hydroanomaly 1.0.0__py3-none-any.whl → 1.2.1__py3-none-any.whl

hydroanomaly/__init__.py

@@ -5,17 +5,20 @@ A simple Python package with just 3 modules:
 1. USGS turbidity data retrieval (returns data and site coordinates)
 2. Sentinel satellite bands retrieval  
 3. Time series visualization
+4. Machine learning anomaly detection (One-Class SVM and Isolation Forest)
 
 That's it - nothing else!
 """
 
-__version__ = "1.0.0"
+__version__ = "1.2.1"
 __author__ = "Ehsan Kahrizi (Ehsan.kahrizi@usu.edu)"
 
 # Import the 3 simple modules
 from .usgs_turbidity import get_turbidity, get_usgs_turbidity
 from .sentinel_bands import get_sentinel_bands, get_satellite_data, get_sentinel, get_sentinel_bands_gee, show_sentinel_ndwi_map
 from .visualize import plot_timeseries, plot_turbidity, plot_sentinel, plot_comparison, plot, visualize
+from .ml import run_oneclass_svm, run_isolation_forest
+
 
 # Export everything
 __all__ = [
@@ -30,14 +33,17 @@ __all__ = [
     'get_sentinel',
     'show_sentinel_ndwi_map',
 
-
     # Visualization functions
     'plot_timeseries',
     'plot_turbidity',
     'plot_sentinel', 
     'plot_comparison',
     'plot',
-    'visualize'
+    'visualize',
+    
+    # Machine learning functions
+    'run_oneclass_svm',
+    'run_isolation_forest'  
 ]
 
 print(f"HydroAnomaly v{__version__} - Simple Water Data Package")

