Open AccessArticle

Calibration and Validation of MODIS-Derived Ground-Level Air Temperature Models by Means of Ground Measurements

Marica Teresa Rocca

Marica Franzini

and

Vittorio Marco Casella

Department of Civil Engineering and Architecture, University of Pavia, 27100 Pavia, Italy

Author to whom correspondence should be addressed.

Appl. Sci. 2025, 15(1), 184; https://doi.org/10.3390/app15010184

Submission received: 15 November 2024 / Revised: 24 December 2024 / Accepted: 26 December 2024 / Published: 28 December 2024

(This article belongs to the Special Issue Remote Sensing Applications in Agricultural, Earth and Environmental Sciences)

Download

Browse Figures

Versions Notes

Abstract

The research initiatives envisaged by the PNRR (Italian National Recovery and Resilience Plan) include the creation of innovation ecosystems to promote collaboration between universities, research centers, and local institutions with a focus on territorial integration and sustainability. The NODES Project (Nord-Ovest Digitale E Sostenibile) is part of this research. In this context, the Laboratory of Geomatics of the University of Pavia, in collaboration with other partners, deals with the study of the suitability maps for the renowned Pinot Noir wine. To achieve this, we considered different thematic input layers: elevation, slope, aspect, soil depth and type, Land Use Land Cover maps, NDVI, and current and forecast climatic aspects. An important thematic layer is concerned with the air temperature, which requires high spatial and temporal resolution. In the selected study area, the Lombardy Region has some accurate and reliable weather stations with high temporal resolution but low spatial resolution (7 stations in 648.5 square kilometers, i.e., one every 92 square kilometers). In addition, we considered Land Surface Temperature (LST) MODIS maps: these maps have good spatial resolution but present some voids and low temporal resolution. From the first evaluations made, the temperatures reported by MODIS are not always in excellent agreement with the ones from monitoring stations. To evaluate MODIS as a data source, we decided to use Kriging spatio-temporal interpolation. Starting from multitemporal MODIS data matrices, we interpolate them to estimate the temperature of the weather stations, in order to compare the estimation with the real weather station data, thus allowing the validation of MODIS data.

Keywords:

MODIS LST maps; weather stations; temperature data validation; NODES project

1. Introduction

Temperature assessment currently relies largely on remote sensing tools, especially for the acquisition of Land Surface Temperature (LST) data. Many studies use MODIS LST products due to their high spatial resolution, making them significant for climate assessments.

Analyzing the differences between LST data and air temperature obtained from in-situ measurements can help determine whether LST products can serve as a reliable substitute for in situ measurements, given the strong correlation between the two [1].

In the case study analyzed in this article, in situ measures, provided by ARPA’s monitoring stations (Agenzia Regionale per la Protezione Ambientale), offer high temporal granularity (hourly) but lack spatial detail. Conversely, satellite data provide high spatial granularity but with lower temporal resolution.

To address the gaps in data from ARPA [2] monitoring stations, this study aims to estimate the correlation between air temperature (Tair) and Land Surface Temperature (LST) to achieve finer spatial granularity in temperature data [3].

1.1. NODES Project

NODES (Nord-Ovest Digitale E Sostenibile) [4] is an ecosystem that involves the territories of North-West Italy: it is composed of 24 partners, including public and private universities, research and competences centers, innovation clusters, incubators, and accelerators. The main objective of the project is to enhance the competitiveness of industries and research institutions, making the region appealing to high-skilled talents and private investments. In accordance with the objectives outlined in Cluster 4 of the Horizon Europe Program and aligning with the priorities of the National Recovery and Resilience Plan (PNRR), NODES aims to play a role in advancing the primary goals of the “Digital, Industry, Aerospace” sector to respond to the challenges posed by the digital and ecological transition. In particular, NODES’ research and innovation initiatives support the adoption of digital technologies by enterprises, enhance digital skills among citizens and workers, and facilitate their access to digital services, especially for vulnerable social groups.

Within the ecosystem, in the framework of Spoke 6 (primary agroindustry) two flag-ship projects have been identified:

VINO (VIneyard management for viNeprOduction);
FORMIDABILÆ (Forage system to make resilient Maize, Dairy, and Biogas supply chains for a Lasting Agricultural Ecosystem).

In this paper, our focus will be on the VINO flagship project which concerns the development of a dedicated platform for the management of Pinot Noir, as a case study.

1.2. Flagship Project VINO

VINO promotes collaboration between stakeholders, focusing on three pillars: technological development, modeling, and application. In fact, the use of remote sensing and big data proves useful for precision agriculture [5], together with sustainable and organic practices. The project addresses the impact of climate change on viticulture in North-West Italy, focusing on areas such as Oltrepò Pavese (Figure 1), Langhe, and Colli Piacentini: climate change, with projected increases in temperature and altered precipitation patterns posing significant challenges to vineyard performance [6]. The project seeks sustainable solutions to adapt viticulture to climate change, including mapping vocational features, evaluating alternative crops, and valorizing waste materials.

Obtaining temperature maps is important to optimize viticultural strategies, as wines are sensitive to microclimatic variations. Temperature maps allow for tailored practices such as irrigation and harvest timing. This targeted approach is useful to enhance grape quality.

1.3. Suitability Map

The natural environment in which a plant thrives is the result of the combination of numerous factors such as soil composition, temperature, humidity, light exposure, and wind. Assessing the suitability of a study area for wine production involves the integration of knowledge across disciplines to understand these environmental elements.

Suitability maps aim to identify areas in which the environmental conditions allow the optimal development of the crop. The choices and weight of the different factors evaluated must be defined for each specific case study, also considering the accuracy of the available data [7]. The use of a Geographic Information System (GIS) is of great support for the creation of a suitability map [8], using both collected and processed data in vector and raster formats. To create a suitability map for the cultivation of Pinot Noir, the following thematic layers were identified as most relevant:

Land morphology, including elevation, slope, and aspect;
Soil characteristics and pedology;
Land-use models;
Historical and forecast climate data.

