ritviksahajpal / contiguous_SIF

this contains the code for generating the CSIF dataset and the NN parameters

Contiguous_SIF

This contains the code for generating the CSIF dataset and the NN parameters

NN parameters

We tested a variety of NN architecture (1-3 layers and 2-9 neurons for each layer), the best formed NN was selected (1 layer 5 neurons). The NN_parameter folder contains the weights and biases for each layer. This folder also contains the two files required for the normalization of reflectances (.nc files).

preprocessing of MCD43C4

We fellowed Zhang et al (2017) and generate the 4-day reflectance. Three steps corresponds to the three python files:

1. aggreated the daily reflectance files to 4day using their median values.
2. get a mean seasonal cycle as a reference.
3. reconstruct the SIF for each year.

generate the CSIF product

Several products are needed to generate the CSIF (clear-day and all-daily)

1. 4day refletance dataset from the previous step.
2. mean and standard deviation of reflectance for each band (in NN parameter folder).
3. solar zenith angle calcualted based on the solar time and latitude.
4. BESS daily PAR product (only required if all-daily SIF is calculated).
5. DEM product to calcualt the clear-sky PAR.

The clear-inst and clear-daily SIF data is first calculated using the neuron network. The all-daily SIF is then calcualted using the clear-inst SIF and the daily PAR from BESS.