Can gridded real‐time weather data match direct ground observations for irrigation decision‐support?

Abstract

Agricultural decision‐support systems are commonplace in extension and outreach. These systems typically rely on either historical or direct ground observations to make grower recommendations. Sensor data create many challenges for application developers, though, including managing device‐level characteristics, ensuring observation data quality, and handling missing data. In many data flows for decision support, encapsulation is a best practice development approach where data collection and storage are isolated from application development by application programming interfaces (APIs). Here, we consider the data quality of gridded and non‐gridded weather data types in agricultural modeling for predicting evapotranspiration (ET) and growing degree days (GDD). We compare API‐accessible gridded datasets from GEMS Exchange to MESONET (mesoscale network of weather and climatological stations) data from the Minnesota Department of Agriculture (MDA). We evaluate the data sources directly for goodness‐of‐fit for solar radiation, temperature (min and max), dew point, and wind speed, as well as downstream predictions of reference ET (ETref) and GDD. Our findings show that gridded data, despite its tendency to overestimate solar radiation, does not significantly impact the accuracy of ET (R² = 0.92 for 2022 and 0.93 for 2023; root mean square error [RMSE] = 0.55 mm for 2023) or GDD predictions (R² = 0.99 for 2022 and 0.98 for 2023; RMSE = 0.53°C [2022], RMSE = 0.70°C [2023]). This suggests that application programming interface (API)‐based gridded data, accessible for all locations, can be reliably used for ETref and GDD modeling for decision support and complements MESONET measures by providing developers with standard software interfaces for real‐time weather information.

Abbreviations

• ET
• evapotranspiration
• ETref
• reference evapotranspiration
• FAO
• Food and Agriculture Organization of the United Nations
• GDD
• growing degree days
• GEMS
• genetics, environment, management, and socioeconomics
• IMA
• irrigation management assistant
• MAE
• mean absolute error
• MDA
• Minnesota Department of Agriculture
• RMSE
• root mean square error

INTRODUCTION

Big data have enormous potential to address the complex challenges of agriculture for a range of different stakeholders (Coble et al., 2018; Hubbard et al., 1983). However, major challenges remain to operationalize data-driven decision-making ranging from data quality and interoperability to model performance and user trust (Runck, Joglekar, et al., 2022). There are different types of data that are available, and it is important to consider what these weather and climate data products are to make the best use of them.

The first type of data is weather station observations, and these are fundamental measurements and their direct use in decision support has some advantages. However, they face various challenges including limited spatial representativeness of individual stations, particularly in areas with varied landscapes (Pielke et al., 2007). Despite this, weather stations are reliable sources of data, and their use in validating gridded datasets helps ensure consistency and accuracy. Non-gridded point datasets from weather stations are typically unevenly spaced, thus offering flexibility in capturing phenomena, but issues related to individualized maintenance and sparse coverage can gradually undermine data quality, presenting challenges in analysis and interpretation (Blankenau et al., 2020 Hughes et al., 2009; Mendelsohn et al., 2007).

In contrast, gridded datasets interpolate weather station data in combination with other sources of information such as elevation, surrounding land cover, and satellite observations into a regular pattern. This broad coverage makes them easily accessible for modeling/irrigation and crop growth (Araghi, Martinez, Olesen, 2022; Chakraborty et al., 2018). Gridded datasets are used to offer a more spatially complete representation, serving as a partial solution to address issues such as missing data and spatial bias (Robeson & Ensor, 2006). Gridded weather data are typically less accurate than measured weather data for predicting key agricultural factors like humidity and rainfall (Mourtzinis et al., 2017), but have utility in modeling agricultural systems (Araghi, Martinez, Olesen, Hoogenboom, 2022). Comparisons between different types of gridded and point datasets tend to show variable relationships between the datasets, with gridded datasets overestimating reference evapotranspiration (ETref) (Blankenau et al., 2020; Hughes et al., 2009).

This discrepancy stems from a tendency to overestimate parameters such as wind speed, solar radiation, and air temperature, while simultaneously underestimating vapor pressure in comparison to readings from agricultural weather stations (Blankenau et al., 2020; Battisti et al., 2019; Monteiro et al., 2021). However, gridded data often provide an important source where there is no point data (Blankenau et al., 2020; Volk et al., 2023). Both types of data face limitations, from data resolution and interpolation errors in gridded datasets to difficulties in spatial analysis with non-gridded data. Understanding these problems is crucial for effectively using and interpreting the data in scientific and practical applications. Here we build on prior work by exploring the relationship between gridded and non-gridded agricultural weather datasets for 13 locations in Minnesota (Figure 1,2).

[IMAGE OMITTED. SEE PDF]

Understanding weather-related variables (Environment—E) and how they interact with genetic variation (G) and management approaches (M) (G × E × M) is critical for agricultural sustainability (Fischer, 2015; Hawkesford & Riche, 2020). Many components of weather can be measured with high fidelity and can then be used for modeling (Supit, 1986). The length of any time series and the data quality impacts the ability to use data to predict plant productivity with accuracy (Tsuji et al., 1998; Van Ittersum et al., 2013; Whisler et al., 1986). Having accurate data is essential for retrospective analysis (Fischer, 2015), in-season decision support (Frisvold & Murugesan, 2013), and future prediction (Hoogenboom, 2000). High quality data are particularly useful for simulation, being necessary for complex crop models (Hoogenboom, 2000). In addition to these generic uses, there are also commodity-specific uses such as vineyard management (McNew et al., 1991) and dairy farm management (Jones et al., 1989).

Specific measurements have different uses and require different levels of data quality. For example, evapotranspiration (ET) is a crucial component for irrigation scheduling. By integrating real-time weather data with ET models, farmers and irrigation managers can make informed decisions on irrigation (Akintuyi, 2024; Liang & Shah, 2023). For ET, gridded weather datasets have more utility than non-gridded datasets, and have a long history of use for individual farms and remote sensing (Allen et al., 1991; Baldocchi et al., 2001; Bastiaanssen et al., 2005; Senay et al., 2022). Gridded ETref data products are widely used in remote sensing ET models such as METRIC (Mapping EvapoTranspiration at high Resolution with Internalized Calibration) and EEFlux (Earth Engine Evapotranspiration Flux), which are based on Landsat imagery (Allen et al., 2007; Allen et al., 2015).