1.4. Bioclimatic Indices

Bioclimatic indices combine atmospheric parameters to provide an analytical indication of climatic features and their evolution in space and time that can directly relate to natural resource availability, distribution, and related bio-physical processes [9].

Temperatures and climatic trends strongly influence the production of wine, together with other factors such as rain, drought, hail, cold and heat, humidity, and aridity. There is therefore no doubt about the importance of the link between climate and the qualitative characteristics of wine production [10,11].

Climate change, in recent years, has had considerable effects on vineyard management [12], for this reason, bioclimatic indices [13] are increasingly used to evaluate the best time to harvest grapes [14]. Bioclimatic indices are useful for creating a thematic layer to assess the suitability of the study area for the cultivation of Pinot Noir.

Winkler Index

Among the ripening indices for grape, Winkler Index (WI from now on), despite its simplicity, is the most used: it is also known as “Thermal sum” [15].

This index is calculated by adding up the average daily air temperature (Tair) above 10° C (temperature above which the vine has vegetative activity) over the period from 1 April to 31 October, see Equation (1). Growing Degree-Days (GDD) are summed and used to assign a region to one of five climatic zones, ranging from the coolest to the warmest (see Table 1).

W I = \sum_{1 A p r}^{31 O c t} (T_{a v g} - 10) [° C]

(1)

The index, based solely on temperature, was developed by Albert Winkler and Maynard Amerine in the 1940s, with the purpose of encouraging and guiding the choice of the grape variety best suited to the region’s climate. The WI is a standard for describing regional climates for viticulture in the United States, specifically aiming to correlate wine quality with climate, originally focusing on the California climatic region [16].

Parts of California, particularly Napa Valley and Sonoma County, are known for their Mediterranean climate and vineyards, resembling Italian landscapes. These regions, located along the 45th parallel, share this characteristic with renowned wine-growing areas such as Piedmont and Oltrepò Pavese in Italy, Bordeaux in France, and Georgia, one of the oldest wine-producing countries in the world. Given their common latitude, the application of WI can also be extended to Italy, in particular to our study area.

2. Materials

2.1. Study Area

The NODES project involves territories in north-western Italy, with a focus on the Oltrepò Pavese region, which has been identified as a study area for its renowned wine production. The vineyard area of this zone extends for approximately 13.270 hectares, representing about 55% of the entire viticultural area of the Lombardy region: of which the main protagonist is Pinot Noir [17].

Oltrepò Pavese has long been a reference point for Italian winemaking, thanks to its favorable climate, the variety of soils, and the experience in growing high-quality grapes.

Located in the southern part of the province of Pavia, Figure 1a, the Oltrepò Pavese is on the border with Piedmont and Emilia Romagna. It has a total area of almost 100,000 hectares and a dry climate in winter and ventilated in summer, with high-temperature variations: its characteristics have made it an ideal place for the cultivation of vines and the production of wine. Altitudes of the selected area, as shown in Figure 1b, range between about 57 m and 1.460 m.

2.2. ARPA Monitoring Stations

ARPA Lombardia [2] monitors the temperature and other climatic variables through a system made up of fixed monitoring stations, see Figure 2. The list of these stations can be downloaded from the data portal of the Lombardy Region: Meteorological Stations Map [18]. The list includes over 1230 sensors, providing information on the Sensor ID, type (water level, snow height, precipitation, temperature, relative humidity, global radiation, wind speed and direction), unit of measurement, altitude, province, start date and end of the observations and the coordinates. For this study, data relating to temperature monitoring stations, located within the study area or adjacent to it, were identified as shown in Figure 3. The data were acquired from the ARPA website via a request form [19].

ARPA monitoring stations measure air temperature on an hourly basis: meteorological measurements provide temporally accurate information about the Tair.

The main problem that occurs is that sparse coverage of stations is known to introduce uncertainty in the interpolation of temperature and related data sets [20,21].

Consequently, the development of methods to improve the accuracy of interpolated surfaces based on sparsely distributed point measurements is an area of active research in geospatial studies [22].

To evaluate the temperature of the study area so that its spatial heterogeneity can be fully represented, a greater density of monitoring stations would be necessary [23]. These stations have a high temporal resolution (hourly measurements) but a low spatial resolution [24] (about one monitoring station per 100 km² in the Lombardy Region, and about one station per 92 km² in the study area).

Table 2 shows the control units within the study area, with the indication of the sensor ID, name of the monitoring station, altitude above sea level, and data availability of the measurement detection.

Figure 4 shows an example of the time series of the average hourly air temperature (Tair) recorded by the ARPA Monitoring Station located in Fortunago (PV), with ID 8007. The measurements refer to the selected period, ranging from January 2018 to December 2022.

2.3. MODIS Land Surface Temperature Data

Land Surface Temperature (LST) is one of the products of the MODerate Resolution Imaging Spectroradiometer (MODIS). MODIS sensor, aboard NASA’s Terra and Aqua satellites, captures data in 36 spectral bands, including those related to the temperature of the Earth’s surface [25].

Specifically, MODIS LST data are derived from algorithms taking into account different factors like surface temperature, surface reflectivity, atmospheric emission, and solar radiation.

The Terra MODIS LST (MOD11A2) and Aqua MODIS LST (MYD11A2) version 6.1, provide an average 8-day per-pixel Land Surface Temperature/Emissivity with a 1 km spatial resolution. Each pixel value is an average of all the corresponding MOD11A1 and MYD11A1 LST pixels collected within that 8-day period [26,27].

Both satellites cover the entire Earth’s surface every one or two days: their orbits are set so that, relative to the study area, MODIS LST Terra data are available during day and nighttime around 9:06–12:18 and 18.30–23:48 local time, and MODIS LST Aqua around 9:42–15:12 and 0:18–5:30. The details are summarized in Table 3.

The satellites collect daily and 8-day data, but daily data contain many empty pixels, as shown in Figure 5, so we focused on the latter. An example of the MODIS 8-day data is shown in Figure 6.

3. Methods

We managed to download all the MODIS 8-day LST raster from the Aqua and Terra satellites, for the years from 01/01/2018 to 31/12/2022, using a custom Google Engine script. The script allows, by setting the geometry of the study area and the selected bands, to collect the MODIS georeferenced images in the timeframe of interest.

