SOIL MOISTURE MODELING USING ARTIFICIAL NEURAL NETWORKS

Similar documents
Global Climates. Name Date

Research Article Weather Forecasting Using Sliding Window Algorithm

The Climate of Bryan County

The Climate of Payne County

The Climate of Kiowa County

The Climate of Marshall County

UWM Field Station meteorological data

LONG TERM LOAD FORECASTING OF POWER SYSTEMS USING ARTIFICIAL NEURAL NETWORK AND ANFIS

DESIGN AND DEVELOPMENT OF ARTIFICIAL INTELLIGENCE SYSTEM FOR WEATHER FORECASTING USING SOFT COMPUTING TECHNIQUES

Jackson County 2013 Weather Data

A Support Vector Regression Model for Forecasting Rainfall

The Climate of Murray County

2003 Water Year Wrap-Up and Look Ahead

The Climate of Texas County

The Climate of Pontotoc County

The Climate of Grady County

Jackson County 2018 Weather Data 67 Years of Weather Data Recorded at the UF/IFAS Marianna North Florida Research and Education Center

Jackson County 2014 Weather Data

CoCoRaHS Monitoring Colorado s s Water Resources through Community Collaborations

Prediction of Monthly Rainfall of Nainital Region using Artificial Neural Network (ANN) and Support Vector Machine (SVM)

The Climate of Seminole County

Climatography of the United States No

The Climate of Haskell County

Table of Contents. Page

A comparative study of ANN and angstrom Prescott model in the context of solar radiation analysis

Comparison of Adaline and Multiple Linear Regression Methods for Rainfall Forecasting

Study of Hydrometeorology in a Hard Rock Terrain, Kadirischist Belt Area, Anantapur District, Andhra Pradesh

A Feature Based Neural Network Model for Weather Forecasting

Weather Forecasting using Soft Computing and Statistical Techniques

Memo. I. Executive Summary. II. ALERT Data Source. III. General System-Wide Reporting Summary. Date: January 26, 2009 To: From: Subject:

Jackson County 2019 Weather Data 68 Years of Weather Data Recorded at the UF/IFAS Marianna North Florida Research and Education Center

Variability of Reference Evapotranspiration Across Nebraska

Modelling and Prediction of 150KW PV Array System in Northern India using Artificial Neural Network

What is the difference between Weather and Climate?

Chiang Rai Province CC Threat overview AAS1109 Mekong ARCC

Remote Sensing Applications for Land/Atmosphere: Earth Radiation Balance

Geostatistical Analysis of Rainfall Temperature and Evaporation Data of Owerri for Ten Years

EE-588 ADVANCED TOPICS IN NEURAL NETWORK

2016 Meteorology Summary

Climate Change Impact Assessment on Indian Water Resources. Ashvin Gosain, Sandhya Rao, Debajit Basu Ray

Applications/Users for Improved S2S Forecasts

Climatography of the United States No

Climatography of the United States No

Climatography of the United States No

Climatography of the United States No

Tracking the Climate Of Northern Colorado Nolan Doesken State Climatologist Colorado Climate Center Colorado State University

Climatography of the United States No

Climatography of the United States No

Chapter-1 Introduction

Short Term Load Forecasting Using Multi Layer Perceptron

OVERVIEW OF IMPROVED USE OF RS INDICATORS AT INAM. Domingos Mosquito Patricio

2003 Moisture Outlook

Climatography of the United States No

Climatography of the United States No

The Climate of Oregon Climate Zone 5 High Plateau

Comparison of meteorological data from different sources for Bishkek city, Kyrgyzstan

Development of Innovative Technology to Provide Low-Cost Surface Atmospheric Observations in Data-sparse Regions

Climatography of the United States No

Climatography of the United States No

Climatography of the United States No

Climatography of the United States No

Climatography of the United States No

Climatography of the United States No

Climatography of the United States No

Climatography of the United States No

Site Description: Tower Site

Constructing a typical meteorological year -TMY for Voinesti fruit trees region and the effects of global warming on the orchard ecosystem

