MODIS data availability and applicability for land cover change analysis of the Darhad Valley, Mongolia F. Aaron Rains Biology 515 Fall 2009 Abstract. The quantification of land-cover change due to human and climate related pressures is an important variable in the management of coupled human/natural systems. In the Darhad Valley of Northern Mongolia, overgrazing and erosion are causing changes in the land cover types and abundances found there. Observing, analyzing and quantifying these changes via geospatial analysis of remotely sensed data: MODIS Terra products, NDVI, VCF and Land cover composites) was the primary goal of this research. Secondarily, the utility of these MODIS data was evaluated as a tool for fine to meso (defined here as 250m -1000m, and sub decadal, respectively) spatial and temporal scale changes. The scale used in this research was essentially defined by the availability of data Overview Observing and quantifying land-cover change in remote areas of the world is often a challenging task. Many factors such as data availability, resolution, cost, and an applicable temporal range of data are common obstacles researchers face. In this study I looked at the applicability of the MODIS terra products for detecting a fine to meso scale land cover change in the remote Darhad Valley of Mongolia. A nomadic-pasturalistic culture dominates in this area, with traditional, annual migrations of thousands of animals being cycled through various seasonal feeding grounds. The ecological impacts of these migrations are showing as erosion of grasslands gives way to migrating and accumulating sand dunes (Montagne 2009). The detection and potential monitoring of these changes using available MODIS Terra data were the focus of this research. Materials and Methods The area of interest (AOI) was initially defined by previous researched performed in the area by Dr. Cliff Montage and the Bioregions team. To set the physical parameters, a digital topographical map obtained from Google Earth (www.google.com) was used to roughly estimate the watershed boundaries of the Darhad Valley. Once the AOI had been defined, a subset of data was created using the Oak Ridge National Laboratory’s interactive Distributed Active Archive Center website (/daac.ornl.gov/MODIS/). The temporal range of the data was also chosen here, based on availability. The desired data output was chosen from the suite of MODIS data indices. Several outputs were chosen based on applicability including a sixteen day composite of normal deviated vegetation index (NDVI) (MOD13Q1, 250m) data for the same two week period leading up to July 11 in 2000 and 2009 (Figure 3). This date range was chosen based on availability of data and an estimated occurrence of high primary production that would occur in this time period. The mild summer weather likely to occur during this time would also limit cloudobscurity occurrences. A land cover (LC) classification layer (Figure 1) was also included in the data set. This was created as an annual composite of land cover (MOD12C1, 1000m) for the year 2005. Other annual LC data were unavailable from this source. Other data were obtained from another source, the National Aeronautical and Space Administration’s (NASA) data bank website (lpdaac.usgs.gov/). A broader temporal range of the LC data (MOD12C1, 1000m) were obtained (2001-2004) and utilized in the final analysis. Vegetation continuous fields (VCF MOD44b, 500m resolution) (Figure 2) and Vegetation coverage conversion (VCC, 250m) data were also reviewed and deemed either applicable or not applicable based on coverage or temporal ranges. The VCF data were available as a composite of 2001 through 2005. This data layer is composed of values ranging from 0-100% for three class types: Barren ground, herbaceous cover and tree cover. Most of these data were requested of the respective websites, downloaded in hdf. format, converted to tiff. format and imported into a GIS using ArcGIS software. Within the GIS spatial analysis, re-projecting, creating and comparing attribute tables, manipulating the classification and display schemes and reclassifying certain data were part of the analysis performed. A final comparison using the created attribute tables was performed on the available data to determine whether land cover change could be detected using these data. Results The MODIS data that I was able to obtain included NDVI data for two, 16 day composite time periods in July 2000 and July 2009. The NDVI data was reclassified and analyzed using Arcmap software. The results of this temporal comparison revealed a strong difference in NDVI for the two time periods. With the July 2000 NDVI showing much higher values in general. The results of my NDVI comparison would prove to be inconclusive based on the nature of NDVI and its variability. A complete VCF data set (including all three classification types) was available only for 2000 through 2005. Attempts were made at retrieving complete VCF data for a later time period from the Land processes data active archive center’s (LPDAAC) WIST warehouse website but had not been processed and made available as of December, 2009. According to a representative from the LPDAAC, complete annual sets of VCF were being processed and should be available sometime in the future. The latter VCF data that was available (years after 2005) included only percent tree cover data at this time with the other classifications to follow. This data was proven limited based on availability of temporal range and completeness of data offered to the public. The LC data could prove to be a relatively useful to for the comparison of land cover type averages on an annual basis. Unfortunately, I was unable to convert the 2001 LC or latter VCF data from hdf. format to tif. format. This was due to the availability of the tool and a lack of technical proficiency on behalf of the