The downloaded data resulted in 229 images from the Aqua satellite and 230 images from the Terra satellite, relating to a polygon that circumscribes the study area with coordinates: [8.998, 44.750], [9.412, 44.749], [9.415, 45.081], [8.998, 45.082], [8.998, 44.750], and with pixels of size equal to 1 km².

For our work, we selected 4 bands:

Day Temperature;
Night Temperature;
Time day (time of acquisition of the day temperature);
Time night (time of acquisition of the night temperature).

Starting from the downloaded images, we developed several MATLAB version 24.2 (R2024b) scripts to harmonize the data, fill voids, and perform analyses. These scripts operated a regression on both ARPA and MODIS data, to derive raster images of air temperature from Land Surface Temperature.

For the analysis of the data, we divided the workflow into different Tasks (see Figure 7), described as follows.

Task 1—Harmonization of downloaded data;
Task 2—Data stack of raster images;
Task 3—Fill-missing of the voids in the raster images;
Task 4—Interpolation of MODIS LST data at the geographic coordinates of ARPA Monitoring Stations;
Task 5—Creation of a table containing MODIS LST data and ARPA Monitoring Stations Tair data averaged over 8-days;
Task 6—Comparison of MODIS LST and Monitoring Stations Tair data;
Task 7—Linear regression for the estimation of Tair from MODIS satellite data.

3.1. Task 1—Harmonization of Downloaded Data

The bands related to temperatures of every raster image have been harmonized: the pixels with a value equal to 0 were replaced with NaN values, then the pixel values were scaled by multiplying by 0.02 (a scale factor applied to the original downloaded images as reported in the MODIS metadata) in order to obtain the temperature in degrees Kelvin. Subsequently, we converted the temperature to degree Celsius, as shown in Formulas (2) and (3).

H a r m o n i z e d P i x e l v a l u e = (P i x e l v a l u e \cdot 0.02) - 273.15 [° C]

(2)

e . g ., H a r m o n i z e d P i x e l v a l u e = (14,136 \cdot 0.02) - 273.15 = 9.57 [° C]

(3)

The two bands related to the acquisition time are originally stored as a decimal number, so the pixel value has been scaled by multiplying by 0.1 and then converted to sexagesimal format, see Formulas (4) and (5).

H a r m o n i z e d P i x e l v a l u e = P i x e l v a l u e \cdot 0.1 = H . d d \overset{c o n v e r s i o n}{\to} H : m m

(4)

e . g ., H a r m o n i z e d P i x e l v a l u e = 128 \cdot 0.1 = 12.8 \overset{c o n v e r s i o n}{\to} 12 h 48^{'}

(5)

3.2. Task 2—Data Stack of Raster Images

To efficiently manage and process the data, we created an array structure, in which to store the four bands and other useful information such as the acquisition date and the georeferencing of each raster image for both the satellites. This structure allows for easy access to the data.

The first check we carried out was to verify whether raster images were missing for any date. In this case, a new raster layer, filled with NaN values, was created. This approach, maintaining data integrity, ensured to appropriately handle missing data for further processing.

3.3. Task 3—Fill-Missing of the Voids in the Raster Images

To handle missing values in the structure created in the previous task, we implemented a fill-missing operation in two directions, one we called horizontal fill-missing, and subsequently the vertical fill-missing. Figure 8 shows an excerpt of the structure we created, composed of one raster image per date, with each pixel representing the Land Surface Temperature (LST) value. The white pixels in the figure represent NaN values, indicating areas where the data are unavailable or undefined.

Now we explore in more detail what we refer to as horizontal and vertical fill-missing operations.

In order to understand the horizontal fill-missing operation, assume that the red pixel in Figure 9 has a NaN value.

To determine its value, we performed a spatial interpolation by calculating the average of the pixels within a 5 × 5 window, centered on the aforementioned red pixel. It is, in effect, a spatial interpolation over an area of 5 × 5 km².

After the horizontal fill, the raster images still contain some fillable voids: therefore, we applied the vertical fill-missing operation which is, in fact, a temporal interpolation over a time-span of about 3 weeks.

Vertical fill-missing computes the mean value by considering the column of 3 layers of pixels (one per image) aligned with the red one (NaN value) as shown in Figure 10.

3.4. Task 4—Interpolation of MODIS LST Data at the Geographic Coordinates of ARPA Monitoring Stations

To compare in situ Tair measurements from ARPA Monitoring Stations with MODIS LST data, we used the coordinates of the Monitoring Stations within the study area, obtained from the data downloaded from Geoportale Lombardia.

We wanted to obtain the time series of the daytime and nighttime LST at these locations, by interpolating values of the MODIS raster images of both Aqua and Terra satellites. The following Figure 11 shows the time series of MODIS temperatures recorded by the Aqua and Terra satellites, daytime and nighttime, from January 2018 to December 2022. In this specific case, the temperatures are interpolated at the coordinates of the Fortunago ARPA Monitoring Station-ID 8007. Daytime temperatures for Aqua (in blue) and Terra (in red) indicate that the Aqua sensor typically records higher values than the Terra sensor. For nighttime temperatures, the opposite occurs: Terra temperatures (in orange) are overall higher than Aqua temperatures (in light blue).

For a comprehensive view of the MODIS LST time series at the coordinates of all the considered ARPA Monitoring Stations, please refer to Appendix A (Figure A1, Figure A2, Figure A3, Figure A4, Figure A5 and Figure A6).

3.5. Task 5—Creation of a Table Containing MODIS LST Data and ARPA Monitoring Stations Tair Data Averaged over 8-Days

The data downloaded from the ARPA website are available in csv format, one file per Monitoring Station. The data contain the following information: sensor ID, acquisition Date-Time, and the average temperature measured on an hourly basis.

A function was used to load and harmonize all the Monitoring Station tables, setting the value −999 to NaN.

Subsequently, starting from the collected data, we compute the average over an 8-day period by sensor ID. The 8-day average has been chosen to align with the temporal resolution of the satellite data, to obtain a meaningful comparison between in-situ and satellite-derived measurements.