Another common use of gridded and non-gridded data is to calculate growing degree days (GDD). This metric has been used across a broad range of crops (e.g., soybean, corn, sorghum, forages) to quantify growth and help in decision support (Gerber et al., 2024; Kessler et al., 2020; Mourtzinis et al., 2017; X. Xu & Lan, 2017). GDD has been a standard method for creating cropping calendars and understanding the rate at which field crops develop (Dass & Rai, 2013; Hortic & Arnold; Liu et al., 2022; Madariaga & Knott, 1951; McMaster & Wilhelm, 1997; Miller et al., 2001). At the global scale, gridded weather data are not as reliable as measured weather data for precise farming uses, especially when predicting crop water needs, yields, and GDD and suggests regularly checking gridded weather data accuracy in agricultural research (Mourtzinis et al., 2017). The variability and trends of GDD across different crop growth stages enable more accurate predictions and adaptations in farming practices to meet the increasing demands of food production (Anandhi, 2016; Neild & Seeley, 1977; Ojeda-Bustamante et al., 2004).

Accurate data are essential for creating reliable agricultural statistics, and many regional extension decision-support tools rely on data products driven by the nearest weather station to a farm (Frisvold & Murugesan, 2013). These data have greatly benefited from retrospectively understanding field seasons, but suffer from having limited spatial extent and thus being of less use in forecasting (Mourtzinis et al., 2017). The lack of detailed daily weather data at the right scale makes it difficult to forecast the weather's impact on crops (Van Wart et al., 2013). Although there are many weather stations in the US Corn Belt, most do not record all the necessary parameters (e.g., solar radiation, ET) for predictive modeling (Grassini et al., 2015; Hoogenboom et al., 2019).

Here, we consider weather and climate data products. These real-time data streams play a foundational role in driving digital agriculture models. These data are typically either gridded or non-gridded. Each type of data has characteristics that make them suitable for different kinds of applications. This limited number of comparisons, in the US Midwest, motivated our comparison of gridded and non-gridded data on ETref and GDD modeling. By employing both gridded and non-gridded datasets, we capture a broader spectrum of environmental data, enhancing the robustness and applicability of our model for reference evapotranspiration from Penman–Monteith equation (Food and Agriculture Organization of the United Nations 56 [FAO-56]; Monteith, 1965).

In addition to evaluating weather datasets for ETref and GDD calculations, we are incorporating these datasets into the irrigation management assistant (IMA) tool, a free online platform designed to optimize water use in agriculture, for Minnesota farmers (ima.respec.com). IMA tool utilizes weather data, crop type, and user input to provide automated daily recommendations for efficient irrigation scheduling. Currently, the tool supports irrigation scheduling for five different crops—corn, soybeans, alfalfa, potatoes, and edible beans—across approximately 6500 acres, with ongoing expansion across Minnesota (B. Runck, Sharma, et al., 2022). As part of our ongoing efforts to improve irrigation decision-making, we are working to refine ET and GDD recommendations on corn within the IMA tool. ETref, derived from weather data (FAO-56; Monteith, 1965), is integral to estimating crop water demand and optimizing irrigation scheduling. Similarly, GDD from minimum and maximum temperature data along with crop base temperature is essential for tracking crop growth stages, ensuring irrigation aligns with key developmental phases (McMaster & Wilhelm, 1997). By enhancing these calculations and integrating more precise datasets, we aim to improve the accuracy and effectiveness of irrigation recommendations, ultimately helping growers make better-informed decisions.

Our main objective was to ascertain the best type of weather data product for use in agricultural research and extension. We seek to provide insights into the implications of dataset selection on model performance, offering guidance for researchers and practitioners in selecting the most appropriate dataset type for their specific needs. This work contributes to the knowledge of weather data analysis but also has practical implications for improving the accuracy and efficiency of environmental and agricultural models and tools.

MATERIALS AND METHODS

Data acquisition

There were two different platforms used to gather data in this study: (1) the GEMS (genetics, environment, management, and socioeconomics) Exchange was used to obtain gridded weather datasets. GEMS Exchange is a service of the University of Minnesota that provides application programming interface access to data products and models. The platform uses the EASE-Grid 2.0 projection (Equal-Area Scalable Earth Grids), a hierarchical grid system with multiple levels of resolution, ranging from fine-scale grids (1 m) to coarse-scale grids (36 km) (Thompson et al., 2022). Spatial interpolation in GEMS Exchange is performed using best practices in the domain for combining multiple data sources and is proprietary in nature. GEMS Exchange is currently proprietary and relies on station data combined with ERA-5 renalysis plus other observational datasets to enhance accuracy and completeness, especially in areas with sparse station coverage. These methods estimate values for grid cells by analyzing spatial relationships between known data points and applying mathematical functions to fill gaps. The second data source was the Minnesota Department of Agriculture (MDA) weather network from which we obtained non-gridded point data from 13 weather stations from 13 locations of Minnesota (Figure 1; Table 1). This data acquisition involved accessing MDA's publicly available meteorological data, selecting relevant stations to ensure geographic coverage pertinent to our study, and integrating these site-specific measurements into our dataset (Minnesota Department of Agriculture, n.d.). This process allowed us to use high-accuracy weather observations, which is crucial for evaluating and contrasting with the interpolated data from gridded sources. The weather variables used for both gridded and non-gridded datasets included minimum/maximum temperature, wind speed, dew point, and solar radiation from the same latitude and longitude locations to ensure direct comparability.

TABLE 1 Weather data sources and access dates used in this study.

