view backend/doc/schema/oracle-sinfo-uinfo.sql @ 9482:e5cb5efbc43e

standard veg'zone data cage fix
author gernotbelger
date Wed, 12 Sep 2018 12:50:28 +0200
parents 361de818f76e
children 69a7edd7d1e1
line wrap: on
line source
SET AUTOCOMMIT ON;


--Add new column wsts.sinfo_selection

ALTER TABLE wsts ADD (
  sinfo_selection CHAR(1) CHECK((sinfo_selection IS NULL) OR (sinfo_selection IN ('W','Q')))
);
COMMENT ON COLUMN wsts.sinfo_selection IS 'Whether and how the WST series is selected within S-INFO calculation types ("W"=without-Q-group, "Q"=with-Q-group, NULL=not-selectable)' ;


--Add new columns bed_height.sounding_width_info and bed_height.notes

ALTER TABLE bed_height ADD (
  sounding_width_info VARCHAR2(256),
  notes VARCHAR2(256)
);
COMMENT ON COLUMN bed_height.sounding_width_info IS 'File header line info "ausgewertete Peilbreite"' ;
COMMENT ON COLUMN bed_height.notes IS 'File header line info "weitere Bemerkungen"' ;


--Add new columns bed_height_values.min_height (Talweg) and bed_height_values.max_height (Kammlage) (same type as existing height column)

ALTER TABLE bed_height_values ADD (
  min_height FLOAT(126),
  max_height FLOAT(126),
  height01 FLOAT(32),
  height02 FLOAT(32),
  height03 FLOAT(32),
  height04 FLOAT(32),
  height05 FLOAT(32),
  height06 FLOAT(32),
  height07 FLOAT(32),
  height08 FLOAT(32),
  height09 FLOAT(32),
  height10 FLOAT(32)
);
COMMENT ON COLUMN bed_height_values.min_height IS 'Minimum bed height (Talweg) in m' ;
COMMENT ON COLUMN bed_height_values.max_height IS 'Maximum bed height (Kammlage) in m' ;
COMMENT ON COLUMN bed_height_values.height01 IS 'Bed height of section 1 in m' ;
COMMENT ON COLUMN bed_height_values.height02 IS 'Bed height of section 2 in m' ;
COMMENT ON COLUMN bed_height_values.height03 IS 'Bed height of section 3 in m' ;
COMMENT ON COLUMN bed_height_values.height04 IS 'Bed height of section 4 in m' ;
COMMENT ON COLUMN bed_height_values.height05 IS 'Bed height of section 5 in m' ;
COMMENT ON COLUMN bed_height_values.height06 IS 'Bed height of section 6 in m' ;
COMMENT ON COLUMN bed_height_values.height07 IS 'Bed height of section 7 in m' ;
COMMENT ON COLUMN bed_height_values.height08 IS 'Bed height of section 8 in m' ;
COMMENT ON COLUMN bed_height_values.height09 IS 'Bed height of section 9 in m' ;
COMMENT ON COLUMN bed_height_values.height10 IS 'Bed height of section 10 in m' ;


--Bed mobility

CREATE TABLE bed_mobility (
    id  NUMBER(9,0) PRIMARY KEY,
    river_id  NUMBER(38,0) NOT NULL CONSTRAINT cBedMobilityRivers REFERENCES rivers(id) ON DELETE CASCADE,
    kmrange_info  VARCHAR2(32),
    filename  VARCHAR2(256) NOT NULL,
    notes  VARCHAR2(256)
);
COMMENT ON TABLE bed_mobility IS 'Longitudinal section of the bed mobility of a river' ;
COMMENT ON COLUMN bed_mobility.kmrange_info IS 'File header line info "Strecke"' ;
COMMENT ON COLUMN bed_mobility.filename IS 'Name without type extension of the imported file' ;
COMMENT ON COLUMN bed_mobility.notes IS 'File header line info "weitere Bemerkungen"' ;

CREATE SEQUENCE BED_MOBILITY_ID_SEQ ;