hydroanomaly/ml.py

@@ -0,0 +1,170 @@
+import numpy as np
+import pandas as pd
+from sklearn.preprocessing import StandardScaler
+from sklearn.svm import OneClassSVM
+from sklearn.ensemble import IsolationForest
+from sklearn.metrics import f1_score, recall_score, precision_score
+import matplotlib.pyplot as plt
+
+# ============= Helper Functions =========================================================================
+def match_nearest(row, usgs):
+    target_time = row['datetime']
+    same_day = usgs[usgs['datetime'] == target_time.datetime()]
+    if same_day.empty:
+        return np.nan
+    delta = (same_day['datetime'] - target_time).abs()
+    return same_day.loc[delta.idxmin(), 'turbidity']
+
+
+
+# ============= Preprocessing and Feature Engineering ========================================================
+def preprocess_data(sentinel, usgs):
+    # Add matched turbidity
+    sentinel['turbidity'] = sentinel.apply(lambda row: match_nearest(row, usgs), axis=1)
+    df = sentinel.dropna(subset=['turbidity'])
+
+    # Water pixels filtering
+    if 'SCL' in df.columns and (df['SCL'] == 6).sum() > 0:
+        df = df[df['SCL'] == 6].drop_duplicates(subset=['B2', 'B3', 'B4'])
+
+    # Feature engineering
+    bands = ['B2','B3','B4','B5','B6','B7','B8','B8A','B9','B11','B12']
+    df['NDVI'] = (df['B8'] - df['B4']) / (df['B8'] + df['B4'])
+    df['NDWI'] = (df['B3'] - df['B8']) / (df['B3'] + df['B8'])
+    df['NDSI'] = (df['B3'] - df['B11']) / (df['B3'] + df['B11'])
+
+    df = df.sort_values('datetime').reset_index(drop=True)
+    df['turbidity_diff1'] = df['turbidity'].diff()
+    df['turbidity_diff2'] = df['turbidity_diff1'].diff()
+    thresh = 2 * df['turbidity_diff2'].std()
+    df['spike'] = (df['turbidity_diff2'].abs() > thresh).astype(int)
+    df = df.dropna()
+
+    # Class label
+    df['Classe'] = (df['turbidity'] > 20).astype(int)
+    return df, bands
+
+
+# ============= Anomaly Detection Methods ================================================================
+def run_oneclass_svm(sentinel, usgs, plot=True):
+    """
+    Apply One-Class SVM anomaly detection on Sentinel/USGS data.
+    Returns: DataFrame with predictions, and best model parameters.
+    """
+    df, bands = preprocess_data(sentinel, usgs)
+    features = bands + ['NDVI','NDWI','NDSI','turbidity_diff1','turbidity_diff2','spike']
+    X = df[features].fillna(df[features].mean()).values
+    y = df['Classe'].values
+
+    scaler = StandardScaler()
+    X_scaled = scaler.fit_transform(X)
+
+    X_class0 = X_scaled[y == 0]
+    X_class1 = X_scaled[y == 1]
+
+    train_size = max(1, int(0.8 * len(X_class0)))
+    X_train = X_class0[:train_size]
+    X_test = np.vstack([X_class0[train_size:], X_class1])
+    y_test = np.array([0]*(len(X_class0)-train_size) + [1]*len(X_class1))
+
+    best_f1 = -1
+    best_model, best_y_pred, best_params = None, None, None
+    for gamma in ['auto', 'scale']:
+        for nu in [0.01, 0.05, 0.1, 0.2]:
+            model = OneClassSVM(kernel='rbf', gamma=gamma, nu=nu)
+            model.fit(X_train)
+            y_pred = np.where(model.predict(X_test) == 1, 0, 1)
+            if len(np.unique(y_pred)) > 1:
+                f1 = f1_score(y_test, y_pred)
+                if f1 > best_f1:
+                    best_f1 = f1
+                    best_model = model
+                    best_y_pred = y_pred
+                    best_params = {'gamma': gamma, 'nu': nu}
+
+    if best_f1 > -1:
+        df_out = df.iloc[-len(y_test):].copy()
+        df_out['predicted'] = best_y_pred
+        if plot:
+            plt.figure(figsize=(15,6))
+            plt.plot(df_out['datetime'], df_out['turbidity'], label='turbidity', color='blue')
+            plt.scatter(df_out[df_out['Classe']==1]['datetime'], df_out[df_out['Classe']==1]['turbidity'],
+                        color='red', marker='x', label='True Anomaly', s=100)
+            plt.scatter(df_out[df_out['predicted']==1]['datetime'], df_out[df_out['predicted']==1]['turbidity'],
+                        edgecolors='orange', facecolors='none', marker='o', label='Predicted Anomaly', s=80)
+            plt.title("True vs Predicted Anomalies (OneClassSVM)")
+            plt.xlabel("datetime")
+            plt.ylabel("turbidity")
+            plt.legend()
+            plt.grid(True)
+            plt.tight_layout()
+            plt.show()
+        return df_out, best_params, best_f1
+    else:
+        print("Could not find a good model. Try different hyperparameters.")
+        return None, None, None
+
+
+# ============= Isolation Forest Method ================================================================
+def run_isolation_forest(sentinel, usgs, plot=True):
+    """
+    Apply Isolation Forest anomaly detection on Sentinel/USGS data.
+    Returns: DataFrame with predictions, and best model parameters.
+    """
+    df, bands = preprocess_data(sentinel, usgs)
+    features = bands + ['NDVI','NDWI','NDSI','turbidity_diff1','turbidity_diff2','spike']
+    X = df[features].fillna(df[features].mean()).values
+    y = df['Classe'].values
+
+    scaler = StandardScaler()
+    X_scaled = scaler.fit_transform(X)
+
+    X_class0 = X_scaled[y == 0]
+    X_class1 = X_scaled[y == 1]
+
+    train_size = max(1, int(0.8 * len(X_class0)))
+    X_train = X_class0[:train_size]
+    X_test = np.vstack([X_class0[train_size:], X_class1])
+    y_test = np.array([0]*(len(X_class0)-train_size) + [1]*len(X_class1))
+
+    best_f1 = -1
+    best_model, best_y_pred, best_params = None, None, None
+    for contamination in [0.01, 0.05, 0.1, 0.15, 0.2, 0.3]:
+        model = IsolationForest(
+            n_estimators=100,
+            contamination=contamination,
+            max_samples='auto',
+            bootstrap=True,
+            random_state=42
+        )
+        model.fit(X_train)
+        y_pred = np.where(model.predict(X_test) == 1, 0, 1)
+        if len(np.unique(y_pred)) > 1:
+            f1 = f1_score(y_test, y_pred)
+            if f1 > best_f1:
+                best_f1 = f1
+                best_model = model
+                best_y_pred = y_pred
+                best_params = {'contamination': contamination}
+
+    if best_f1 > -1:
+        df_out = df.iloc[-len(y_test):].copy()
+        df_out['predicted'] = best_y_pred
+        if plot:
+            plt.figure(figsize=(15,6))
+            plt.plot(df_out['datetime'], df_out['turbidity'], label='turbidity', color='blue')
+            plt.scatter(df_out[df_out['Classe']==1]['datetime'], df_out[df_out['Classe']==1]['turbidity'],
+                        color='red', marker='x', label='True Anomaly', s=100)
+            plt.scatter(df_out[df_out['predicted']==1]['datetime'], df_out[df_out['predicted']==1]['turbidity'],
+                        edgecolors='orange', facecolors='none', marker='o', label='Predicted Anomaly', s=80)
+            plt.title("True vs Predicted Anomalies (Isolation Forest)")
+            plt.xlabel("datetime")
+            plt.ylabel("turbidity")
+            plt.legend()
+            plt.grid(True)
+            plt.tight_layout()
+            plt.show()
+        return df_out, best_params, best_f1
+    else:
+        print("Could not find a good model. Try different hyperparameters.")
+        return None, None, None

hydroanomaly/sentinel_bands.py

@@ -79,14 +79,14 @@ def get_sentinel_bands_gee(
     s2_masked = s2.map(dynamic_scl_mask)
 
     def extract_features(image):
-        date = image.date().format('YYYY-MM-dd HH:mm:ss')
+        datetime = image.date().format('YYYY-MM-dd HH:mm:ss')
         values = image.reduceRegion(
             reducer=ee.Reducer.mean(),
             geometry=buffered_point,
             scale=20,
             maxPixels=1e8
         )
-        return ee.Feature(None, values.set('date', date))
+        return ee.Feature(None, values.set('datetime', datetime))
 