Data source	Variable	Accessed	Link
GEMS	Temperature (max/min)	March 24, 2024	GEMS exchange
MDA	Temperature (max/min)	April 3, 2024	MDA data
GEMS	Wind speed	March 24, 2024	GEMS exchange
MDA	Wind speed	April 3, 2024	MDA data
GEMS	Solar radiation	March 24, 2024	GEMS exchange
MDA	Solar radiation	April 3, 2024	MDA data
GEMS	Dew point	March 24, 2024	GEMS exchange
MDA	Dew point	April 3, 2024	MDA data

Data cleaning

Data loading

The initial step in our data cleaning involved loading the datasets for each location and year. This was accomplished using the read_csv function from the Pandas library, which allowed us to efficiently import our weather data stored in CSV format from the “GEMS Exchange” and “MDA” directories.

Conversion of units

To ensure consistency across datasets, we converted the units of key meteorological variables. Specifically, wind speed values at 3 m height, originally recorded in miles per hour (mph) in the MDA dataset, were converted to meters per second (m/s) using a conversion factor (1 mph = 0.44704 m/s). These measurements were then standardized to a height of 2 m using FAO-56 report formula U2 = U × [4.87/ln(67.8 × h − 5.42)] where U is the wind speed at height h (Allen, 1998). Wind speed measurements from the GEMS Exchange dataset, recorded at a height of 10 m and already in units of m/s, were similarly adjusted to 2 m. This standardization of units was crucial for accurate cross-dataset comparisons. Then, we compared the wind speed values at heights of 2 m to check for any differences and make sure our comparisons were accurate. Additionally, for solar radiation data, we applied a conversion factor of 0.0864 to convert from watts per square meter (W/m²) to megajoules per square meter per day (MJ/m²/day) in the GEMS Exchange datasets. This means that 1 W/m² is approximately equal to 0.0864 MJ/m²/day. This conversion factor is in accordance with the FAO-56 report, which outlines methodologies for meteorological data processing and analysis (Allen, 1998).

Handling missing data

During our data assessment, we found the datasets to be quite robust, with no instances of zero values detected, indicating the overall quality of the data. However, we did encounter two missing rows in the month of December 2022 within the MDA's Pine Point location datasets. These missing rows were documented and addressed as part of our data evaluation process. Additionally, to maintain consistency, we also removed the corresponding dates from the GEMS 2022 dataset. Despite these missing rows, the datasets remained reliable for our analyses, ensuring the integrity of our findings.

Data alignment

Aligning data from different sources was essential for our comparative analysis. We ensured that both the GEMS Exchange and MDA datasets had an equal number of rows by matching data entries based on corresponding dates and locations. This alignment was critical for conducting pairwise comparisons and statistical analyses.

Analysis

To determine how the choice between gridded and non-gridded data influences model accuracy, particularly in light of cumulative errors, and how these choices intersect with the availability of labor for dataset management and model application. We evaluated these datasets based on their practicality—whether they simplify or complicate the process of data handling (ease vs. expense)—and their reliability and precision (data accuracy). The first part was to examine the suitability of these datasets for modeling GDD and ETref, critical parameters in agricultural and environmental studies. GDD assigns a heat value to each day by adding the daily maximum and minimum temperatures, dividing by two to get an average, and subtracting the plant's base temperature, the temperature below which development stops (McMaster & Wilhelm, 1997; Miller et al., 2001; S. Yang et al., 1995).

This process began with the construction of combined data frames for each specific location under study (i.e., 13 locations for 2022 and 2023 years) from the MDA weather network, and we chose the gridded data from the exact same locations. Each combined data frame was structured to include two main columns: one representing the MDA dataset (non-gridded) and the other representing the GEMS Exchange dataset (gridded), thereby facilitating direct comparisons. The primary analytical technique employed was linear regression, which served to quantify and model the relationship between the datasets from the two sources. This approach was specifically chosen to compare the gridded GEMS data with the point-specific MDA measurements under the hypothesis that a predictable relationship exists between them.

Using the linear regression class from scikit-learn, we calculated the best-fit line that minimized the squared differences between the observed GEMS values and the predicted values derived from the MDA dataset. To quantify the effectiveness of our model and the level of agreement between the datasets, we computed the following metrics: R-squared (R²), which measures the proportion of variance in the dependent variable that is predictable from the independent variable; root mean square error (RMSE), which provides the average magnitude of the regression errors, highlighting the deviation of predicted values from actual values; and mean absolute error (MAE), which averages the absolute differences between predictions and actual outcomes, giving a clear measure of prediction accuracy.

In our research, we developed a detailed ET model by integrating critical atmospheric and geographic parameters to provide a thorough understanding of the ET process. This model includes essential climate variables such as temperature extremes, solar radiation, dew point temperature, and wind speed, along with geographic details like sunset hour angle, solar declination, latitude, and elevation. We used the well-established Penman–Monteith equation (FAO-56) to ensure accurate and reliable ET estimations.

To evaluate the data accuracy for wind speed, solar radiation, minimum temperature, maximum temperature, and dew point using RMSE, MAE, and R², we assessed the linear correlation between the datasets using the Pearson correlation coefficient. Furthermore, we evaluated the Penman–Monteith model's performance for reference evapotranspiration (ET ref) and GDD modesl in comparison with the MDA ET values and GDD values. Our approach involved utilizing GEMS Exchange data to compute reference ET and GDD, allowing us to compare the values with those derived from MDA.

We selected corn to calculate GDD because it is one of the most widely grown crops in Minnesota, making it highly relevant for regional agricultural studies (Bundy et al., 2024). The GDD estimates the growth and development of corn by calculating the accumulation of heat units over time. GDD = [(T MAX + T MIN)/2] − T BASE where TMax = daily maximum temperature, TMIN = daily minimum temperature, and TBASE = base temperature of crop. Next, the daily minimum and maximum temperatures are adjusted. The minimum temperature is set to either the daily minimum temperature or 10°C, whichever is higher. The maximum temperature is set to either the daily maximum temperature or 30°C, whichever is lower (Cross & Zuber, 1972). The mean temperature for the day is then calculated by averaging the adjusted minimum and maximum temperatures. The daily GDD is computed by subtracting the base temperature of 10°C for corn that is commonly used in literature from the mean temperature (McMaster & Wilhelm, 1997; Darby & Lauer, 2002; Abendroth et al., 2019).