CREATE TABLE bed_mobility_values (
    id  NUMBER(9,0) PRIMARY KEY,
    bed_mobility_id  NUMBER(9,0) NOT NULL CONSTRAINT cBedMobilityValuesBedMobility REFERENCES bed_mobility(id) ON DELETE CASCADE,
    station  NUMBER(7,3) NOT NULL,
    moving  NUMBER(1,0) DEFAULT 0 NOT NULL CHECK(moving IN (0,1))
);
COMMENT ON TABLE bed_mobility_values IS 'Bed mobility of a km' ;
COMMENT ON COLUMN bed_mobility_values.station IS 'River km' ;
COMMENT ON COLUMN bed_mobility_values.moving IS 'Whether the river bed at the station is moving' ;

CREATE SEQUENCE BED_MOBILITY_VALUES_ID_SEQ ;


--Infrastructure

CREATE TABLE infrastructure (
    id  NUMBER(9,0) PRIMARY KEY,
    river_id  NUMBER(38,0) NOT NULL CONSTRAINT cInfrastructureRivers REFERENCES rivers(id) ON DELETE CASCADE,
    annotation_type_id  NUMBER(38,0) NOT NULL CONSTRAINT cInfrastructureAnnotationType REFERENCES annotation_types(id),
    year  NUMBER(4,0) CHECK((year >= 1700) AND (year <= 2199)),
    dataprovider  VARCHAR2(256),
    evaluation_by  VARCHAR2(256),
    kmrange_info  VARCHAR2(32),
    filename  VARCHAR2(256) NOT NULL,
    notes  VARCHAR2(256)
);
COMMENT ON TABLE infrastructure IS 'Longitudinal section of infrastructures of a river and a type' ;
COMMENT ON COLUMN infrastructure.year IS 'File header line info "Stand"' ;
COMMENT ON COLUMN infrastructure.dataprovider IS 'File header line info "Datenherkunft"' ;
COMMENT ON COLUMN infrastructure.evaluation_by IS 'File header line info "Auswerter"' ;
COMMENT ON COLUMN infrastructure.kmrange_info IS 'File header line info "Strecke"' ;
COMMENT ON COLUMN infrastructure.filename IS 'Name without type extension of the imported file' ;
COMMENT ON COLUMN infrastructure.notes IS 'File header line info "weitere Bemerkungen"' ;

CREATE SEQUENCE INFRASTRUCTURE_ID_SEQ ;


CREATE TABLE infrastructure_values (
    id  NUMBER(9,0) PRIMARY KEY,
    infrastructure_id  NUMBER(9,0) CONSTRAINT cInfrastructureValuesInfrastru REFERENCES infrastructure(id) ON DELETE CASCADE,
    station  NUMBER(7,3) NOT NULL,
    attribute_id  NUMBER(38,0) CONSTRAINT cInfrastructureValuesAttrib REFERENCES attributes(id),
    height  NUMBER(6,2)
);
COMMENT ON TABLE infrastructure_values IS 'Infrastructure at a river station' ;
COMMENT ON COLUMN infrastructure_values.station IS 'River km' ;
COMMENT ON COLUMN infrastructure_values.height IS 'Geodetic height of the top of the infrastructure in m' ;

CREATE SEQUENCE INFRASTRUCTURE_VALUES_ID_SEQ ;


--Channel

CREATE TABLE channel (
    id  NUMBER(9,0) PRIMARY KEY,
    river_id  NUMBER(38,0) NOT NULL CONSTRAINT cChannelRivers REFERENCES rivers(id) ON DELETE CASCADE,
    kmrange_info  VARCHAR2(32),
    filename  VARCHAR2(256) NOT NULL,
    notes  VARCHAR2(256),
    year_from  NUMBER(4,0),
    year_to  NUMBER(4,0)
);
COMMENT ON TABLE channel IS 'Longitudinal section of the navigable channel of a river' ;
COMMENT ON COLUMN channel.kmrange_info IS 'File header line info "Strecke"' ;
COMMENT ON COLUMN channel.filename IS 'Name without type extension of the imported file' ;
COMMENT ON COLUMN channel.notes IS 'File header line info "weitere Bemerkungen"' ;
COMMENT ON COLUMN channel.year_from IS 'Start year of the period for which the channel values are valid, NULL when valid for the whole past';
COMMENT ON COLUMN channel.year_to IS 'End year of the period for which the channel values are valid, NULL when unlimited';