In the end, we obtained a table, consisting of 1610 rows and 13 columns, containing both satellite and Monitoring Stations data. An illustrative excerpt of this table is shown in the following Table 4. Latitude (Lat) and longitude (Lon) represent the coordinates, in decimal degrees, of the considered monitoring station (Sensor ID). MODIS date refers to the date of acquisition of raster data (8-day span). Tday and Tnight correspond to the Land Surface Temperatures measured by MODIS Aqua and Terra satellites, measured twice a day. Tair ARPA is the air temperature recorded by the ARPA monitoring station, calculated on an 8-day average to ensure comparability with satellite data.

3.6. Task 6—Comparison of MODIS LST and Monitoring Stations Tair Data

The following Figure 12 shows the behavior of the 8-day average temperature of the ARPA Monitoring Station located in Fortunago, ID 8007, compared to the day and night LST of the MODIS Aqua and Terra sensors. As shown in the figure, the behavior of ARPA Tair (gray dotted line) falls within the range of the temperatures measured by MODIS satellites, even if the measured entities are different. This indicates a general agreement between the ground-based ARPA measurements and the satellites’ data, reinforcing the reliability of using satellite observations for calibrating air temperatures. Appendix A (Figure A7, Figure A8, Figure A9, Figure A10, Figure A11 and Figure A12) provides the entire overview of the comparison of ARPA Tair and MODIS LST.

3.7. Task 7—Linear Regression for the Estimation of Tair from MODIS Satellite Data

Using the created dataset mentioned above, containing both MODIS surface temperatures (LST) and ARPA air temperature (Tair) data, we have performed a linear regression analysis to better understand the correlation between the two sources of temperature measurement [1]. It should be noted that the linear regression is not performed individually for each monitoring station, but rather as an areal regression considering all stations within the study area, with some exceptions explained below. We assumed as explanatory variables the temperatures from ARPA, which represent the “true” air temperatures. To evaluate the linear regression coefficients, we decided to exclude two monitoring stations: the one with ID 8002 since the dataset is not of great interest as it presents values for a small part of the dates considered (i.e., from January 2018 to September 2018) and the one with ID 8007, excluded in order to use its data for the validation of the model with the aim to test its the predictive capacity.

The linear regression has a dual objective: firstly, to evaluate the agreement between MODIS LST and ARPA Tair, and secondly, to estimate in a spatially continuous way the air temperature from available MODIS data.

The process of regression involves minimizing the residuals, which are the differences between MODIS-estimated temperatures and the ones actually recorded by ARPA stations. By building a linear regression model, we establish a mathematical equation that allows us to predict air temperature from MODIS surface temperature. The entire analysis was carried out thanks to a customized MATLAB version 24.2 (R2024b) script we specifically designed for this purpose.

4. Results

Using the linear regression model obtained from Task 7, we could derive the following equation:

{\hat{T}}_{a i r} = 2.8447 + 0.09588 \cdot T_{d A} + 0.38685 \cdot T_{n A} + 0.14467 \cdot T_{d T} + 0.34914 \cdot T_{n T}

(6)

where

{\hat{T}}_{a i r}

represents the estimated temperature of a given pixel, using inputs from the MODIS satellite. Specifically:

$T_{d A}$ = LST day temperature from MODIS—Aqua,
$T_{n A}$ = LST night temperature from MODIS—Aqua,
$T_{d T}$ = LST day temperature from MODIS—Terra,
$T_{n T}$ = LST night temperature from MODIS—Terra.

From the analysis of the coefficients of the equation, we can observe that nighttime temperatures (

T_{n A}

and

T_{n T}

) have a stronger influence on the estimation of air temperature compared to daytime temperatures. With a Root Mean Squared Error (RSME) of 1.11 °C, an R-squared of 0.981, and a near-zero p-value, the model performance has overall a good fit [28]. Using this calibration, we were able to estimate air temperatures from MODIS data at the coordinates of ARPA monitoring stations. These estimations, as shown in Figure 13, were then compared with the actual air temperature measurements recorded by ARPA.

The continuous line represents the air temperature (Tair) measured by the ARPA Monitoring Station located in Fortunago (ID 8007), while the dotted line shows the estimated air temperature derived from MODIS after applying the linear regression. The gray line represents the difference (ΔT) between MODIS-derived estimated

{\hat{T}}_{a i r}

and ARPA

T_{a i r}

Appendix A (Figure A13, Figure A14, Figure A15, Figure A16, Figure A17 and Figure A18) provides a comprehensive comparison of MODIS-derived and ARPA-measured Tair.

Figure 14 shows an extract of the results obtained, with reference to the data of the 8-day period from 20 July 2022 to 28 July 2022. By applying the coefficients derived from the linear regression model to each pixel of the four MODIS-LST images related to the same time, we generated raster maps that estimate air temperature across the entire study area with a resolution of 1 km². In this way, we can obtain a detailed spatial representation of the air temperature, which would not be possible with in situ measurements alone. By increasing the spatial granularity of temperature data, this technique allows for better estimation of bioclimatic indicators, enabling informed decisions related to resource management, land use planning, and assessment of climate impact on agriculture.

5. Discussion

The approach applied for this study integrates in situ observations with remote sensing data, as in situ measurements provide air temperature at high temporal but low spatial resolution, while satellite-derived data offer surface temperature estimated with higher spatial but lower temporal resolution. In this context, linear regression allows calibration of Land Surface Temperatures from MODIS satellites against air temperature data collected by ARPA, so that the predicted temperature reflects real conditions as closely as possible. Despite the simplicity of the statistical model used as an initial hypothesis for this calibration, we chose to apply linear regression because, as a preliminary method, it aligns with established approaches in the literature for temperature data calibration [29].

We could achieve high accuracy, as the dataset used to calibrate the data has an r² higher than 0.981 and RMSE of 1.11 °C, while the comparison of estimated

\hat{T}

air and the Tair measured by the monitoring station excluded from the calibration (ID 8007) shows a r² of 0.979 and a RMSE of 1.03 °C

The monitoring station with the lowest performance, located in Canevino—ID 9019—see Figure A17 in Appendix A, has an r² of 0.976 and a RMSE of 1.21 °C probably due to particular features in the station’s surroundings, as it is quite isolated from the urban fabric. The average annual uncertainty, considering all monitoring stations in the study area, has a range between 0.78 and 0.97 °C.

