IoT | Die TensorFlow-Wetterstation sagt die Niederschlagsintensität voraus

⭐ Geben Sie die Tabellenkalkulations-ID und den Bereich ein, um die neuesten Ergebnisse der Wetterstation an die Tabellenkalkulation anzuhängen.

https://docs.google.com/spreadsheets/d/spreadsheetId/edit#gid=0

// Geben Sie Ihre SpreadsheetId ein:
$spreadsheetId ='';
// Geben Sie den Bereich (die erste Zeile) ein, unter dem neue Werte angehängt werden (8 Zeilen) :
$range ='A1:H1';
// Aktuelle Ergebnisse der Wetterstation an die Tabelle anhängen.
$values ​​=[
 [$variables_from_module["wd" ], $variables_from_module["a_ws"], $variables_from_module["m_ws"], $variables_from_module["1_rf"], $variables_from_module["24_rf"], $variables_from_module["tem"], $variables_from_module["hum"], $variables_from_module["b_pr"]]
];
$body =new Google_Service_Sheets_ValueRange([
 'values' => $values
]);
$params =[
 'valueInputOption' => "RAW"
];

...

 $result =$service->spreadsheets_values->append( $spreadsheetId, $range, $body, $params);
 printf("

%d Zellen angehängt.", $result->getUpdates()->getUpdatedCells()); 

💻 account_verification_token.php

⭐ Nach Autorisierung der Webanwendung durch die erstellte AuthUrl , geben Sie den angegebenen Bestätigungscode ein, um ein Zugriffstoken festzulegen.

$account_verification_token =""; // Geben Sie den Bestätigungscode nach der Autorisierung ein. 

Schritt 3.1:Einrichten der Webanwendung auf dem Raspberry Pi (oder einem beliebigen Server)

Nachdem ich die PHP-Webanwendung erstellt hatte, entschied ich mich, sie auf meinem Raspberry Pi auszuführen, aber Sie können die Anwendung auf jedem Server ausführen, solange es sich um einen PHP-Server handelt.

Wenn Sie Raspberry Pi verwenden möchten, aber nicht wissen, wie Sie einen LAMP-Webserver auf dem Raspberry Pi einrichten, können Sie sich dieses Tutorial ansehen.

⭐ Verschieben Sie zunächst den Anwendungsordner (remote_weather_station) mit dem Terminal auf den Apache-Server (/var/www/html), da der Apache-Server ein geschützter Ort ist.

sudo mv /home/pi/Downloads/remote_weather_station /var/www/html/

⭐ Die Webanwendung gibt beim ersten Öffnen aufgrund des fehlenden Verifizierungscodes einen Fehler aus. Um den Bestätigungscode zu erhalten, rufen Sie den von der Anwendung generierten Autorisierungslink auf.

⭐ Klicken Sie dann auf Gehe zur Remote-Wetterstation (unsicher) .

⭐ Gewähren Sie der Anwendung die erforderlichen Berechtigungen.

⭐ Kopieren Sie nun den Bestätigungscode und fügen Sie ihn in account_verification_token.php ein .

⭐ Kehren Sie zur Webanwendung zurück. Es sollte die Bestätigungsmeldung angezeigt werden:Erfolgreich! Aktualisieren Sie die Seite .

⭐ Nach dem Aktualisieren der Seite verwendet die Webanwendung den Verifizierungscode, um ein Zugriffstoken zu erhalten und speichert das Zugriffstoken in einer Datei namens token.json . Somit wird kein Fehler mehr ausgegeben.

⭐ Um zu testen, ob die PHP-Webanwendung neue Daten korrekt an die angegebene Tabelle anhängt:

http://localhost/remote_weather_station/?wd=12&a_ws=8&m_ws=11&1_rf=14&24_rf=84&tem=24&hum=32&b_pr=57

Schritt 4:Wetterdaten lesen und senden mit ESP8266 (WiFi)

Um Wetterdaten auf meinem Balkon zu sammeln, habe ich ein NodeMCU ESP-12E (ESP8266) Entwicklungsboard mit der Wetterstation verwendet.

Ich habe es so programmiert, dass es alle fünf Minuten Wetterdaten an die PHP-Webanwendung sendet.

⭐ Fügen Sie die erforderlichen Bibliotheken hinzu und definieren Sie die WLAN-Einstellungen.