CREATE SEQUENCE CHANNEL_ID_SEQ ;


CREATE TABLE channel_values (
    id  NUMBER(9,0) PRIMARY KEY,
    channel_id  NUMBER(9,0) CONSTRAINT cChannelValuesChannel REFERENCES channel(id) ON DELETE CASCADE,
    station  NUMBER(7,3) NOT NULL,
    width  NUMBER(6,2),
    depth  NUMBER(6,2)
);
COMMENT ON TABLE channel_values IS 'Nominal size of the navigable channel at a river station' ;
COMMENT ON COLUMN channel_values.station IS 'River km' ;
COMMENT ON COLUMN channel_values.width IS 'Nominal width of the channel in m' ;
COMMENT ON COLUMN channel_values.depth IS 'Nominal depth of the channel in m' ;

CREATE SEQUENCE CHANNEL_VALUES_ID_SEQ ;


--Collision

CREATE TABLE collision_type (
    id  NUMBER(9,0) PRIMARY KEY,
    name  VARCHAR2(64) NOT NULL UNIQUE
);
COMMENT ON TABLE collision_type IS 'Type of a ship collision' ;
COMMENT ON COLUMN collision_type.name IS 'Name of the collision type' ;

CREATE SEQUENCE COLLISION_TYPE_ID_SEQ ;


CREATE TABLE collision (
    id  NUMBER(9,0) PRIMARY KEY,
    river_id  NUMBER(38,0) NOT NULL CONSTRAINT cCollisionRivers REFERENCES rivers(id) ON DELETE CASCADE,
    year  NUMBER(4,0) NOT NULL,
    kmrange_info  VARCHAR2(32),
    filename  VARCHAR2(256) NOT NULL,
    notes  VARCHAR2(256)
);
COMMENT ON TABLE collision IS 'Longitudinal section of ship collisions with the river bed' ;
COMMENT ON COLUMN collision.year IS 'File header line info "Jahr"' ;
COMMENT ON COLUMN collision.kmrange_info IS 'File header line info "Strecke"' ;
COMMENT ON COLUMN collision.filename IS 'Name without type extension of the imported file' ;
COMMENT ON COLUMN collision.notes IS 'File header line info "weitere Bemerkungen"' ;

CREATE SEQUENCE COLLISION_ID_SEQ ;


CREATE TABLE collision_values (
    id  NUMBER(9,0) PRIMARY KEY,
    collision_id  NUMBER(9,0) NOT NULL CONSTRAINT cCollisionValueCollision REFERENCES collision(id) ON DELETE CASCADE,
    station  NUMBER(7,3) NOT NULL,
    event_date  DATE NOT NULL,
    gauge_w  NUMBER(6,2),
    gauge_name  VARCHAR2(64),
    collision_type_id  NUMBER(9,0) NOT NULL CONSTRAINT cCollisionValuesCollisionType REFERENCES collision_type(id)
);
COMMENT ON TABLE collision_values IS 'Collision event' ;
COMMENT ON COLUMN collision_values.station IS 'River km' ;
COMMENT ON COLUMN collision_values.event_date IS 'Date of the collision' ;
COMMENT ON COLUMN collision_values.gauge_w IS 'Waterlevel during the collision' ;
COMMENT ON COLUMN collision_values.gauge_name IS 'Name of the gauge' ;

CREATE SEQUENCE COLLISION_VALUES_ID_SEQ ;


