About

Dr. David Mulla is a Professor and Larson Endowed Chair in Soil and Water Resources in the Department of Soil, Water, and Climate at the University of Minnesota. He received his BS from the University of California, Riverside, his MS and PhD from Purdue University, with his doctoral emphasis in soil physics. His research emphasizes AI-guided modeling and remote sensing in agriculture, non-point source transport and modeling of water contaminants such as nitrates, phosphorus, trace metals, and organic chemicals in soil, surface, and ground water, as well as the impacts of biofuel and alternative crop production systems, agricultural best management practices, and green infrastructure practices at solar farms on ecosystem services. Additionally, his work includes the measurement, modeling, and management of soil erosion and runoff, and soil spatial variability, landscape, and terrain modeling for precision agriculture and conservation. Dr. Mulla has contributed significantly to the adoption of precision agriculture in the US and globally, fostering economic and environmental efficiencies in agriculture. His extensive experience includes modeling erosion and nutrient losses, studying pollution of Minnesota’s water resources, and evaluating BMPs for water quality improvement. He has served on various national committees, including those of the National Academy of Sciences, and has been recognized with awards such as the Pierre C. Robert Precision Agriculture Research Award. Dr. Mulla has authored over 240 publications, received over $57 million in funding, and has worked internationally, including consulting projects in Morocco.

Research topics

• Computer Science
• Artificial Intelligence
• Geography
• Biology
• Agronomy
• Remote sensing
• Mathematics
• Computer vision
• Environmental science
• Chemistry
• Engineering
• Botany
• Economics
• Ecology
• Meteorology
• Agricultural economics
• Statistics

Selected publications

• Linking remote sensing with crop modeling for yield and nitrate leaching predictions in Minnesota

  Journal of Environmental Quality · 2026-01-01 · 1 citations

  articleOpen accessSenior authorCorresponding

  Abstract Upscaling crop yield and nitrate‐N leaching loss from experimental sites to large areas under alternative crop rotations is crucial for assessing strategies and setting goals to protect groundwater quality at a regional scale. Nitrogen (N) rate field trials were used to calibrate the Environmental Policy Integrated Climate (EPIC) model for continuous‐corn ( Zea mays L.) (C‐C), corn‐soybean ( Glycine max L.) (C‐Sb), and alfalfa ( Medicago sativa L.)‐corn (A‐C), with or without rye ( Secale cereale L.) cover crop. Satellite estimates of crop evapotranspiration (ET c ) were used to upscale the EPIC model for crop yield and nitrate‐N leaching, using the irrigation‐water permitting data from 2010 to 2017 for 13,375 ha of sandy soils in Bonanza Valley, central Minnesota. Four alternative management scenarios were evaluated with EPIC: (1) reducing N fertilizer rate from the maximum return to N value (MRTN) (of 0.05 to a value of 0.1 (for the N price/crop value ratio), (2) adding rye cover crop at MRTN of 0.1, (3) irrigating with EPIC auto‐trigger in scenario 2, and (4) converting 50% of C‐C acreage in scenario 3 to A‐C. Nash‐Sutcliffe coefficients, normalized root‐mean‐square error, and R 2 values based on ET c /crop yield for calibration and validation of the EPIC model ranged 0.95–0.54, 4.67–19.4, and 0.96–0.74; and 0.74–0.41, 7.99–23.4, and 0.88–0.55, respectively. Results indicate that corn yield at MRTN of 0.05 averaged 12.5, 13.2, and 13.4 t ha −1 under C‐C, C‐Sb, and A‐C rotations, while yields at MRTN of 0.1 were reduced by 4.1%, 3.5%, and 3.3%, respectively. The baseline scenario of C‐C, C‐Sb, and A‐C rotations at MRTN of 0.05 had annual nitrate‐N leaching losses of 51.8, 45.5, and 31.4 kg ha −1 , while MRTN of 0.1 reduced these losses by 9.1%, 5.0%, and 3.8%, respectively. Rye after corn and soybean reduced nitrate‐N leaching losses in the MRTN of 0.1 scenario by 5.8% and 13.6%, respectively. EPIC auto‐irrigation of corn, soybean, and alfalfa at MRTN of 0.1 reduced nitrate‐N leaching losses with rye (relative to conventional irrigation) by 9.6%, 9.1%, and 8.5%, respectively. Further, replacing half of the C‐C acreage with A‐C rotation would provide a 6.1% reduction, resulting in a total reduction of 27.4% in nitrate‐N leaching to groundwater when all alternative practices are combined. Overall, augmenting EPIC model with field‐observed ancillary data and remote sensing successfully predicted the yield and NO 3 ‐N leaching losses under different crop rotations, indicating opportunities to upscale field‐scale agroecosystem simulations, particularly if used to calculate NO 3 ‐N leaching on a long‐term basis at the regional scales.

