Unexpected results for EVI process graph

[m-mohr]

I have the following process graph, which is meant to compute a monthly EVI value over a single year, which I'd like to have in a single file.

{
  "process_graph": {
    "1": {
      "arguments": {
        "data": {
          "from_node": "reduce_dimension_TDTKX9963G"
        },
        "period": "month",
        "reducer": {
          "process_graph": {
            "2": {
              "arguments": {
                "data": {
                  "from_parameter": "data"
                }
              },
              "process_id": "mean",
              "result": true
            }
          }
        }
      },
      "process_id": "aggregate_temporal_period"
    },
    "2": {
      "arguments": {
        "data": {
          "from_node": "3"
        },
        "dimension": "t",
        "process": {
          "process_graph": {
            "2": {
              "arguments": {
                "data": {
                  "from_parameter": "data"
                }
              },
              "process_id": "array_create",
              "result": true
            }
          }
        },
        "target_dimension": "bands"
      },
      "process_id": "apply_dimension"
    },
    "3": {
      "arguments": {
        "data": {
          "from_node": "1"
        },
        "label": "1",
        "name": "bands",
        "type": "bands"
      },
      "process_id": "add_dimension"
    },
    "load_collection_LUCZT2231M": {
      "arguments": {
        "bands": [
          "B08",
          "B04",
          "B02"
        ],
        "id": "SENTINEL2_L2A",
        "properties": {},
        "spatial_extent": {
          "east": 6.646557,
          "north": 50.82171,
          "south": 50.811346,
          "west": 6.631536
        },
        "temporal_extent": [
          "2018-01-01",
          "2019-01-01"
        ]
      },
      "description": "B08 = NIR, B04 = RED, B02 = BLUE",
      "process_id": "load_collection"
    },
    "reduce_dimension_TDTKX9963G": {
      "arguments": {
        "data": {
          "from_node": "load_collection_LUCZT2231M"
        },
        "dimension": "bands",
        "reducer": {
          "process_graph": {
            "6rgsd17e7": {
              "arguments": {
                "x": -7.5,
                "y": {
                  "from_node": "e5n3z9bta"
                }
              },
              "process_id": "multiply"
            },
            "72xvdpkex": {
              "arguments": {
                "x": {
                  "from_node": "v8dynjn0c"
                },
                "y": {
                  "from_node": "axbzqj0r0"
                }
              },
              "process_id": "add"
            },
            "7rz81lmvj": {
              "arguments": {
                "data": {
                  "from_parameter": "data"
                },
                "index": 0
              },
              "process_id": "array_element"
            },
            "8ftvtyz1w": {
              "arguments": {
                "x": {
                  "from_node": "7rz81lmvj"
                },
                "y": {
                  "from_node": "lg832392l"
                }
              },
              "process_id": "subtract"
            },
            "axbzqj0r0": {
              "arguments": {
                "x": 6,
                "y": {
                  "from_node": "lg832392l"
                }
              },
              "process_id": "multiply"
            },
            "e5n3z9bta": {
              "arguments": {
                "data": {
                  "from_parameter": "data"
                },
                "index": 2
              },
              "process_id": "array_element"
            },
            "el6umyy2u": {
              "arguments": {
                "x": 2.5,
                "y": {
                  "from_node": "8ftvtyz1w"
                }
              },
              "process_id": "multiply"
            },
            "lg832392l": {
              "arguments": {
                "data": {
                  "from_parameter": "data"
                },
                "index": 1
              },
              "process_id": "array_element"
            },
            "pzpe4qrxw": {
              "arguments": {
                "x": {
                  "from_node": "72xvdpkex"
                },
                "y": {
                  "from_node": "6rgsd17e7"
                }
              },
              "process_id": "add"
            },
            "v8amok3l5": {
              "arguments": {
                "x": {
                  "from_node": "el6umyy2u"
                },
                "y": {
                  "from_node": "pzpe4qrxw"
                }
              },
              "process_id": "divide",
              "result": true
            },
            "v8dynjn0c": {
              "arguments": {
                "x": 1,
                "y": {
                  "from_node": "7rz81lmvj"
                }
              },
              "process_id": "add"
            }
          }
        }
      },
      "description": "Formula is:\n2.5 * (NIR - RED) / (1 + NIR + 6*RED + -7.5*BLUE)",
      "process_id": "reduce_dimension"
    },
    "save_result_ZZGVZ2715G": {
      "arguments": {
        "data": {
          "from_node": "2"
        },
        "format": "GTiff"
      },
      "process_id": "save_result",
      "result": true
    }
  }
}

There are some issues and strange things happening:

1. The resulting values are not only between -1 and 1, but are usually between -0.1 and 3. Why?
2. The resulting metadata has only a single band in the metadata although it should have 12.
3. Why do I need to use add_dimension although I would expect that apply_dimension creates it? See apply_dimension: 'No band dimension' although meant to be created automatically #169

[jdries]

EVI expects bands to be in the 0..1 range, wheras our digital numbers are 0..10000.
Could you try dividing band values by 10000 before computing EVI?

[m-mohr]

That should explain issue 1, but haven't checked yet. We should figure out a way to document this so that people are aware of the value ranges. We discussed this a couple of times already, so finally opened a dedicated issue for it: https://github.com/openEOPlatform/architecture-docs/issues/237

[soxofaan]

related: last week I updated the "Getting started" page for python client, which originally also included wrong EVI calculation (because of missing digital number rescaling).
Proper rescaling is mentioned here: https://open-eo.github.io/openeo-python-client/basics.html#band-math