#include 
#include  
#include 
#include 
 #include 

// Definieren Sie Ihre WLAN-Einstellungen.
const char *ssid ="";
const char *password =" "; 

⭐ Definieren Sie die Wetterstationseinstellungen und die seriellen Anschlusspins - D6, D5.

// Wetterstationseinstellungen definieren:
char databuffer[35];
double temp;
int transfer =0;

// Define die seriellen Anschlusspins - RX und TX.
SoftwareSerial Serial_1(D6, D5); // (Rx, Tx) 

⭐ Im getBuffer() Funktion, Daten von der Wetterstation abrufen.

void getBuffer(){
 int index;
 for (index =0;index <35;index ++){
 if(Serial_1.available()){ 
 databuffer[index] =Serial_1.read();
 if (databuffer[0] !='c'){
 index =-1;
 }
 }
 else{
 index --;
 }
 }
} 

⭐ Debuggen Sie die Wetterdaten und erstellen Sie den Link.

String weather_data ="wd=" + String(WindDirection()) + "&a_ws=" + String(WindSpeedAverage()) + "&m_ws=" + String(WindSpeedMax()) + "&1_rf=" + String(RainfallOneHour()) + "&24_rf=" + String(RainfallOneDay()) + "&tem=" + String(Temperature()) + "&hum=" + String(Feuchtigkeit()) + "&b_pr=" + String( BarPressure());
String server ="http://192.168.1.24/remote_weather_station/?;

...

int WindDirection(){ return transCharToInt(Datenpuffer,1,3); } // Windrichtung (Grad)

float WindSpeedAverage(){ temp =0.44704 * transCharToInt(databuffer,5,7); Rücklauftemperatur; } // Durchschnittliche Fluggeschwindigkeit (1 Minute)

float WindSpeedMax(){ temp =0.44704 * transCharToInt(databuffer,9,11); Rücklauftemperatur; } //Max. Luftgeschwindigkeit (5 Minuten)

float Temperature () {temp =(transCharToInt(databuffer,13,15) - 32.00) * 5.00 / 9.00; Rücklauftemperatur; } // Temperatur ("C")

float RainfallOneHour(){ temp =transCharToInt(databuffer,17,19) * 25.40 * 0.01; Rücklauftemperatur; } // Niederschlag (1 Stunde)

float RainfallOneDay(){ temp =transCharToInt(databuffer,21,23) * 25.40 * 0.01; Rücklauftemperatur; } // Niederschlag (24 Stunden)

int Humidity(){ return transCharToInt(databuffer,25,26); } // Feuchtigkeit (%)

float BarPressure(){ temp =transCharToInt(databuffer,28,32); Rücklauftemperatur / 10,00; } // Barometrischer Druck (hPA) 

⭐ Sende alle 5 Minuten Datenpakete an Raspberry Pi (oder einen beliebigen Server).

übertragen++; Serial.println("Time =>" + String(übertragen) + "s / " + String(int(5*60)) + "s\n\n");
 if(übertragen ==5* 60){
 // Erstellen Sie das HTTP-Objekt, um eine Anfrage an den Server zu senden. 
 HTTPClient http; 
 http.begin(Server + Wetterdaten);
 int httpCode =http.GET(); 
 String payload =http.getString(); 
 Serial.println("Data Send...\nHTTP Code => " + String(httpCode) + "\nServer Response => " + payload + "\n\n"); 
 http.end();
 transferring =0; 
 }
 // Wait 1 second...
 delay(1000); 

⭐ After uploading the code to the NodeMCU ESP-12E (ESP8266) development board, it displays weather data every second on the serial monitor and sends data packets every five minutes (300 seconds) to the PHP web application.

Connection is successful!

...

Weather Data => wd=0&a_ws=0.00&m_ws=0.00&1_rf=0.00&24_rf=0.00&tem=21.67&hum=29&b_pr=1016.70
Buffer => c000s000g000t071r000p000h29b10167*3
Time => 299s / 300s


Weather Data => wd=0&a_ws=0.00&m_ws=0.00&1_rf=0.00&24_rf=0.00&tem=21.67&hum=29&b_pr=1016.70
Buffer => c000s000g000t071r000p000h29b10167*3
Time => 300s / 300s

... 

⭐ Then, it shows the response from the server.

...

Data Send...
HTTP Code => 200
Server Response => Token Found!

8 cells appended.

... 

