Date: Aug 17, 2017
Software: FUSION
References:
FUSION Manual: http://forsys.cfr.washington.edu/fusion/FUSION_manual.pdf
The lidar metrics I want to compute include Elev_mode, Elve_P95 and Total_retu.
1. Results got from FUSION
2. Eliminate points with value -9999. This means these points are not covered by lidar data.
3. Eliminate points with Elev_95 > 500 m. These points are located around the edges. Their heights are not properly corrected. Delete 231 out of 4865. Now have 4634 left for building lidar and Landsat model.
4. For overlapped points. There are overlaps between two adjacent lidar images. For the overlapped points, I chose the one with the larger total return. This is done in Python.
Ground-lidar model
1. Calculate lidar-biomass
Stepwise lidar-ground model:
LOG(BIOMASS)= -1.05151 + 0.99383*LOG(ELEV_MODE) + 0.32649*LOG(TOTAL_RETU) + 0.74666*LOG(ELEV_P95)
In Arcmap field calculator:
LOGAGB = -1.05151 + 0.32649*math.log10( !Total_retu! ) + 0.99383*math.log10( !Elev_mode! )+0.74666*math.log10( !Elev_P95! )
LIDARAGB = 10** !logAGB!
2. Comparison of ground biomass and lidar biomass.
Lidar - Landsat models
1. Landsat variables
Band 1-5, 7; RVI; DVI; NDVI; SAVI; MSAVI2
2. Building models
PROC REG DATA = Lidar_bio;
MODEL AGB_lidar = RVI DVI NDVI SAVI MSAVI2 B1 B2 B3 B4 B5 B7 / SELECTION = RSQUARE;
RUN;
PROC REG DATA = Lidar_bio;
MODEL AGB_lidar = RVI DVI NDVI SAVI MSAVI2 B1 B2 B3 B4 B5 B7 / SELECTION = STEPWISE;
RUN;
RSQUARE result:
Stepwise result:
3. LOG transformation for just Y.
From results of STEPWISE, R square improves from 0.3059 to 0.3659.
4. Log both Xs and Y.
5. Comparison of R square for raw data, log transform Y, both log transform Xs and Y.
Underlined numbers are results from stepwise selection method.
I am going to use the model highlighted in green. It has 8 variables and has not much difference with the full model.
6. Lidar-Landsat model
