Research

Research and Data Science Projects

Research Scientist at MSU and Great Lakes Bioenergy Research Center:

Big data and machine learning in agriculture, Geospatial modeling, Data-driven agriculture.

1. - Integrating satellite and drone imagery, machine learning algorithms and process-based crop models to understand agricultural systems. Overall goal is to enable smarter solutions for precision agriculture in a resource-constrained and rapidly changing world.
  - Developing and applying image classification and multi-level (supervised and unsupervised) machine learning algorithms for cloud and shadow detection in satellite imagery.
  - Developing machine learning models to predict the impact of climate, soil and management on crop yield, nutrient uptake, and water use efficiency, and ultimately, to inform management decisions for sustainable agricultural production.
  - Developing algorithms for canopy cover estimation and plant growth prediction from greenhouse (leafy greens) images using image segmentation and machine learning.
  - Mapping crop growth, spacing and vegetation health from UAV imagery using image segmentation and edge detection algorithms.

Consultant Data Scientist at Croptix:

Quantification of water- and nutrient-stress using a handheld spectrophotometer and machine learning models.

1. - Processed Croptix spectrophotometer data and developed machine learning classification models to detect early drought stress, nutrient (nitrogen, phosphorus) deficiency, and disease pressure in agricultural crops.
  - ML models implemented for classifying spectral features are weighted k-nearest neighbor, decision tree, linear and quadratic discriminant analysis, support vector machine, ensemble models, and artificial neural network models.
  - Developed GIS-based visualizations and customized software tools to draw insightful conclusions from the agricultural data to inform water and nutrient management decisions.

3. Postdoctoral Research:

Impact of climate change and agricultural water use on groundwater resources in the United States and India.

- - Processed geospatial data and ground-based observations to estimate the effects of climate variability, crop irrigation demand and streamflow on groundwater level fluctuations and future water availability across major agricultural regions of United States (Collaboration with RDCEP, University of Chicago).
  - Developed predictive models based on machine learning algorithms (neural networks, genetic algorithm), classification and regression models, spectral analysis, mutual information and Monte Carlo uncertainty analysis.
  - Simulation modeling using high performance computing including parallel simulation and large (netcdf) dataset analysis and manipulation.
  - Simulation of groundwater depletion model for India using physical and economic factors with a particular focus on impacts of policy and possible reform pathways (Collaboration with Johns Hopkins, Energy, Resources and Environment, Washington, DC).

4. Ph. D. Research:

Assessment of groundwater resources and simulation-optimization modeling in Deltaic-aquifer systems.

- - Application of neural networks, genetic algorithms, self-organizing feature map in subsurface characterization.
  - Numerical modeling of groundwater flow in a coastal aquifer of India and simulation of salient groundwater management scenarios using Visual MODFLOW and GIS.
  - Groundwater potential assessment using GIS, multi-criteria decision analysis (analytic hierarchy process) and probabilistic modeling (Frequency ratio and Weight- of-evidence).
  - Development of integrated simulation-optimization model for optimal land and water utilization in a coastal basin of India.

5. Masters Research:

Statistical and ANN modeling of groundwater fluctuations in an alluvial unconfined aquifer system.

- - Application of multiple linear regression (MLR), artificial neural network (ANN), fuzzy logic (ANFIS), and support vector machine (SVM) to predict transient water levels of an unconfined aquifer in Japan.

Software/Programming Skill

- Python, MATLAB, TensorFlow, ArcGIS, QGIS, Visual MODFLOW, SWAT, STATISTICA, RockWorks, Erdas Imagine, NeuroSolutions, Aquifer Test and HydroDesktop.
- Satellite based products (Planet, GRACE, MODIS, TRMM, CHIRPS) and various USGS models for data analysis, processing and modeling for agricultural and hydrological applications.