Climatography of the United States No

Climatography of the United States No

Climatography of the United States No

Climatography of the United States No

Climatography of the United States No

Climatography of the United States No

Climatography of the United States No

YACT (Yet Another Climate Tool)? The SPI Explorer

Climatography of the United States No

Climatography of the United States No

Prediction of Global Solar Radiation in UAE

Thunderstorm Forecasting by using Artificial Neural Network

Climatography of the United States No

Climatography of the United States No

APPENDIX G-7 METEROLOGICAL DATA

Climatography of the United States No

Climatography of the United States No

Climatography of the United States No

Colorado s 2003 Moisture Outlook

Climatography of the United States No

Final Report. COMET Partner's Project. University of Texas at San Antonio

Climatography of the United States No

Climate Variability in South Asia

Site Description: Tower Site

Local Prediction of Precipitation Based on Neural Network

Climatography of the United States No

SOUTH MOUNTAIN WEATHER STATION: REPORT FOR QUARTER 2 (APRIL JUNE) 2011

Radial basis function neural networks model to estimate global solar radiation in semi-arid area

Climatography of the United States No

Interannual variation of MODIS NDVI in Lake Taihu and its relation to climate in submerged macrophyte region

SHORT-TERM FORECASTING OF WEATHER CONDITIONS IN PALESTINE USING ARTIFICIAL NEURAL NETWORKS

Application of Fully Recurrent (FRNN) and Radial Basis Function (RBFNN) Neural Networks for Simulating Solar Radiation

Solar Irradiance Prediction using Neural Model

Transcription:

Int'l Conf. Artificial Intelligence ICAI'17 241 SOIL MOISTURE MODELING USING ARTIFICIAL NEURAL NETWORKS Dr. Jayachander R. Gangasani Instructor, Department of Computer Science, jay.gangasani@aamu.edu Dr. Monday Mbila (Contact Author) Associate Professor, Department of Soil Science, monday.mbila@aamu.edu Dr. Yinshu Wu Assistant Professor, Department of Mathematics, yinshu.wu@aamu.edu Dr. Jian Fu Professor, Department of Computer Science, jian.fu@aamu.edu ABSTRACT An artificial neural network (ANN) based algorithm was implemented, tested, and compared with regression models for soil moisture estimation. The research focused on developing soil moisture estimation capability by using an Artificial Neural Network (ANN) based model. The ANN model was calibrated (trained) and validated (tested) with soil moisture profiles measured from the Alabama Mesonet (ALMNet). The performance of the ANN model was evaluated by direct comparison between the soil moisture calculated from regression models, and Mesonet soil moisture measurements. Correlations were observed between the ANN estimates and Mesonet annual measurements. Less strong correlations were observed between the ANN estimates and Mesonet monthly measurements, while the least correlations were observed for the regression model. KEY WORDS: Alabama Mesonet, Artificial Neural Networks, Soil moisture prediction, and Regression models Type of the submission: Regular Research Paper

242 Int'l Conf. Artificial Intelligence ICAI'17 INTRODUCTION Soil moisture measurements in agricultural settings provide important information for drought early warning (Thober et al, 2015) as well as predicting floods (Zink et al., 2016). But soil moisture measurement methods are laborious, time consuming, expensive, and limited in application due to point-sampling in spite of heterogeneous distribution of moisture in soils. On the other hand, largescale satellite-based remote measurement techniques do not yet provide the required resolution to resolve spatiotemporal variability present at the scale needed. Therefore exploring the application of predictive models to soil moisture studies is imperative. The use of Artificial Neural Network for modeling soil moisture is a potential solution to this problem, and has already provided promising results (Pandey et al., 2010). Artificial Neural Networks (ANN) is a part of Computer Science that is analogous with artificial intelligence. ANNs are considered to be more relevant and useful than genetic algorithms and fuzzy logic system in dealing with soil and agricultural issues (Jain et al., 1996). Artificial neural networks is the type of network that sees the nodes as artificial neurons thus, these are called artificial neural networks (ANNs). An artificial neuron is a computational model inspired in the natural neurons. Natural neurons receive signals through synapses located on the dendrites or membrane of the neuron. When the signals received are strong enough (surpass a certain threshold), the neuron is activated and emits a signal though the axon. This signal might be sent to another synapse, and might activate other neurons (Carlos, 2003). Neural networks in general use machine learning based on the concept of self-adjustment of internal control parameters. Artificial neural networks are pliable mathematical structures that are capable of identifying complex non-linear relationships among input and output data sets. The principal differences between the various types of ANNs are arrangement of neurons and the many ways to assess the weights and functions for inputs and neurons (training). There are a variety of ANN architectures, such as multi-layer perceptron. The multilayer perceptron (MLP) neural network has been designed to function well for non-linear phenomena. A feed forward MLP network consists of a layer of input neurons and output layer with selected number of input and output neurons, respectively with one or more hidden layers in between the input and the output layer with some number of neurons on each (Melesse, 2006). The objectives of this research were: a) to develop an ANN model that estimates soil moisture content based on data from meteorological stations in Alabama such as precipitation(x0), atmospheric temperature(x1), solar radiation(x2), and wind speed (xp) as the input variables; and b) to evaluate the performance of the ANN model by comparison with other soil moisture models, and with measured soil moisture. DESIGN AND METHODOLOGY Location of the Study Area and Research Sites The study was carried out with data from the Alabama A&M University Main Campus Weather Station. This station is located about three hundred feet from the campus East gate at Latitude: 34 deg; 47 min N; and Longitude: 86 deg; 33 min W. The Station is recognized by the USDA Soil Climate Analysis Network (SCAN) Data & Products resources as: AAMU-JTG; Madison County.

