The analysis happens at the level of the three basins of interest: The Xingu, Tocantins and Sâo Francisco basins, shown in figure A below.
The sources of data used in the analysis are available at several spatial scales, including at the ERA5 gridcell level (Figure B), and at the municipality level (Figures C and D). In Figure C, the borders of the municipalities involved are displayed along the borders of the basins; in Figure D, the share of area of each municipality in each basin is shown.
The data available from the National Water Agency is at the microbasin level, not shown in the figures below. The Tocantins basin is divided in around 22 thousand microbasins, the Xingu around 8 thousand, and the São Francisco 50 thousand microbasins.
If we assume that the proportion of a certain activity or land use in each basin is proportional to the amount of area of that municipality that is in this basin, how much error can be generated?
In the case of pastures, the shaer of pastures in each municipality and basin was estimated, and then we analyzed how the share of pasture area in each basin is correlated to the share of municipality area in each basin.
As the graph below shows, in most cases that is a good assumption in general, but it can be quite misleading for some municipalities.
This was estimated by ANA for the year 2017.
Below I suggest two different ways of estimating mean irrigation ET per month: First, considering all data points. Second, considering only data points in which delta_ET > 0
## Joining with `by = join_by(TRASE_ID, YEAR)`
## `summarise()` has grouped output by 'basinname', 'YEAR'. You can override using
## the `.groups` argument.