  PublisherOA PDFDOI

• Impact of Irrigation Scheduling on Corn Yield and Water Balance in Sandy Soils of Central Minnesota Using the EPIC Model.

  Irrigation and Drainage · 2026-02-17

  articleOpen accessSenior authorCorresponding

  ABSTRACT Efficient irrigation scheduling is critical in sandy soils to achieve optimum yield and maximum net benefit with minimum environmental concerns. The Environmental Policy Integrated Climate (EPIC) model was used to study the impacts of autoirrigation of corn on crop yield and soil water balance during 2019–2021 at −300 and −450 kPa for loamy sand (Becker) and sandy loam (Westport) sites in Central Minnesota, respectively. Moreover, four irrigation scheduling methods were compared, i.e., checkbook (CB), EPIC‐autoirrigation (EPIC), irrigation management assistant (IMA) and soil‐moisture‐based (SM). The simulation results indicate that irrigation water requirements under CB, EPIC, IMA and SM averaged 155.6, 141.2, 76.1 and 151.6 mm, respectively, on top of 627.5‐mm annual precipitation. EPIC simulations indicate that the CB, EPIC, IMA and SM methods, averaged two sites, produced corn yields of 12.3, 12.6, 11.1 and 12.2 t ha −1 , with deep percolation losses of 198.5, 171.9, 163.3 and 189.2 mm, respectively. EPIC appeared to be an effective alternative to the labour‐intensive SM method for irrigation management to obtain maximum corn yield and economic benefits, while the CB and IMA methods had higher deep percolation and soil water stress, respectively.

  PublisherOA PDFDOI

• Reducing Hydrological Uncertainty: A Multi-Variable SWAT Calibration Using Global AET Products

  2026-03-13

  articleOpen access

  Parameter equifinality remains a central challenge in hydrological modeling, limiting the reliability of process-based tools such as the Soil and Water Assessment Tool (SWAT). This study evaluates how multi-variable calibration strategies that combine in-situ streamflow with five remote sensing global actual evapotranspiration (RSAET) products (GLEAM v3.6, GLEAM v4.2, ETMonitor, PML, and SSEBop) can reduce equifinality, using the Essaouira watershed (Morocco) as a study case. A total of 10,000 Monte Carlo simulations were performed, from which the 100 best-performing parameter sets were selected for posterior uncertainty assessment. A Composite Identifiability Score (CIS) was developed by integrating normalized metrics of standard deviation, entropy, peak-to-width ratio, and Kullback-Leibler divergence to quantify parameter identifiability.Results show that streamflow-only calibration (S0) yields the highest CIS, confirming the strong constraining power of discharge on routing and runoff parameters. However, multi-variable calibration further reduces equifinality for several soil–plant–atmosphere parameters, with the Streamflow + GLEAM v3.6 configuration achieving the highest multi-source CIS, followed by SSEBop and GLEAM v4.2. In terms of performance, streamflow-only scenarios achieve the highest NSE and lowest PBIAS, while hybrid streamflow–AET calibrations maintain strong predictive skill and improve the physical consistency of ET-related processes. In contrast, AET-only calibrations exhibit poor runoff-volume accuracy and large water-balance inconsistencies.Overall, integrating complementary AET datasets with discharge observations enhances parameter identifiability, constrains key hydrological processes, and mitigates equifinality. This demonstrates a practical pathway to strengthen SWAT model robustness in data-scarce regions, as is the case for many African basins. These results are preliminary, and ongoing work aims to consolidate them by extending the calibration and identifiability framework to include soil-moisture remote-sensing products, with the goal of further constraining soil-water dynamics and reducing remaining model uncertainties.

  PublisherDOI

• Towards Fine-Tuning-Based Site Calibration for Knowledge-Guided Machine Learning: A Summary of Results

  ArXiv.org · 2025-12-17

  preprintOpen access

  Accurate and cost-effective quantification of the agroecosystem carbon cycle at decision-relevant scales is essential for climate mitigation and sustainable agriculture. However, both transfer learning and the exploitation of spatial variability in this field are challenging, as they involve heterogeneous data and complex cross-scale dependencies. Conventional approaches often rely on location-independent parameterizations and independent training, underutilizing transfer learning and spatial heterogeneity in the inputs, and limiting their applicability in regions with substantial variability. We propose FTBSC-KGML (Fine-Tuning-Based Site Calibration-Knowledge-Guided Machine Learning), a pretraining- and fine-tuning-based, spatial-variability-aware, and knowledge-guided machine learning framework that augments KGML-ag with a pretraining-fine-tuning process and site-specific parameters. Using a pretraining-fine-tuning process with remote-sensing GPP, climate, and soil covariates collected across multiple midwestern sites, FTBSC-KGML estimates land emissions while leveraging transfer learning and spatial heterogeneity. A key component is a spatial-heterogeneity-aware transfer-learning scheme, which is a globally pretrained model that is fine-tuned at each state or site to learn place-aware representations, thereby improving local accuracy under limited data without sacrificing interpretability. Empirically, FTBSC-KGML achieves lower validation error and greater consistency in explanatory power than a purely global model, thereby better capturing spatial variability across states. This work extends the prior SDSA-KGML framework.