Int'l Conf. Artificial Intelligence ICAI'17 243 The weather station has the following sensors installed for measuring the variables: soil hydra probes for measuring soil moisture and temperature; propeller-type anemometer for measuring wind speed and direction; pyranometer for solar radiation; tipping bucket rain gauge for measuring precipitation; thermometer for measuring air temperature; and humidity probe for measuring relative humidity. Selection of input variables We evaluated the correlation among the variables and soil moisture based on the following conceptual model: precipitation (x0), soil temperature (x1), solar radiation (x2), wind speed (xp) and soil moisture (y +k). Data Preprocessing After data download, three data preprocessing procedures were conducted to train the ANNs more efficiently for: a) solving the problem of missing data; b) Normalizing the data; and c) Randomizing the data. The missing data was replaced by the average of neighboring values during the same week. Normalization procedure before presenting the input data to the network was needed since mixing variables with large magnitudes and small magnitudes will confuse the learning algorithm on the importance of each variable. Mixing large and small magnitudes could force the procedure to finally reject the variable with the smaller magnitude (Tymvios et al., 2008). Building the Artificial Neural Network At this stage, building the program required specifying the number of hidden layers, neurons in each layer, transfer function in each layer, training function, weight/bias learning function, and performance function. Since this Program aimed to predict the seasonal/monthly soil moisture, the idea was to design the program into a multi-class classification problem using neural network. The Procedure: Training Data: Year 2004-2013 Test Data: Year 2014 Training the Network During the training process, the weights of the variables were adjusted in order to make the actual outputs (predicted) close to the target (measured) outputs of the network. In this study, the 10- year data period (from 2004 to 2013), 7-year data period (from 2005 to 2011), and 5-year data period (from 2005 to 2009) from the ALABAMA MESONET were used for training. Programming the Actual Neural Network Model For this work, MATLAB (R2013b) was used to write script files for developing MLP and RBF ANN models and performance functions for calculating the model performance error statistics such as R2, RMSE and MBE. RESULTS AND DISCUSSION Soil Moisture Prediction Using Artificial Neural Network ANN learns to process data by arbitrary classification of the data and comparing the data with known actual classification of the data. The errors from the initial classification of the data is fed

244 Int'l Conf. Artificial Intelligence ICAI'17 back into the network, and used to modify the networks algorithm the second time around, and so on for many iterations. Table 1: Results of the ANN prediction Year Precipitation (inches) Soil Temp ( F) Radiation (watts) Wind Speed (mph) Soil Moisture (%) Class 1 2004 0.0245536 12.7 89.6 2.6 25.4 Class 2 2005 0.0426729 41.4 131.6 2.4 23.7 Class 3 2006 0.0191899 25.6 153.7 2.5 19.0 Class 4 2007 0.0208852 18.6 149.7 2.3 18.7 Class 5 2008 0.0215465 23.9 137.6 2.3 21.2 Class 6 2009 0.0267366 26.5 131.6 2.1 24.7 Class 7 2010 0.1618504 24.1 126.2 1.9 26.2 Class 8 2011 0.1111749 22.7 146.0 4.6 26.4 Class 9 2012 0.0076516 28.5 163.9 1.9 26.0 Class 10 2013 0.007349 29.4 178. 8 4.2 30.5 Class #? 2014 0.0057565 12. 7 167.5 4.3 27.8 Table 2: Comparison of the Results of the Artificial Neural Network Predicted and Measured Monthly Soil Moisture Discussion Months Measured Predicted % Error Jan 31.07 28.86 7 Feb 34.82 33.62 3 Mar 33.03 34.85 6 Apr 35.05 31.13 11 May 30.15 27.77 8 Jun 36.52 30.71 16 Jul 24.55 23.29 5 Aug 24.52 27.91 14 Sep 24.52 27.91 14 Oct 24.65 20.03 19 Soil Moisture Prediction Using Multiple Regression Models Artificial Neural Network Model Prediction

