Abstract
Remote sensing-based data on vegetation conditions provide important information for agriculture. In this study, the potential uses of the freely available High-Resolution Vegetation Phenology and Productivity Product (HR-VPP) are tested. This test examines the 2018 drought year in the state of Saxony, Germany, and the capabilities and limitations of the HR-VPP product in use with Integrated Administration and Control System (IACS) data. The results show that field and crop type-specific spatial (re)analyses of a drought are possible and that there is still great potential in this data analysis. Using the data in a new proposed VPP-based Farm-Level Temporal Comparison Indicator (VPP-FLTCI), it was not possible to tease out patterns in why farms applied for state drought aid in 2018 compared to other farms. In the future, even better and more detailed analyses based on the HR-VPP can be expected, as the data series with now a total of 5 years is still very short to generate sufficient references, especially in Central European agriculture, which is characterized by crop rotation.
HIGHLIGHTS
Unique analysis brought official IACS data together with open CLMS HR-VPP product.
Detailed 2018 drought reanalysis was possible on a crop type scale.
Detailed 2018 drought reanalysis was possible on a regional and spatial scale.
VPP data could not explain farms’ drought aid applications alone.
INTRODUCTION
A large number of Earth observation satellites collect today information on vegetation. This information supports decision-making processes in agriculture (Segarra et al. 2020), forestry (Lastovicka et al. 2020; Philipp et al. 2021), nature conservation (Lastovicka et al. 2020; Smith et al. 2021), climate impact (Gouveia et al. 2017) and climate change monitoring studies (Chen et al. 2019). Satellite-measured vegetation information is often calculated and presented with vegetation indices (Xue & Su 2017).
Climate change is making monitoring fire (Brigitte et al. 2012) and drought events (Gouveia et al. 2017) increasingly important globally and regionally. Agricultural production in Europe repeatedly suffers losses (Reinermann et al. 2019; Beillouin et al. 2020) due to such drought events, which are as high as in 2018 that government aid for farmers was made possible (D'Agostino 2018). In Germany alone, 292 million euros was paid out to farmers as part of drought aid (Handelsblatt 2022). The Federal State of Saxony accounted for around 33 million euros of this (10.3%) (Media Service Saxony 2018). However, the monitoring of agricultural drought is important not only on a regional scale, because on a global scale, there are high financial losses on agriculture due to drought events that affect millions of people (Hazaymeh & Hassan 2016; Sutanto et al. 2019).
To measure agricultural drought (as a differentiation from hydrological drought, meteorological drought and socio-economic drought (Hazaymeh & Hassan 2016)), knowledge of soil water deficit in particular plant available water, which in turn leads to the loss of agricultural yields (Liliane & Charles 2020), is particularly necessary. The opposite approach is taken by remote sensing data on vegetation since the condition of this can be used to indirectly infer the supply of soil water. Existing products to estimate soil moisture in the area of interest are for example the ‘Soil Moisture Traffic Light’ (https://life.hydro.tu-dresden.de/BoFeAm/dist/index.html (Kronenberg et al. 2022)) which used the water budget model LWF-Brook90R (Schmidt-Walter et al. 2020). Six classes of soil drought intensity are offered each by the ‘UFZ Drought Monitor’ (https://www.ufz.de/index.php?de=37937) and the ‘BKG Drought Atlas’ (https://atlas.bkg.bund.de/webapps/duerreatlas/), as both are based on the mesoscale hydrologic model (mHM) (Samaniego et al. 2010; Kumar et al. 2013). In the neighboring Czech Republic, there was the Czech Drought Monitor System, which not only depicts current and modeled conditions, but also weekly reports on vegetation conditions based on satellite images (Trnka et al. 2020). Here, drought forecast (+9 days and +2 months) calculated by numerical weather prediction models were implanted too. Global or continental drought monitoring systems (e.g. the European Drought Observatory) also exist, but will not be mentioned further here.
Using in-situ drought indicators like the Standardized Precipitation Index (SPI) (McKee et al. 1993), Standardized Soil Moisture Index (SSI) (Hao & AghaKouchak 2013) and Multivariate Standardized Drought Index (MSDI) (Hao & AghaKouchak 2014) were also common. They could connected in models like the Global Integrated Drought Monitoring and Prediction System (GIDMaPS) (Hao et al. 2014), too.