Past Projects (Descriptions and Findings)

1. Impact of climate change and agricultural water use on groundwater storage change in the United States (Penn State University and Center for Robust Decision Making on Climate and Energy Policy (RDCEP), University of Chicago)

Warming temperatures, changes in precipitation patterns, and subsequent increasing agricultural water demands are creating stress on groundwater resources in the United States. Many agricultural regions have experienced groundwater decline in the past three decades, which has consequently affected food production and ecosystem health. This study investigates how projected temporal variability in precipitation, temperature, stream discharge and irrigation demand will influence groundwater availability in the United States. Large raw datasets include time series point measurements of groundwater level available at nearly 900,000 wells from the US Geological Survey, and gridded processed climate data and simulated irrigation demand at 5 arcminute spatial resolution. Overall, the groundwater model integrates an artificial neural network (ANN) model with input data preprocessing using single spectrum analysis, mutual information, and genetic algorithms. The model is calibrated using 33 years of climate, streamflow and ocean temperature observations, and simulated crop water demand. Model runs using projected environmental parameters and irrigation demand are used to simulate changes in future groundwater storage.

Climate data from two GCMs each running RCP 4.5 and 8.5 are used to generate scenario input parameters; the climate data is used directly in the ANN model, and also in a land surface hydrology model to predict streamflow, and in a crop model to predict irrigation demand. The models are run in a high performance parallel computing environment to obtain estimates of future groundwater level change for thousands of wells. Based on this combined climate-agriculture-groundwater model, changes in future groundwater storage for several major agricultural regions are projected up to 2049. These results will be useful for identifying the locations of future groundwater stress, which will have implications for sustainable agricultural production, and will help inform management decisions in a rapidly changing and resource-constrained world.

(a) High Plains Aquifer and (b) Mississippi River Valley Alluvial Aquifer showing observation wells (red circles), stream gauges (blue stars), and DSSAT-simulated annual average irrigation demand at 5 arc min resolution. Link to pdf

Temporal pattern of precipitation, temperature, streamflow and irrigation in High Plains Aquifer.

HANN-modeled cumulative groundwater level change (2003–2012) for (a) HPA and (c) MRVA, and residual groundwater level change for (b) HPA and (d) MRVA. Link to pdf

2. Using spectral signatures and machine learning algorithms to detect multiple stress conditions in agricultural crops (Collaboration with Croptix, Inc., State College, PA)

This study investigates the potential and feasibility of a portable, smartphone-based visible-near infrared sensor to gain better insight into the health of crops. We capture optical measurements from the leaf of a plant using smartphone-based plant health sensor. These data are then processed (smoothed, normalized) and then fed into machine learning-based models to identify various aspects of plant health, including presence of disease, nutrient deficiency or water stress.

Principal components with explained variances for three classes (CSD, Healthy, and HLB) of citrus leaves

Confusion matrix from WKNN model for true and predicted classes of CSD, Healthy and HLB citrus leaves in (a) training and (b) testing.

3. Groundwater depletion model for India using physical and economic factors (Collaboration with Initiative for Sustainable Energy Policy (ISEP) at the Johns Hopkins School of Advanced International Studies)

This study develops new methods to estimate groundwater depletion across India, evaluates the impacts of changes in electricity prices and agricultural minimum support prices, and investigates the distributional consequences of policy reforms to assess the political feasibility of reform strategies. The effects of environmental and economic factors on groundwater levels and irrigation water use are investigated.

PhD Dissertation Research

1. Statistical and artificial neural network (ANN) modeling of groundwater fluctuations in an alluvial unconfined aquifer system of Japan

Groundwater modelling has emerged as a powerful tool to develop a sustainable management plan for efficient groundwater utilization and protection of this vital resource. This study deals with the development of five hybrid artificial neural network (ANN) models and their critical assessment for simulating spatio-temporal fluctuations of groundwater in an alluvial aquifer system. Unlike past studies, in this study, all the relevant input variables having significant influence on groundwater have been considered, and the hybrid ANN-Genetic Algorithm (GA) has been used to simulate groundwater levels at 17 sites over the study area. The parameters of the ANN models were optimized using a GA optimization technique. The predictive ability of the five hybrid ANN models developed for each of the 17 sites was evaluated using six goodness-of-fit criteria and graphical indicators, together with adequate uncertainty analyses. The analysis of the results of this study revealed that the multilayer perceptron Levenberg–Marquardt model is the most efficient in predicting monthly groundwater levels at almost all of the 17 sites, while the radial basis function model is the least efficient. The GA technique was found to be superior to the commonly used trial-and-error method for determining optimal ANN architecture and internal parameters. Of the goodness-of-fit statistics used in this study, only root-mean-squared error and Nash–Sutcliffe efficiency were found to be more powerful and useful in assessing the performance of the ANN models. It can be concluded that the hybrid ANN modelling approach can be effectively used for predicting spatio-temporal fluctuations of groundwater at basin or subbasin scales.