The result shows the strong physical relation between surface and air temperature, enhancing the accuracy of the MODIS LST products [30]. The higher influence of LST Night temperature on the

\hat{T}

air can be explained by the lack of solar radiation and its effects on the thermal infrared signal [31].

From the analysis of ΔT (

\hat{T}

air-Tair), see Figure 13, we could detect a pattern that shows an overall overestimation of the temperature in summer and an underestimation in winter. This occurs for all monitoring stations, with no significant relation detected between elevation and errors [30]. Further work is needed to thoroughly evaluate the patterns detected, applying more complex regression methods such as global non-linear models [32] and Geographically Weighted Regression (GWR) [33].

6. Conclusions

This study concerns the analysis of the correlation between Land Surface Temperature (LST) retrieved from MODIS satellite data and air temperature (Tair) measured by ARPA monitoring stations in the Oltrepò Pavese area. Since the Oltrepò Pavese is renowned for its significant wine production and extensive vineyards, understanding the impacts of climate change on viticulture is of great importance to adopting solutions to adapt viticulture to evolving climate conditions. This can be achieved thanks to suitability maps that can be useful to identify areas where environmental conditions allow optimal vine growth.

Among the thematic layers considered, such as land morphology and land use, bioclimatic indices play a crucial role.

Temperature and climatic patterns directly influence wine production; therefore, we focused on air temperature, which is the key data for evaluating the Winkler Index that is useful to guide decisions related to vineyard cultivation. Our goal is to create annual raster maps of the Winkler Index. By exploiting these annual raster layers, we aim to capture temporal variations in the Winkler Index, ensuring a complete understanding of trends over time.

While monitoring stations measure air temperature with high temporal resolution, their spatial distribution is sparse. Therefore, it is essential to increase the density of temperature data. For this reason, we investigated the possibility of using LST as an input variable to estimate Tair [34], to obtain a higher spatial resolution than the in-situ measurements provided by ARPA monitoring stations. The 8-day mean of Tair can be estimated using remote sensing techniques, as from MODIS LST products we can retrieve four temperatures (day and night LST temperature from Terra and Aqua MODIS satellites). We were able to determine that there is a strong correlation between LST and near-surface air temperature (Tair), although the two temperatures have different physical meanings and responses to atmospheric conditions [34].

By applying a linear regression model on the initial dataset, we estimated air temperature from MODIS LST data with a temporal granularity of 8 days and a spatial granularity of 1 km². This method has potential application in regions with low density of monitoring stations, in search of sustainable solutions to adapt viticulture to climate change, including mapping of vocational features, evaluation of alternative crops, and valorization of waste materials. Furthermore, this approach can be useful in cases where monitoring stations do not cover the entire time span considered.

The simplicity of the employed models, the accessibility of input data at a global scale, and the good performances obtained suggest that this methodology could be applied to other regions.

Future developments could explore the improvement of Tair estimates from remote sensing data by applying alternative regression algorithms incorporating both temporal and spatial factors and evaluate other remote sensing temperature data such as those from the COPERNICUS program.

Author Contributions

Writing—original draft, M.T.R.; Writing—review & editing, M.F.; Supervision, V.M.C. All authors have read and agreed to the published version of the manuscript.

Funding

Funded by the European Union—NextGenerationEU, Mission 4 Component 1.5—ECS00000036—CUP F17G22000190007.

Institutional Review Board Statement

Not applicable.

Informed Consent Statement

Not applicable.

Data Availability Statement

The original contributions presented in this study are included in the article. Further inquiries can be directed to the corresponding author.

Acknowledgments

The present paper has been prepared in the framework of PhD Course XXXVIII cycle.

Conflicts of Interest

The authors declare no conflict of interest.

Appendix A

Figure A1. Comparison between daytime and nighttime temperatures detected from MODIS Aqua and Terra satellites, interpolated to the Varzi Nivione ARPA Monitoring Station coordinates (ID 2082).

Figure A2. Comparison between daytime and nighttime temperatures detected from MODIS Aqua and Terra satellites, interpolated to the Varzi v. Mazzini ARPA Monitoring Station coordinates (ID 8002).

Figure A3. Comparison between daytime and nighttime temperatures detected from MODIS Aqua and Terra satellites, interpolated to the Voghera v. Cambiaso ARPA Monitoring Station coordinates (ID 8191).

Figure A4. Comparison between daytime and nighttime temperatures detected from MODIS Aqua and Terra satellites, interpolated to the Santa Margherita di Staffora Casanova ARPA Monitoring Station coordinates (ID 8202).

Figure A5. Comparison between daytime and nighttime temperatures detected from MODIS Aqua and Terra satellites, interpolated to the Canevino ARPA Monitoring Station coordinates (ID 9019).

Figure A6. Comparison between daytime and nighttime temperatures detected from MODIS Aqua and Terra satellites, interpolated to the Broni ARPA Monitoring Station coordinates (ID 17432).

Figure A7. Time-series of the 8-day average air temperature of the ARPA Monitoring Station in Varzi Nivione, and of the LST detected by the MODIS sensors, interpolated at the location coordinates.

Figure A8. Time-series of the 8-day average air temperature of the ARPA Monitoring Station in Varzi v. Mazzini, and of the LST detected by the MODIS sensors, interpolated at the location coordinates.

Figure A9. Time-series of the 8-day average air temperature of the ARPA Monitoring Station in Voghera v. Cambiaso, and of the LST detected by the MODIS sensors, interpolated at the location coordinates.

Figure A10. Time-series of the 8-day average air temperature of the ARPA Monitoring Station in Santa Margherita di Staffora Casanova, and of the LST detected by the MODIS sensors, interpolated at the location coordinates.

Figure A11. Time-series of the 8-day average air temperature of the ARPA Monitoring Station in Canevino, and of the LST detected by the MODIS sensors, interpolated at the location coordinates.