--Tkh

CREATE TABLE tkh (
    id  NUMBER(9,0) PRIMARY KEY,
    river_id  NUMBER(38,0) NOT NULL CONSTRAINT cTkhRivers REFERENCES rivers(id) ON DELETE CASCADE,
    day  DATE,
    kmrange_info  VARCHAR2(32),
    filename  VARCHAR2(256) NOT NULL,
    groupname  VARCHAR2(32) NOT NULL CHECK(groupname IN ('messung','berechnung')),
    sounding_info  VARCHAR2(64),
    evaluation_by  VARCHAR2(256),
    notes  VARCHAR2(256)
);
COMMENT ON TABLE tkh IS 'Longitudinal section of computed Transportkoerperhoehe of a river' ;
COMMENT ON COLUMN tkh.day IS 'File header line info "Datum"' ;
COMMENT ON COLUMN tkh.kmrange_info IS 'File header line info "Strecke"' ;
COMMENT ON COLUMN tkh.filename IS 'Name without type extension of the imported file' ;
COMMENT ON COLUMN tkh.groupname IS 'Group the file belongs to' ;
COMMENT ON COLUMN tkh.sounding_info IS 'File header line info "Peilung"' ;
COMMENT ON COLUMN tkh.evaluation_by IS 'File header line info "Auswerter"' ;
COMMENT ON COLUMN tkh.notes IS 'File header line info "weitere Bemerkungen"' ;

CREATE SEQUENCE TKH_ID_SEQ ;


CREATE TABLE tkh_column (
    id  NUMBER(9,0) PRIMARY KEY,
    tkh_id  NUMBER(9,0) NOT NULL CONSTRAINT cTkhColumnTkh REFERENCES tkh(id) ON DELETE CASCADE,
    name  VARCHAR2(64) NOT NULL
);
COMMENT ON TABLE tkh_column IS 'Longitudinal section of computed Transportkoerperhoehe for a waterlevel series' ;
COMMENT ON COLUMN tkh_column.name IS 'Name of the tkh computation column' ;

CREATE SEQUENCE TKH_COLUMN_ID_SEQ ;


CREATE TABLE tkh_values (
    id  NUMBER(9,0) PRIMARY KEY,
    tkh_column_id  NUMBER(9,0) NOT NULL CONSTRAINT cTkhValuesTkhColumn REFERENCES tkh_column(id) ON DELETE CASCADE,
    station  NUMBER(7,3) NOT NULL,
    tkheight  NUMBER(7,3)
);
COMMENT ON TABLE tkh_values IS 'Transportkoerperhoehe of a river station and referenced file column' ;
COMMENT ON COLUMN tkh_values.station IS 'River km' ;
COMMENT ON COLUMN tkh_values.tkheight IS 'Transportkoerperhoehe of a river station computed for a waterlevel in  m' ;

CREATE SEQUENCE TKH_VALUES_ID_SEQ ;


--Flow depth

CREATE TABLE flow_depth (
    id  NUMBER(9,0) PRIMARY KEY,
    river_id  NUMBER(38,0) NOT NULL CONSTRAINT cFlowDepthRivers REFERENCES rivers(id) ON DELETE CASCADE,
    year  NUMBER(4,0) NOT NULL,
    kmrange_info  VARCHAR2(32),
    filename  VARCHAR2(256) NOT NULL,
    sounding_info  VARCHAR2(64),
    evaluation_by  VARCHAR2(255),
    notes  VARCHAR2(256)
);

COMMENT ON TABLE flow_depth IS 'Longitudinal section of computed flow depth of a river' ;
COMMENT ON COLUMN flow_depth.year IS 'File header line info "Bezugsjahr"' ;
COMMENT ON COLUMN flow_depth.kmrange_info IS 'File header line info "Strecke"' ;
COMMENT ON COLUMN flow_depth.filename IS 'Name without type extension of the imported file' ;
COMMENT ON COLUMN flow_depth.sounding_info IS 'File header line info "Peilung"' ;
COMMENT ON COLUMN flow_depth.evaluation_by IS 'File header line info "Auswerter"' ;
COMMENT ON COLUMN flow_depth.notes IS 'File header line info "weitere Bemerkungen"' ;

CREATE SEQUENCE FLOW_DEPTH_ID_SEQ ;


CREATE TABLE flow_depth_column (
    id  NUMBER(9,0) PRIMARY KEY,
    flow_depth_id  NUMBER(9,0) NOT NULL CONSTRAINT cFlowDepthColumnFlowDepth REFERENCES flow_depth(id) ON DELETE CASCADE,
    name  VARCHAR2(64) NOT NULL
);

