EMSC3025-6025

Assignment II — Total Water Storage

- / -

Total Water Storage: Soil Moisture and Groundwater

Space gravity tells you how much water a basin holds, but not where the water is. This assignment takes the total water storage measured by GRACE over the Murray-Darling Basin, subtracts a satellite estimate of the water held in the soil, and calls what is left groundwater. Building the two time series is the routine part. The interesting part is that the answer depends on a number nobody measured — how deep into the soil the satellite estimate is assumed to reach — and most of this assignment is about finding out how much it depends on it.

Total: 15 marks.

Computer labs for this assignment

The same three labs that supported Assignment I are used again here. If you have not run them, do that first.

The idea

Total water storage is the sum of every store of water in a column of ground: surface water, soil moisture, groundwater, snow and ice, and the water held in vegetation. GRACE measures the sum. If you assume that the only two stores that change over the Murray-Darling Basin are soil moisture and groundwater, then

\Delta \text{Groundwater} = \Delta \text{TWS} - \Delta \text{Soil Moisture}

Question 5 asks you to take that assumption apart.

The two datasets

GRACE and GRACE-FO total water storage

Monthly mass anomaly grids from the Center for Space Research (CSR) at the University of Texas at Austin, release RL06.3. Download the single file with all corrections applied:

CSR RL06.3 mascons, all corrections

The file is CSR_GRACE_GRACE-FO_RL0603_Mascons_all-corrections.nc, about 110 MB on disk and about 1 GB once the array is in memory. Read the description on that page before you use it.

ESA CCI combined soil moisture

The ESA Climate Change Initiative combined active-plus-passive surface soil moisture record, version 09.1, produced by TU Wien and partners. The full record is the daily product and is distributed by the CEDA archive:

ESA CCI soil moisture at CEDA

You do not need the whole archive. A cut-down copy of the daily record, covering the Murray-Darling Basin from 2003-01-01 to 2023-12-31, is on the course data server:

https://data.gadopt.org/water-course/esacci_sm_combined_v09-1_MDB_daily_2003-2023.nc

It is 170 MB. Download it once into your session and keep it; do not put the download inside a cell that you re-run.

Five ways this will silently go wrong

1. The GRACE time axis does not decode. CSR writes the attribute as Units with a capital U. xarray looks for a lowercase units, does not find it, and leaves time as plain floating-point numbers: 107.0, 129.5, 227.5 and so on. Those are days since 2002-01-01, as the file’s time_epoch attribute states. Convert them yourself. The same applies to lwe_thickness, whose Units attribute is cm.

2. The three datasets in this course use three different coordinate conventions. The rainfall file in Assignment I runs latitude -10 down to -44. The soil moisture file also runs north to south, -23.125 down to -38.875. The GRACE file runs the other way, -89.875 up to 89.875, and its longitude runs 0 to 360 rather than -180 to 180. A slice given in the wrong order returns an empty array and no error. Print the shape of every subset before you use it.

3. GRACE anomalies are relative to a fixed baseline. The file’s time_mean_removed attribute reads 2004.000 to 2009.999: the mean of those six years has already been subtracted from every grid cell. Your soil moisture series is an absolute depth and has had nothing subtracted. Before you take the difference in , remove the 2004 to 2009 mean from the soil moisture series as well. If you skip this, every groundwater number you report is wrong by a constant of several centimetres. Note also what those six years were in the Murray-Darling Basin.

4. Months are missing, and one gap is a year long. About twenty-five individual months are absent from the GRACE record, and there is no data at all between June 2017 and June 2018, when GRACE had ended and GRACE-FO had not yet started. The epochs are also irregular dates near the middle of each month, not the first of the month. Do not interpolate across the mission gap. Average both products to calendar months, match on year and month, and leave the missing months as gaps in the plot.

5. Averaging daily soil moisture over a changing set of cells introduces a trend that is not there. Compute the monthly mean cell by cell first, then average over the basin, and weight the average by cell area as in Lab 4. Record how many valid days went into each monthly mean and say what minimum you required.

Questions

  1. (3 marks) Build a time series of the change in total water storage averaged over the Murray-Darling Basin, from April 2002 to the end of 2022, in millimetres of equivalent water. Use the union of the north and south boundaries from MDB_boundaries.zip, the same pair of shapefiles as Assignment I. Plot it, and state the wettest and driest month in the record and the full range in millimetres.