Remote sensing data provided a large number of agricultural drought monitoring indices, similar to in-situ indices, for this purpose (Hazaymeh & Hassan 2016). In contrast to commonly used in-situ-based indices like SPI, the remote sensing indices were calculated mainly by band recombination. Satellites from which spectral bands are used for this purpose are currently often Sentinel-2, Landsat 8 or Moderate Resolution Imaging Spectroradiometer (MODIS). Remote sensing indices can be used, for example, to measure vegetation health (Trnka et al. 2020) like the Normalized Difference Vegetation Index (NDVI) or to calculate drought conditions with the Perpendicular Drought Index (PDI) (Ghulam et al. 2008). Also, a monitoring of the vegetation water content with the Simple Ration Water Index (SRWI) (Zarco-Tejada et al. 2003) or Normalized Different Water Index (NDWI) (Gao 1996), was possible. Kloos et al. (2021) figured out with correlation analysis over the growing season from 2001 to 2020 in Bavaria (Germany), that the Temperature Condition Index (TCI) and the Vegetation Health Index (VHI) correlate strongly with soil moisture agricultural yield anomalies. They derived from this that these indices had the potential to detect agricultural drought in their study area. A soil moisture deficit in the growing season was the basis of agricultural drought, so soil moisture-based indices described agricultural drought better than meteorological and hybrid drought indices (Chatterjee et al. 2022). There were also various satellite image-based products and services for soil moisture as an indicator of drought. For Europe, these would include for example the following products: Copernicus Global Land Service–Soil Surface Moisture (CGLS-SSM) (Bauer-Marschallinger & Massart 2022), Soil Moisture and Ocean Salinity (SMOS), ASCAT or Soil Moisture Active Passive (SMAP) (El Hajj et al. 2018; Portal et al. 2020). However, the spatial resolution of these products of at least 1 km is not sufficient for many applications, such as field-based evaluations. Remote sensing can also contribute to the understanding and measurement of drought events using indirect methods by looking at the dynamics of dams and lakes and their water surface extent (Wieland & Martinis 2020; Büttig et al. 2022; Dehkordi et al. 2022). There are also indirect in-situ indicators, such as low water levels at river gauges.
Multi-criteria approaches (MCAs), such as those by Ihinegbu & Ogunwumi (2021) combined in-situ observations (precipitation) with satellite imagery (land surface temperature, NDVI). Using this MCA, they classified the neighboring federal state to the north of Saxony into three zones of drought prevalence with a focus on agricultural fields and found that more than 77% of the fields were in the high drought zone in 2018. Philipp et al. (2021) said by using the harmonic modeling method of NDVI, that in August 2018 the models indicated in Germany severe drought conditions across all forests. A strong correlation between in-situ and remote sensing data was observed for soil moisture and the self-calibrated Palmer Drought Severity Index (Philipp et al. 2021). As will be shown in the next section, the HR-VPP is an aggregation of indices to a new product, too.
Not only is the current monitoring of a drought important, but also the forecasting of such events is essential. Bouras et al. (2021) analyzed, that models using multiple data sources outperformed models based on a single dataset in cereal yield forecasting. Two months before harvest, for yield prediction satellite drought indices (e.g. TCI) were a major source of information by contributing up to 73% to the prediction accuracy in Marocco. For forecasting, models like LISFLOOD using standardized meteorological drought indices (e.g. SPI or Standardized Precipitation Evaporation Index (SPEI)) were used with a higher skill prediction for longer drought events than for short ones (Sutanto et al. 2019).
By providing services, such as the Copernicus Land Monitoring Service (CLMS, https://land.copernicus.eu/) provide ready-made products to the data user. No in-house processing resources, ground truth surveys and algorithms need to be applied to compute these products. There are problems associated with such simplification, since technical questions call for specific products that cannot answered satisfactorily with products ‘for the general public’. In order to evaluate whether these products are usable at all, they have first been tested with regard to their limits and possibilities. Especially institutions from the areas of public authorities and administration can benefit from such ready products since participation in the use of Copernicus data is associated with expenses that this user group often cannot afford.
