Journal Paper Digests

Reading time ~8 minutes

Journal Paper Digests 2023 #24

  • Physics-Informed Neural Networks for solving transient unconfined groundwater flow
  • A copula-based parametric composite drought index for drought monitoring and applicability in arid Central Asia
  • Visible and near infrared spectroscopy for predicting soil nitrogen mineralization rate: Effect of incubation period and ancillary soil properties
  • Real-Time Irrigation Scheduling Based on Weather Forecasts, Field Observations, and Human-Machine Interactions
  • A Complete Water Balance of a Rain Garden
  • Agricultural value chains and food security in the Pacific: Evidence from Papua New Guinea and Solomon Islands

Agricultural value chains and food security in the Pacific: Evidence from Papua New Guinea and Solomon Islands

Small island developing states in the Pacific face multiple development challenges driven by rapid population growth and high transportation costs due to remoteness and isolation. Combined with the adverse consequences of extreme weather events and climate change, these challenges exacerbate poverty and food insecurity. Agricultural value chain development presents a pathway to poverty reduction and food security. In this paper, we assess the impacts of two value chain development projects in Papua New Guinea and Solomon Islands on dietary diversity and food security of small-scale producers. Project impacts on dietary diversity are positive and significant in both countries, but improved food security is only observed in Solomon Islands. These impacts are mainly driven by crop yields, value of crop production and sales, crop diversification and share of crop sales. We find that treatment households are more likely to consume less nutritious foods such as sweets and oils. Our findings expand the literature in a data-scarce region and caution that value chain interventions without nutrition-focused components to induce behavioral change may have unintended impacts on healthy diets.

A Complete Water Balance of a Rain Garden

A bioinfiltration rain garden was retrofitted from an existing traffic island at Villanova University in 2001. It has been monitored continuously since 2003 at a 5-min timeseries resolution and with instrumentation that would enable a water balance calculation. This 20-year data set allows for an in-depth analysis of the hydrologic pathways and management in the rain garden. Using physical equations and modeled data (based on real-time measurements), a balance of all influent, stored, and effluent water within the rain garden was constructed. Analysis shows the rain garden captures 73.5% of runoff, resulting in a post-implementation management of 86.2% of all rainfall in its watershed. In comparison to the hydrology of other land covers, implementing the rain garden resulted in the management of 37.6% more rainfall than pre-implementation, producing a hydrological signature similar to that of cultivated land or low development levels (e.g., 30% impervious). Additionally, with the long data record, several statistical techniques were applied to determine the amount of monitoring needed for a certain level of precision in system performance assessment. For 5% uncertainty, approximately 3 years of continuous data is needed to assess performance. This analysis not only facilitates understanding the function of rain garden systems, but also provides conclusions and methodology for understanding the uncertainty associated with the extent of monitoring performed on these green stormwater infrastructure systems. These findings provide practical knowledge as monitoring of stormwater management infrastructures is becoming a more standard part of their operation.

Real-Time Irrigation Scheduling Based on Weather Forecasts, Field Observations, and Human-Machine Interactions

Real-time irrigation schedules have been shown to outperform predetermined irrigation schedules that do not consider the present state and requirements. However, implementing real-time irrigation scheduling requires reliable present soil-crop-atmosphere dynamics and weather predictions; moreover, enabling farmers to adopt recommended water applications remains challenging as they rely on personal experience and knowledge. Farmers and computer-based tools are rarely connected in a closed-loop and farmers’ feedback are usually not incorporated into a real-time modeling procedure. To resolve these critical issues, this paper addresses the feasibility of a real-time irrigation scheduling tool (RTIST) based on weather forecasts, field observations, and human-machine interactions. RTIST integrates a simulation & optimization model, a data assimilation (DA) technique, and a human-computer interaction method, and enables optimality, accuracy, and applicability of the tool. The principle of the RTIST is to engage farmers directly into computer modeling, and support irrigation scheduling decisions jointly based on model provided information and farmers’ own justification. The optimization and simulation are validated by running the tool on two crop fields, showing the accuracy of present estimation and future prediction of soil moisture and leaf area index, taking advantage of field observation and DA. The applicability of RTIST is tested via virtual irrigation exercises with a group of farmers for a corn field in Eastern Nebraska. RTIST with farmers’ direct engagement shows increased productivity in comparison to traditional practices. Especially, farmers’ feedbacks show interest in using the tool in real-world irrigation scheduling and providing meaningful suggestions to improve the tool for real-world application.

