def test_amax_long_records(self): catchment = Catchment("Aberdeen", "River Dee") catchment.amax_records = [AmaxRecord(date(1999, 12, 31), 5.0, 0.5), AmaxRecord(date(2000, 12, 31), 1.0, 0.5), AmaxRecord(date(2001, 12, 31), 4.0, 0.5), AmaxRecord(date(2002, 12, 31), 2.0, 0.5), AmaxRecord(date(2003, 12, 31), 3.0, 0.5)] self.assertEqual(QmedAnalysis(catchment).qmed(method='amax_records'), 3)
def test_best_method_amax_over_pot(self): catchment = Catchment("Aberdeen", "River Dee") catchment.amax_records = [AmaxRecord(date(1999, 12, 31), 2.0, 0.5), AmaxRecord(date(2000, 12, 31), 1.0, 0.5)] catchment.pot_dataset = PotDataset(start_date=date(1999, 1, 1), end_date=date(1999, 12, 31)) catchment.pot_dataset.pot_records = [PotRecord(date(1999, 1, 1), 3.0, 0.5), PotRecord(date(1999, 2, 1), 2.0, 0.5), PotRecord(date(1999, 12, 31), 1.0, 0.5)] self.assertAlmostEqual(catchment.qmed(), 1.5)
def test_record_start_end(self): catchment = Catchment("Aberdeen", "River Dee") catchment.amax_records = [ AmaxRecord(date(1999, 12, 31), 3.0, 0.5), AmaxRecord(date(2000, 12, 31), 2.0, 0.5), AmaxRecord(date(2001, 12, 31), 1.0, 0.5) ] self.assertEqual(catchment.amax_records_start(), 1999) self.assertEqual(catchment.amax_records_end(), 2001)
def test_dimensionless_flows(self): analysis = GrowthCurveAnalysis(self.catchment) self.catchment.amax_records = [ AmaxRecord(date(1999, 12, 31), 3.0, 0.5), AmaxRecord(date(2000, 12, 31), 2.0, 0.5), AmaxRecord(date(2001, 12, 31), 1.0, 0.5) ] result = analysis._dimensionless_flows(self.catchment) expected = np.array([1.5, 1, 0.5]) assert_almost_equal(result, expected)
def test_best_method_order(self): catchment = Catchment("Aberdeen", "River Dee") catchment.channel_width = 1 catchment.amax_records = [AmaxRecord(date(1999, 12, 31), 1.0, 0.5), AmaxRecord(date(2000, 12, 31), 1.0, 0.5)] catchment.descriptors = Descriptors(dtm_area=1, bfihost=0.50, sprhost=50, saar=1000, farl=1) self.assertEqual(catchment.qmed(), 1.0)
def test_add_catchment_with_amax(self): catchment = Catchment("Aberdeen", "River Dee") catchment.amax_records = [ AmaxRecord(date(1999, 12, 31), 3.0, 0.5), AmaxRecord(date(2000, 12, 31), 2.0, 0.5), AmaxRecord(date(2001, 12, 31), 1.0, 0.5) ] self.db_session.add(catchment) result = self.db_session.query(Catchment).filter_by( location="Aberdeen", watercourse="River Dee").one() self.assertEqual(catchment, result) self.assertEqual(catchment.amax_records, result.amax_records)
def test_all(self): catchment = Catchment("Aberdeen", "River Dee") catchment.channel_width = 1 catchment.amax_records = [AmaxRecord(date(1999, 12, 31), 1.0, 0.5), AmaxRecord(date(2000, 12, 31), 1.0, 0.5)] catchment.descriptors = Descriptors(dtm_area=1, bfihost=0.50, sprhost=50, saar=1000, farl=1, urbext2000=0) qmeds = QmedAnalysis(catchment).qmed_all_methods() self.assertEqual(qmeds['amax_records'], 1) self.assertEqual(qmeds['channel_width'], 0.182) self.assertEqual(qmeds['area'], 1.172) self.assertAlmostEqual(qmeds['descriptors_1999'], 0.2671, 4)
def OnSave(self,event): self.p.temporary_amax_records = [] user_enterer_data_series = self.data_series_entry.GetValue() date_column = int(self.date_column.GetValue())-1 stage_column = int(self.stage_data_column.GetValue())-1 separator = self.delimitor.GetValue() flow_column = int(self.flow_data_column.GetValue())-1 lines = user_enterer_data_series.split('\n') year = 1900 for line in lines: data_entry = line.split(separator) year = year +1 stage_value = -9999.9 flow_value = -9999.9 if len(data_entry) > flow_column: flow_value=float(data_entry[flow_column]) if len(data_entry) > stage_column: stage_value = float(data_entry[stage_column]) record = AmaxRecord(date=date(year,1,1), flow=flow_value, stage=stage_value) self.p.temporary_amax_records.append(record) self.p.refreshAmaxTable() self.Destroy()