For monitoring and assessing drought, data on vegetation conditions can be useful (Reinermann et al. 2019), as described above. Among the products offered is the High-Resolution Vegetation Phenology and Product (HR-VPP) (Smets et al. 2021). However, whether these data are actually useful in the case of agriculture, has yet to be evaluated. Therefore, this study aimed to resolve the following questions:
- 1.
Is the annual HR-VPP product generally suitable to trace the impact of the 2018 drought in Saxony as a reanalysis?
- 2.
Can regional differences of the drought effect be worked out (spatial level of municipalities/counties)?
- 3.
Can crop-specific differences of the drought effect of 2018 be worked out (spatial level of municipalities/counties)?
- 4.
Can the available data be used to explain why which ever farms applied for drought aid and why which ever did not?
The results of this study can provide important information on whether and under what circumstances users from the field of agriculture or agricultural administration can expect added value from these HR-VPP products.
Previous work with the HR-VPP product examines greenness in urban areas and concludes that the HR-VPP leads to large overestimates of urban greenness (Borgogno-Mondino & Fissore 2022). Bojanowski et al. (2022) conclude that the parameters of vegetation phenology have a lower performance in crop recognition and recommend the use of temporal datasets, which are more powerful.
To date, there are fewer studies on the evaluation of HR-VPP data in relation to field crops in case of drought and dry condition productivity. This is certainly related to the obstacle that Integrated Administration and Control System (IACS) data were not freely available. However, these official data allow insights into location and crop type that have previously been hidden from the public. In addition, the core question around the effect of the 2018 drought on agriculture in Saxony is a significant use case that is being addressed.
STUDY AREA
Cereals were the most important field crop in Saxony. The area under cultivation fluctuated by several thousand hectares from year to year. Root crops initially lost importance in the period under consideration, but the area under cultivation was increasing again (SMEKUL 2022).
The drought year 2018 was particularly noteworthy: the average annual temperature in 2018 in Saxony was 10.3 °C. This is 2.2 Kelvin above the multi-year average (1961–1990), which is 8.1 °C. It was the warmest year since 1881. With almost 2,060 h of sunshine (multi-year average 1961–1990: 1,549 h) and with about 475 l/m2 of precipitation (multi-year average 1961–1990: 699 l/m2), 2018 was one of the years with the lowest precipitation since 1881 (SMEKUL 2019). As a result, the harvest volume, which varied from region to region, was significantly lower than usual, especially for winter cereals, but also for potatoes. Corn yielded only half the previous year's level. This and dried-up pastures caused fodder shortages on livestock farms (SMEKUL 2019).
The German State and the Federal State of Saxony drought aid program supported most farmers and livestock keepers to maintain their liquidity despite the high losses and necessary and the need to make additional purchases. More than 30 million euros were available for this purpose in Saxony (SMEKUL 2019; Media Service Saxony 2018).
DATASETS AND PRE-PROCESSING
In this study, four main data sources were used, for 2017, 2018, 2019 and 2020, respectively.
High-Resolution Vegetation Phenology and Product (HR-VPP)
The first data source was the HR-VPP (Smets et al. 2021), which can be obtained from the CLMS. This in turn links to the WEkEO portal to obtain data: https://www.wekeo.eu/ (login required). The HR-VPP was provided by tile, following the common provision of Sentinel-2 tiles (110 × 110 km with 10 km overlap at the edges). With 10 m × 10 m pixel resolution, there was no scale mismatch between the agricultural fields and the product used (Bouras et al. 2021). As a dimensionless value per pixel, the HR-VPP mapped a value that was, at its core, the length of the growing season and the amplitude (intensity) of it. Thirteen parameters are used to calculate this value (Table 1).
