Introduction

Earth observation (EO) data provides an affordable, reliable, and timely source of information that can be used to enhance crop yield forecasting. The freely available data from the EO Copernicus Programme allows high-quality observations with frequent revisits, enhancing yield forecasting models' effectiveness.

Operational yield forecasting approaches are often based on field surveys, empirical regression models, or crop simulation models. The most straightforward approach to link remote sensing to yield forecasts establishes an empirical relationship between in situ yield statistics and remote sensing parameters.

Global and regional crop yield forecasting systems currently exist to provide up-to-date information on food production. Many of these studies contains vast datasets, complex methods, and are performed over many countries at national and sub-national scales. These take a considerable amount of time to develop, test, and publish. Some studies use data that is not clearly and freely available

Objectives

ARGANS have been tasked with testing a range of machine learning (ML) and other methods to predict crop yields at end of season (EOS) across Europe.

• Can a crop yield prediction, using only open-source data at no cost to the user, and easily accessible?
• Can the prediction be made with a high degree of accuracy?
• Can the prediction be made in a timely manner, using only pre-processed remote sensing and climatic data?

Methodology

Monthly averaged timeseries for a range of climatological and biophysical variables were ingested into four models: multi-linear regression (MLR), support vector regression (SVR), random forest regression (RFR), and long short-term memory (LSTM). These were selected due to their varying degrees of complexity and popularity within the literature. The timeseries ingested for each crop was dependent on the crop calendar, defined by the United States Department of Agriculture (USDA). The accuracy of each model was tested using the root mean square error (RMSE) and r ². The models were trained and validated using in situ crop data, collected from the agricultural ministries in Poland and Spain.

Results

The LSTM model produced the most accurate predictions for wheat in Spain, with a RMSE of 0.5 and an r ² of 0.7. The SVR model produced the most accurate predictions for sunflower in Spain, with a RMSE of 0.5 and an r ² of 0.6. The RFR model produced the most accurate predictions for maize in Spain, with a RMSE of 0.5 and an r ² of 0.6. The MLR model produced the most accurate predictions for barley in Poland, with a RMSE of 0.5 and an r ² of 0.6.

Conclusions

It was found that the LSTM model produced the most accurate predictions for wheat in Spain, with a moderate correlation between modelled and in situ crop yields. It was found that Gross Primary Productivity (GPP) held the greatest influence on accurate prediction when predicting wheat in an the LSTM model, with temperature and Fraction of absorbed photosynthetically active radiation (FAPAR) being of approximately equal importance, but to a lesser degree.

close

Introduction

As part of the ESA Coastal Erosion from Space Project (Ref:SO-RP-ARG-003-055), a methodology was developed and applied to track the extent of the coastline through the derivation of waterlines and shorelines. Waterlines are land/sea boundaries derived from remote sensing images, that are transformed into shorelines by applying corrections to account for tide, beach slope and wave conditions.

Objectives

This study focuses on the quantification of coastal physical change and in particular in the type of change studies commonly referred to as shoreline change analysis (SCA).

The objectives of this study are to:

• Develop a methodology to quantify changes in coastal ecosystems using Earth Observation data.
• Apply the methodology to a case study area to quantify changes in coastal ecosystems.

Methodology

An open-source digital shoreline analysis system ODSAS integrates SAGA, an open-source GIS, the R programming language and CoastCR, that coastline time series and trend analyses using the baseline and transect approach.

ODSAS is a tool that allows the user to process and analyse shoreline data in a consistent and repeatable manner. The tool is designed to be user-friendly and to provide a range of statistical metrics that can be used to quantify the rate of coastal change.

The Coast Rates function is central to estimating five main metrics for each individual transect and generating aggregates for all transects, mirroring the functionality found in the USGS Digital Shoreline Analysis System DSAS tool. These metrics include Net Shoreline Movement (NSM), End Point Rate (EPR), Shoreline Change Envelope (SCE), Linear Regression Rate (LRR), and Weighted Linear Regression Rate (WLR), providing a comprehensive analysis of coastal dynamics.