The architecture for the multi-layer feed-forward network used in the simulation of groundwater levels in an unconfined aquifer system of Japan. Link to pdf

Scatter plots of observed and predicted groundwater-levels by ANN and MLR models with ± 2% error band at Site GH-4.5 for the testing period (2003-2004). Link to pdf

Observed and predicted groundwater levels using five ANN models at Site D-6 for the test sets (2003-2004) Link to pdf

Box-whisker plots of observed groundwater levels and groundwater levels simulated by the five ANN models at Site D-6 during testing. Link to pdf

2. Application of soft-computing tools (neural networks, genetic algorithms, self-organizing feature map) in subsurface characterization

Effective characterization of lithology is vital for the conceptualization of complex aquifer systems, which is a prerequisite for the development of reliable groundwater flow and contaminant transport models. However, such information is often limited in most groundwater basins. This study explores the usefulness and potential of hybrid soft-computing framework using traditional ANN (ANN-GDM), GA-based ANN (ANN-GA) and a novel SOM-based ANN (SOM-ANN-GA) techniques for the prediction of lithology at a basin scale. This framework is demonstrated through a case study involving a complex multi-layered aquifer system, where well-log sites were clustered on the basis of sand frequencies and within each cluster, subsurface layers were reclassified into four depth classes based on the maximum drilling depth. Of the three ANN models, the hybrid SOM-ANN-GA models were able to recognize incomplete geologic pattern more reasonably followed by ANN-GA and ANN-GDM models. It is concluded that the hybrid soft-computing framework can serve as a promising tool for characterizing lithology in groundwater basins with missing lithologic patterns.

Prediction of subsurface formation by combined neural network, genetic algorithm and self-organizing map (SOM) approach. Link to pdf

Structure of a self-organizing map neural network (SOM-ANN-GA) for the prediction of subsurface layers.

3. Numerical modeling of groundwater flow in a Deltaic aquifer of India and simulation of salient groundwater management scenarios using Visual MODFLOW and GIS

Process-based groundwater models are useful to understand complex aquifer systems and make predictions about their response to hydrological changes. A conceptual model for evaluating responses to environmental changes is presented, considering the hydrogeologic framework, flow processes, aquifer hydraulic properties, boundary conditions, and sources and sinks of the groundwater system. Based on this conceptual model, a quasi-three-dimensional transient groundwater flow model was designed using MODFLOW to simulate the groundwater system of Mahanadi River delta, eastern India. The model was constructed in the context of an upper unconfined aquifer and lower confined aquifer, separated by an aquitard. Hydraulic heads of 13 shallow wells and 11 deep wells were used to calibrate transient groundwater conditions during 1997–2006, followed by validation (2007–2011). The aquifer and aquitard hydraulic properties were obtained by pumping tests and were calibrated along with the rainfall recharge. The statistical and graphical performance indicators suggested a reasonably good simulation of groundwater flow over the study area. Sensitivity analysis revealed that groundwater level is most sensitive to the hydraulic conductivities of both the aquifers, followed by vertical hydraulic conductivity of the confining layer. The calibrated model was then employed to explore groundwater-flow dynamics in response to changes in pumping and recharge conditions. The simulation results indicate that pumping has a substantial effect on the confined aquifer flow regime as compared to the unconfined aquifer. The results and insights from this study have important implications for other regional groundwater modelling studies, especially in multi-layered aquifer systems.