Step 4.1:Reading and sending weather data with SIM808 (GPRS)

To collect weather data in my backyard, I used a SIM808 shield for Arduino Uno if the distance between the weather station and my router is too far away.

Most of the code is the same and covered in the previous step aside from the parts below.

⭐ Include required libraries and define the sim808 object.

For SIM808 GPS/GPRS/GSM Shield | Download

#include 
#include 

// Define the sim808.
DFRobot_SIM808 sim808(&Serial); 

⭐ Initialize the SIM808 module and continue if it is working accurately.

//******** Initialize sim808 module *************
 while(!sim808.init()) {
 delay(1000);
 Serial.print("Sim808 init error\r\n");
 }
 delay(2000);
 // Continue if the SIM808 Module is working accurately.
 Serial.println("Sim808 init success");
 delay(5000); 

⭐ Send data packets every 5 minutes to the server by establishing a TCP connection to make a GET request.

transferring++; Serial.println("Time => " + String(transferring) + "s / " + String(int(5*60)) + "s\n\n");
 if(transferring ==5*60){
 //*********** Attempt DHCP *******************
 while(!sim808.join(F("cmnet"))) {
 Serial.println("Sim808 join network error!");
 delay(2000);
 }
 //************ Successful DHCP ****************
 delay(5000);
 Serial.println("Successful DHCP");
 //*********** Establish a TCP connection ************
 if(!sim808.connect(TCP,"192.168.1.24", 80)) { // Change it with your server.
 Serial.println("Connection Error");
 }else{
 Serial.println("Connection OK");
 }
 delay(2000);
 
 ... 

⭐ After creating the line string, convert it from string to char array to make an HTTP Get Request with the SIM808.

...
 
 String line ="GET /remote_weather_station/?" + weather_data_1 + weather_data_2 + weather_data_3 + " HTTP/1.0\r\n\r\n";
 Serial.println(line);
 char buffer[512];
 // Convert the line from string to char array to make an HTTP Get Request with the SIM808.
 char web_hook[110];
 String_to_Char(line, 110, web_hook);
 sim808.send(web_hook, sizeof(web_hook)-1);
 while (true) {
 int ret =sim808.recv(buffer, sizeof(buffer)-1);
 if (ret <=0){
 Serial.println("Fetch Over...");
 break; 
 }
 
 ... 

⭐ After uploading the code to the SIM808 shield, it displays weather data every second on the serial monitor and sends data packets every five minutes (300 seconds) to the PHP web application.

Step 4.2:Creating the weather data set for more than three months

After finishing coding, I started to collate weather data on Google Sheets every five minutes for more than three months to build a neural network model and make predictions on the rainfall intensity.

Collected Weather Data:

• Wind Direction (deg)
• Average Wind Speed (m/s)
• Max Wind Speed (m/s)
• One-Hour Rainfall (mm)
• 24-Hours Rainfall (mm)
• Temperature (°C)
• Humidity (%)
• Barometric Pressure (hPa)

Then, I downloaded the spreadsheet as Remote Weather Station.csv , consisting of 32219 rows as my preliminary local weather data set. I am still collating weather data to improve my data set and model :)

Step 5:Building an Artificial Neural Network (ANN) with TensorFlow

When I completed collating my preliminary local weather data set, I started to work on my artificial neural network (ANN) model to make predictions on the rainfall intensity.

I decided to create my neural network model with TensorFlow in Python. So, first of all, I followed the steps below to grasp a better understanding of the weather data:

• Data Visualization
• Data Scaling (Normalizing)
• Data Preprocessing
• Data Splitting

After applying these steps, I decided to use my neural network model to classify different rainfall intensity classes theoretically assigned as labels (outputs) for each input (row). I created my classes according to the rate of precipitation, which depends on the considered time.

The following categories are used to classify rainfall intensity by the rate of precipitation (rainfall):

• Light rain — when the precipitation rate is <2.5 mm per hour
• Moderate rain — when the precipitation rate is between 2.5 mm and 7.6 mm per hour
• Heavy rain — when the precipitation rate is between 7.6 mm and 50 mm per hour
• Violent rain — when the precipitation rate is> 50 mm per hour

According to the precipitation (rainfall) rates, I preprocessed the weather data to assign one of these five classes for each input as its label:

• 0 (None)
• 1 (Light Rain)
• 2 (Moderate Rain)
• 3 (Heavy Rain)
• 4 (Violent Rain)

After scaling (normalizing) and preprocessing the weather data, I elicited seven input variables and one label for each reading, classified with the five mentioned classes. Then, I built an artificial neural network model with TensorFlow to obtain the best possible results and predictions with my preliminary data set.

