November 20, 2014 Mapping croplands using Landsat data with generalized classifier over large areas…

November 20, 2014

Mapping croplands using Landsat data with generalized classifier over large

areas

Aparna Phalke and Prof. Mutlu Ozdogan

Nelson Institute for Environmental Studies University of Wisconsin - Madison

Updates on following

LDA model training and tuning

LDA results of sample footprints

Basic Definitions• Model tuning is the process in which one or more parameters of a device or model are adjusted upwards or downwards to achieve improved or specified results

• The aim of LDA model tuning is to calibrate the parameters of propagation models and improve the key performance indicators.

MethodologyR algorithm

• Divide training data in 75%-25% splits• LDA model trained on 75% data and tested on 25% data

• This procedure repeated 1000 times with random sets of train and test

• LDA model accuracy check with train and test within scene or within footprint.

ResultsGroup 1:

ResultsGroup 2:

ResultsGroup 3:

ResultsGroup 4:

ResultsGroup 5:

ResultsGroup 6:

ResultsLDA model accuracy at different

levels

LDA classified image sample result: Mea

nStd

Min Variance

Range

Counts

Slope Elevation

MaxInputs

LDA classified image results

turkey Group/zone Allmean 1.11E-03 6.10E-04 5.97E-04

sd -1.83E-03 -1.58E-03 -1.62E-03max -8.13E-05 -1.21E-04 -1.23E-04min -3.44E-05 -3.46E-04 -2.09E-04var 1.34E-07 1.82E-07 1.37E-07

range -7.63E-05 -5.60E-05 1.13E-05count 1.99E-02 -1.67E-02 1.02E-02slope 7.91E-02 9.21E-02 1.39E-01

elevation 2.82E-03 1.03E-03 1.88E-04

LDA coefficients of sample example

Own Within

group/zone All

LDA classified image sample results

Conclusion Model tuning helped us in understanding dynamics of whole process, which gives a more accurate picture of how the model is behaving.

Thank [email protected]