Parameter . | Parameter description . | Unit . |
---|---|---|
SOSD | Day of start-of-season | day-of-year |
EOSD | Day of end-of-season | |
MAXD | Day of maximum-of-season | |
SOSV | Vegetation index value at SOSD | PPI |
EOSV | Vegetation index value at EOSD | |
MINV | Average vegetation index value of minima on left and right sides of each season | |
MAXV | Vegetation index value at MAXD | |
AMPL | Season amplitude (MAXV – MINV) | |
LENGTH | Length of season (number of days between start and end) | day |
LSLOPE | Slope of the greening-up period | PPI × day−1 |
RSLOPE | Slope of the senescent period | |
SPROD | Seasonal productivity. The growing season integral is computed as the sum of all daily values between SOSD and EOSD | PPI × day |
TPROD | Total productivity. The growing season integral is computed as the sum of all daily values minus their base-level value |
Parameter . | Parameter description . | Unit . |
---|---|---|
SOSD | Day of start-of-season | day-of-year |
EOSD | Day of end-of-season | |
MAXD | Day of maximum-of-season | |
SOSV | Vegetation index value at SOSD | PPI |
EOSV | Vegetation index value at EOSD | |
MINV | Average vegetation index value of minima on left and right sides of each season | |
MAXV | Vegetation index value at MAXD | |
AMPL | Season amplitude (MAXV – MINV) | |
LENGTH | Length of season (number of days between start and end) | day |
LSLOPE | Slope of the greening-up period | PPI × day−1 |
RSLOPE | Slope of the senescent period | |
SPROD | Seasonal productivity. The growing season integral is computed as the sum of all daily values between SOSD and EOSD | PPI × day |
TPROD | Total productivity. The growing season integral is computed as the sum of all daily values minus their base-level value |
PPI, plant phenology index (Smets et al. 2021).
Measurements are taken for these annual products from 01 January to 31 December of each calendar year. Currently, the product is provided for the years 2017 to 2021. The tiles used for this investigation in Saxony can be taken from Table 2.
Number . | Product Name . |
---|---|
1 | VPP_2017_S2_T32UQA-010m_V101_s1_TPROD |
2 | VPP_2017_S2_T32UQB-010m_V101_s1_TPROD |
3 | VPP_2017_S2_T33UUT-010m_V101_s1_TPROD |
4 | VPP_2017_S2_T33UVT-010m_V101_s1_TPROD |
5 | VPP_2017_S2_T33UUS-010m_V101_s1_TPROD |
6 | VPP_2017_S2_T33UVS-010m_V101_s1_TPROD |
7 | VPP_2018_S2_T32UQA-010m_V101_s1_TPROD |
8 | VPP_2018_S2_T32UQB-010m_V101_s1_TPROD |
9 | VPP_2018_S2_T33UUT-010m_V101_s1_TPROD |
10 | VPP_2018_S2_T33UVT-010m_V101_s1_TPROD |
11 | VPP_2018_S2_T33UUS-010m_V101_s1_TPROD |
12 | VPP_2018_S2_T33UVS-010m_V101_s1_TPROD |
13 | VPP_2019_S2_T32UQA-010m_V101_s1_TPROD |
14 | VPP_2019_S2_T32UQB-010m_V101_s1_TPROD |
15 | VPP_2019_S2_T33UUT-010m_V101_s1_TPROD |
16 | VPP_2019_S2_T33UVT-010m_V101_s1_TPROD |
17 | VPP_2019_S2_T33UUS-010m_V101_s1_TPROD |
18 | VPP_2019_S2_T33UVS-010m_V101_s1_TPROD |
19 | VPP_2020_S2_T32UQA-010m_V101_s1_TPROD |
20 | VPP_2020_S2_T32UQB-010m_V101_s1_TPROD |
21 | VPP_2020_S2_T33UUT-010m_V101_s1_TPROD |
22 | VPP_2020_S2_T33UVT-010m_V101_s1_TPROD |
23 | VPP_2020_S2_T33UUS-010m_V101_s1_TPROD |
24 | VPP_2020_S2_T33UVS-010m_V101_s1_TPROD |
Number . | Product Name . |
---|---|
1 | VPP_2017_S2_T32UQA-010m_V101_s1_TPROD |
2 | VPP_2017_S2_T32UQB-010m_V101_s1_TPROD |
3 | VPP_2017_S2_T33UUT-010m_V101_s1_TPROD |
4 | VPP_2017_S2_T33UVT-010m_V101_s1_TPROD |
5 | VPP_2017_S2_T33UUS-010m_V101_s1_TPROD |
6 | VPP_2017_S2_T33UVS-010m_V101_s1_TPROD |
7 | VPP_2018_S2_T32UQA-010m_V101_s1_TPROD |
8 | VPP_2018_S2_T32UQB-010m_V101_s1_TPROD |
9 | VPP_2018_S2_T33UUT-010m_V101_s1_TPROD |
10 | VPP_2018_S2_T33UVT-010m_V101_s1_TPROD |
11 | VPP_2018_S2_T33UUS-010m_V101_s1_TPROD |
12 | VPP_2018_S2_T33UVS-010m_V101_s1_TPROD |
13 | VPP_2019_S2_T32UQA-010m_V101_s1_TPROD |
14 | VPP_2019_S2_T32UQB-010m_V101_s1_TPROD |
15 | VPP_2019_S2_T33UUT-010m_V101_s1_TPROD |