COMMENT ON TABLE flow_depth_column IS 'Longitudinal section of computed flow depth for a waterlevel series' ;
COMMENT ON COLUMN flow_depth_column.name IS 'Name of the flow depth computation column' ;

CREATE SEQUENCE FLOW_DEPTH_COLUMN_ID_SEQ ;


CREATE TABLE flow_depth_values (
    id  NUMBER(9,0) PRIMARY KEY,
    flow_depth_column_id  NUMBER(9,0) NOT NULL CONSTRAINT cFlowDepthValuesFlowDepthCol REFERENCES flow_depth_column(id) ON DELETE CASCADE,
    station  NUMBER(7,3) NOT NULL,
    depth  NUMBER(7,3)
);

COMMENT ON TABLE flow_depth_values IS 'Flow depth of a river station and referenced file column' ;
COMMENT ON COLUMN flow_depth_values.station IS 'River km' ;
COMMENT ON COLUMN flow_depth_values.depth IS 'Flow depth of a river station computed for a waterlevel in  m' ;

CREATE SEQUENCE FLOW_DEPTH_VALUES_ID_SEQ ;


--Depth evolution

CREATE TABLE depth_evolution (
    id  NUMBER(9,0) PRIMARY KEY,
    river_id  NUMBER(38,0) NOT NULL CONSTRAINT cDepthEvolutionRivers REFERENCES rivers(id) ON DELETE CASCADE,
    reference_year  NUMBER(4,0) NOT NULL,
    start_year  NUMBER(4,0) NOT NULL,
    curr_sounding  VARCHAR2(64) NOT NULL,
    old_sounding  VARCHAR2(64) NOT NULL,
    kmrange_info  VARCHAR2(32),
    curr_glw  VARCHAR2(64) NOT NULL,
    old_glw  VARCHAR2(64) NOT NULL,
    filename  VARCHAR2(256) NOT NULL,
    groupname  VARCHAR2(32) NOT NULL CHECK(groupname IN ('aktuell','etappe')),
    notes  VARCHAR2(256)
);
COMMENT ON TABLE depth_evolution IS 'Longitudinal section of the evolution of the flow depth of a river' ;
COMMENT ON COLUMN depth_evolution.reference_year IS 'File header line info "Bezugsjahr" (GlW)' ;
COMMENT ON COLUMN depth_evolution.start_year IS 'File header line info "Ausgangsjahr" (GlW)' ;
COMMENT ON COLUMN depth_evolution.curr_sounding IS 'File header line info "Aktuelle Peilung / Epoche"' ;
COMMENT ON COLUMN depth_evolution.old_sounding IS 'File header line info "Historische Peilung / Epoche"' ;
COMMENT ON COLUMN depth_evolution.kmrange_info IS 'File header line info "Strecke"' ;
COMMENT ON COLUMN depth_evolution.curr_glw IS 'File header line info "Aktuelle Wasserspiegellage"' ;
COMMENT ON COLUMN depth_evolution.old_glw IS 'File header line info "Historische Wasserspiegellage"' ;
COMMENT ON COLUMN depth_evolution.filename IS 'Name without type extension of the imported file' ;
COMMENT ON COLUMN depth_evolution.groupname IS 'Group the file belongs to' ;
COMMENT ON COLUMN depth_evolution.notes IS 'File header line info "weitere Bemerkungen"' ;

CREATE SEQUENCE DEPTH_EVOLUTION_ID_SEQ ;


CREATE TABLE depth_evolution_values (
    id  NUMBER(9,0) PRIMARY KEY,
    depth_evolution_id  NUMBER(9,0) NOT NULL CONSTRAINT cDepthEvolutionValuesDepthEvol REFERENCES depth_evolution(id) ON DELETE CASCADE,
    station  NUMBER(7,3) NOT NULL,
    total_change  NUMBER(8,4),
    change_per_year  NUMBER(8,4)
);
COMMENT ON TABLE depth_evolution_values IS 'Evolution of the flow depth of a river station in a referenced period of time' ;
COMMENT ON COLUMN depth_evolution_values.station IS 'River km' ;
COMMENT ON COLUMN depth_evolution_values.total_change IS 'Flow depth change of the whole period of time in m' ;
COMMENT ON COLUMN depth_evolution_values.change_per_year IS 'Average yearly flow depth change in the referenced period of time in m' ;

