Ecological observations
We used complementary sources of remote sensing to characterize
structural and functional vegetation changes in the catchment. In first
place, we combined our field surveys of the region with detailed
satellite imagery from Google Earth (2000-2020) and aerial photographs
(1962, see Bogino & Jobbágy, 2011) to describe the advent of new
wetlands, lagoons, sediment deposits, and incisions/streams in the
landscape. All these features were easily identified with high spatial
detail imagery but given its associated poor temporal resolution (up to
10 years in some cases), we narrowed down their timing with a
complementary analysis of Landsat imagery. In the case of wetlands, and
in order to obtain a single date for their advent to be used in the
regional water table depth analysis described above, we used the mean
between the latest date without an earliest date with evident surface
signals, as the time of the observation. In the case of stream
incisions, we adjusted this date to the most likely year of occurrence
based on the precipitation series and local observations by farmers.
Based on our experience observing wetland vegetation and groundwater
levels in the field we were able to distinguish sites with <0
(surface water) and 0-1 m of water table depth in the images, using
these ranges as an estimate for the long-term regional analysis
described above.
To obtain relative estimates of water uptake trends for the different
types of vegetation or vegetation change trajectories we used greenness
indexes. Although real evapotranspiration estimates are available from
platforms like MODIS, they tend to have very negatively biased estimates
which end being extremely dependent on surface greenness (Normalized and
Enhanced Vegetation Indexes, i.e. NDVI and EVI). In previous studies
across neighboring regions we found that the direct use of these indexes
provides better estimate of relative transpiration rates for cultivated
and native vegetation (Contreras, Jobbágy, Villagra, Nosetto, &
Puigdefabregas, 2011; Nosetto, Paez, Ballesteros, & Jobbágy, 2015).
Using 16-day and 250 m resolution NDVI MODIS data and a supervised
classification of the current (year 2018) vegetation into forests,
grasslands and pastures, croplands and wetlands together with
field/ground truth control points, we estimated greenness trends from
2000 to 2020 for all the pixels corresponding to each one of this cover
types. We used two criteria to illustrate greenness trends. In the first
one we obtained an average decadal trend comparing the first and the
second decade mean values for each pixel. In the second one, we obtained
linear regressions of mean annual NDVI in response to the number years
elapsed since 2000 for each individual pixel, computing the proportion
of them that showed significant trends (p<0.05) and averaging
the slope of that subgroup for each vegetation type. Since the low
spatial resolution of MODIS imagery (250 x 250 m) did not allow for a
precise description of wetlands and sediment deposits and comparisons
with their adjacent cropland or forest stands, we performed a
complementary analysis using LANDSAT 8 images with a monthly and a
higher spatial resolution (30 x 30 m) for the years before and after the
last erosion episode (2013-2020). We analyzed three stable vegetation
types (forests, croplands and wetlands) and three vegetation transitions
during that period (croplands to wetlands, croplands to deposits,
incised wetlands). In all cases we had at least 6 stands except in the
case of incised wetlands where a single site was considered.
Statistical greenness differences among the first five cases within each
year of analysis were evaluated using ANOVA.