class TestQmedDonor(unittest.TestCase): catchment = Catchment("Dundee", "River Tay") catchment.country = 'gb' catchment.descriptors = Descriptors(dtm_area=2.345, bfihost=1e-4, sprhost=100, saar=2000, farl=0.5, urbext2000=0, centroid_ngr=Point(276125, 688424)) # QMED descr = 0.61732109 donor_catchment = Catchment("Aberdeen", "River Dee") donor_catchment.country = 'gb' donor_catchment.descriptors = Descriptors(dtm_area=1, bfihost=0.50, sprhost=50, saar=1000, farl=1, urbext2000=0, centroid_ngr=Point(276125, 688424)) donor_catchment.amax_records = [AmaxRecord(date(1999, 12, 31), 1.0, 0.5), AmaxRecord(date(2000, 12, 31), 1.0, 0.5)] # donor QMED descr = 0.59072777 # donor QMED amax = 1.0 @classmethod def setUpClass(cls): settings.config['nrfa']['oh_json_url'] = \ 'file:' + pathname2url(os.path.abspath('./floodestimation/fehdata_test.json')) cls.db_session = db.Session() @classmethod def tearDownClass(cls): db.empty_db_tables() def tearDown(self): self.db_session.rollback() def test_donor_adjustment_factor(self): # 1.0/0.59072777 = 1.69282714 self.assertAlmostEqual(exp(QmedAnalysis(self.catchment). _lnqmed_residual(self.donor_catchment)), 1.69282714) def test_lnqmed_residual_one_donor(self): # ln(1.0 / 0.59072777) self.assertAlmostEqual(QmedAnalysis(self.catchment). _lnqmed_residual(self.donor_catchment), 0.5264, 4) def test_model_error_corr(self): # because we're at zero distance, error correlation = 1 self.assertAlmostEqual(QmedAnalysis(self.catchment). _model_error_corr(self.catchment, self.donor_catchment), 1) def test_vector_b_one_donor(self): assert_almost_equal(QmedAnalysis(self.catchment). _vec_b([self.donor_catchment]), [0.1175]) def test_matrix_sigma_eta_one_donor(self): result = QmedAnalysis(self.catchment)._matrix_sigma_eta([self.donor_catchment]) assert_almost_equal(result, [[0.1175]]) def test_beta_one_donor(self): result = QmedAnalysis(self.catchment)._beta(self.donor_catchment) self.assertAlmostEqual(result, 0.30242554) def test_matrix_sigma_eps_one_donor(self): # 4 * 0.30242554**2 / 2 = 0.18292242 result = QmedAnalysis(self.catchment)._matrix_sigma_eps([self.donor_catchment]) assert_almost_equal(result, [[0.18292242]]) def test_matrix_omega_one_donor(self): # 0.1175 + 0.18292242 = 0.30042242 result = QmedAnalysis(self.catchment)._matrix_omega([self.donor_catchment]) assert_almost_equal(result, [[0.30042242]]) def test_vector_alpha_one_donor(self): # 1/0.30042242 * 0.1175 = 0.39111595 result = QmedAnalysis(self.catchment)._vec_alpha([self.donor_catchment]) assert_almost_equal(result, [0.39111595]) def test_qmed_one_donor(self): # 0.61732109 * 1.69282714**0.39111595 = 0.75844685 result = QmedAnalysis(self.catchment).qmed(method='descriptors', donor_catchments=[self.donor_catchment]) self.assertAlmostEqual(result, 0.75844685, places=4) def test_distance_two_donors(self): analysis = QmedAnalysis(self.catchment, CatchmentCollections(self.db_session), year=2000) donors = analysis.find_donor_catchments()[0:2] # 17001, 10001 result = [d.distance_to(self.catchment) for d in donors] assert_almost_equal(result, [5, 183.8515], decimal=4) def test_lnqmed_residuals_two_donors(self): analysis = QmedAnalysis(self.catchment, CatchmentCollections(self.db_session), year=2000) donors = analysis.find_donor_catchments()[0:2] # 17001, 10001 qmed_amax = [QmedAnalysis(d).qmed() for d in donors] qmed_descr =[QmedAnalysis(d, year=2000).qmed(method='descriptors') for d in donors] assert_almost_equal(qmed_amax, [90.532, 50.18]) # not verified assert_almost_equal(qmed_descr, [51.73180402, 48.70106637]) # not verified result = [analysis._lnqmed_residual(d) for d in donors] assert_almost_equal(result, [0.55963062, 0.02991561]) def test_model_error_corr_two_donors(self): analysis = QmedAnalysis(self.catchment, CatchmentCollections(self.db_session), year=2000) donors = analysis.find_donor_catchments()[0:2] # 17001, 10001 result = [analysis._model_error_corr(self.catchment, d) for d in donors] assert_almost_equal(result, [0.352256808, 0.002198921]) def test_model_error_corr_between_two_donors(self): analysis = QmedAnalysis(self.catchment, CatchmentCollections(self.db_session), year=2000) donors = analysis.find_donor_catchments()[0:2] # 17001, 10001 dist = donors[0].distance_to(donors[1]) self.assertAlmostEqual(dist, 188.8487072) # not verified result = analysis._model_error_corr(donors[0], donors[1]) self.assertAlmostEqual(result, 0.001908936) def test_vector_b_two_donors(self): analysis = QmedAnalysis(self.catchment, CatchmentCollections(self.db_session), year=2000) donors = analysis.find_donor_catchments()[0:2] # 17001, 10001 # [0.352256808, 0.002198921] * 0.1175 = [0.041390175, 0.000258373] result = analysis._vec_b(donors) assert_almost_equal(result, [0.041390175, 0.000258373]) def test_matrix_sigma_eta_two_donors(self): analysis = QmedAnalysis(self.catchment, CatchmentCollections(self.db_session), year=2000) donors = analysis.find_donor_catchments()[0:2] # 17001, 10001 result = analysis._matrix_sigma_eta(donors) # 0.1175 * 0.001908936 = 0.00022430 assert_almost_equal(result, [[0.1175, 0.00022430], [0.00022430, 0.1175]]) def test_beta_two_donors(self): analysis = QmedAnalysis(self.catchment, CatchmentCollections(self.db_session), year=2000) donors = analysis.find_donor_catchments()[0:2] # 17001, 10001 result = [analysis._beta(d) for d in donors] assert_almost_equal(result, [0.16351290, 0.20423656]) def test_lnqmed_corr(self): analysis = QmedAnalysis(self.catchment, CatchmentCollections(self.db_session), year=2000) donors = analysis.find_donor_catchments()[0:2] # 17001, 10001 results = analysis._lnqmed_corr(donors[0], donors[1]) self.assertAlmostEqual(results, 0.133632774) def test_matrix_sigma_eps_two_donors(self): analysis = QmedAnalysis(self.catchment, CatchmentCollections(self.db_session), year=2000) donors = analysis.find_donor_catchments()[0:2] # 17001, 10001 record0 = [donors[0].amax_records_start(), donors[0].amax_records_end()] record1 = [donors[1].amax_records_start(), donors[1].amax_records_end()] self.assertEqual(record0, [1969, 2005]) # n=37 self.assertEqual(record1, [1939, 1984]) # n=46, 16 years overlapping result = analysis._matrix_sigma_eps(donors) # 4 * 0.16351290**2 / 37 = 0.00289043 # 4 * 0.16351290 * 0.20423656 * 16 / 37 / 46 * 0.133632774 = 0.00016781 # 4 * 0.20423656**2 / 46 = 0.00362718 assert_almost_equal(result, [[0.00289043, 0.00016781], [0.00016781, 0.00362718]]) def test_matrix_omega_two_donors(self): analysis = QmedAnalysis(self.catchment, CatchmentCollections(self.db_session), year=2000) donors = analysis.find_donor_catchments()[0:2] # 17001, 10001 result = analysis._matrix_omega(donors) assert_almost_equal(result, [[0.12039043, 0.00039211], [0.00039211, 0.12112718]]) def test_vector_alpha_two_donors(self): analysis = QmedAnalysis(self.catchment, CatchmentCollections(self.db_session), year=2000) donors = analysis.find_donor_catchments()[0:2] # 17001, 10001 result = analysis._vec_alpha(donors) assert_almost_equal(result, [0.34379622, 0.00102012]) # calculated in Excel def test_qmed_two_donors(self): analysis = QmedAnalysis(self.catchment, CatchmentCollections(self.db_session), year=2000) donors = analysis.find_donor_catchments()[0:2] # 17001, 10001 result = analysis.qmed(method='descriptors', donor_catchments=donors) # exp(ln(0.61732109) + 0.34379622 * 0.55963062 + 0.00102012 * 0.02991561) = # exp(ln(0.61732109) + 0.192429411) = 0.748311028 self.assertAlmostEqual(result, 0.748311028, places=5)
def test_best_method_amax(self): catchment = Catchment("Aberdeen", "River Dee") catchment.amax_records = [AmaxRecord(date(1999, 12, 31), 1.0, 0.5), AmaxRecord(date(2000, 12, 31), 1.0, 0.5)] self.assertEqual(catchment.qmed(), 1.0)
def test_amax_rejected_record(self): catchment = Catchment("Aberdeen", "River Dee") catchment.amax_records = [AmaxRecord(date(1999, 12, 31), 2.0, 0.5), AmaxRecord(date(2000, 12, 31), 1.0, 0.5), AmaxRecord(date(2000, 12, 31), 99.0, 0.5, flag=2)] self.assertEqual(QmedAnalysis(catchment).qmed(method='amax_records'), 1.5)