The ETref function calculates key meteorological and agricultural parameters using FAO-56 guidelines (Allen, 1998). The equation we used was:$$\;{\mathrm{ET}} = (0.408{{\Delta}}( {{\mathrm{Rn}} - G} ) + {{\gamma}}( {900/( {T + 273} ){\mathrm{\;U}}2 ( {{\mathrm{es}} - {\mathrm{ea}}} )} )/( {{{\Delta}} + {{\gamma}}( {1 + 0.34{\mathrm{U}}2} )} )$$ where Rn = net radiation at the crop surface (MJ/m²/day), G = soil heat flux density (MJ/m²/day), T = air temperature at 2 m height (°C), U2 = wind speed at 2 m height (m/s), es = saturation vapor pressure (kPa), ea = actual vapor pressure (kPa), es − ea = saturation vapor pressure deficit (kPa), D = slope vapor pressure curve (kPa/°C), $$\gamma $$ = psychrometric constant (kPa/°C). It starts by determining the julian date (day of the year) and adjusts wind speed to 2 m using a logarithmic formula. Solar radiation is converted from watts per square meter to megajoules per square meter per day. The function then calculates the average temperature from daily minimum and maximum temperatures and uses this to find the latent heat of vaporization. Atmospheric pressure is derived from elevation, and from this, the psychrometric constant and the slope of the saturation vapor pressure curve are calculated. Saturation vapor pressures at daily maximum and minimum temperatures are used to find the average saturation vapor pressure. Actual vapor pressure is computed using the dew point temperature, and vapor pressure deficit is calculated as the difference between the average saturation vapor pressure and actual vapor pressure. The function also calculates the inverse relative distance of Earth–Sun, solar declination, and sunset hour angle to determine extraterrestrial radiation. It then computes clear sky solar radiation, net solar radiation, and net longwave radiation. Net radiation is obtained by subtracting net longwave radiation from net solar radiation. Finally, the function calculates reference evapotranspiration (ET) using net radiation, temperature, wind speed, and vapor pressure deficit, providing essential data for agricultural planning and water management.

This comparative analysis provided valuable insights into the strength and direction of the relationship between the datasets across the 13 locations studied, enhancing our understanding of their interrelation. Additionally, we plotted a 1:1 line on each graph to visually assess alignment between the datasets. This line represents a perfect agreement, serving as a benchmark for evaluating how closely the data points from the two datasets correspond to each other. The presence of this line in our plots enabled a straightforward visual comparison, supplementing our statistical analysis by visually demonstrating the degree to which the datasets agree or differ. Data and analysis materials can be found at .

RESULTS AND DISCUSSION

Relationship between parameters

Our regression analysis between gridded and non-gridded weather datasets for 2022 and 2023 revealed new insights (Figures S1–S14). Maximum temperature measurements showed high model accuracy (R² = 0.99, RMSE = 1.13°C, MAE = 0.79°C in 2022 and R² = 0.99, RMSE = 1.27°C, MAE = 0.85°C in 2023), indicating robust performance across both types of datasets. Minimum temperature measurements showed high model accuracy (R² = 0.96, RMSE = 2.84°C, MAE = 1.34°C in 2022 and R² = 0.83, RMSE = 4.93°C, MAE = 2.38°C in 2023), indicating good performance across both types of datasets. Despite the fact that wind speed was originally measured at different heights (10 m for GEMS exchange gridded data and 3 m for MDA non-gridded data), after adjusting the measurements to a consistent height of 2 m using FAO-56 formula, the wind speed values exhibited moderate accuracy. In 2022, the model achieved an R² of 0.76, RMSE of 0.61 m/s, and MAE of 0.46 m/s. In 2023, the model showed an R² of 0.69, RMSE of 0.57 m/s, and MAE of 0.42 m/s. These results reflect some challenges in modeling wind speed across different terrains, even after height adjustment. Further refinement of the models and consideration of terrain-specific factors may enhance the accuracy of wind speed predictions.

Daily net solar radiation presented the lowest accuracy (R² = 0.58, RMSE = 5.64 MJ/m²/day, MAE = 4.58 MJ/m2/day in 2022 and R² = 0.61, RMSE = 5.68 MJ/m²/day, MAE = 4.69MJ/m2/day in 2023), highlighting a critical area where both gridded and non-gridded data struggle, likely due to atmospheric conditions like cloud cover not being uniformly captured. Dew point data showed excellent consistency and accuracy (R² = 0.98, RMSE = 1.66°C, MAE = 0.91°C in 2022 and R² = 0.99, RMSE = 1.21°C, MAE = 0.63°C in 2023), demonstrating reliable modeling across both datasets (Figure 2; Table 2).

[IMAGE OMITTED. SEE PDF]

TABLE 2 Regression between different parameters among different datasets.

	2022	2023	2022	2023
Parameters	Root mean square error	Mean absolute error	Root mean square error	Mean absolute error	R2 score	Correlation coefficient	R2 score	Correlation coefficient
Min temp (°C)	2.84	1.34	4.93	2.38	0.96	0.97	0.83	0.91
Max temp (°C)	1.13	0.79	1.27	0.85	0.99	0.99	0.99	0.99
Wind Speed(m/s)	0.61	0.46	0.57	0.42	0.76	0.87	0.69	0.83
Solar radiation MJ/m2/day	5.64	4.58	5.68	4.69	0.58	0.75	0.61	0.78
Dew point(°C)	1.66	0.91	1.21	0.63	0.98	0.99	0.99	0.99

While gridded data perform adequately and offer the advantage of consistent spatial coverage, it shares similar challenges with non-gridded data in accurately capturing parameters such as solar radiation. The findings indicate that while temperature and dew point data exhibit high accuracy, wind speed and solar radiation measurements require further refinement. Enhancing the models and incorporating more terrain-specific factors and improved atmospheric modeling can lead to better predictions and support agricultural and climatological applications. Many places have daily records of high and low temperatures and rainfall. However, there are only a few sites that provide accurate measurements of solar radiation (Hunt et al., 1998; Nonhebel, 1993). Additionally, the complex interactions between atmospheric components, such as clouds, aerosols, and gases, further complicate accurate measurements and estimations (Almorox et al., 2011; K. Yang et al., 2006; Quej et al., 2016). This emphasizes the need for enhanced measurement or modeling techniques, particularly for solar radiation, to improve the overall reliability of weather predictions from both gridded and non-gridded sources.