16 | VPP_2019_S2_T33UVT-010m_V101_s1_TPROD |
17 | VPP_2019_S2_T33UUS-010m_V101_s1_TPROD |
18 | VPP_2019_S2_T33UVS-010m_V101_s1_TPROD |
19 | VPP_2020_S2_T32UQA-010m_V101_s1_TPROD |
20 | VPP_2020_S2_T32UQB-010m_V101_s1_TPROD |
21 | VPP_2020_S2_T33UUT-010m_V101_s1_TPROD |
22 | VPP_2020_S2_T33UVT-010m_V101_s1_TPROD |
23 | VPP_2020_S2_T33UUS-010m_V101_s1_TPROD |
24 | VPP_2020_S2_T33UVS-010m_V101_s1_TPROD |
Data source: https://www.wekeo.eu/.
The selected product differed in up to a maximum of two main growing seasons. The decision was made to use the first growing season (‘s1’ = Season 1) total productivity (‘TPROD’), as this represented the main vegetation phase in Central Europe. In addition, this selection minimized the effects of subsequent crops (winter crops seeding in autumn, catch crops), which are still cultivated and grown in the same year of investigation but represented not the researched crop in this year.
Integrated Administration and Control System (IACS)
The second data source was non-freely available data from agricultural subsidies (IACS). These were collected as part of the farmers' application for agricultural subsidies in the European Union and were also available in GIS-readable data for the years 2017 to 2020. For Saxony, these contain a total of around 165,000 to 170,000 digitized fields each year with information on up to 166 notifiable crop types. This was information on field areas around 900,000 ha depending on the year. Due to data protection regulations and the fact that these data were farm business secrets, they were not freely available. To fill the data gap for non-governmental researchers here, one could also alternatively use satellite-based crop type mapping (Bojanowski et al. 2022) instead of IACS data.
Application data of farmers for drought aid in Saxony 2018
Also, and for the same reasons, not freely available were the 2018 farm drought aid applications. These applications were documents and contained only economic information. However, the farm identifier number could be used to identify the farm and thus associated farmland in IACS. HR-VPP values for this farmland could now be compared between applicants and comparable land from farms that did not apply. Applications from around 310 farms with a farmland area of 150,000 ha in 2018 were available for this study. For comparison, a total of around 900,000 ha were registered in IACS in Saxony in the same year. Thus, the area of drought applications corresponded to a share of circa 17%. This was a one-time government aid to date, so no other data was available for comparable years. The condition for a successful application is that farms submit this application for which the annual production from land production was at least 30% below the average of the previous 3 years and whose existence is threatened.
CORINE Land Cover dataset 2018
To initially investigate whether the HR-VPP dataset was sensitive enough to different categories of land cover and land use to measure the 2018 drought, the 2018 CORINE Land Cover (CLC) dataset is used: https://land.copernicus.eu/pan-european/corine-land-cover. CLC divided the earth's surface into 44 classes as land cover/land use mapping (LULC), but not all classes were relevant for the study area of Saxony. Of particular note was the minimum mapping unit of 25 ha per object. Subsequent alterations must have a minimum extent of 5 ha. CLC often had to work with generalizations in order to comply with the minimum mapping unit. These generalizations contradicted the high spatial resolution of the VPP product and associated land cover mapping capabilities (MMU = 100 m2 VPP versus 250,000 m2 CLC). For this reason, the linking of both products (VPP with CLC) could only serve to provide a first general overview of the possibilities and limitations of the VPP product.