Figure A12. Time-series of the 8-day average air temperature of the ARPA Monitoring Station in Broni, and of the LST detected by the MODIS sensors, interpolated at the location coordinates.

Figure A13. Comparison between MODIS-derived (dotted line) and ARPA-measured (continuous line) Tair at the coordinates of Varzi Nivione Monitoring Station (ID 2082), while the gray line depicts their difference (ΔT).

Figure A14. Comparison between MODIS-derived (dotted line) and ARPA-measured (continuous line) Tair at the coordinates of Varzi via Mazzini Monitoring Station (ID 8002), while the gray line depicts their difference (ΔT).

Figure A15. Comparison between MODIS-derived (dotted line) and ARPA-measured (continuous line) Tair at the coordinates of Voghera via Cambiaso Monitoring Station (ID 8191), while the gray line depicts their difference (ΔT).

Figure A16. Comparison between MODIS-derived (dotted line) and ARPA-measured (continuous line) Tair at the coordinates of Santa Margherita di Staffora Casanova Monitoring Station (ID 8202), while the gray line depicts their difference (ΔT).

Figure A17. Comparison between MODIS-derived (dotted line) and ARPA-measured (continuous line) Tair at the coordinates of Canevino Monitoring Station (ID 9019), while the gray line depicts their difference (ΔT).

Figure A18. Comparison between MODIS-derived (dotted line) and ARPA-measured (continuous line) Tair at the coordinates of Broni Monitoring Station (ID 17432), while the gray line depicts their difference (ΔT).