Layers:

• 7 [Input]
• 16 [Hidden]
• 32 [Hidden]
• 64 [Hidden]
• 128 [Hidden]
• 5 [Output]

To execute all steps above, I created a class named Weather_Station in Python after including the required libraries:

import tensorflow as tf
from tensorflow import keras
import matplotlib.pyplot as plt
import numpy as np
import pandas as pd 

Subsequently, I will discuss coding in Python for each step I mentioned above.

Also, you can download IoT_weather_station_neural_network.py to inspect coding.

Step 5.1:Visualizing and scaling (normalizing) the weather data

Before diving in to build a model, it is important to understand the weather data to pass the model appropriately formatted data.

So, in this step, I will show you how to visualize weather data and scale (normalize) in Python.

⭐ First of all, read the weather data from Remote Weather Station.csv .

csv_path ="E:\PYTHON\Remote Weather Station.csv"
df =pd.read_csv(csv_path) 

⭐ In the graphics() function, visualize the requested columns from the weather data set by using the Matplotlib library.

def graphics(self, column_1, column_2, xlabel, ylabel):
 # Show requested columns from the data set:
 plt.style.use("dark_background")
 plt.gcf().canvas.set_window_title('IoT Weather Station')
 plt.hist2d(self.df[column_1], self.df[column_2])
 plt.colorbar()
 plt.xlabel(xlabel)
 plt.ylabel(ylabel)
 plt.title(xlabel)
 plt.show() 

⭐ In the data_visualization() function, inspect all columns before scaling weather data to build a model with appropriately formatted data.

def data_visualization(self):
 # Inspect requested columns to build a model with appropriately formatted data:
 self.graphics('WD', '1h_RF', 'Wind Direction (deg)', 'One-Hour Rainfall (mm)')
 self.graphics('Av_WS', '1h_RF', 'Average Wind Speed (m/s)', 'One-Hour Rainfall (mm)')
 self.graphics('Mx_WS', '1h_RF', 'Maximum Wind Speed (m/s)', 'One-Hour Rainfall (mm)')
 self.graphics('24h_RF', '1h_RF', '24-Hour Rainfall (mm)', 'One-Hour Rainfall (mm)')
 self.graphics('Tem', '1h_RF', 'Temperature (°C)', 'One-Hour Rainfall (mm)')
 self.graphics('Hum', '1h_RF', 'Humidity (%)', 'One-Hour Rainfall (mm)')
 self.graphics('b_PR', '1h_RF', 'Barometric Pressure (hPA)', 'One-Hour Rainfall (mm)')
 

After visualizing weather data, I scaled (normalized) each column to format it properly.

Normally, each row (reading) looked like this before scaling:

• 315, 0.45, 0, 0.51, 0.51, 22.78, 65, 1005.5

⭐ In the scale_data() function:

⭐ The wind direction in units of degrees and angles do not make good model inputs. 360° and 0° should be close to each other and wrap around smoothly. The direction should not matter if the wind is not blowing. Thus, convert the wind direction and velocity columns to a wind vector to interpret them easily with a neural network model.

def scale_data(self):
 # Wind Direction and Speed:
 wv =self.df.pop('Av_WS')
 max_wv =self.df.pop('Mx_WS')
 # Convert to radians.
 wd_rad =self.df.pop('WD')*np.pi / 180
 # Calculate the wind x and y components.
 self.df['scaled_WX'] =wv*np.cos(wd_rad)
 self.df['scaled_WY'] =wv*np.sin(wd_rad)
 # Calculate the max wind x and y components.
 self.df['scaled_max_WX'] =max_wv*np.cos(wd_rad)
 self.df['scaled_max_WY'] =max_wv*np.sin(wd_rad)
 
 ... 

⭐ For the remaining columns except for 1h_RF , divide them into average reading values to scale (normalize) and format.

...