METHODS
After determining whether a potential could generally be seen in the use of the data (section ‘Results 1’, Figure 2), the detailed analysis was continued. For this purpose, the zonal statistics for the crops were calculated from the IACS data for each year. The sum of the VPP value per field was collected as the main statistical value. Since the area size of the field as well as the regional location were also known, a wide variety of crop-specific and spatial evaluations could now be performed. To obtain mean values per crop, VPP totals were summed by spatial study level (county, municipality, farm) and divided by the associated area. Predominantly mean values were calculated for the selected crop in the selected spatial unit (section ‘Results 2’, Figure 2). Welch's t-test was used to demonstrate whether VPP productivity significantly changed in 2018 compared to available years 2017, 2019 or 2020. The six most widely grown crops in Saxony were used for this purpose. In 2018, these were winter wheat (20.67%), winter rapeseed (13.85%), mowing pastures (12.73%), winter barley (9.9%), silage maize (7.3%) and meadows (6.6%). These crops together covered almost 72% of the total cultivated area in Saxony (about 650,000 ha) and thus could be seen as a representative analysis for the whole state area, but also for other crops.
Bringing in information on which farms have applied for drought aid allowed for further analysis in this direction on a regional or local scale (section ‘Results 3’, Figure 2).
In the last step, the farm-specific disadvantage in productivity triggered by the 2018 drought was calculated. The average values of the HR-VPP per field were used as the data basis for this and extrapolated to the farm. Due to the crop rotation in cultivation and the short time series of only 4 years of HR-VPP, only a few reference areas of the farm itself were available to validate the influence of the drought.
This lack of reference data in productivity was resolved by using the multi-year mean of the respective field crop at the spatial level of the municipalities. If a farm had areas in different communities, they were each compared to the mean typical for those communities. The distribution of the farm's areas was intended to better resolve the different natural features of the state of Saxony. The very heterogeneous site conditions for agriculture in Saxony, ranging from river valleys to heath landscapes and from loess areas to the crest of the Ore Mountains, were thus minimized in their influence on the calculation of the value of the farms. The comparison of farms with land on good soils versus on poor soils was omitted. The influence of drought on vegetation productivity thus took into account individual farm site conditions.
The proposed ‘VPP-based Farm-Level Temporal Comparison Indicator (VPP-FLTCI)’ calculated here (Table 3) was a dimensionless value that allows comparison of the impact of drought on and between farms. Comparability was ensured because the comparison was always made within the cultivated crop types of a farm. This means that it was not an arable farm that was compared with a forage (grassland) farm, but the respective influences of the drought on the farm-specific crops.
. | . | . | . | . | F . | G . | . | I H*E . |
---|---|---|---|---|---|---|---|---|
A Farm . | B Crop . | C Municipal . | D = 20 Area in ha . | E20/D Partial area . | VPP mean for farm per crop and municipal in 2018 . | VPP mean at municipal in 2017, 2019 and 2020 . | H F − G VPP difference . | VPP difference partial at area . |
Jon Doe | Corn | Warendorf | 10 | 0.5 | 80 | 90 | −10 | −5 |
Jon Doe | Corn | Versmold | 5 | 0.25 | 110 | 105 | 5 | 1.25 |
Jon Doe | Wheat | Versmold | 5 | 0.25 | 70 | 100 | −30 | −7.5 |
VPP Farm-Level Temporal Comparison Indicator (FLTCI) | −11.25 |
. | . | . | . | . | F . | G . | . | I H*E . |
---|---|---|---|---|---|---|---|---|
A Farm . | B Crop . | C Municipal . | D = 20 Area in ha . | E20/D Partial area . | VPP mean for farm per crop and municipal in 2018 . | VPP mean at municipal in 2017, 2019 and 2020 . | H F − G VPP difference . | VPP difference partial at area . |
Jon Doe | Corn | Warendorf | 10 | 0.5 | 80 | 90 | −10 | −5 |
Jon Doe | Corn | Versmold | 5 | 0.25 | 110 | 105 | 5 | 1.25 |
Jon Doe | Wheat | Versmold | 5 | 0.25 | 70 | 100 | −30 | −7.5 |
VPP Farm-Level Temporal Comparison Indicator (FLTCI) | −11.25 |
Since the FLTCI was always determined by dividing with the area shares, the size of the farm and the number or size of fields were not relevant in the collection of the final value. The consideration of the spatial cultivation conditions was an important contribution to this, since one could not compare arable farms with good conditions with arable farms with poor natural conditions. However, one could then very well compare, measured against the respective baseline, how strongly the drought affects productivity (section ‘Results 3’, Figure 2).