Locations of 108 well log sites (a) with 16 geologic cross-sections (nine E–W sections, seven N–S sections). Geologic profile of the study area along the N–S D–D′ cross-section (b) drawn using RockWorks.

Pre-monsoon hydraulic head contour maps with groundwater flow vectors for the (a) unconfined aquifer and (b) confined aquifer. Post-monsoon hydraulic head contour maps with groundwater flow vectors for the (c) unconfined aquifer and (d) confined aquifer. Link to pdf

Measuring groundwater level in a well using water level indicator during a field experiment in India.

4. Development of integrated simulation-optimization model for efficient land and water utilization and optimal crop planning in India

Sustainable management of freshwater resources has become a key concern throughout the world owing to shrinking water availability, growing freshwater needs, and unsustainable use of surface water and groundwater resources. To address this issue, in this study, groundwater flow simulation models have been integrated with optimization techniques to determine optimal management strategies for a complex groundwater system. Given a set of hydrologic and management constraints, two optimization models were developed and applied to a portion of the Mahanadi River basin in Odisha state, India. First optimization model links a groundwater-flow simulation model MODFLOW to an optimization model SOMO1 to estimate maximum permissible groundwater pumpage, subject to various constraints that protect the aquifer from seawater intrusion. Second optimization model was developed to: a) optimize a water resources allocation scheme considering the conjunctive use of surface water and groundwater, and b) to determine a suitable cropping pattern to maximize net annual returns from crop yield. This model was formulated taking into account suitability of available cultivable land, type of seasons, and the existing crops grown in the basin. Both linear programming (LP) and genetic algorithm (GA) was employed to solve the second optimization model. Overall, the performance of the GA optimization model was found superior to that of the LP optimization model in the optimal allocation of available land and water resources in the basin. Based on the GA optimization results, it is recommended that paddy cultivation by the farmers needs to be minimized and the adoption of crop diversification practice should be encouraged in order to improve livelihoods of the farmers as well as to ensure sustainable utilization of water resources in the study area.

Schematic representation of integrated modeling framework for optimal land and water allocation in a Deltaic aquifer system.

Optimal allocation of land by LP and GA optimization Model.

5. Groundwater potential assessment using GIS, multi-criteria decision analysis (analytic hierarchy process) and probabilistic modeling (Frequency ratio and Weight-of-evidence)

Quantification of groundwater resources is indispensable for developing an efficient strategy for sustainable groundwater management. Integration of remote sensing (RS) and geographical information system (GIS) techniques with multicriteria decision analysis (MCDA) has emerged as a powerful tool for the economical and rapid assessment of groundwater resources at a macro scale. The main intent of this study is to evaluate the performance of two GIS-based approaches, namely multicriteria decision analysis (MCDA) as Approach I and probabilistic modeling as Approach II for groundwater prospecting. In Approach I, the thematic layers and their features relevant to groundwater prospect were extracted using RS and GIS, and appropriate weightages were assigned to individual layers and their features based on the analytic hierarchy process (AHP) scale. After the normalization of these weights, the selected thematic layers were integrated in the GIS environment to generate a groundwater prospect map. In Approach II, two probabilistic models, viz. frequency ratio (FR) and weight of evidence (WOE) were used. The frequency ratio and WOE probability values were calculated for each of the selected themes and then groundwater prospect maps were generated by overlaying the themes in GIS. The groundwater prospect maps thus obtained by the two approaches were classified into four distinct groundwater potential zones. These maps were verified using the available well-yield data. The verification results indicated that out of the AHP, FR and WOE techniques, the AHP technique is superior (prediction accuracy of 77%) to the probabilistic models (FR and WOE), though the WOE model also performed reasonably well with a prediction accuracy of 73%. It is concluded that for more reliable results, the AHP technique can be used for assessing groundwater potential in a given area/region. The findings of this study are useful for the cost-effective identification of suitable well locations as well as for the efficient planning and development of groundwater resources.

Groundwater prospect map based on Weight of Evidence (WOE), Frequency Ratio (FR) and Analytical Hierarchy Process (AHP) models. Link to pdf

Google Sites

Report abuse