These statistical metrics were used to provide an indication of the rate of coastal advance or retreat. These have been used to populate ecosystem accounting tables to determine the importance of regulating services in the coastal zone. Areas with faster rates of retreat can be determined to provide a low service value to regulating erosion as these areas are not resistant to erosion.

Results

It was found that regions of the coastline have been retreating at a high rate, and have low ecosystem value, whereas there are other regions more resistent to coastal change and providing more ecosystem value, to assist policy-makers assess the return on investment of planned intervention, such as beach nourishment or sea defenses.

close

Introduction

Monitoring and managing the Worldwide water use has never been more important. Increasing global population is placing unprecedented pressure on water resources, with global freshwater use increasing six-fold since 1900. As water use is increasing, the quantity of water per capita, available from renewable sources (river basins and recharging aquifers) is decreasing. Agriculture and irrigation play a significant role in this decrease, with the sector responsible for 24% of all water abstraction within the European Union (European Court of Auditors, 2021).

Water management is such a critical topic that the United Nations have designated it a Sustainable Development Goal (SDG). SDG 6.4 relates to increasing water use efficiency (WUE) across all sectors, including agriculture (United Nations, 2024).

Within agriculture, WUE is used to monitor the efficiency of irrigation systems. Under irrigated systems, water use efficiency increases as plants are held under mild water stress. This stress causes plants to decrease transpiration and maintain productivity. Furthermore, fertilizers are added within the water in irrigation systems, so it pays to determine how effective the irrigation systems are, to minimise fertilizer leaching. Improving water use efficiency can also increase crop yield, especially in drier and semi-arid conditions.

WUE is one of many indices and other geophysical products derived from Earth Observation satellite remote sensing images; data now in abundance, with timeseries spanning over two decades, with near global coverage and frequent repeat times. These timeseries provide opportunities to observe long-term patterns. It is of interest to test the usefulness of these datasets, regarding water use in agriculture.

Objectives

The purpose of this work is an initial investigation into water use efficiency and water management of agricultural land, within Europe. This purpose will be reached by achieving the following objectives:

• Develop a water use efficiency product from remotely sensed imagery
• Acquire a pre-processed product that describes soil moisture
• Acquire and pre-process crop yield data for an area of interest in Europe
• Correlate crop yield data with water use efficiency and soil moisture products

Methodology

This study is demonstrating the use of WUE products constructed from remotely sensed imagery. The “earth observation” methodology of integrated WUE will be used in this study. This was calculated as:

\[ \textit{WUE} = \frac{\textit{GPP}}{\textit{ET}} \]

Where GPP is Gross Primary Productivity (kg carbon /m²), and ET is evapotranspiration (kg water /m²). GPP and ET were collected from the Moderate Resolution Imaging Spectroradiometer (MODIS), at a spatial resolution of 500 metres, and a temporal resolution of every 8 days.

Volumetric Soil Water (VSW) was accessed from ECMWF ERA-5 Re-analysis. Level 4, at a depth of 100-289 cm, was used due to the dependency that vines have on soil moisture at this depth. Crop yield data was obtained from the Ministry of Agriculture and Fisheries (MAPA) for Spain (MAPA, 1999-2022).

The focus of this study will be to demonstrate the capability of Earth Observation data to assess how vineyards use water throughout the growing season. Spanish vineyards are being investigated due to their unique sensitivity to weather patterns. The homogeneity of their extent throughout the study area also aids this analysis due to the relatively low-resolution of earth observation products used within this study.

While this study remains a broad investigation into water management, there are three key questions that will be explored:

1. Are irrigated crop yields more stable than non-irrigated crop yields over long (~20 year) timeframes?
2. Do irrigated plots have a higher WUE than non-irrigated plots?
3. Is a change in VSW reflected in a change in the proportion of irrigated and non-irrigated yield?

