULE X: Guess a riddle! -- We defined 3 different epc files with differences in a few parameters give invented names to them (HARMONY, LEAFAREA, NITROGEN) -- We run carbon simulations and get the outputs -- We defined these unreal-data as observation datasets in the project database give invented names to them (HUMUS, AIR, NINE) The question is: Which observation comes from which ecophysiological parametrization? KEY: Use the GLUE analysis! Group work: Creat 3 groups, each select one obs dataset
Canopy average specific leaf area Leaf N in Rubisco HARMONY NITROGEN LEAFAREA Differencies in the defined epc-s
Already exist Monte Carlo Experiments based on the 3 different epc…
The question is: Which observation comes from which ecophysiological parametrization? Exercise: 1) Choose one TEST DEMO Daily Observation dataset (i.e. NINE) 2) Run a GLUE workflow at the portal compare your obs data to the different MCE results (i.e. NINE X LEAFAREA; NINE X HARMONY …) 3) Download (from database) and check the results in Excel! 4) Draw GLUE plots and try to answer the question!
How to run GLUE? GO TO PORTAL: LOG IN! GO TO ECOSYSTEM MODELING!
SELECT GLUE WORKFLOW TO RUN…
GIVE A NAME…START RUN… …WAIT FOR THE INTERACTION PAGE…
SET YOUR RUN ON THE INTERACTION PAGE… GIVE A NAME TO THE RUN… Biome-BGC MuSo 2.2 TEST DEMO HHS (HU) [855]
SET YOUR RUN… AIR, NINE OR HUMUS HARMONY, NITROGEN OR LEAFAREA TEST DEMO HHS MCE FIVEPARAMS ….HARMONY [1145] TEST DEMO Daily observation data – AIR [1164]
CHECK THE STATUS OF YOUR RUN AT PROJECT DATABASE …GET THE RESULTS
GLUE RESULTS: 2 FILES LOOK THE randinputs_likelihoods.csv Every line refers to one parameter set. Last column is the likelihood value. plot likelihood vs parameters one-by-one
EPC files HARMONY NITROGEN LEAFAREA Observations AIR HUMUS NINE ? Canopy average specific leaf area Leaf N in Rubisco HARMONY NITROGEN LEAFAREA Differencies in the defined epc-s Match the pairs!
Answers…
HUMUS observation compared to MCE Canopy average specific leaf areaLeaf N in Rubisco Find max Likelihood… Canopy average specific leaf area Leaf N in Rubisco HARMONY Likelihood
AIR observation compared to MCE Canopy average specific leaf areaLeaf N in Rubisco Find max Likelihood… Canopy average specific leaf area Leaf N in Rubisco LEAFAREA Likelihood
NINE observation compared to MCE Canopy average specific leaf areaLeaf N in Rubisco Find max Likelihood… Canopy average specific leaf area Leaf N in Rubisco NITROGEN Likelihood
EPC files HARMONY NITROGEN LEAFAREA Observations AIR HUMUS NINE Canopy average specific leaf area Leaf N in Rubisco HARMONY NITROGEN LEAFAREA Differencies in the defined epc-s Match the pairs!
Thank you for your attention!