CREATE SEQUENCE DEPTH_EVOLUTION_VALUES_ID_SEQ ;


--Daily discharge

CREATE TABLE daily_discharge (
    id  NUMBER(9,0) PRIMARY KEY,
    gauge_id  NUMBER(38,0) NOT NULL CONSTRAINT cDailyDischargeGauge REFERENCES gauges(id) ON DELETE CASCADE,
    filename  VARCHAR2(256) NOT NULL
);

COMMENT ON TABLE daily_discharge IS 'Daily discharge value series' ;
COMMENT ON COLUMN daily_discharge.filename IS 'Name without type extension of the imported file' ;

CREATE SEQUENCE DAILY_DISCHARGE_ID_SEQ ;


CREATE TABLE daily_discharge_values (
    id  NUMBER(9,0) PRIMARY KEY,
    daily_discharge_id  NUMBER(9,0) NOT NULL CONSTRAINT cDailyDischargeValueDailyDisch REFERENCES daily_discharge(id) ON DELETE CASCADE,
    day  DATE NOT NULL,
    discharge  NUMBER(8,3) NOT NULL
);

COMMENT ON TABLE daily_discharge_values IS 'Mean daily discharge of a date and a gauge' ;
COMMENT ON COLUMN daily_discharge_values.day IS 'Date of the discharge value' ;
COMMENT ON COLUMN daily_discharge_values.discharge IS 'Mean daily discharge in m^3/s' ;

CREATE SEQUENCE DAILY_DISCHARGE_VALUES_ID_SEQ ;


--Salix line

CREATE TABLE salix (
    id  NUMBER(9,0) PRIMARY KEY,
    river_id  NUMBER(38,0) NOT NULL CONSTRAINT cSalixRivers REFERENCES rivers(id) ON DELETE CASCADE,
    evaluation_by  VARCHAR2(255),
    kmrange_info  VARCHAR2(32),
    filename  VARCHAR2(256) NOT NULL,
    notes  VARCHAR2(256)
);

COMMENT ON TABLE salix IS 'Longitudinal section of the salix line parameters of a river' ;
COMMENT ON COLUMN salix.evaluation_by IS 'File header line info "Auswerter"' ;
COMMENT ON COLUMN salix.kmrange_info IS 'File header line info "Strecke"' ;
COMMENT ON COLUMN salix.filename IS 'Name without type extension of the imported file' ;
COMMENT ON COLUMN salix.notes IS 'File header line info "weitere Bemerkungen"' ;

CREATE SEQUENCE SALIX_ID_SEQ ;


CREATE TABLE salix_values (
    id  NUMBER(9,0) PRIMARY KEY,
    salix_id  NUMBER(9,0) NOT NULL CONSTRAINT cSalixValuesSalix REFERENCES salix(id) ON DELETE CASCADE,
    station  NUMBER(7,3) NOT NULL,
    factor  NUMBER(6,2) NOT NULL,
    mnw_mw_diff  NUMBER(6,2)
);

COMMENT ON COLUMN salix_values.station IS 'River km' ;
COMMENT ON COLUMN salix_values.factor IS 'Salix "factor" of the station in m' ;
COMMENT ON COLUMN salix_values.mnw_mw_diff IS 'Difference between MNW and MW in m (less than zero)' ;

CREATE SEQUENCE SALIX_VALUES_ID_SEQ ;