Exploration of evapotranspiration (ET) gridded and point data products

In our comparative analysis, we utilized consistent parameters as the parameter we used to compare gridded and non-gridded data (Table 2): wind speed, dew point, minimum/maximum temperature, and solar radiation—to model reference evapotranspiration (ETref). The performance of the Penman–Monteith ET model between MDA and GEMS Exchange across the years 2022 and 2023 demonstrated high accuracy and consistency. In 2022, the model achieved an RMSE of 0.55 mm and a MAE of 0.037 mm, with an R² score of 0.92, indicating that 92% of the variance in ET could be accurately predicted. In 2023, there was a similar RMSE of 0.55 mm and MAE also remained stable with 0.38 mm, and the R² score of 0.93. These results suggest minimal deviations in prediction errors and an enhancement in the model's accuracy over time (Figure 3). The high accuracy and consistency observed in the Penman–Monteith ET model performance between point data from MDA and GEMS exchange data highlight reliability of these data sources for modeling reference evapotranspiration. The stable RMSE, MAE, and R² score across both years show the robustness of the gridded datasets used in the model and its potential for effective application in agricultural planning, water resource management, and climate studies.

[IMAGE OMITTED. SEE PDF]

Exploration of growing degree day (GDD) gridded and point data products

In our analysis of GDD for corn, we utilized both gridded and point data products with a base temperature of 10°C (50°F). Minimum and maximum temperatures were sourced from the MDA and GEMS Exchange, as detailed in Table 1. The performance comparison of the GDD model between MDA and GEMS data for corn demonstrated high accuracy and consistency. Specifically, in 2022, the GDD values comparison between MDA and GEMS exhibited an RMSE of 0.52°C, an R² score of 0.99, and an MAE of 0.29°C. For 2023, the comparison between GEMS gridded data and MDA point datasets showed an R² score of 0.98, an RMSE of 0.70°C, and an MAE of 0.35°C. The high accuracy and consistency observed in the GDD model performance between the MDA point data and GEMS gridded data underscore the reliability of both data sources for agricultural applications (Figure 4). The low RMSE and MAE values, coupled with the high R² scores for both 2022 and 2023, indicate that GEMS gridded data can serve as a viable alternative to MDA point data, particularly in scenarios where broad spatial coverage is required. The slight increase in RMSE and MAE in 2023 compared to 2022 might be attributed to variations in weather patterns or potential discrepancies in the granularity of the gridded data. Nonetheless, the R² score of 0.99 and 0.98 for both years suggests that the model consistently captures the GDD trends effectively (see Table 3).

[IMAGE OMITTED. SEE PDF]

TABLE 3 Regression analysis between evapotranspiration (ET) and growing degree days (GDD) models among different datasets.

	2022	2023	2022	2023
Model	Root mean square error	Mean absolute error	Root mean square error	Mean absolute error	R2 score	Correlation coefficient	R2 score	Correlation coefficient
ET(mm)	0.55	0.37	0.55	0.38	0.92	0.95	0.93	0.96
GDD(°C)	0.52	0.29	0.70	0.35	0.99	0.99	0.98	0.99

Limitations

Our study demonstrates that GEMS Exchange data show a good fit with MDA non-gridded datasets, indicating robust performance in predicting both GDD and ET. However, we have not evaluated different interpolation methods away from the specific measure locations. Uncertainty in interpolation methods can affect the reliability of predictions, especially in areas with sparse weather station coverage or heterogeneous environmental conditions. Our current tool (IMA) uses nearest-neighbor interpolation from the nearest weather station, and this work shows that the gridded sources are comparable in performance. Future work should compare this with MESONET (mesoscale network of weather and climatological stations)-derived interpolation methods, such as inverse distance weighting and kriging, against the gridded product.

CONCLUSION AND RECOMMENDATIONS

Our analysis demonstrated that these datasets are effective in accurately modeling key agricultural phenomena such as ET and GDD, as indicated by the high R² values and low error metrics for the years 2022 and 2023. These models play a critical role in crop yield predictions by accurately estimating heat accumulation and water requirements, both of which are fundamental for crop growth and productivity. For instance, GDD calculations enable the prediction of phenological stages, allowing farmers to optimize planting schedules and resource allocation. Similarly, precise ET estimations support efficient irrigation management, reducing water stress and enhancing overall yield stability. By addressing yield deviations caused by climatic variability, these models can inform proactive decision-making to mitigate potential yield losses. While there is a tendency for gridded data to overestimate parameters like solar radiation, this discrepancy did not significantly impact the accuracy of our ET models. This suggests that despite variations in solar radiation estimates, the gridded data from GEMS Exchange remain reliable for calculating ET, which is crucial for agricultural planning and water resource management.

On the other hand, the non-gridded data from the MDA provide highly localized weather information, essential for site-specific agricultural decisions and precision farming. While offering detailed, ground-level insights into environmental conditions, the coverage of non-gridded data is less extensive than that of gridded datasets. This precision comes with higher maintenance costs and potential spatial discrepancies, as weather stations are often not located directly on farms. These limitations may affect the data's ability to represent specific farm conditions, posing challenges for broader spatial applications and long-term agricultural studies. These data have implications for real-time decision support. As gridded data products become more accurate, and as longer time series are established, management timing can be refined improving both profitability and ecosystem services.

