LMF (Local Maximum Fitting)
for LINUX & Windows

HONDA Kiyoshi, Yann Chemin, RS&GIS, Asian Institute of Technology
last update 2005/05/29 since 28 May 2005

We have been developing algorithms such as for calibrating crop models using RS data through data assimilation, monitoring actual irrigation, counting number of cropping, discriminating crop types and etc (publication). The RS data used is mainly NDVI time series data from low to moderate resolution satellites which is useful to monitor and analyze crop dynamics, such as 10days SPOT Vegetation Instrument Data, 8 days or 16days MODIS data.

However, even the data is the composite of several or more days' data, the data is still not free from the influence of cloud cover. Pixels may have been covered by clouds partially or totally covered by thin clouds which usually reduces the value of NDVI.

ANDES project (AFFRC 1998, JST 1998 ) as a part of "Research and Development for Applying Advacned Computational Science and Technology" or ACT-JST has developed LMF algorithm ( LMF stands for Local Maximum Fitting ) (SAWADA,H. 2001, NAGATANI et. al. 2002 ). The detail algorithm can be found in SAWADA 2002. The algorithm firstly finds local maximums within a certain time window, and regenerate the time series data as a sum of harmonic curves. The number of harmonic curves is limited by AIC ( Akaike Information Criterion ) to avoid over-fitting. The detail is in Lecture note on LMF and NN of Advanced RS course of AIT.

An example of LMF processed MODIS 8 days NDVI at Suphanburi Province in Thailand.
column=22, line=35 in the sample image included in the distribution.
NDVI is being scaled from 0-2000. The base wavelenth of the harmonic curve is 1 year, thus the same pattern repeats 3 times for 3 years data like above (The trend is being considered). The window size of the local maxim filter is 6(NWIN=3).

Almost of source codes ( FORTRAN ) are available from the project home page (in Japanese), however, the codes are for unix+Fortran77(OpenMP) and the binaries are for SGI. We have modified the program for NDVI by adding a subroutine and removing interfaces to multiple processors. Now, the code can be compiled under common environment; LINUX(f77) and Windows (Visual Fortran). We have obtained an approval from JST, and made the code and executable downloadable from this site.

Download Source Code and Executable: ( LINUX , Windows )
The software is being distributed as it is. "ACT-JST" must be referred for re-distribution and publishing the result.
The zip file includes source, executable and sample data. You will find data directory where you can just run the executable to test the LMF using small 8 days MODIS NDVI data although it may take 20 minutes on Pentium M 1.1 GHz. You have to change cntl.dat to process your own data.

We have applied the LMF on 8 days NDVI of MODIS in a project "Near Real Time Agriculture Activity Monitoring using Multi-Temporal MODIS Earth Observation Satellite Data" supported by RTG (Royal Thai Government) - AIT joint research fund. We have organized a seminar to disseminate research result including LMF to Thai researchers on 28th-29th April, 2005 at Faculty of Engineering, Kasetsart University, Bangkok, Thailand.( photos )
Download Distributed Material (Coming Soon).

Our team is utilizing cluster computers for our research especially data assimilation. If you are interested in installing clusters, or GRID, we may have collaboration.

REFERENCES
SAWADA, H., SAWADA, Y., 2002, Development of Seasonal Change Model of Vegetation from High-frequency Observation Satellite Data, Environmental Information Science, No. 16, pp.181- 185, 200, Center for Environmental Information Science, Tokyo ( in Japanese )

SAWADA, H., 2001, Asia-Pacific network for disaster mitigation using Earth Observation Satellite (ANDES), October 1, 1998-September 30, 2001. Department of Global, Forest Research Forestry and Forest Products Research Institute, Ibaraki, Japan., Available online at: http://www.affrc.go.jp/ANDES/sawady/index.html

NAGATANI, I., SAITO, G., TORITANI, H. and SAWADA, H., 2002, Agricultural map of Asian region using time series AVHRR NDVI data. In Proceedings Online of the 23rd Asian Conference on Remote Sensing, 25-29 November 2002, Birendra International Convention Centre in Kathmandu, Nepal. Available online at: http://www.gisdevelopment.net/aars/acrs/2002/pos2/184.pdf

AFFRC, 1998, ANDES Asia-Pacific Network for Disaster Mitigation Using Earth Observation Satellite, Available online at: http://www.affrc.go.jp/ANDES/ ( AFFRC: Agriculture, Forestry and Fisheries Research Council )

JST, 1998, Asia-Pacific Network for Disaster Mitigation Using Earth Observation Satellite (ANDES), Available online at: http://act.jst.go.jp/content/h10/ter_cosm/T04/PageMain_e.html

statistics

LMF (Local Maximum Fitting) for LINUX & Windows

HONDA Kiyoshi, Yann Chemin, RS&GIS, Asian Institute of Technology last update 2005/05/29 since 28 May 2005

LMF (Local Maximum Fitting)
for LINUX & Windows

HONDA Kiyoshi, Yann Chemin, RS&GIS, Asian Institute of Technology
last update 2005/05/29 since 28 May 2005