The same method was used to calculate ‘baseline’ values for comparably normal years with less drought impact. This was the average per farm for the years 2017, 2019 and 2020 based on crops grown in that period and regionality based on the municipalities. For Table 3, this meant that column F no longer uses the VPP average for 2018, but instead uses the farm VPP average for the reference years 2017, 2019 and 2020. By using the reference, farm performance was captured. The underlying question here was whether farms from a lower VPP baseline were more likely to apply for drought assistance than farms with higher productivity in the reference years.
There were 419 municipalities as administrative units in Saxony in 2018. The use of a smaller spatial administrative area, like local districts, as a reference for averaging was not recommended. Many communal districts would not be able to collect reference datasets due to the short time series if this crop was not otherwise cultivated in this area and time period.
By using the FLTCI on a drought year like 2018 compared to non-drought or less drought years as in the example in Table 3, it became a farm-level based drought indicator, too.
RESULTS
General suitability for 2018 drought reanalysis in Saxony
The results of the significance test showed (Table 4) that the VPP values of 2018 were mostly significantly different from those of other years (2017, 2019 or 2020) for the six crop types studied. No significant difference existed only for winter barley between 2017 and 2018, and maize from 2018 to 2019 and 2020. However, the latter 2 years were also not notable for predominantly humid conditions during the growing season. Especially high water-consuming crops such as maize thus suffered from continued topsoil water depletion in subsequent years. This was also reflected in the drought-influenced productivity in 2018 and was continued in the following years, which was why the drought year can no longer be significantly separated from these within the VPP values. It could be seen that many crops also show significant differences between comparison years when compared to each other.
. | p-value . | Winter wheat . | Winter barley . | Winter rapeseed . | ||||||
---|---|---|---|---|---|---|---|---|---|---|
. | 2017 . | 2019 . | 2020 . | 2017 . | 2019 . | 2020 . | 2017 . | 2019 . | 2020 . | |
2018 | Winter wheat | 0.0 | 0.0 | 0.0 | 0.0 | 0.0 | 0.0 | 0.0 | 0.0 | 0.0 |
Winter barley | 0.4 | 1.7 | 0.0 | 1.3 | 0.0 | 0.0 | 0.0 | 0.0 | 0.0 | |
Winter rapeseed | 0.0 | 0.0 | 0.0 | 0.0 | 5.8 | 1.6 | 0.0 | 0.0 | 0.0 | |
Maize | 0.0 | 0.0 | 0.0 | 0.0 | 0.0 | 0.0 | 0.0 | 0.0 | 0.0 | |
Mowing pastures | 0.0 | 0.0 | 0.0 | 0.0 | 1.2 | 2.0 | 0.0 | 0.0 | 0.0 | |
Pastures | 0.0 | 0.0 | 0.0 | 0.0 | 7.6 | 8.3 | 0.0 | 0.0 | 0.0 | |
. | Maize . | Mowing pastures . | Pastures . | |||||||
p-value . | 2017 . | 2019 . | 2020 . | 2017 . | 2019 . | 2020 . | 2017 . | 2019 . | 2020 . | |
2018 | Winter wheat | 0.0 | 7.2 | 2.6 | 0.0 | 0.0 | 0.0 | 0.0 | 0.0 | 0.0 |
Winter barley | 5.8 | 4.8 | 9.7 | 0.0 | 0.0 | 0.0 | 0.0 | 0.0 | 0.0 | |
Winter rapeseed | 0.0 | 0.0 | 0.0 | 0.0 | 8.4 | 3.0 | 0.0 | 3.2 | 0.0 | |
Maize | 0.0 | 1.8 | 3.1 | 0.0 | 0.0 | 0.0 | 0.0 | 0.0 | 0.0 | |
Mowing pastures | 0.0 | 0.0 | 0.0 | 0.0 | 0.0 | 0.0 | 0.0 | 0.0 | 0.0 | |
Pastures | 0.0 | 0.0 | 0.0 | 0.0 | 0.0 | 0.0 | 0.0 | 0.0 | 0.0 |
. | p-value . | Winter wheat . | Winter barley . | Winter rapeseed . | ||||||
---|---|---|---|---|---|---|---|---|---|---|
. | 2017 . | 2019 . | 2020 . | 2017 . | 2019 . | 2020 . | 2017 . | 2019 . | 2020 . | |