Results

The dataset of available images and crop yield data was limited to 20 years and insufficient to resolve trends with any degree of certainty. Nonetheless, the study confirmed expectations that irrigation is effective in increasing crop yields. The z-score, a measure of variability within a sample, seems to indicate that irrigation promotes stability in the face of water stress, i.e. the distribution from the mean in irrigated plots is less than non-irrigated, but the sample size is too small for confidence.

Monitoring soil moisture in the most critical months of a vine growth demonstrates some impact on the final yield of a crop. VSW between February and May decreases dramatically in years when crop yield is also low. As expected, VSW does appear to mirror crop yield, especially between the years of 2012 and 2017.

Spring WUE appears to fluctuate considerably more than VSW, from year to year. It does appear that in drought years (such as 2012 and 2017), WUE increases, as would be expected based on the known physiological response when under drought stress, vines reduce evapotranspiration and, hence, increase in WUE. This begs the question of why, in 2002, a similar response wasn’t detected. However, in 2002 the water stress occurred in December, perhaps too early to affect the vines, but the detrimental effect on the cop yield was during harvest.

Discussion

The investigation of vineyards throughout La Rioja has demonstrated a proof-of-concept for analysing integrated WUE in irrigation using freely available and post-processed earth observation data. This study posed three questions to capture the breadth of what was being explored. From this analysis, the answers to those questions are as follows:

• Are irrigated crop yields more stable than non-irrigated crop yields over long (~20 year) timeframes?

Irrigated and non-irrigated crop yields demonstrate the same pattern over the past 20 years. However, irrigated crop yields are consistently higher than non-irrigated yields. A lower coefficient of variance (CV) for irrigated crops, compared to non-irrigated crops was apparent.

• Do irrigated plots have a higher WUE than non-irrigated plots?

WUE is higher in irrigated plots than non-irrigated plots. This may sound counter intuitive. However, if vines are kept under a very small amount of manageable water-stress, productivity in the vine can continue, while evapotranspiration decreases, increasing WUE.

• Is a change in VSW reflected in a change in the proportion of irrigated and non-irrigated yield?

Decreased VSW, as expected, lead to a similar response in crop yield, and irrigated yield constituting as a larger proportion of the overall yield. This change was most evident in weather-stressed years, such 2002, 2012, and 2017. This is as expected, as the decrease in VSW suggests a lack of moisture available to vines’ relatively deep routes, and hence the onset of drought-like conditions. In these conditions, irrigated plots are more suited to yield, and hence the difference between irrigated and non-irrigated yield increases.

Conclusions

WUE derived from EO sensors has been demonstrated, if only in a limited sample, to have potential use in monitoring water stress and its impact on crop yield. However, it is only one of many indices or variables that are available from EO remote sensing, many having likely relevance to monitoring agricultural water use. And viniculture, like all agriculture, has many significant factors to consider in addition to environmental factors. Soil type, topography, root stock and grape varieties, all contribute to productivity and, hence, yield.

The analysis suggests that irrigated yields are more consistent throughout weather stress, and that VSW is in-part related to a successful crop yield. This work also suggests that during weather-stressed years the ratio between irrigated and non-irrigated yield volume is in favour of irrigated yields. This suggests a further reliance upon irrigated yields in times of low soil moisture and low WUE. A preliminary plot-level study also suggests that irrigated plots have a higher WUE than non-irrigated plots.

This work would be improved upon by the access of higher resolution imagery, and detailed plot-level data. This would allow a more investigative approach which would consider multiple irrigation systems and wine varieties, in the three distinct regions of La Rioja. Specifically, advancements in Sentinel GPP and Sentinel ET would aid the production of a higher-resolution WUE product. A drought index, such as TVDI, would also more clearly quantify areas that are experiencing drought-induced stress.

close