References

Mihăilă, D.; Bistricean, P.-I.; Sfîcă, L.; Horodnic, V.-D.; Prisăcariu, A.; Amihăesei, V.-A. Summer Discrepancies between 2 m Air Temperature and Landsat LST in Suceava City, Northeastern Romania. Remote. Sens. 2024, 16, 2967. [Google Scholar] [CrossRef]
ARPA Lombardia Agenzia Regionale per La Protezione Dell’Ambiente. Available online: https://www.arpalombardia.it (accessed on 14 November 2024).
Magarreiro, C.; Gouveia, C.M.; Barroso, C.M.; Trigo, I.F. Modelling of Wine Production Using Land Surface Temperature and FAPAR—The Case of the Douro Wine Region. Remote. Sens. 2019, 11, 604. [Google Scholar] [CrossRef]
HUB NODES, S.c.a.r.l. NODES—Nord Ovest Digitale e Sostenibile. Available online: https://www.ecs-nodes.eu/ (accessed on 14 November 2024).
Mihailescu, E.; Soares, M.B. The Influence of Climate on Agricultural Decisions for Three European Crops: A Systematic Review. Front. Sustain. Food Syst. 2020, 4, 64. [Google Scholar] [CrossRef]
Santos, J.A.; Fraga, H.; Malheiro, A.C.; Moutinho-Pereira, J.; Dinis, L.-T.; Correia, C.; Moriondo, M.; Leolini, L.; Dibari, C.; Costafreda-Aumedes, S.; et al. A Review of the Potential Climate Change Impacts and Adaptation Options for European Viticulture. Appl. Sci. 2020, 10, 3092. [Google Scholar] [CrossRef]
Pastore, V.; Arous, A.; Palese, A.M.; Persiani, A.; Celano, G. Tecnologie avanzate in viticoltura e enologia per un vino innovativo ottenuto dal vitigno Aglianicone. In Carta Di Attitudine Della Produzione Viticola Nell’Area Cilento, Alburni e Vallo Di Diano; Celano, G., Palese, A.M., Piccolo, A., Eds.; Chapter VII; ATS De Conciliis: Potenza, Italy, 2015; pp. 107–112. [Google Scholar]
Cristóbal, J.; Ninyerola, M.; Pons, X. Modeling air temperature through a combination of remote sensing and GIS data. J. Geophys. Res. Atmos. 2008, 113. [Google Scholar] [CrossRef]
Passarella, G.; Bruno, D.; Lay-Ekuakille, A.; Maggi, S.; Masciale, R.; Zaccaria, D. Spatial and temporal classification of coastal regions using bioclimatic indices in a Mediterranean environment. Sci. Total. Environ. 2020, 700, 134415. [Google Scholar] [CrossRef]
Massano, L.; Fosser, G.; Gaetani, M.; Bois, B. Assessment of climate impact on grape productivity: A new application for bioclimatic indices in Italy. Sci. Total. Environ. 2023, 905, 167134. [Google Scholar] [CrossRef]
Alba, V.; Gentilesco, G.; Tarricone, L. Climate change in a typical Apulian region for table grape production: Spatialisation of bioclimatic indices, classification and Future Scenarios. OENO One 2021, 55, 317–336. [Google Scholar] [CrossRef]
Comte, V.; Schneider, L.; Calanca, P.; Rebetez, M. Effects of climate change on bioclimatic indices in vineyards along Lake Neuchatel, Switzerland. Theor. Appl. Clim. 2022, 147, 423–436. [Google Scholar] [CrossRef]
Blanco-Ward, D.; Monteiro, A.; Lopes, M.; Borrego, C.; Silveira, C.; Viceto, C.; Rocha, A.; Ribeiro, A.; Andrade, J.; Feliciano, M.; et al. Climate change impact on a wine-producing region using a dynamical downscaling approach: Climate parameters, bioclimatic indices and extreme indices. Int. J. Clim. 2019, 39, 5741–5760. [Google Scholar] [CrossRef]
Ferroni, F. Grape Ripening Indices for Smart Vineyard Management. Available online: https://www.agricolus.com/en/grape-ripening-indices-for-smart-vineyard-management/ (accessed on 5 July 2024).
del Río, M.S.; Raventós, L.; Garza, V. Zoning of the Querétaro Wine Region. In Proceedings of the BIO Web of Conferences, Cádiz/Jerez, Spain, 5–9 June 2023; EDP Sciences: Les Ulis, France, 2023; Volume 68. [Google Scholar]
Shabam, P.L. The Limitations of the Winkler Index. Available online: https://winebusinessanalytics.com/sections/printout_article.cfm?article=feature&content=208245 (accessed on 12 July 2024).
Guzzon, F.; Ardenghi, N.M.G.; Bodino, S.; Tazzari, E.R.; Rossi, G. Guida All’Agrobiodiversità Vegetale Della Provincia Di Pavia; Pavia University Press: Pavia, Italy, 2019. [Google Scholar]
Regione Lombardia—Open Data Mappa Stazioni Meteorologiche Lombardia. Available online: https://www.dati.lombardia.it/Ambiente/Mappa-Stazioni-Meteorologiche/8ux9-ue3c (accessed on 12 July 2024).
ARPA Form Richiesta Dati. Available online: https://www.arpalombardia.it/temi-ambientali/meteo-e-clima/form-richiesta-dati/ (accessed on 12 July 2024).
Diego Díaz, M.; Luis Morales, S.; Giorgio Castellaro, G.; Fernando Neira, R. Topoclimatic Modeling of Thermopluviometric Variables for the Bío Bío and La Araucanía Regions, Chile. Chil. J. Agric. Res. 2010, 70, 604–615. [Google Scholar] [CrossRef]
Morales-Salinas, L.; Castellaro, G.; Frederiksen, N.; Oosrio, L.F.R.; Roman, J.N.; Jaque, G.F.; Avaria, C.E.; Morales, F. Spatial characterization of climatic variables for Arica-Parinacota and Tarapacá, Chile using topoclimatic analysis. Cuad. Investig. Geogr. 2023, 49, 39–53. [Google Scholar] [CrossRef]
Njoku, E.A.; Akpan, P.E.; Effiong, A.E.; Babatunde, I.O. The effects of station density in geostatistical prediction of air temperatures in Sweden: A comparison of two interpolation techniques. Resour. Environ. Sustain. 2023, 11, 100092. [Google Scholar] [CrossRef]
Hofstra, N.; New, M.; McSweeney, C. The influence of interpolation and station network density on the distributions and trends of climate variables in gridded daily data. Clim. Dyn. 2010, 35, 841–858. [Google Scholar] [CrossRef]
Naserikia, M.; Hart, M.A.; Nazarian, N.; Bechtel, B.; Lipson, M.; Nice, K.A. Land surface and air temperature dynamics: The role of urban form and seasonality. Sci. Total. Environ. 2023, 905, 167306. [Google Scholar] [CrossRef]
NASA Official MODIS—MODerate Resolution Imaging Spectroradiometer. Available online: https://modis.gsfc.nasa.gov/about/ (accessed on 17 July 2024).
Wan, Z.; Hook, S.; Hulley, G. MODIS/Terra Land Surface Temperature/Emissivity 8-Day L3 Global 1km SIN Grid V061 [Data Set]. Available online: https://lpdaac.usgs.gov/products/mod11a2v061/ (accessed on 19 April 2024).
Wan, Z.; Hook, S.; Hulley, G. MODIS/Aqua Land Surface Temperature/Emissivity 8-Day L3 Global 1km SIN Grid V061 [Data Set]. Available online: https://lpdaac.usgs.gov/products/myd11a2v061/.
Galdón-Ruíz, A.; Fuentes-Jaque, G.; Soto, J.; Morales-Salinas, L. A simple method for the estimation of minimum and maximum air temperature monthly mean maps using MODIS images in the region of Murcia, Spain. Rev. Teledeteccion 2023, 2023, 59–71. [Google Scholar] [CrossRef]
Hereher, M.E.; El Kenawy, A. Extrapolation of daily air temperatures of Egypt from MODIS LST data. Geocarto Int. 2022, 37, 214–230. [Google Scholar] [CrossRef]
Benali, A.; Carvalho, A.; Nunes, J.; Carvalhais, N.; Santos, A. Estimating air surface temperature in Portugal using MODIS LST data. Remote. Sens. Environ. 2012, 124, 108–121. [Google Scholar] [CrossRef]
Vancutsem, C.; Ceccato, P.; Dinku, T.; Connor, S.J. Evaluation of MODIS land surface temperature data to estimate air temperature in different ecosystems over Africa. Remote. Sens. Environ. 2010, 114, 449–465. [Google Scholar] [CrossRef]
Guo, Y.; Unger, J.; Khabibolla, A.; Tian, G.; He, R.; Li, H.; Gál, T. Modeling urban air temperature using satellite-derived surface temperature, meteorological data, and local climate zone pattern—A case study in Szeged, Hungary. Theor. Appl. Clim. 2024, 155, 3841–3859. [Google Scholar] [CrossRef]
Qin, Y.; Ren, G.; Huang, Y.; Zhang, P.; Wen, K. Application of geographically weighted regression model in the estimation of surface air temperature lapse rate. J. Geogr. Sci. 2021, 31, 389–402. [Google Scholar] [CrossRef]
Mutiibwa, D.; Strachan, S.; Albright, T. Land Surface Temperature and Surface Air Temperature in Complex Terrain. IEEE J. Sel. Top. Appl. Earth Obs. Remote. Sens. 2015, 8, 4762–4774. [Google Scholar] [CrossRef]

Figure 1. Location of the Study Area in the province of Pavia and the Lombardy Region (a), with a focus on the Digital Terrain Model (DTM) to highlight altitude variations across the Oltrepò Pavese (b).

Figure 2. ARPA Weather Stations distribution in the Lombardy Region.

Figure 3. ARPA weather stations within the study area and its vicinity, along with the corresponding sensor IDs.

Figure 4. Series of the average hourly air temperature measured by the Fortunago Monitoring Station—ID 8007.

Figure 5. MODIS data in the study area: raster with 1 km² tiles depicting the daily average daytime temperature, captured by the Aqua satellite on 30 March 2018.

Figure 6. MODIS data in the study area: raster with 1 km² tiles depicting the 8-day average daytime temperature, captured by the Aqua satellite from 30 March to 6 April 2018.

Figure 7. Workflow diagram illustrating the sequential process for the analysis of MODIS Land Surface Temperature (LST) and ARPA Monitoring Stations’ air temperature (Tair) data, to estimate Tair from satellite data.

