Iterative Approach

Software Used

1. Google Earth Engine:

   • An account is needed to view and use Google Earth Engine. Link to sign up
   • After registration the code editor can be viewed at Link.

2. Python:

   • Python3 can be installed via this link.

3. Matplotlib:

   • Library Link

Importing and Processing Datasets and County Geometries

Three datasets were used to gather NDVI values per plot in a specific county.

1. LandIQ from 2014 with metadata of what the plot is growing at that year and polygon outline of the plot.
2. Landsat 7 NDVI has precomputed NDVI values from Landsat images. Link to Dataset
3. GeoJSON with geometries outlining counties. Link to GeoJSON

var table: Table users/dantran2016/atlas_i15_CropMapping2014;
var landSatNDVI =  ee.ImageCollection("LANDSAT/LE07/C01/T1_8DAY_NDVI");

Landsat 7 data is collected in ImageCollections and contain different images with data below.

{
  "type": "Image",
  "bands": [
    {
      "id": "NDVI",
      "data_type": {
        "type": "PixelType",
        "precision": "float",
        "min": -1,
        "max": 1
      },
      "crs": "EPSG:4326",
      "crs_transform": [
        1,
        0,
        0,
        0,
        1,
        0
      ]
    }
  ],
  "version": 1557407867588447,
  "id": "LANDSAT/LE07/C01/T1_8DAY_NDVI/19990101",

LandIQ has metadata defining what the plot was growing. For example, "Crop2014": "Miscellaneous Grain and Hay", "Crop2014": "Urban" We discarded the Urban classified land since our focus is to classify fallowed agricultural land.

var nonUrban = table.filter(ee.Filter.neq("Crop2014","Urban"));

Further filtering can be done to focus on a county by filtering the LandIQ dataset by the polygon of a county.

var kernCounty = ee.Geometry.MultiPolygon(
    [[[ -118.140074, 34.820780 ], [ -118.854253, 34.817772 ], [ -118.854576, 34.803027 ], [ -118.877289, 34.803212 ], [ -118.881729, 34.817802 ], [ -118.894474, 34.817972 ], [ -118.881364, 34.790629 ], [ -118.976723, 34.790660 ], [ -118.976721, 34.812199 ], [ -119.243645, 34.814178 ], [ -119.243645, 34.857576 ], [ -119.278346, 34.857276 ], [ -119.276946, 34.879675 ], [ -119.382451, 34.879675 ], [ -119.382154, 34.900936 ], [ -119.442352, 34.901274 ], [ -119.472754, 34.901174 ], [ -119.472719, 35.076885 ], [ -119.490709, 35.077208 ], [ -119.490632, 35.091805 ], [ -119.560975, 35.087673 ], [ -119.553641, 35.179975 ], [ -119.667056, 35.174809 ], [ -119.666663, 35.262527 ], [ -119.809449, 35.263584 ], [ -119.809346, 35.350865 ], [ -119.880172, 35.351211 ], [ -119.880045, 35.439133 ], [ -119.997392, 35.439495 ], [ -119.997382, 35.468702 ], [ -120.015659, 35.469039 ], [ -120.014602, 35.483652 ], [ -120.033314, 35.483648 ], [ -120.033241, 35.498642 ], [ -120.051050, 35.498627 ], [ -120.051237, 35.512695 ], [ -120.068905, 35.512779 ], [ -120.068657, 35.526320 ], [ -120.086674, 35.526554 ], [ -120.085922, 35.614524 ], [ -120.193918, 35.614359 ], [ -120.193892, 35.726513 ], [ -120.194146, 35.789204 ], [ -119.538116, 35.789567 ], [ -119.214033, 35.790489 ], [ -118.507224, 35.789711 ], [ -118.464791, 35.792637 ], [ -118.067719, 35.791537 ], [ -118.008043, 35.789161 ], [ -118.000908, 35.789488 ], [ -117.923120, 35.786812 ], [ -117.924459, 35.798149 ], [ -117.632996, 35.797251 ], [ -117.634251, 35.709927 ], [ -117.651986, 35.709934 ], [ -117.652319, 35.680782 ], [ -117.616195, 35.680856 ], [ -117.616395, 35.651755 ], [ -117.633830, 35.651569 ], [ -117.634771, 35.564109 ], [ -117.630126, 35.564071 ], [ -117.630216, 35.451041 ], [ -117.633659, 35.450997 ], [ -117.633290, 35.097558 ], [ -117.632011, 34.822270 ], [ -117.667292, 34.822526 ], [ -118.130847, 34.820938 ], [ -118.132940, 34.820739 ], [ -118.140074, 34.820780 ]]]
  );
nonUrban = nonUrban.filterBounds(kernCounty);

Parameters:

• Months of Interest
• Years of Interest
• Can be a single time period or multiple

var yearsOfInterest = ee.List.sequence(1999,2018);
var monthsOfInterest = ee.List.sequence(1,12);
var monthsOfInterest = ee.List([7,12]);

Warning: There are cases where at a point in time there is not a NDVI value for a plot and an error will be thrown.