 # Temperature:
 tem =self.df.pop('Tem')
 self.df['scaled_Tem'] =tem / 25
 # Humidity:
 hum =self.df.pop('Hum')
 self.df['scaled_Hum'] =hum / 70 
 # Barometric Pressure:
 bPR =self.df.pop('b_PR')
 self.df["scaled_bPR"] =bPR / 1013 
 # 24 Hour Rainfall (Approx.)
 rain_24 =self.df.pop('24h_RF')
 self.df['scaled_24h_RF'] =rain_24 / 24 

After completing scaling (normalizing), I extracted these new columns from the weather data set:

• scaled_WX
• scaled_WY
• scaled_max_WX
• scaled_max_WY
• scaled_Tem
• scaled_Hum
• scaled_bPR
• scaled_24h_RF

Step 5.2:Training the model (ANN) on the rainfall intensity classes

Before building and training a neural network model in TensorFlow, I needed to create the input array and the label array by preprocessing the scaled and normalized weather data set.

⭐ In the create_input_and_label() function:

⭐ Firstly, append each input element as a NumPy array to the input array and convert it to a NumPy array by using the asarray() function.

⭐ Each input element includes seven variables [shape=(7, )]:

• [scaled_WX, scaled_WY, scaled_max_WX, scaled_max_WY, scaled_Tem, scaled_Hum, scaled_bPR]

def create_input_and_label(self):
 n =len(self.df)
 # Create the input array using the scaled variables:
 for i in range(n):
 self.input.append(np.array([self.df['scaled_WX'][i], self.df['scaled_WY'][i], self.df['scaled_max_WX'][i], self.df['scaled_max_WY'][i], self.df['scaled_Tem'][i], self.df['scaled_Hum'][i], self.df['scaled_bPR'][i]]))
 self.input =np.asarray(self.input)
 
 ... 

⭐ Then, evaluate the approximate rainfall (precipitation) rate for each reading (row).

approx_RF_rate =(1h_RF + scaled_24h_RF) * 100

⭐ According to the rainfall rate, assign a class [0 - 4] for each input element and append them to the label array.

...
 
 for i in range(n):
 _class =0
 # Evaluate the approximate rainfall rate:
 approx_RF_rate =(self.df['1h_RF'][i] + self.df['scaled_24h_RF'][i]) * 100
 # As labels, assign classes of rainfall intensity according to the approximate rainfall rate (mm):
 if approx_RF_rate ==0:
 _class =0
 elif approx_RF_rate <2.5:
 _class =1
 elif 2.5  _class =2
 elif 7.6  _class =3
 else:
 _class =4
 self.label.append(_class)
 self.label =np.asarray(self.label) 

After preprocessing the scaled weather data to create input and label arrays, I split them as training (60%) and test (40%) data sets:

def split_data(self):
 n =len(self.df)
 # (60%, 40%) - (training, test)
 self.train_input =self.input[0:int(n*0.6)]
 self.test_input =self.input[int(n*0.6):]
 self.train_label =self.label[0:int(n*0.6)]
 self.test_label =self.label[int(n*0.6):] 

Then, I built my artificial neural network (ANN) model by using Keras and trained it with the training data set for nineteen epochs.

You can inspect these tutorials to learn about activation functions, loss functions, epochs, etc.

def build_and_train_model(self):
 # Build the neural network:
 self.model =keras.Sequential([
 keras.Input(shape=(7,)),
 keras.layers.Dense(16, activation='relu'),
 keras.layers.Dense(32, activation='relu'),
 keras.layers.Dense(64, activation='relu'),
 keras.layers.Dense(128, activation='relu'),
 keras.layers.Dense(5, activation='softmax')
 ])
 # Compile:
 self.model.compile(optimizer='adam', loss="sparse_categorical_crossentropy", metrics=['accuracy'])
 # Train:
 self.model.fit(self.train_input, self.train_label, epochs=19)
 
 ... 

After training with the preliminary training data set, the accuracy of the model is between 0.83 and 0.85 .

Step 5.3:Evaluating the model and making predictions on the rainfall intensity

After building and training my artificial neural network model, I tested its accuracy by using the preliminary testing data set.

For now, the evaluated accuracy of the model is between 0.72 and 0.73 due to inadequate testing data set overfitting the model. However, I am still collecting weather data to improve the model accuracy.

...
 
 # Test the accuracy:
 print("\n\nModel Evaluation:")
 test_loss, test_acc =self.model.evaluate(self.test_input, self.test_label) 
 print("Evaluated Accuracy:", test_acc) 

Then, I used my neural network model to make predictions on the rainfall intensity with a given prediction array consisting of readings from the weather station after the training of the model. As a starting point, the model works fine :)

The model predicts possibilities of labels for each input element as an array of 5 numbers. They represent the model's "confidence" that the given input element corresponds to each of the five different classes of rainfall intensity [0 - 4].

