Changes between Version 43 and Version 44 of DevelopmentActivities/ORCHIDEE-DOFOCO

Timestamp:

2013-01-22T13:46:52+01:00 (11 years ago)

Author:

mmcgrath

Comment:

--

DevelopmentActivities/ORCHIDEE-DOFOCO

@@ -2 +2 @@
 
 == APPROACH ==
-The two stream radiation transfer model of Pinty et al. 2006 (J. Geophys. Res., 111, D02116, doi:10.1029/2005JD005952) is implemented in ORCHIDEE. This routine uses leaf scatter and canopy structure to calculate the radiation transfer and as a result albedo can be calculated.  We are also making it into a multi-layer model.  The calculation of the effective LAI now uses the Pgap model of Haverd et al described in their 2012 paper (Agricultural and Forest Meteorology, Vol. 160, pp 14-35) to take the canopy structure into account.
+The two stream radiation transfer model of Pinty et al. 2006 (J. Geophys. Res., 111, D02116, doi:10.1029/2005JD005952) is implemented in ORCHIDEE. This routine uses leaf scatter and canopy structure to calculate the radiation transfer and as a result albedo can be calculated.  We are also making it into a multi-layer model with an arbitrary number of layers.  The single scattering and forward efficiency values of the canopy are changed to match the diffuse and direct top of the canopy albedos for the single layer case.  The calculation of the effective LAI now uses the Pgap model of Haverd et al described in their 2012 paper (Agricultural and Forest Meteorology, Vol. 160, pp 14-35) to take the canopy structure into account.
 
 == TASKS ACCOMPLISHED ==
@@ -9 +9 @@
          * A framework capable of using an arbitrary number of canopy layers is in place, although the tests are still not satisfactory
          * The snow calculation has been modified to include snow in the background reflectance of Pinty's scheme, effectively putting snow under the canopy
+         * Bernard's two stream solver routine has been modified to have both a diffuse and direct background reflectance value
 
 == ROUTINES CHANGED ==
@@ -34 +35 @@
         * The new scheme was compared against the old one in v1.9.6.0, and the results are qualitatively similar, but not identical
         * A 20 layer model was tested against the single layer model, giving small differences (a few percent, relative, at midday)
-        * We tried to optimize the single scattering and forward efficiency values in the n-layer case to reproduce the average albedo in the 1-layer case, but a simple optimization scheme is only able to make very small gains and takes an order of magnitude more time...it appears that we might be able to optimize the diffuse and direct TOC albedo to match the 1-layer case, but it could take thousands of steps for a simple compass search (four function evals per step), even adjusting the step length during the optimization....maybe try a generalized set search and vary the step directions as well?
+        * We tried to optimize the single scattering and forward efficiency values in the n-layer case to reproduce the average albedo in the 1-layer case, and using a generalised set search a full optimization with 20 layers can take thousands of steps and 200 times longer than doing no-optimization, and only a third of the points are optimised completely (for two layers, it takes about ten times longer than the no-opt case and optimises almost 70% of the timesteps)...using only 10-50 optimization steps results in 1-10% of the points being fully optimized, but looking at a graph of the two runs shows that the differences are really minimal, so we need to test the partial optimization in the new energy budget to see if it makes a real difference
         * Inclusion of the new snow albedo gives realistic trends, with the total albedo spiking after snowfalls and showing diurnal variation with leaves are present, but being flat at other times (the solar angle is incorporated through the use of Pgap, which mostly depends on the canopy)
         * I made use of single scattering albedos that Juliane got from Bernard's inversion scheme for 20 tree species, running tests with the min and max values for single scattering albedo and forward scattering efficiency for both VIS and NIR bands...the MUE for the overal albedo varies by 0.01 to 0.04 albdo units depending on the spectrum, showing that we should take this into account