Using sampleRegions with randomPoints samples less points than what is provided

Question

Link for the code: https://code.earthengine.google.com/89616ca3147334915eca9f3957bb778f

I am using GEE in order to access image, generate random points on top of it and then sample it using the random points.

I have found out unexpected behavior : I generate 200 random points, and then when I sample the region, I get back only 192 points. This happens for different numbers of sample points: (500 random points ->472 sampled
100 random points -> 95 sampled
1000 random points -> 946 sampled etc...)

var smap=ee.ImageCollection('NASA_USDA/HSL/SMAP10KM_soil_moisture').filterBounds(geometry).filterDate('2020-01-01', '2021-01-01').median().clip(geometry);

var randomPoints = ee.FeatureCollection.randomPoints(
    {region:geometry, points: 1000});
    
var fcPointSamp = smap.sampleRegions({
  collection: randomPoints,
  scale: 10000
});

Map.addLayer(smap,{},'SMAP');

Map.addLayer(randomPoints,{},'random_points')
Map.centerObject(geometry,8)

print('Number of points', randomPoints.size());
print('Number of sampled points', fcPointSamp.size());

Why does this happen? Is there any way I can control the number of points that are being sampled?

Link for the code: https://code.earthengine.google.com/89616ca3147334915eca9f3957bb778f

EDIT:
based on Daniel's answer: workaround for geemap:

dummy_band=img.addBands(ee.Image(-9999).rename('temp')).clip(fc)

img_replaced_masked = img.unmask(dummy_band.select('temp'))

###replace the masked values in img with the -9999 values
img_mask_masked_values = img.unmask(dummy_band.select('temp'))

then, when I used sampleRegions, the no value pixels got value of "-9999"

Answer 1

You miss samples because there are areas in geometry not covered by smap. So, randomPoints will contain samples in this area, but not fcPointSamp.

A fix to this is to use stratifiedSample() instead. It will give you a random sampling, and you will not end up with missing values. Since you don't actually want to stratify on anything, you can create a dummy constant image, and use that band to stratify, so you end up with a single stratum.

var fcPointSamp = smap
  .addBands(ee.Image(1).rename('temp')) // Add dummy band
  .stratifiedSample({ 
    numPoints: 1000, 
    classBand: 'temp', // We stratify on the dummy band - a single class
    scale: 1000, // Since your geometry is so small, you cannot fit 1,000 samples with a scale of 10,000
    geometries: true // Just included so it can be put on a map, to highlight that this is random sampling
  });

https://code.earthengine.google.com/c79d76884a132f92f670a97fc8155023