⭐ In the make_prediction() function, make a prediction for each input element in a given array and get the most accurate label [0 - 4] by using the argmax() function to display its class name.

def make_prediction(self, pre_array):
 print("\n\nModel Predictions:\n")
 prediction =self.model.predict(pre_array)
 for i in range(len(pre_array)):
 print("Prediction => ", self.class_names[np.argmax(prediction[i])]) 

Prediction Inputs:

• [0, 0, 0.31819805, 0.31819805, 0.6988, 0.81498571, 0.99349753]
• [0, -0, 0, -0, 0.8444, 1, 0.96835143]
• [0, 0, 0.45, 0, 0.87577, 0.95857143, 1.00128332]
• [-0, -0, -0, -0, 0.8224, 1.05714286, 0.99279368]

Prediction Outputs:

• 0 [None]
• 3 [Heavy Rain]
• 4 [Violent Rain]
• 4 [Violent Rain]

Connections and Adjustments (ESP8266)

// Connections
// NodeMCU ESP-12E (ESP8266) :
// Weather Station
// VV --------------------------- 5V
// D5 --------------------------- RX
// D6 --------------------------- TX
// G --------------------------- GND 

I connected the sensor (converter) board to the NodeMCU ESP-12E (ESP8266) development board and fastened them to a plastic box while collecting weather data at my balcony.

Connections and Adjustments (SIM808)

// Connections
// Arduino Uno:
// SIM808 GPS/GPRS/GSM Shield For Arduino
// D0 --------------------------- RX
// D1 --------------------------- TX
// D12 --------------------------- POWER 
// Weather Station
// 5V --------------------------- 5V
// D5 --------------------------- RX
// D6 --------------------------- TX
// GND --------------------------- GND 

⭐ Note:D0, D1, D12 pins are occupied by the SIM808 GPS/GPRS/GSM Shield.

⭐ Connect an external battery (7-23V) for the SIM808 module to work properly.

⭐ Attach the GPS antenna and the GSM antenna to the SIM808 shield.

⭐ Insert a SIM card into the SIM slot on the SIM808 shield.

⭐ Before uploading the code, set the function switch on the shield to None (1).

⭐ Upload the code.

⭐ Then, set the function switch to Arduino (3).

⭐ Press the Boot button on the shield until seeing the Net indicator LED flashing every 1 second and wait for the SIM card to register the network - the Net indicator LED will slowly flash every 3 seconds.

⭐ Click here to get more information about the SIM808 GSM/GPS/GPRS Shield.

After setting up the SIM808 shield with Arduino Uno, I connected the sensor (converter) board to it.

I placed all components into a plastic box to collect weather data outside, even in extreme weather conditions, since the sensor board is not weather-proof.

Then, I fastened the plastic box to the weather station.

Videos and Conclusion

After completing coding and assembling the weather station, I collated local weather data for more than three months every five minutes to build my neural network model successfully.

I am still collecting weather data at my balcony and backyard to improve my neural network model and its accuracy :)

Further Discussions

☔ 💧 Since we need local weather data sets to get more accurate predictions on weather phenomena, budget-friendly NN-enabled weather stations like this can be placed on farms and greenhouses to avert the detrimental effects of excessive rainfall on agriculture.

☔ 💧 As early warning systems working with neural networks, we can use them simultaneously to create a swarm of weather stations communicating and feeding each other to forecast rainfall intensity precisely in local areas.

☔ 💧 Also, we can utilize that information for assessing:

• vital water resources,
• agriculture,
• crop productivity,
• ecosystems,
• hydrology.

References

[1] Extreme weather - heavy rainfall , NIWA, https://niwa.co.nz/natural-hazards/extreme-weather-heavy-rainfall

[2] University of Illinois at Urbana-Champaign, News Bureau. "Excessive rainfall as damaging to corn yield as extreme heat, drought. " ScienceDaily. ScienceDaily, 30 April 2019. www.sciencedaily.com/releases/2019/04/190430121744.htm.

[3] Andrew Culclasure, Using Neural Networks to Provide Local Weather Forecasts , Georgia Southern University, Spring 2013, 29https://digitalcommons.georgiasouthern.edu/cgi/viewcontent.cgi?article=1031&context=etd.