AUTHOR CONTRIBUTIONS

Samikshya Subedi: Formal analysis; methodology; visualization; writing—original draft; writing—review and editing. Ayoub Kechchour: Formal analysis; methodology; writing—review and editing. Michael Kantar: Project administration; writing—review and editing. Vasudha Sharma: Conceptualization; methodology; project administration; writing—review and editing. Bryan C. Runck: Conceptualization; formal analysis; methodology; project administration; resources; supervision; writing—review and editing.

ACKNOWLEDGMENTS

Funding for this project was provided by the Minnesota Environment and Natural Resources Trust Fund as recommended by the Legislative-Citizen Commission on Minnesota Resources (LCCMR) project ML 2021, Chp6, Art6, Sec 2, 04e-E812SIM 2021–266. We would like to thank Ann Piotrowski, Bobby Schulz, Logan Gall, Paul Senne, Christopher Anderson for helpful comments and suggestions.

References

Abendroth, L. J., Miguez, F. E., Castellano, M. J., & Hatfield, J. L. (2019). Climate warming trends in the US Midwest using four thermal models. Agronomy Journal, 111(6), 3230–3243.

Akintuyi, O. B. (2024). Adaptive AI in precision agriculture: A review: Investigating the use of self‐learning algorithms in optimizing farm operations based on real‐time data. Research Journal of Multidisciplinary Studies, 7(02), 016–030.

Allen, R. G. (1998). Crop evapotranspiration—Guidelines for computing crop water requirements‐FAO irrigation and drainage (Paper 56, pp. 60–64). Food and Agricultural Organization of the United Nations.

Allen, R. G., Howell, T. A., Pruitt, W. O., Walter, I. A., & Jensen, M. E. (1991). Lysimeters for evapotranspiration and environmental measurements. ASCE.

Allen, R. G., Morton, C., Kamble, B., Kilic, A., Huntington, J., Thau, D., Gorelick, N., Erickson, T., Moore, R., Trezza, R., Ratcliffe, I., & Robison, C. (2015). EEFlux: A Landsat‐based evapotranspiration mapping tool on the Google Earth Engine. In 2015 ASABE/IA Irrigation Symposium: Emerging Technologies for Sustainable Irrigation—A Tribute to the Career of Terry Howell, Sr. Conference Proceedings (pp. 1–11). American Society of Agricultural and Biological Engineers. https://doi.org/10.13031/irrig.20152143511

Allen, R. G., Tasumi, M., & Trezza, R. (2007). Satellite‐based energy balance for mapping evapotranspiration with internalized calibration (METRIC)—Model. Journal of Irrigation and Drainage Engineering, 133(4), 380–394. https://doi.org/10.1061/(ASCE)0733-9437(2007)133:4(380)

Almorox, J., Hontoria, C., & Benito, M. (2011). Models for obtaining daily global solar radiation with measured air temperature data in Madrid (Spain). Applied Energy, 88(5), 1703–1709.

Anandhi, A. (2016). Growing degree days–Ecosystem indicator for changing diurnal temperatures and their impact on corn growth stages in Kansas. Ecological Indicators, 61, 149–158.

Araghi, A., Martinez, C. J., & Olesen, J. E. (2022). Evaluation of multiple gridded solar radiation data for crop modeling. European Journal of Agronomy, 133, Article 126419.

Araghi, A., Martinez, C. J., Olesen, J. E., & Hoogenboom, G. (2022). Assessment of nine gridded temperature data for modeling of wheat production systems. Computers and Electronics in Agriculture, 199, 107189.

Baldocchi, D., Falge, E., Gu, L., Olson, R., Hollinger, D., Running, S., Anthoni, P., Bernhofer, C., Davis, K., Evans, R., Fuentes, J., Goldstein, A., Katul, G., Law, B., Lee, X., Malhi, Y., Meyers, T., Munger, W., Oechel, W., … Wofsy, S. (2001). FLUXNET: A new tool to study the temporal and spatial variability of ecosystem‐scale carbon dioxide, water vapor, and energy flux densities. Bulletin of the American Meteorological Society, 82(11), 2415–2434.

Bastiaanssen, W. G. M., Noordman, E. J. M., Pelgrum, H., Davids, G., Thoreson, B. P., & Allen, R. G. (2005). SEBAL model with remotely sensed data to improve water‐resources management under actual field conditions. Journal of irrigation and drainage engineering, 131(1), 85–93.

Battisti, R., Bender, F. D., & Sentelhas, P. C. (2019). Assessment of different gridded weather data for soybean yield simulations in Brazil. Theoretical and Applied Climatology, 135, 237–247. https://doi.org/10.1007/s00704-018-2383-y

Blankenau, P. A., Kilic, A., & Allen, R. (2020). An evaluation of gridded weather data sets for the purpose of estimating reference evapotranspiration in the United States. Agricultural Water Management, 242, 106376.

Bundy, L. R., Gensini, V. A., Ashley, W. S., Haberlie, A. M., & Changnon, D. (2024). United States crop conditions: 1986–2022. Agronomy Journal, 116, 1397–1416. https://doi.org/10.1002/agj2.21558

Chakraborty, D., Sehgal, V. K., Dhakar, R., & Das, D. K. (2018). Comparison of gridded weather datasets with point datasets over three locations of India. Journal of Agricultural Physics, 18(2), 218–223.

Coble, K. H., Mishra, A. K., Ferrell, S., & Griffin, T. (2018). Big data in agriculture: A challenge for the future. Applied Economic Perspectives and Policy, 40(1), 79–96. https://doi.org/10.1093/aepp/ppx056

Cross, H. Z., & Zuber, M. S. (1972). Prediction of flowering dates in maize based on different methods of estimating thermal units. Agronomy Journal, 64(3), 351–355. https://doi.org/10.2134/agronj1972.00021962006400030029x

Darby, H. M., & Lauer, J. G. (2002). Harvest date and hybrid influence on corn forage yield, quality, and preservation. Agronomy Journal, 94(3), 559–566. https://doi.org/10.2134/agronj2002.5590

Dass, B., & Rai, A. K. (2013). Growing degree days (GDD) measurement system to predict plant stages. Jawaharlal Nehru Krishi Vishwa Vidyalaya Research Journal, 47(3), 346–349.