  PublisherOA PDFDOI

• Impact of water regimes on the chemical composition, antioxidant activity and quality of olive oil (Olea europaea L.) under semi-arid conditions in Tunisia

  Notulae Botanicae Horti Agrobotanici Cluj-Napoca · 2025-06-30

  articleOpen access

  Water scarcity is the main challenge in irregular olive oil production in Tunisia. In olive orchards the use of deficit irrigation technique is the main method to optimize water saving and achieve sustainable olive production. The aim of this study was to assess four levels of drip irrigation (100% ETc, 60% ETc, 50% ETc and 40% ETc (evapotranspiration)) on ‘Chemlali’ and ‘Koroneiki’ olive oil varieties, in semi-arid conditions, for two consecutive crop years. The quality parameters, antioxidant compounds, and antioxidant properties against DPPH and ABTS cation radicals of oils, were evaluated. Moreover, the fatty acid composition of ‘Chemlali’ and ‘Koroneiki’ oils was determined through gas chromatographic analysis. The result showed that irrigation regimes had a moderate effect on the standard quality parameters (free fatty acids, peroxide value, K232 and K270) of virgin olive oil as well as on the composition of the fatty acids. The oil obtained from trees treated with 50% ETc irrigation strategy was characterized by a high content of oleic acid (approximately 61% for ‘Chemlali’ and 78% for ‘Koroneiki’), a low level of palmitic acid, a high content of phenolic compounds (217.44 and 198.99 mg of eq catechin kg-1 of oil for ‘Chemlali’ and ‘Koroneiki’, respectively), O-diphenols, chlorophyll and carotenoids and high antioxidant properties for both olive cultivars. Therefore, this method is considered the best irrigation strategy to optimize water management and improve the quality, the antioxidant content and properties of oil.

  PublisherOA PDFDOI

• Subterra: Estimating Soil Moisture at Root Zone Depths Using Science-Guided Learning

  2025-05-05 · 2 citations

  article

  Accurate prediction of soil moisture (SM) is crucial for applications in agriculture, hydrology, and climate modeling. Traditional data-driven machine learning (ML) approaches often require extensive labeled datasets and fail to incorporate the physical principles governing SM dynamics. In this study, we propose a novel science-guided learning framework to predict SM in the top 20 cm of soil using deep neural networks (DNNs). By integrating physical equations, such as Richards' equation, into the learning process, our approach ensures scientifically consistent predictions while improving model generalizability. Our empirical results show that the proposed graph-based model outperforms traditional ML approaches by more than 30% in accuracy, while predicting SM with an error of less than 5% compared to ground-truth in-situ measurements.

  PublisherDOI

• Nitrogen and phosphorus removal from agricultural drainage water by a modular bioreactor

  Journal of Environmental Management · 2025-06-09 · 2 citations

  article
  PublisherDOI

• EPIC model prediction of winter rye cover crop effects on crop yield and nitrate‐N leaching in Minnesota

  Agronomy Journal · 2025-04-25 · 5 citations

  articleOpen accessSenior authorCorresponding

  Abstract Coarse‐textured soils in central Minnesota cultivated with corn ( Zea mays L.) and soybean ( Glycine max L.) exhibit good productivity, however, are vulnerable to nitrate‐N leaching losses. In such circumstances, winter rye ( Secale cereale L.) as a cover crop may reduce nitrate‐N leaching by scavenging soil nitrogen (N) in late‐fall and early‐spring fallow period. The Environmental Policy Integrated Climate (EPIC) model was used for decadal‐scale (2010–2020) simulation of yield/biomass and nitrate‐N leaching in corn– (C–C) and corn–soybean/soybean–corn (C–Sb/Sb–C) rotations, with and without winter rye, under different fertilizer N rates applied to corn (0, 100, 200, 250, and 300 kg ha −1 ) on irrigated coarse‐textured soils in central Minnesota. Model efficiency calculated based on Nash–Sutcliffe coefficient, relative root mean square error, and R 2 statistics indicate that EPIC assessment for calibration and validation treatments was excellent‐good for corn/soybean yield, and good‐satisfactory for rye biomass and NO 3 ‐N leaching losses. Results indicate that N fertilizer rates up to 250 kg N ha −1 applied to corn had a positive impact on rye biomass; however, large crop‐rotation and climate‐induced variations were observed. Annual nitrate‐N leaching losses at maximum return to nitrogen rates at a 0.05 N price to crop value ratio for corn under C–C (250 kg N ha −1 ) and C–Sb/Sb–C (200 kg N ha −1 ) with no‐rye averaged 61.5, 47.4, and 41.8 kg ha −1 , while grain yield averaged 12.5, 12.3, and 4.0 t ha −1 for corn (C–C), corn (C–Sb/Sb–C), and soybean (C–Sb/Sb–C), respectively. Planting rye under these rotations gave annual average reductions in nitrate‐N losses relative to corresponding no‐rye treatments of 2.9 (4.7%), 3.4 (7.3%), and 6.5 kg ha −1 (15.6%), with rye N uptake of 10.3, 12.1, and 33.5 kg ha −1 ; and rye biomass production of 0.61, 0.74, and 2.0 t ha −1 , respectively. EPIC assessment indicates that winter rye as cover crop did not negatively impact the subsequent corn/soybean yield and proved to be an effective strategy for reducing nitrate‐N losses, particularly following the soybean crop.