Visible and near infrared spectroscopy for predicting soil nitrogen mineralization rate: Effect of incubation period and ancillary soil properties

Soil nitrogen mineralization rate (SNMR) influences crop N uptake and nitrate leaching leading to environmental pollution. This study aims at (i) examining whether visible and near-infrared reflectance spectroscopy (vis-NIRS) can predict SNMR and (ii) investigating if incubation periods and ancillary soil attributes can improve the prediction accuracy. Total 133 soil samples collected from seven fields were incubated under aerobic conditions for 60 days with seven batches of sub-samples. Mineral N was measured at regular time intervals and soil samples were scanned using a vis-NIRS sensor (Tec5 Technology, Germany) parallelly. SNMR was determined by fitting a zero-order kinetic to the net mineralized N as a function of the incubation time. Soil total nitrogen (TN), total carbon (TC) and electrical conductivity (EC) were determined once. Partial least squares regression (PLSR) models were calibrated individually for each field both for vis-NIR spectra and its combinations with TN, TC and EC. Six out of seven batches of sub-samples were used for calibrating PLSR when remaining one batch was used for model validation, and it rotated across all seven batches. Vis-NIRS alone predicted SNMR with moderate accuracy in five of seven fields (coefficient of determination, 0.53 ≤ R2 ≥ 0.66, ratio of prediction to deviation, 1.51 ≤ RPD ≥ 1.76), while models were poor in two fields (R2 = 0.23–0.26, RPD = 1.18 – 1.20). Inclusion of soil TC, TN and/or EC was expected to improve accuracy, but improvements varied across fields (R2 = 0.23–0.79, RPD = 1.18 – 2.26). Similarly, the incubation period increased vis-NIRS prediction accuracy, but frequently occurred among 2nd to 6th batches (R2 = 0.35–0.82, RPD = 1.28 – 2.44). Even incorporating secondary properties and increasing incubation duration hardly improved predictions, improvement can be compromised since it is not significant mostly and often underperformed or remained unchanged. Considering the time and effort required to incubate and analyze soil properties, this study suggests using a vis-NIRS sensor to estimate SNMR in fresh soil conditions i.e., without incubation and incorporation of secondary properties.

A copula-based parametric composite drought index for drought monitoring and applicability in arid Central Asia

Due to the complexity of meteorological and hydrological conditions in a changing environment, previous drought indices for monitoring a specific drought type do not reflect the overall regional situation of water scarcity. Therefore, in order to obtain accurate and reliable drought monitoring, a more integrated drought index should be developed to identify drought events comprehensively. In this paper, a non-linear trivariate drought index (NTDI) was constructed based on the joint probability distribution of parametric copulas, combining precipitation (P), potential evapotranspiration (PET), and root zone soil moisture (SM) variables. Subsequently, it was respectively compared with four drought indices, SPEI, SSMI, SC-PDSI and TVDI, and cross-validated with actual recorded drought events and annual crop yield to evaluate its applicability in arid Central Asia (ACA). The results indicated that: (1) Frank copula (1-,3-month scale) and Gumbel copula (6-,12-month scale) were considered to be the best-fitted copula functions for constructing joint probability distributions in the ACA. (2) The NTDI integrated the P-PET and SM drought signals to sensitively capture drought onset and duration, reflecting the combined characteristics of meteorological and agricultural drought. (3) The drought information expressed by NTDI was generally consistent with recorded drought events, and the monitoring results are accurate. (4)The NTDI performed better in agricultural drought monitoring than other drought indices. This study provides a reliable multivariate composite indicator which is significant for drought monitoring, prevention and risk assessment in ACA.

Physics-Informed Neural Networks for solving transient unconfined groundwater flow

Neural networks excel in various machine learning applications; however, they lack the physical interpretability and constraints crucial for numerous scientific and engineering problems. This limitation hinders their ability to accurately capture and predict complex physical systems’ behavior, potentially yielding inaccurate or unreliable results. Physics-Informed Neural Networks (PINNs) are a class of machine learning models that integrate the power of neural networks with the physical laws governing natural phenomena. PINNs provide an effective tool for solving intricate physical problems, ranging from fluid dynamics to materials science, by incorporating physical constraints into the neural network architecture. PINNs can substantially enhance the accuracy and efficiency of model predictions, even in data-limited situations. This work offers insight into recent developments in the PINN field, including their mathematical formulation and training algorithms, and emphasizes their application in solving transient unconfined groundwater flow. In this context, the phreatic surface acts as a spatiotemporally varying boundary condition, and properly accounting for its position is vital for precise predictions of unconfined groundwater flow and related environmental and engineering applications. The study’s objective is to develop a reliable model for estimating the phreatic surface and the spatiotemporal distribution of piezometric heads in a vertical cross-section of an unconfined aquifer. Two cases are examined: the first involves a homogeneous and isotropic aquifer, while the second comprises a mildly heterogeneous and anisotropic one. The challenges and opportunities arising from this emerging research area are also explored, and essential directions for future research are underscored.

Journal Paper Digests

## Journal Paper Digests 2024 #5* Global cropland nitrous oxide emissions in fallow period are comparable to growing-season emissions* So...… Continue reading

Journal Paper Digests

Published on February 28, 2024

Journal Paper Digests

Published on February 27, 2024