Fischer, R. A. (2015). Definitions and determination of crop yield, yield gaps, and of rates of change. Field Crops Research, 182, 9–18. https://doi.org/10.1016/j.fcr.2014.12.006

Frisvold, G. B., & Murugesan, A. (2013). Use of weather information for agricultural decision making. Weather, Climate, and Society, 5, 55–69. https://doi.org/10.1175/WCAS‐D‐12‐00022.1

Gerber, J. S., Ray, D. K., Makowski, D., Butler, E. E., Mueller, N. D., West, P. C., Johnson, J. A., Polasky, S., Samberg, L. H., Siebert, S., & Sloat, L. (2024). Global spatially explicit yield gap time trends reveal regions at risk of future crop yield stagnation. Nature Food, 5(2), 125–135. https://doi.org/10.1038/s43016‐023‐00913‐8

Grassini, P., Van Bussel, L. G. J., Van Wart, J., Wolf, J., Claessens, L., Yang, H., Boogaard, H., De Groot, H., Van Ittersum, M. K., & Cassman, K. G. (2015). How good is good enough? Data requirements for reliable crop yield simulations and yield‐gap analysis. Field Crops Research, 177, 49–63. https://doi.org/10.1016/j.fcr.2015.03.004

Hawkesford, M. J., & Riche, A. B. (2020). Impacts of G × E × M on nitrogen use efficiency in wheat and future prospects. Frontiers in Plant Science, 11, 1157. https://doi.org/10.3389/fpls.2020.01157

Hoogenboom, G. (2000). Contribution of agrometeorology to the simulation of crop production and its applications. Agricultural and Forest Meteorology, 103(1–2), 137–157. https://doi.org/10.1016/S0168‐1923(00)00108‐8

Hoogenboom, G., Porter, C. H., Boote, K. J., Shelia, V., Wilkens, P. W., Singh, U., White, J. W., Lizaso, J. I., Moreno, L. P., Pavan, W., Ogoshi, R., Hunt, L. A., Tsuj, G. Y., & Jones, J. W. (2019). The DSSAT crop modeling ecosystem. In Advances in crop modelling for a sustainable agriculture (pp. 173–216). Burleigh Dodds Science Publishing.

Hubbard, K. G., Rosenberg, N. J., & Nielsen, D. C. (1983). Automated weather data network for agriculture. Journal of Water Resources Planning and Management, 109(3), 213–222. https://doi.org/10.1061/(ASCE)0733‐9496(1983)109:3(213)

Hughes, S. L., Holliday, N. P., Colbourne, E., Ozhigin, V., Valdimarsson, H., Østerhus, S., & Wiltshire, K. (2009). Comparison of in situ time‐series of temperature with gridded sea surface temperature datasets in the North Atlantic. ICES Journal of Marine Science, 66(7), 1467–1479. https://doi.org/10.1093/icesjms/fsp041

Hunt, L. A., Kuchar, L., & Swanton, C. J. (1998). Estimation of solar radiation for use in crop modelling. Agricultural and Forest Meteorology, 91(3‐4), 293–300. https://doi.org/10.1016/S0168‐1923(98)00055‐0

Jones, E., Batte, M. T., & Schnitkey, G. D. (1989). The impact of economic and socioeconomic factors on the demand for information: A case study of Ohio commercial farmers. Agribusiness, 5, 557–571. https://doi.org/10.1002/1520‐6297(198911)5:6%3c557::AID‐AGR2720050603%3e3.0.CO;2‐D

Kessler, A., Archontoulis, S. V., & Licht, M. A. (2020). Soybean yield and crop stage response to planting date and cultivar maturity in Iowa, USA. Agronomy Journal, 112(1), 382–394. https://doi.org/10.1002/agj2.20053

Liang, C., & Shah, T. (2023). IoT in agriculture: The future of precision monitoring and data‐driven farming. Eigenpub Review of Science and Technology, 7(1), 85–104.

Liu, L. W., Lu, C. T., Wang, Y. M., Lin, K. H., Ma, X., & Lin, W. S. (2022). Rice (Oryza sativa L.) growth modeling based on growth degree day (GDD) and artificial intelligence algorithms. Agriculture, 12(1), 59. https://doi.org/10.3390/agriculture12010059

Madariaga, F. J., & Knott, J. E. (1951). Temperature summations in relation to lettuce growth. Proceedings of the American Society for Horticultural Science, 58, 147–152.

Mcmaster, G., & Wilhelm, W. W. (1997). Growing degree‐days: One equation, two interpretations. Agricultural and Forest Meteorology, 87(4), 291–300. https://doi.org/10.1016/S0168‐1923(97)00027‐0

McNew, K. P., Mapp, H. P., Duchon, C. E., & Merritt, E. S. (1991). Sources and uses of weather information for agricultural decision makers. Bulletin of the American Meteorological Society, 72, 491–498.

Mendelsohn, R., Kurukulasuriya, P., Basist, A., Kogan, F., & Williams, C. (2007). Climate analysis with satellite versus weather station data. Climatic Change, 81(1), 71–83. https://doi.org/10.1007/s10584‐006‐9139‐x

Miller, P., Lanier, W., & Brandt, S. (2001). Using growing degree days to predict plant stages. Ag/Extension communications coordinator, communications services, Montana State University‐Bozeman.

Minnesota Department of Agriculture. (n.d.). Minnesota Ag weather network (MAWN). https://www.mda.state.mn.us/minnesota‐ag‐weather‐network

Monteiro, A. F. M., Martins, F. B., Torres, R. R., Marrafon de Almeida, V. H., Carvalho Abreu, M., & Vieira Mattos, E. (2021). Intercomparison and uncertainty assessment of methods for estimating evapotranspiration using a high‐resolution gridded weather dataset over Brazil. Theoretical and Applied Climatology, 146, 583–597. https://doi.org/10.1007/s00704-021-03747-1

Monteith, J. L. (1965). Evaporation and environment. In Symposia of the society for experimental biology (Vol. 19, pp. 205–234). Cambridge University Press.