    Save the resulting series to a text file, so that you do not have to repeat the integration every time you re-run the notebook. Submit that file.

  2. (3 marks) Build the matching time series of soil moisture over the basin, as a monthly mean, and convert it from a volume fraction to an equivalent water depth in millimetres:

    d = \theta \times L

    where \theta is the volumetric soil moisture and L is the thickness of the soil layer the estimate is taken to represent. For this question take L = 20 cm.

    Report the mean, the standard deviation and the range of the resulting depth. Compare the standard deviation with the standard deviation of the total water storage series from Question 1, and say what that comparison already tells you about how much of the basin’s storage variability the soil moisture can account for.

  3. (2 marks) Download one daily file of the combined product directly from CEDA and plot it globally. The daily files are public and about 1.4 MB each; the address follows the pattern

    https://dap.ceda.ac.uk/neodc/esacci/soil_moisture/data/daily_files/COMBINED/v09.1/<year>/ ESACCI-SOILMOISTURE-L3S-SSMV-COMBINED-<yyyymmdd>000000-fv09.1.nc

    Then explain, in a short paragraph each, how the combined product is made and what its limitations are. Your answer must cover which sensors contribute and how active and passive retrievals are merged, how deep into the soil the microwave signal actually senses and why, and where on the map the retrieval fails and for what physical reason. The flag variable in the file you downloaded names those failure modes; use it. See the ESA Soil Moisture Project and the references given there.

  4. (4 marks) Combine Questions 1 and 2 with to produce a time series of the change in groundwater storage over the basin, from 2003 to 2022, in millimetres. The soil moisture record begins in January 2003, which sets the start; of the 240 calendar months in that period, only about 207 have a GRACE solution.

    Repeat the calculation for the northern and the southern basin separately, using the two shapefiles individually rather than their union.

    Then interpret the result. Identify the Millennium Drought and the 2010-2011 recovery in your series, and quantify them: how much storage did the basin lose, and how much of it came back? Do the northern and southern basins behave the same way? GRACE resolves features of about 300 km, and each half of the basin is not much larger than that, so say what you think the north-south comparison can and cannot support.

  5. (3 marks) The 20 cm in Question 2 is not a measurement. It is a choice, and the answer moves with it. Repeat Question 4 for L = 5 cm, L = 20 cm and L = 100 cm, and report for each

    • the standard deviation of the soil moisture anomaly in millimetres;
    • the fraction of the variance of the total water storage series that the soil moisture accounts for;
    • the correlation between your groundwater series and the raw total water storage series.

    You will find that these three columns behave very differently, and the point of the question is to explain why. At the shallow end, what does your groundwater series amount to? At the deep end, what exactly have you multiplied, and by how much? State which value of L, if any, you would defend, and on what evidence.

    Then answer the question the assignment is really asking: are these estimates of groundwater change reliable? A good answer names the water that quietly assigns to groundwater but that is not groundwater. Consider at least the soil water below the depth the satellite senses, and the water held in the basin’s reservoirs — the Murray-Darling storages hold of order twenty thousand gigalitres when full, and you are in a position to work out what that is in millimetres spread over the basin and to compare it with the soil moisture correction you applied.

EMSC6025 requirements

EMSC6025 students must also:

  • Justify each choice made in bringing the two datasets onto a common footing: the spatial averaging, the monthly aggregation, the treatment of missing months and the alignment of the reference periods.
  • Attach an uncertainty to the groundwater series. The soil moisture files carry an sm_uncertainty variable; the GRACE literature gives an error for basin-scale mass estimates at this size. Say which of the two dominates.
  • Compare your groundwater series against at least one published estimate for the Murray-Darling Basin and account for the differences.

These requirements form part of the 15 marks. They do not add marks.

Submission Instructions:
  • Submit your answers through the Canvas portal.
  • Submit one Jupyter notebook containing your code, your figures and your written answers. It must run end to end from a fresh kernel without errors.
  • Submit the text files of the time series you saved along the way.
  • If your files are too large for Canvas, send them via AARNet FileSender and note in your Canvas submission that you have done so.