     features = s2_masked.map(extract_features)
     fc = ee.FeatureCollection(features).filter(ee.Filter.notNull(['B2']))
@@ -95,9 +95,9 @@ def get_sentinel_bands_gee(
     rows = [f['properties'] for f in data['features']]
     df = pd.DataFrame(rows)
     if not df.empty:
-        df['date'] = pd.to_datetime(df['date'])
-        df = df.sort_values('date')
-        df = df.set_index('date')
+        df['datetime'] = pd.to_datetime(df['datetime'])
+        df = df.sort_values('datetime')
+        df = df.set_index('datetime')
     return df
 
 

{hydroanomaly-1.0.0.dist-info → hydroanomaly-1.2.1.dist-info}/METADATA

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: hydroanomaly
-Version: 1.0.0
+Version: 1.2.1
 Summary: A Python package for hydro anomaly detection with simple USGS data retrieval
 Author-email: Ehsan Kahrizi <ehsan.kahrizi@usu.edu>
 License: MIT License
@@ -28,7 +28,7 @@ License: MIT License
 Project-URL: Homepage, https://github.com/yourusername/hydroanomaly
 Project-URL: Bug Reports, https://github.com/yourusername/hydroanomaly/issues
 Project-URL: Source, https://github.com/yourusername/hydroanomaly
-Keywords: python,package,hydro,anomaly,detection
+Keywords: python,package,hydrology,anomaly detection,remote sensing
 Classifier: Programming Language :: Python :: 3
 Classifier: Operating System :: OS Independent
 Requires-Python: >=3.6

hydroanomaly-1.2.1.dist-info/RECORD

@@ -0,0 +1,10 @@
+hydroanomaly/__init__.py,sha256=HgiUaNbWp_hAHaOvzCJXRBDwe89orkscx3Sd7lvgnQo,1664
+hydroanomaly/ml.py,sha256=X2bYinNwRzQz0yFtRSzmN8OIZaX7vfT4BPSE6KMoYAM,7066
+hydroanomaly/sentinel_bands.py,sha256=XdpXUsJ8VeRQp9akDeQaVBefuuMrQIabslu8tg_FTpk,5399
+hydroanomaly/usgs_turbidity.py,sha256=k0cXRXpTe1YgjfR0Htw77SLD8hM--43jiEiJwx1vRg0,5664
+hydroanomaly/visualize.py,sha256=d_Ou1sTr648TdAW-94NXwNbLPL4rvYVYb5pw4Xux3aE,7228
+hydroanomaly-1.2.1.dist-info/licenses/LICENSE,sha256=OphKV48tcMv6ep-7j-8T6nycykPT0g8ZlMJ9zbGvdPs,1066
+hydroanomaly-1.2.1.dist-info/METADATA,sha256=keF0Y92CYoZxOnU9vYaGfnefWr8s9mN7ncA2Qn2bp4I,12981
+hydroanomaly-1.2.1.dist-info/WHEEL,sha256=_zCd3N1l69ArxyTb8rzEoP9TpbYXkqRFSNOD5OuxnTs,91
+hydroanomaly-1.2.1.dist-info/top_level.txt,sha256=t-5Lc-eTLlkxIhR_N1Cpp6_YZafKS3xLLk9D2CtbE7o,13
+hydroanomaly-1.2.1.dist-info/RECORD,,

hydroanomaly-1.0.0.dist-info/RECORD

@@ -1,9 +0,0 @@
-hydroanomaly/__init__.py,sha256=PTaMZkFNuJnuA9v6tPIll0NoELl3AKJqpi5R-Y_kW9A,1442
-hydroanomaly/sentinel_bands.py,sha256=Y6RAunVJDYLs13WemSSQNEu07GqmhR64fC2mLPxwh2k,5371
-hydroanomaly/usgs_turbidity.py,sha256=k0cXRXpTe1YgjfR0Htw77SLD8hM--43jiEiJwx1vRg0,5664
-hydroanomaly/visualize.py,sha256=d_Ou1sTr648TdAW-94NXwNbLPL4rvYVYb5pw4Xux3aE,7228
-hydroanomaly-1.0.0.dist-info/licenses/LICENSE,sha256=OphKV48tcMv6ep-7j-8T6nycykPT0g8ZlMJ9zbGvdPs,1066
-hydroanomaly-1.0.0.dist-info/METADATA,sha256=yij922cCRRXtvyel5t1r45B5pNkg71h8jufNu07T-14,12962
-hydroanomaly-1.0.0.dist-info/WHEEL,sha256=_zCd3N1l69ArxyTb8rzEoP9TpbYXkqRFSNOD5OuxnTs,91
-hydroanomaly-1.0.0.dist-info/top_level.txt,sha256=t-5Lc-eTLlkxIhR_N1Cpp6_YZafKS3xLLk9D2CtbE7o,13
-hydroanomaly-1.0.0.dist-info/RECORD,,