researcher. Discussion Monitoring spatially fine scale and relatively short-term temporal changes in land cover in a remote region with the current suite of MODIS Terra products has several pros and cons. Current resolutions of the data used in this study are a major limiting factor. The date range for which many of the applicable data are available also limits the temporal frame that a researcher may study. Obvious benefits of using MODIS data include the availability to the general public as well as being free of charge. In general MODIS data ranges from 250m to 1000m based on the bands with in spectral range of the sensor (Zhan et al., 2000). The VCF data is offered in a 500m resolution, but the temporal range needed for a comparison of land cover change over time is lacking. When the VCF data has time to mature and a complete set from 2001 through the present is available, this could prove to be a powerful tool for monitoring general land cover change at the meso scale. With a finer resolution than the other data and broader temporal range of availability, NDVI can be a powerful tool for monitoring the trend of vegetation abundance and richness over time (Ranjeet et al., 2007). The difference in values from the two singular 16 day time period composites, one from 2000 and the other from 2009 that I obtained and compared, could be attributed to climate variations leading up to the time when the data was captured or a host of phenological variables (Lunetta et al., 2005). To use NDVI as an accurate comparison tool for land-cover change, annual averages or a complete set of yearly composites for the time of study would need to be obtained and statistically analyzed for any trending behavior of the values. These procedures fell beyond the scope of this research technically and temporally. The land cover change (MCD12Q1) data could prove to be useful for a general comparison of land cover class type averages on an annual basis. When this data set grows beyond its current temporal range and resolution, it could prove to be highly useful for similar applications. Friedl et al., 2010, report on new updates and refinements to previous algorithms and data inputs. This has enhanced the resolution of the collection 5 land cover tool to 500m, four times the resolution of the previous collection 4 (Friedl et al., 2010). When available, collection will likely increase to utility of this data set as a tool for land cover change detection. The VCC data, although unavailable for my AOI, has shown to be a valuable tool for monitoring land-cover changes in other regions of the world. Zhan et al. (2002) have produced and applied the VCC algorithm to MODIS 1b data for several experimental areas that are experiencing rapid and expansive land cover changes due to human activities and extreme natural disturbances. This method is designed to serve as an alarm for early detection of rapid change. These changes can then be monitored using higher resolution sensors (Zhan et al., 2002). Conclusion Observing and analyzing fine scale change using the current suite of MODIS data is difficult at best. Although my data was inconclusive, I feel that I have developed a much broader understanding of the availability and utility of the majority of the available MODIS data. I am also now familiar with the data acquisition process and many valuable yet basic functions of the ArcGIS software. I have also become familiar with some of the work that is being done of the cutting edge of land-cover change detection via remote sensing. I have learned that the researcher must identify the scale of change he/she is intending to study before making a choice of which remotely sensed data to use and from which sensor/satellites. The suite of MODIS data available is in fact highly useful for large-scale studies (Zhan et al., 2001) where the resolution of the data and the scale of the change phenomenon taking place, are in the same realm. Increasing resolutions and algorithm refinements (Friedl et al., 2010) hold much promise for the utility of this data for future applications as well as the strengthening of the current available data pool. Overall, the potential for detecting and monitoring land use change in regions like the Darhad valley exists to a burgeoning degree. As these data sets grow in time and spatial resolutions the applicability of these data for land change science will also increase. MODIS data should prove to be an increasingly useful tool for researchers through the future. Citations Ranjeet, J., Jiquan, C., Nan, L., Gou, K., Cunzhu, L., Yafen, W., Noormets, A., Keping, M., and Xinggou, H., 2007. Predicting plant diversity based on remote sensing products in the semi-arid region of Inner Mongolia. Remote Sensing of Environment 112, 20182032. Lunetta, R., Knight, J., Ediriwickrema, J., Lyon, J., and Worthy, L.D., 2006. Land Cover change detection using multi-temporal MODIS NDVI data. Remote Sensing of Environment 105, 142-154. Zhan, X., Sohlberg, R.A., Townsend, J.R.G., DiMiceli, C., Carroll, M.L., Eastman, J.C., Hansen, M.C., and DeFries, R.S., 2002. Detection of land cover changes using MODIS 250m data. Remote Sensing of Environment 83, 336-350 Zhan, X., DeFries, R.S., Townsend, J.R.G., DiMiceli, C., Hansen, M.C., Huang, C., and Sohlberg, R.A., 2000. The 250m global land cover change product from the Moderate Imaging Resolution Spectroradiometer of NASA’s Earth Observing Satellite. International Journal of Remote Sensing 201, no. 6 & 7, 1433-1460. Friedl, M. A., Sulla-Menashe, D., Tan, B., Schneider, A., Ramankutty, N., Sibley, A., and Huang, X., 2010. MODIS collection 5 land cover: Algorithms refinements and characterizations of new datasets. Remote Sensing of Environment, v. 114 issue 1, p. 168-182. Montagne, Clifford. Professor, Land Resources and Environmental Sciences, personal communication, 2009, Montana State University, Bozeman, MT. END