Mourtzinis, S., Rattalino Edreira, J. I., Conley, S. P., & Grassini, P. (2017). From grid to field: Assessing quality of gridded weather data for agricultural applications. European Journal of Agronomy, 82, 163–172. https://doi.org/10.1016/j.eja.2016.10.013

Neild, R. E., & Seeley, M. W. (1977). Growing degree days predictions for corn and sorghum development and some applications to crop production in Nebraska (Research Bulletin No. 280). University of Nebraska‐Lincoln.

Nonhebel, S. (1993). The importance of weather data in crop growth simulation models and assessment of climatic change effects. Wageningen University and Research.

Ojeda‐Bustamante, W., Sifuentes‐Ibarra, E., Slack, D. C., & Carrillo, M. (2004). Generalization of irrigation scheduling parameters using the growing degree days concept: Application to a potato crop. Irrigation and Drainage, 53(3), 251–261. https://doi.org/10.1002/ird.134

Pielke, R., Nielsen‐Gammon, J., Davey, C., Angel, J., Bliss, O., Doesken, N., Cai, M., Fall, S., Niyogi, D., Gallo, K., Hale, R., Hubbard, K. G., Lin, X., Li, H., & Raman, S. (2007). Documentation of uncertainties and biases associated with surface temperature measurement sites for climate change assessment. Bulletin of the American Meteorological Society, 88(6), 913–928. https://doi.org/10.1175/BAMS‐88‐6‐913

Quej, V. H., Almorox, J., Ibrakhimov, M., & Saito, L. (2016). Empirical models for estimating daily global solar radiation in Yucatán Peninsula, Mexico. Energy Conversion and Management, 110, 448–456.

Robeson, S. M., & Ensor, L. A. (2006). Daily precipitation grids for South America. Bulletin of the American Meteorological Society, 87(8). https://doi.org/10.1175/BAMS-87-8-1095

Runck, B., Sharma, V., Becker, T., Senne, P., & Erdman, J. (2022). Irrigation management tool update: Protecting irrigators' privacy while providing real‐time modeling. University of Minnesota Extension. https://blog‐crop‐news.extension.umn.edu/2022/08/protecting‐privacy‐of‐irrigators‐while.html

Runck, B. C., Joglekar, A., Silverstein, K. A. T., Chan‐Kang, C., Pardey, P. G., & Wilgenbusch, J. C. (2022). Digital agriculture platforms: Driving data‐enabled agricultural innovation in a world fraught with privacy and security concerns. Agronomy Journal, 114(5), 2635–2643. https://doi.org/10.1002/agj2.20873

Senay, G. B., Friedrichs, M., Morton, C., Parrish, G. E. L., Schauer, M., Khand, K., Kagone, S., Boiko, O., & Huntington, J. (2022). Mapping actual evapotranspiration using Landsat for the conterminous United States: Google Earth Engine implementation and assessment of the SSEBop model. Remote Sensing of Environment, 275, 113011. https://doi.org/10.1016/j.rse.2022.113011

Supit, I. (1986). Manual for generation of daily weather data (No. 7). CABO.

Thompson, J. A., Brodzik, M. J., Silverstein, K. A., Hurley, M. A., & Carlson, N. L. (2022). EASE‐DGGS: A hybrid discrete global grid system for Earth sciences. Big Earth Data, 6(3), 340–357.

Tsuji, G. Y., Hoogenboom, G., & Thornton, P. K. (Eds.). (1998). Understanding options for agricultural production (Vol. 7). Springer Science & Business Media.

Van Ittersum, M. K., Cassman, K. G., Grassini, P., Wolf, J., Tittonell, P., & Hochman, Z. (2013). Yield gap analysis with local to global relevance—A review. Field Crops Research, 143, 4–17.

Van Wart, J., van Bussel, L. G., Wolf, J., Licker, R., Grassini, P., Nelson, A., Boogaard, H., Gerber, J., Mueller, N. D., Claessens, L., van Ittersum, M. K., & Cassman, K. G. (2013). Use of agro‐climatic zones to upscale simulated crop yield potential. Field Crops Research, 143, 44–55. https://doi.org/10.1016/j.fcr.2012.11.023

Volk, J. M., Huntington, J., Melton, F. S., Allen, R., Anderson, M. C., Fisher, J. B., Kilic, A., Senay, G., Halverson, G., Knipper, K., Minor, B., Pearson, C., Wang, T., Yang, Y., Evett, S., French, A. N., Jasoni, R., & Kustas, W. (2023). Development of a benchmark eddy flux evapotranspiration dataset for evaluation of satellite‐driven evapotranspiration models over the CONUS. Agricultural and Forest Meteorology, 331, 109307. https://doi.org/10.1016/j.agrformet.2023.109307

Whisler, F. D., Acock, B., Baker, D. N., Fye, R. E., Hodges, H. F., Lambert, J. R., Lemmon, H. E., McKinion, J. M., & Reddy, V. R. (1986). Crop simulation models in agronomic systems. Advances in Agronomy, 40, 141–208. https://doi.org/10.1016/S0065‐2113(08)60282‐5

Xu, H., Twine, T. E., & Girvetz, E. (2016). Climate change and maize yield in Iowa. PLoS One, 11(5), e0156083.

Xu, X., & Lan, Y. (2017). Spatial and temporal patterns of carbon footprints of grain crops in China. Journal of Cleaner Production, 146, 218227. https://doi.org/10.1016/j.jclepro.2016.11.181

Yang, K., Koike, T., & Ye, B. (2006). Improving estimation of hourly, daily, and monthly solar radiation by importing global data sets. Agricultural and Forest Meteorology, 137(1–2), 43–55.

Yang, S., Logan, J., & Coffey, D. L. (1995). Mathematical formulae for calculating the base temperature for growing degree days. Agricultural and Forest Meteorology, 74(1–2), 61–74. https://doi.org/10.1016/0168‐1923(94)02185‐M