Search

This repository contains two arrays of numberssaved in .npy format. The file "pars_save.npy" provides the results of the Markov Chain Monte Carlo runs (i.e. samples from the posterior distributions of parameters given observations). This array has 4dimensions; its shape is(4, 3, 10000, 16). Along the first dimension arethe observation scenarios: HOURLY, 1 AM/PM, 6 AM/PM, 1+6, and the "true" model. Along the second dimension are the 3 MCMC chains for each scenario. The third dimension is the iterations of the MCMC for each chain. Along the fourth dimension are the parameters, in the same order and with the same units as in Table 1 from the paper. To obtain the results in the paper, for each chain the first 6000 iterations were discarded as burn-in, and every 100th sample was taken from the subsequent iterations to avoid autocorrelation. The file "output_quantiles.npy" characterizes the posterior distribution of retrieved land surface states.This array has 4 dimensions; its shape is (5, 5, 113880, 5). Along the first dimension are5 land surface variables, in the following order: leaf water potential (LWP) that has been normalized based on soil water retention curves as discussed in the paper (MPa) column-averaged soil moisture (m^3/m^3) evapotranspiration (ET) (mol H2O/m^2/s) gross primary productivity (GPP) (umol CO2/m^2/s) raw LWP (MPa) Along the second dimension arethe observation scenarios: HOURLY, 1 AM/PM, 6 AM/PM, 1+6, and the "true" model. The third dimension is time, starting at midnight on January 1, 2005 and progressing at a 1-hour time step for 13 years Along the fourth dimension arequantiles of the posterior distribution, namely 0.05, 0.25, 0.5 (median), 0.75, and 0.95 Note that values of the "true" model do not change over the fourth dimension, as it is only a single model run rather than an ensemble/distribution. For example, to extract a time series of the posterior median of the 1 AM/PM retrieval for ET forthe first year of simulations, you could execute the following code in Python: import numpy as np data = np.load("output_quantiles.npy") gpp_1AMPM_med = data[3, 1, 0:(24*365), 2] &nbsp