  PublisherOA PDFDOI

• Quantifying the Role of Agricultural Management Practices on Erosion and Runoff in Minnesota Under Future Climate Scenarios

  JAWRA Journal of the American Water Resources Association · 2025-10-01

  articleOpen accessSenior authorCorresponding

  ABSTRACT There is still uncertainty about how agricultural watersheds will respond to changing future climate. This modeling study aims to evaluate different agricultural strategies for reducing runoff and hillslope soil losses in two Minnesota watersheds. The Water Erosion Prediction Project (WEPP) model was implemented to simulate various strategies and adoption rates of cover crops, reduced tillage systems, and perennial crops across the watersheds. Daily weather was simulated for 1965–2019, 2020–2059, and 2060–2099 in CLImate GENerator (CLIGEN) using climate projections from two Coupled Model Intercomparison Project Phase 5 models and three relative concentration pathway (RCP) emission scenarios. Results indicate that the modeled future scenarios will maintain or slightly decrease both runoff and hillslope soil losses in the two simulated watersheds, while steeper agricultural fields will remain at high risk based on the modeled unsustainable erosion rates. Among the agricultural strategies analyzed, the study revealed that integrating perennials into 50%–70% of fields decreased future watershed runoff by 2%–5% and hillslope soil losses by 24%–46%. Reduced tillage also proved substantial reductions in contrast to the negligible effects observed with cover crops. Given the uncertainty in future climate, further research is warranted to expand this type of scenario analysis to other parts of Minnesota and the Midwest USA, helping build better climate resilience into cropping systems.

  PublisherOA PDFDOI

• Combining Satellite and Weather Data for Crop Type Mapping: An Inverse Modelling Approach

  Society for Industrial and Applied Mathematics eBooks · 2024-01-01 · 6 citations

  book-chapter

  Accurate and timely crop mapping is essential for yield estimation, insurance claims, and conservation efforts. Over the years, many successful machine learning models for crop mapping have been developed that use just the multi-spectral imagery from satellites to predict crop type over the area of interest. However, these traditional methods do not account for the physical processes that govern crop growth. At a high level, crop growth can be envisioned as physical parameters, such as weather and soil type, acting upon the plant leading to crop growth which can be observed via satellites. In this paper, we propose Weather-based Spatio-Temporal segmentation network with ATTention (WSTATT), a deep learning model that leverages this understanding of crop growth by formulating it as an inverse model that combines weather (Daymet) and satellite imagery (Sentinel-2) to generate accurate crop maps. We show that our approach provides significant improvements over existing algorithms that solely rely on spectral imagery by comparing segmentation maps and F1 classification scores. Furthermore, effective use of attention in WSTATT architecture enables detection of crop types earlier in the season (up to 5 months in advance), which is very useful for improving food supply projections. We finally discuss the impact of weather by correlating our results with crop phenology to show that WSTATT is able to capture physical properties of crop growth.

  PublisherDOI

Frequent coauthors

• Prasanna H. Gowda

  46 shared

• B. J. Dalzell

  Agricultural Research Service

  25 shared

• Yuxin Miao

  21 shared

• Jeffrey S. Strock

  University of Minnesota

  14 shared

• P. C. Robert

  Université de Lille

  14 shared

• P. C. Robert

  13 shared

• Patrick L. Brezonik

  University of Minnesota

  12 shared

• Carl J. Rosen

  12 shared

Labs

• Department of Soil, Water, and ClimatePI

Education

• PhD, Soil, Water & Climate

  University of Minnesota

Awards & honors

• Founding Fellow in the University of Minnesota’s Institute o…
• Fellow in the Soil Science Society of America (SSSA)
• Fellow in the Agronomy Society of America
• Pierre C. Robert Precision Agriculture Research Award from t…
• SSSA Soil Science Applied Research Award (2013)