CREATE TABLE salix_rank (
    id  NUMBER(9,0) PRIMARY KEY,
    min_value  NUMBER(6,2),
    max_value  NUMBER(6,2),
    name  VARCHAR2(16) NOT NULL
);
INSERT INTO salix_rank (id, min_value, max_value, name) VALUES (1, NULL, -0.3, 'invalid');
INSERT INTO salix_rank (id, min_value, max_value, name) VALUES (2, -0.3, 0.3, 'very good');
INSERT INTO salix_rank (id, min_value, max_value, name) VALUES (3, 0.3, 0.5, 'good');
INSERT INTO salix_rank (id, min_value, max_value, name) VALUES (4, 0.5, 1.0, 'moderate');
INSERT INTO salix_rank (id, min_value, max_value, name) VALUES (5, 1.0, 1.5, 'bad');
INSERT INTO salix_rank (id, min_value, max_value, name) VALUES (6, 1.5, NULL, 'very bad');


--Vegetation zone

CREATE TABLE vegetation_type (
    id  NUMBER(9,0) PRIMARY KEY,
    name  VARCHAR2(256) NOT NULL
);
INSERT INTO vegetation_type (id, name) VALUES (1, 'zonal forest');
INSERT INTO vegetation_type (id, name) VALUES (2, 'dry hartwood forest floodplain');
INSERT INTO vegetation_type (id, name) VALUES (3, 'wet hartwood forest floodplain');
INSERT INTO vegetation_type (id, name) VALUES (4, 'salix alba forest');
INSERT INTO vegetation_type (id, name) VALUES (5, 'salix shrubs');
INSERT INTO vegetation_type (id, name) VALUES (6, 'reed bed');
INSERT INTO vegetation_type (id, name) VALUES (7, 'bank pioneers');
INSERT INTO vegetation_type (id, name) VALUES (8, 'no vegetation');
INSERT INTO vegetation_type (id, name) VALUES (9, 'water');


CREATE TABLE vegetation (
    id  NUMBER(9,0) PRIMARY KEY,
    river_id  NUMBER(38,0) NOT NULL CONSTRAINT cVegetationRivers REFERENCES rivers(id) ON DELETE CASCADE,
    name  VARCHAR2(256),
    filename  VARCHAR2(256) NOT NULL,
    notes  VARCHAR2(256)
);

COMMENT ON TABLE vegetation IS 'Classification of the vegetation of a river' ;
COMMENT ON COLUMN vegetation.name IS 'File header line info "Einteilung"' ;
COMMENT ON COLUMN vegetation.filename IS 'Name without type extension of the imported file' ;
COMMENT ON COLUMN vegetation.notes IS 'File header line info "weitere Bemerkungen"' ;

CREATE SEQUENCE VEGETATION_ID_SEQ ;


CREATE TABLE vegetation_zone (
    id  NUMBER(9,0) PRIMARY KEY,
    vegetation_id  NUMBER(9,0) NOT NULL CONSTRAINT cVegetationZoneVegetation REFERENCES vegetation(id) ON DELETE CASCADE,
    vegetation_type_id  NUMBER(9,0) NOT NULL CONSTRAINT cVegetationZoneVegetationType REFERENCES vegetation_type(id) ON DELETE CASCADE,
    min_overflow_days  NUMBER(3,0) NOT NULL,
    max_overflow_days  NUMBER(3,0) NOT NULL,
    color_r  NUMBER(3,0) NOT NULL,
    color_g  NUMBER(3,0) NOT NULL,
    color_b  NUMBER(3,0) NOT NULL
);

COMMENT ON TABLE vegetation_zone IS 'Vegetation zone of a river station' ;
COMMENT ON COLUMN vegetation_zone.min_overflow_days IS 'Minimum number (inclusive) of overflow days in a year for the zone type' ;
COMMENT ON COLUMN vegetation_zone.max_overflow_days IS 'Maximum number (inclusive) of overflow days in a year for the zone type' ;
COMMENT ON COLUMN vegetation_zone.color_r IS 'Red value (0-255) of the zone color' ;
COMMENT ON COLUMN vegetation_zone.color_g IS 'Green value (0-255) of the zone color' ;
COMMENT ON COLUMN vegetation_zone.color_b IS 'Blue value (0-255) of the zone color' ;

CREATE SEQUENCE VEGETATION_ZONE_ID_SEQ ;

http://dive4elements.wald.intevation.org