Generating continuous 25 m 8-day actual evaporation grids using spatio-temporal blending of Landsat and MODIS data for the Darwin catchments

[摘要] Remote sensing platforms have varying spatial, temporal, spectral, and radiometric characteristics. Therefore using the same algorithm on remote sensing data from different platforms will result in different characteristics in the modelled output. This research introduces a simple, transparent, and computationally inexpensive framework for blending data having different temporal and spatial characteristics, retaining the optimal spatio-temporal features of the inputs. For the Darwin study site covering 30,200 km2 (comprising the (i) Finniss River; (ii) Adelaide River; (iii) Mary River (NT); and (iv) Wildman River basins), high-spatial-resolution (i.e., 25-m) and high-temporal-frequency (i.e., 8-day) estimates of actual evapotranspiration (ETa) from reflective remote sensing indices and meteorological data were generated. This required blending two remote sensing datasets: (1) high spatial resolution and low temporal frequency Landsat imagery (25 m and 16-day at best if no clouds); and (2) high temporal frequency and low spatial resolution MODIS imagery (250 m and 8-day at best if no clouds). The large study site necessitated computational efficiency, which meant a new blending algorithm had to be developed to produce a model of ETa that retained the optimal characteristics of the inputs: (1) accuracy of the MODIS ETa dataset; (2) temporal variability (variance) of the MODIS ETa dataset; and (3) spatial variability (variance) of the Landsat ETa dataset.ETa accuracy was assessed by comparing modelled results to observed ETa from seven flux towers scatted throughout the Darwin catchments, and long-term average precipitation less long-term average stream flow from 15 small unimpaired catchments. Accuracy assessment results for the new dataset (25 m and 8-day) were comparable to the MODIS data with an order of magnitude better spatial resolution (i.e., 25 m compared to 250 m). This generated ETa output is suitable for tracking water-use of ground-water dependent ecosystems (GDE) in the study site which can be used to constrain hydrological models that are fundamentally mass balance models (so continuous ETa estimates are needed). This was previously not possible at the GDE resolution using either the Landsat imagery or MODIS imagery alone for use in hydrological models. The method can be applied to other environments, and can overcome issues of cloudiness, to generate accurate high-spatial-resolution (i.e., 25 m) and high temporal frequency (i.e., 8-day) estimates of ETa.

[发布日期] 2017-06-28 [发布机构] CSIRO

[学科分类] 地球科学(综合)