Linear downscaling from MODIS to landsat: connecting landscape composition with ecosystem functions
- 18 Downloads
The open and free access to Landsat and MODIS products have greatly promoted scientific investigations on spatiotemporal change in land mosaics and ecosystem functions at landscape to regional scales. Unfortunately, there is a major mismatch in spatial resolution between MODIS products at coarser resolution (≥ 250 m) and landscape structure based on classified Landsat scenes at finer resolution (30 m).
Based on practical needs for downscaling popular MODIS products at 500 m resolution to match classified land cover at Landsat 30 m resolution, we proposed an innovative modelling approach so that landscape structure and ecosystem functions can be directly studied for their interconnections. As a proof-of-concept of our downscaling approach, we selected the watershed of the Kalamazoo River in southwestern Michigan, USA as the testbed.
MODIS products for three fundamental variables of ecosystem function are downscaled to ensure the approach can be extrapolated to multiple functional measurements. They are blue-sky albedo (0–1), evapotranspiration (ET, mm), and gross primary production (GPP, Mg C ha−1 year−1). An object-oriented classification of Landsat images in 2011 was processed to generate a land cover map for landscape structure. The downscaling model was tested for the five Level IV ecoregions within the watershed.
We achieved satisfactory downscaling models for albedo, ET, and GPP for all five ecoregions. The adjusted R2 was > 0.995 for albedo, 0.915–0.997 for ET, and 0.902–0.962 for GPP. The estimated albedo, ET, and GPP values appear different in the region. The estimated albedo was the lowest for water (0.076–0.107) and the highest for cropland (0.166–0.172). Estimated ET was the highest for the built-up cover type (525.6–687.1 mm) and the lowest for forest (209.7–459.7 mm). The estimated GPP was the highest for the build-up cover type (8.65–9.85 Mg C ha−1 year−1) and the lowest for forest.
Estimated values for albedo, ET, and GPP appear reasonable for their ranges in the Kalamazoo River region and are consistent with values reported in the literature. Despite these promising results, the downscaling approach relies on strong assumptions and can carry substantial uncertainty. It is only valid at a spatial scale where similar climate, soil, and landforms exist (i.e., values in isolated patches of the same cover type are similar). Plausibly, the uncertainties associated with each estimation, as well as the model residuals, can be explored for other pattern-process relationships within the landscape.
KeywordsDownscaling MODIS Landsat GPP ET Albedo Kalamazoo River watershed
This study was supported, in part, by the NASA Carbon Cycle & Ecosystems program (NNX17AE16G), the Great Lakes Bioenergy Research Center funded by the U.S. Department of Energy, Office of Science, Office of Biological and Environmental Research under Award Numbers DE-SC0018409 and DE-FC02-07ER64494; and the Long-term Ecological Research Program (DEB 1637653) at the Kellogg Biological Station, and the NASA Science of Terra and Aqua program (NNX14AJ32G). We thank the fruitful discussion at LEES Lab meetings where several members made constructive suggestions for model development. Isabel Arroca assisted in formatting the references. The reviews from two anonymous reviewers helped improving the quality of this manuscript.
- Bass DG (2009) Inferring dissolved phosphorus cycling in a TMDL watershed using biogeochemistry and mixed linear models. PhD disertation, Michigan State University, East Lansing, MichiganGoogle Scholar
- Chapman KA, Brewer R (2008) Prairie and Savanna in southern lower Michigan: history, classification, ecology. Michigan Bot 47:1–48Google Scholar
- Chen J (1991) Edge effects: microclimatic pattern and biological responses in old-growth Douglas-fir forests. PhD dissertation, University of Washington, Seattle, WAGoogle Scholar
- Chen J, Wan S, Henebry G et al (eds) (2013) Dryland east Asia: land dynamics amid social and climate change. DE GRUYTER, BerlinGoogle Scholar
- Cressie N, Wikle CK (2015) Statistics for spatio-temporal data. Wiley, New Jersey, p 624pGoogle Scholar
- Fongers D (2008) Kalamazoo River watershed hydrologic study. Michigan Department of Environmental Quality, Lansing, MI, p 67pGoogle Scholar
- LeMoine JM, Chen J (2003) Placing our research objectives and results in time and space. Acta Phytoecol Sin 27:1–10Google Scholar
- Schaetzl RJ, Darden JT, Brandt DS (2009) Michigan geography and geology. Pearson Learning SolutionsGoogle Scholar
- Turner MG, Gardner RH (2015) Landscape ecology in theory and practice: pattern and process, 2nd edn. Springer, New York, p 482pGoogle Scholar
- Wan Z (2013) MODIS land surface temperature products users’ guide. Institute for Computational Earth System Science. University of California, Santa Barbara, CAGoogle Scholar
- Wu J, Hobbs RJ (eds) (2007) Key topics in landscape ecology. Cambridge University Press, Cambridge, p 297pGoogle Scholar