Figure 1

Oro-hydrography of the Geba catchment and flight lines of Italian aerial photographs"

Figure S1 Proportion of elevation (classified based on traditional agro-ecology) and lithology for the observed and predicted map. For explanation on variables see Table 1."

Figure S2 Geba catchment showing: a) lithology (extracted from Tesfamichael et al., 2010; Tesfaye and Gebretsadik, 1982) and soil suitability for cultivation (based on Tielens, 2012); b) slope gradient extracted from DEM - SRTM (USGS, 2014). AP: Location of Italian aerial photo"

Table 1

Variables and their descriptions"

Table 2

Commission-omission error matrix of LUC point count and mapping"

Figure 2

An example of a point grid superimposed on: a) aerial photograph of January 3, 1936 with coordinate of center of vertical photo 13.561547°N and 39.024014°E (i.e. south of Abiy Adi); b) Google Earth image of January 4, 2014. Significant conversion of LUC occurred at this location as shown by: c) terrestrial photo around grid point N16 where land cover changed from dense forest in 1936 (d) to open forest in 2014 (e), and f) terrestrial photo around grid point F12 where land cover changed from open forest in 1936 (g) to cropland with trees in 2014 (h)"

Figure 3

Areal percentage of different land use/cover types in the 1930s and 2014 (n = 34192)"

Table 3

Land use/cover transformation matrix from the 1930s to 2014 in Geba catchment"

Table 4

Relative importance of variables in the model using three criteria number of subset (the number of model subset that includes the variables), RSS (the scaled summed decrease of residual sum of squares overall subset) and Generalized Cross Validation (GCV). For the explanation on the variables see Table 1"

Figure 4

The spatial distribution and change of three dominant land use/cover of the study area: Cropland (a-c), shrubland (d-f) and forest (g-i). A positive (+) and negative (-) sign in these figures indicates the spatial expansion and shrinkage of LUC types, respectively. Graduated classifications illustrate the percent occurrence of land use type in the 1930s and 2014 that also reveal the degree of spatial changes within each location. Sd, Sf1, Sf2, Sh and Ss are supplementary data at Dergajen, Dessa forest 1, Dessa forest 2, Hadinet and Sinkata respectively."

Table S1

Keys used for the classification of LUC during the point counting on AP and GE"

Table S2

Frequency and percentage of scene at which land use/cover types have shown different changes between the 1930s and 2014; Chi-square test result. n = 139"

Table S3

Equations of three major land use/cover (cropland, shrubland and forest) in the 1930s and 2014 which were developed by Multivariate Adaptive Regression Spline model. C1930s = cropland in the 1930s, C2014 = cropland in 2014, S1930s = shrubland in the 1930s, S2014 = shrubland in 2014, F1930s = forest in the 1930s and F2014 = forest in 2014. h=hinge function with zero and a constant (knot) of a factor. For the explanation see Table 1."

Figure S3 Maps of socio-economic variables"

Figure S4 Frequency of occurrence (absence or presence) of different land use/cover types in the 1930s and 2014 scenes, n = 139"