Figure 8. Excerpt of the stacked structure of the MODIS raster images, the white pixels represent an example of missing values (NaN).

Figure 9. Example of horizontal fill for missing values. In order to retrieve the value of the selected no-value pixel, we performed an interpolation by computing the mean of the pixels within a 5 × 5 window centered on the red pixel.

Figure 10. Example of vertical fill for missing values: after performing the horizontal fill, the selected pixel still has no value. To retrieve its value, we perform an interpolation considering the column of 3 layers of pixels aligned with the red one and computing the mean.

Figure 11. Comparison between daytime and nighttime temperatures detected from MODIS Aqua and Terra satellites, interpolated at the Fortunago ARPA Monitoring Station coordinates (ID 8007).

Figure 12. Time-series of the 8-day average air temperature of the ARPA Monitoring Station in Fortunago, and of the LST detected by the MODIS sensors, interpolated at the location coordinates.

Figure 13. Comparison between MODIS-derived (dotted line) and ARPA-measured (continuous line) Tair at the coordinates of Fortunago Monitoring Station (ID 8007), while the gray line depicts their difference (ΔT).

Figure 14. Data images related to the 8-day data 20/07/2022–28/07/2022. (a) From the combination of the four 8-day MODIS LST images (Aqua day, Aqua night, Terra day and Terra night) with their respective coefficients—deduced from the linear regression—we obtained (b) the estimated air temperature.

Table 1. Classification of wine-producing regions according to the Winkler Index (GDD).

Regions	Heat Summation (GDD) [°C]	Area	Wine Type
Region 1	<1370	Chablis, Champagne, Burgundy.	Light-bodied white wines. e.g., Pinot Noir and Gris.
Region 2	1370–1650	Bordeaux, northern Rhone.	Medium-bodied red wines. e.g., Chardonnay, Pinot Noir, Merlot.
Region 3	1650–1930	Sonoma Valley, southern Rhone.	Full-bodied red wines. e.g., Sauvignon Blanc, Grenache.
Region 4	1930–2200	Napa and Barossa Valley Tuscany.	Fortified wines. e.g., Barbera, Porto, Zinfadel.
Region 5	>2200	Jerez, Madera, Lodi.	Bulk wines, table and drying grapes. e.g., Muscat, Nero d’Avola.

Table 2. List of ARPA Monitoring Stations within the study area.

Sensor ID	Monitoring Station’s Name	MSL [m]	Data Availability
2082	Varzi Nivione	500	From 15 May 2011
8002	Varzi Via Mazzini	485	From 01 January 1998 to 19 September 2018
8007	Fortunago	501	From 01 January 1998
8191	Voghera Via Cambiaso	95	From 04 March 2003
8202	Santa Margherita di Staffora Casanova	575	From 01 January2003 to 30 June 2023
9019	Canevino	455	From 01 January 2004
17432	Broni	77	From 22 March 2018

Table 3. Specification of the MODIS data used to obtain LST maps.

Satellite	Sensor	Version	Bands	Spatial Resolution	Temporal Resolution	Pass Time
MODIS	Terra	MOD11A2 6.1	LST_Day LST_Night Day_view_time Night_view_time	1 km	8 days	9:06–12:18 18.30–23:48
MODIS	Aqua	MYD11A2 6.1	LST_Day LST_Night Day_view_time Night_view_time	1 km	8 days	9:42–15:12 0:18–5:30

Table 4. Table containing LST MODIS and ARPA Tair (8-day average).

Sensor ID	Lat [DD]	Lon [DD]	MODIS Date	Tday MODIS Aqua [°C]	Tnight MODIS Aqua [°C]	Tday MODIS Terra [°C]	Tnight MODIS Terra [°C]	Tair ARPA (8-Day Avg) [°C]
8007	44.9125	9.1950	30/03/2018	20.86	4.68	20.43	6.25	10.27
8007	44.9125	9.1950	07/04/2018	19.57	6.11	17.46	5.11	10.39
8007	44.9125	9.1950	15/04/2018	26.75	10.86	25.34	11.52	17.53
8007	44.9125	9.1950	23/04/2018	24.61	10.63	24.46	12.92	17.59
8007	44.9125	9.1950	01/05/2018	24.99	8.47	22.46	12.42	14.49
8007	44.9125	9.1950	09/05/2018	20.90	8.48	20.20	10.13	14.44
8007	44.9125	9.1950	17/05/2018	23.25	8.69	21.69	11.84	15.74

Disclaimer/Publisher’s Note: The statements, opinions and data contained in all publications are solely those of the individual author(s) and contributor(s) and not of MDPI and/or the editor(s). MDPI and/or the editor(s) disclaim responsibility for any injury to people or property resulting from any ideas, methods, instructions or products referred to in the content.

© 2024 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (https://creativecommons.org/licenses/by/4.0/).

Share and Cite

MDPI and ACS Style

Rocca, M.T.; Franzini, M.; Casella, V.M. Calibration and Validation of MODIS-Derived Ground-Level Air Temperature Models by Means of Ground Measurements. Appl. Sci. 2025, 15, 184. https://doi.org/10.3390/app15010184

AMA Style

Rocca MT, Franzini M, Casella VM. Calibration and Validation of MODIS-Derived Ground-Level Air Temperature Models by Means of Ground Measurements. Applied Sciences. 2025; 15(1):184. https://doi.org/10.3390/app15010184

Chicago/Turabian Style

Rocca, Marica Teresa, Marica Franzini, and Vittorio Marco Casella. 2025. "Calibration and Validation of MODIS-Derived Ground-Level Air Temperature Models by Means of Ground Measurements" Applied Sciences 15, no. 1: 184. https://doi.org/10.3390/app15010184

APA Style

Rocca, M. T., Franzini, M., & Casella, V. M. (2025). Calibration and Validation of MODIS-Derived Ground-Level Air Temperature Models by Means of Ground Measurements. Applied Sciences, 15(1), 184. https://doi.org/10.3390/app15010184

Note that from the first issue of 2016, this journal uses article numbers instead of page numbers. See further details here.

Article Menu