Estimating distributed soil texture using time series of thermal 1 remote sensing – A case study in central Europe 2

For understanding water and solute transport processes knowledge about the respective hydraulic 10 properties is necessary. Commonly, hydraulic parameters are estimated via pedo-transfer functions using soil 11 texture data to avoid cost intensive measurements of hydraulic parameters in the laboratory. Therefore, current 12 soil texture information is only available at coarse spatial resolution of 250 m to1000 m. Here, a method is 13 presented to derive high-resolution (15 m) topsoil texture patterns for the meso-scale Attert catchment 14 (Luxembourg, 288 km2) from 28 images of ASTER thermal remote sensing. A principle component analysis of 15 the images reveals the most dominant thermal patterns (principle components, PCs) that are related to 212 16 fractional soil texture samples. Within a multiple linear regression framework, distributed soil texture information 17 is estimated and related uncertainties are assessed. An overall root mean squared error of 12.7 percentage points 18 (pp) lies well within and even below the range of recent studies on soil texture estimation, while requiring sparser 19 sample setups and a less diverse set of basic spatial input. 20

. If time series of thermal RS data are available, the concept of thermal inertia is applicable 10 to gain information on soil texture. Thermal inertia is the spatially varying tendency of the land surface to resist 11 changes in temperature forced by energy input. Responsible for these spatial differences in inertia are patterns of 12 thermal conductivity, density, and specific heat capacity of the land surface material (Rees and Rees, 2013; 13 Minacapilli et al., 2012). However, thermal observations of land surface are non-linear integrals over all three 14 dimensions in space of the occurring materials (Hall et al., 1995;Betts et al., 1996). These integrals consider spatial 15 averaging, as well as thermal emission and propagation from sub surface thermal sources up to vegetation. 16 Parameters that influence the surface temperature are incoming radiation, land use, albedo, and available water 17 content. Especially the latter is strongly controlled by soil texture, which subsequently should influence the thermal 18 inertia signature as given by the temporal patterns of surface temperature.

26
The rest of the manuscript is organized as follows: Section 2 introduces the test site, the implemented and auxiliary 27 data, as well as methods applied and developed. Section 3 shows and discusses the results of the estimator setups 28 and its cross validations. Finally, section 4 reviews main findings and gives overall conclusions. 29 agriculture (65.4%), followed by forests (29.7%) and settlements and other sealed areas (4.8%) (Corine land cover; 1 EEA, 1995). The monthly mean temperature ranges from 0 °C in January to 18 °C in July (197118 °C in July ( -2000; the climate 2 is pluvial oceanic.

3
The existing agricultural soil map (1:100.000; SPP, 1969; Fig. 2, bottom) lacks of quantitative descriptions but 4 give hints on spatial patterns of soil texture and its systematic distribution: Silt explicitly occurs in four out of the 5 six existent soil classes in the area; clay soils are observed in the North West and sandy soils occur in the South 6 East. Thus, relations between geology and soil can be observed, particularly for schists and clay in the 7 northwestern, and sandstone and sand in the southeastern region.
1 where is the soil data before transformation, ′ after transformation and is a parameter estimated from the data 2 or error distribution to achieve normality. An optimal is estimated by an iterative Monte Carlo procedure

18
The MLRE calibrated with the complete field sample set is then used to calculate fully distributed texture maps.
19 Figure 7 shows the resulting soil texture maps. Each texture fraction is modeled separately with the aforementioned 20 PC combination. Finally, the three texture fractions are translated into USDA soil types, which are then mapped 21 back into the catchment (Fig. 7, lower right). A comparison of predicted and observed texture data shows a large 22 overlap between both (Fig. 7, lower left).

23
The distribution of soil texture conforms to the distribution of the soil characteristics displayed in the available 24 qualitative agricultural soil maps (Fig. 2). Clay is dominant in the north, rather sandy soils can be found in the 25 south and mainly silty soils prevail in the remaining parts of the catchment. Further analysis of the soil texture 26 distribution reveals relations to topographic structures, different land cover types and geology ( Fig. 1 and 2).