Int'l Conf. Artificial Intelligence ICAI'17 245 The error difference between the actual soil moisture value and the predicted soil moisture from the monthly training data ranged for 3% to 19% with an average of 10% error for the year. Therefore the error for the monthly training data, while larger than that of the 10-year training data, was quite variable. Differences in error for the model are probably due to data size and gaps. This observation suggests that yearly soil moisture prediction may be more accurate, because the 10 days of bad data has greater impact in monthly soil moisture prediction than in yearly soil moisture predictions. For instance, 10 days of bad data due to missing data from bad receivers or sensors in one month will have 10/30 error proportion. This is equal to 33% approximately, meaning that 33% of the data is missing or wrong. However, for yearly soil moisture prediction, 10 days of missing or wrong data translates to 10/365 (2.7%). This means that there is only 2.7% error from missing or bad data in yearly soil moisture prediction compared to 33% error in monthly soil moisture prediction.

246 Int'l Conf. Artificial Intelligence ICAI'17 In general, regression models predicted soil moisture variations but those models explained a very low percentage of the soil moisture variability. Multiple regression models explained more of the soil moisture variability (46%) than any of the single variables that were investigated (Table 3). Solar radiation and soil temperature each individually explained about 40% of the variability in soil moisture changes. Wind speed and rainfall explained much less of the soil moisture variability with 20% and 3%, respectively. Artificial neural network predicted soil moisture variations and explained a high percentage (52%) of the soil moisture variability (Figure 1). Figure 1: Comparison of ANN Predicted and Measured Soil Moisture CONCLUSION The study was carried out with data from the Alabama Agricultural and Mechanical University Main Campus Weather Station where a meteorological station is installed to gather information on rain gauge, air temperature and relative humidity, soil moisture and temperature, wind speed and direction, and solar radiation. Data collected for ten years from the station was analyzed and used to develop and train the ANN and regression models to estimate soil moisture content. The impacts of solar radiation, rainfall, wind speed, and soil temperature were considered by using regression and the multiple regression models. Regression models predicted soil moisture variations, but those models explained a low percentage of the soil moisture variability. Multiple regression models explained more of the soil moisture variability than any of the single variables that were investigated (46%). Solar radiation and soil temperature each individually explained about 40% of the variability in soil moisture changes. Wind speed and rainfall explained much less of the soil moisture variability with 20% and 3%, respectively. The ANN model contained between 3% and 19% error difference between the ANN predicted, and the measured soil moisture value. Using the ANN explained 52% of the variability in measured soil moisture content. Overall, the ANN network, show better potential for predicting soil moisture changes using meteorological data generated by the weather stations.

Int'l Conf. Artificial Intelligence ICAI'17 247 REFERENCES Carlos Gershenson - 2003, Neural and Evolutionary Computing (cs.ne); Artificial Intelligence (cs.ai). Jain, A.K., Mao, J., Mohiuddin, K.M., 1996. Artificial neural networks: a tutorial. Comput. IEEE March, 31 44. Melesse Assefa, and Xixi Wang. 2006. Multitemporal Scale Hydrographic Prediction using Artificial Neural Networks, Journal of the American Water resources Association. Pandey, A, S. K. Jha, J. K. Srivastava, and R. Prasad. 2010. Artificial neural network for the estimation of soil moisture and surface roughness. Russian Agricultural. Thober S., R. Kumar, J. Sheffield, J. Mai, D. Schafer, and L. Samaniego. 2015. Seasonal Soil Moisture Drought Prediction over Europe Using the North American Multi-Model Ensemble (NMME). Journal of Hydrometeorology. 16:2329-2344. DOI: 10.1175/JHM- D-15-0053.1. Tymvios, F., Michaelides, S. and C. Skouteli. 2008. Estimation of Surface solar radiation with Artificial neural networks, In: Modeling Solar Radiation at the Earth Surface, Viorel Badescu, pp. (221-256). Springer, ISBN 978-3-540-77454-9, Germany. Zink, M, L. Samaniego, R. Kumar, S. Thober, J. Mai, D. Schafer, and Marx. 2016. The German drought monitor. Environ. Res. Lett. 11:074002. Pp 9. http://dx.doi.org/10.1088/1748-9326/11/7/074002.