Given the parameters a list of Landsat imageCollections filtered by individual month and year can be generated through the function below.

var yearlyCollections  = yearsOfInterest.map(function(i){
    return monthsOfInterest.map(function(j){
      i = ee.Number(i);
      j = ee.Number(j);
      var date = ee.Date.fromYMD(i,j,1);
      return landSatNDVI.filterDate(date, date.advance(1,'month'));
  });}).flatten();

The map function iterates through a data structure's elements. Inside the map function computations can be done based on the element's metadata and appends the modified element to a new data structure.

Link to Script

Gathering Data

Google Earth Engine allows the adding of properties via the set function. Below the mean NDVI value is calculated per plot and set as a property on the plot.

var numberOfEntries = yearlyCollections.size();
var yearlyCollectionIndex = 0;

var checkFallow = function(feature){
    //Selects a filtered Landsat imageCollection
    var convertYearly = ee.ImageCollection(yearlyCollections.get(yearlyCollectionIndex));
    //Filters Landsat by the bounds of a feature's (single plot) geometry
    var plotNDVI = convertYearly.filterBounds(feature.geometry());
    //Computes the mean NDVI value over the entire plot
    plotNDVI = plotNDVI.mean();
    var meanDict = plotNDVI.reduceRegion({
      reducer: ee.Reducer.mean(),
      geometry: feature.geometry(),
      scale: 1,
      bestEffort:true
    });
    //Extracts the NDVI value from the dictonary return value of meanDict
    plotNDVI = ee.Number(meanDict.get("NDVI"));

    //Constructing a dictonary with the index and the NDVI value to be saved to the feature.
    var foo = {};
    var monthYear = yearlyCollectionIndex.toString();
    foo[monthYear] = plotNDVI;

    return feature.set(foo);
};

Using the script below time series NDVI can be stored in feature collection.

var landIQNDVI = ee.FeatureCollection(nonUrban.map(setNDVI));

for(yearlyCollectionIndex = yearlyCollectionIndex+1; yearlyCollectionIndex < numberOfEntries; yearlyCollectionIndex++){
  landIQNDVI = ee.FeatureCollection(landIQNDVI.map(setNDVI));
}

landIQNDVI returns a feature collection with mean NDVI values over the times of interest.

"properties": {
    "0": 0.09877312395773193,
    "1": 0.06276312808948771,
    "2": 0.5253586653470814,
    "3": 0.2979903895553776,
    "4": 0.28399248585516185,
    "5": 0.11432357099110248,
    "6": 0.47526314529567587,
    "7": 0.34818060441157317,
    "8": 0.45370712232626437,
    "9": 0.312030452955916,
    "10": 0.22345866422510072,
    "11": 0.08438224407847458,
    "12": 0.521465526833003,
    "13": 0.204677961783142,
    "14": 0.6150093086963376,
    "15": 0.23447950068881104,
    "16": 0.5725433139818257,
    "17": 0.3034091259869929,
    "18": 0.4643725507406636,
    "19": 0.4620643699828776,
    "20": 0.49508907247464856,
    "21": 0.3243342797228612,
    "22": 0.520806995871174,
    "23": 0.5113243309062386,
    "24": 0.5146000702264357,
    "25": 0.23753865377151034,
    "26": 0.6727596679922516,
    "27": 0.43168228585573964,
    "28": 0.4618968378861816,
    "29": 0.29327033538090086,
    "30": 0.37452569638729827,
    "31": 0.22109621685776032,
    "32": 0.27541618087125747,
    "33": 0.39848670981790774,
    "34": 0.4741756929825517,
    "35": 0.4315363232683603,
    "36": 0.3767581578248459,
    "37": 0.567144975313284,
    "38": 0.4721550143541549,
    "39": 0.281364929728735,

Link to Script

Exporting Data

The function below exports a feature collection to a cloud server, Google Drive, or Google Earth Engine asset.

Export.table.toDrive({
  collection: ee.FeatureCollection(landIQNDVI),
  description: 'LandSat7-Bi-Yearly-NDVI',
  fileFormat: 'CSV',
});

To access the export window on Google Earth Engine, click the task tab on the right panel.

Then click the run button to export the data.

Link to Script

Limitations of Google Earth Engine

Exporting NDVI values for Kern county (359,893 plots) at one time point took 1 to 2 hours to export. The export time scales linearly with the number of time plots. For example, 40 time points took 3 days to compute.

Visualizing Results