2018 | Winter wheat | 0.0 | 0.0 | 0.0 | 0.0 | 0.0 | 0.0 | 0.0 | 0.0 | 0.0 |
Winter barley | 0.4 | 1.7 | 0.0 | 1.3 | 0.0 | 0.0 | 0.0 | 0.0 | 0.0 | |
Winter rapeseed | 0.0 | 0.0 | 0.0 | 0.0 | 5.8 | 1.6 | 0.0 | 0.0 | 0.0 | |
Maize | 0.0 | 0.0 | 0.0 | 0.0 | 0.0 | 0.0 | 0.0 | 0.0 | 0.0 | |
Mowing pastures | 0.0 | 0.0 | 0.0 | 0.0 | 1.2 | 2.0 | 0.0 | 0.0 | 0.0 | |
Pastures | 0.0 | 0.0 | 0.0 | 0.0 | 7.6 | 8.3 | 0.0 | 0.0 | 0.0 | |
. | Maize . | Mowing pastures . | Pastures . | |||||||
p-value . | 2017 . | 2019 . | 2020 . | 2017 . | 2019 . | 2020 . | 2017 . | 2019 . | 2020 . | |
2018 | Winter wheat | 0.0 | 7.2 | 2.6 | 0.0 | 0.0 | 0.0 | 0.0 | 0.0 | 0.0 |
Winter barley | 5.8 | 4.8 | 9.7 | 0.0 | 0.0 | 0.0 | 0.0 | 0.0 | 0.0 | |
Winter rapeseed | 0.0 | 0.0 | 0.0 | 0.0 | 8.4 | 3.0 | 0.0 | 3.2 | 0.0 | |
Maize | 0.0 | 1.8 | 3.1 | 0.0 | 0.0 | 0.0 | 0.0 | 0.0 | 0.0 | |
Mowing pastures | 0.0 | 0.0 | 0.0 | 0.0 | 0.0 | 0.0 | 0.0 | 0.0 | 0.0 | |
Pastures | 0.0 | 0.0 | 0.0 | 0.0 | 0.0 | 0.0 | 0.0 | 0.0 | 0.0 |
The bold marked fields indicate testing of the same fruit types.
Regional and crop-specific VPP differences and drought effect
Peas and spring cereals (oats and barley) showed predominantly low values (Figure 4). Ponds used for pond farming had low values as expected, since this was actually not a comparable crop type in terms of land cover. High VPP values were reserved for grassland and arable feed crops (clover grass or field grass) as year-round land cover. Sugar beets also achieved high values. However, the highest productivity was in winter rapeseed. Winter cereals and corn reached medium values in this consideration. A harvest date in early summer is certainly not insignificant for these values.
These spatial differences could be shown in even greater detail for higher-resolution spatial levels. For example, the VPP mean value per municipality for winter wheat in 2018 could serve as an example (Figure 5). Clearly visible were the municipalities where winter wheat had high productivity. Municipalities with good production conditions were located directly next to municipalities with poor conditions, as could be clearly seen in the counties of North Saxony or Meißen.
These often deviated by another −10 to −20% and even more from the farmers that did not submit an application. Outliers upwards could be found especially in the county towns, since only a small number of fields were included in the statistics. Overall, strong disparities could be seen between the individual counties and crop types, also in combination with both variables.
Explaining each farm application for financial support with VPP data
No recognition of a particular pattern was possible at this point either. Farms that applied for drought aid were distributed to the same extent as those that did not. Farm performance in the reference years (x-axis) did not appear to affect the distribution.
DISCUSSION
For the preliminary investigation of an unknown dataset, the methodology of using CLC data proved sufficient. Here, the analysis immediately yielded results that were within the expected range (Figure 3). Since nationwide testing was performed, positive VPP trends could also have canceled each other out with negative ones. This was a conceivable scenario for events that were not as extreme as 2018, since, for example, the growing season could be extended in the mountains (Trnka et al. 2011), in Saxony the Ore Mountains, which is directly reflected in the VPP value. Why broad-leaved and coniferous forests (CLC codes 311 and 312) did not respond to the 2018 drought cannot be determined at this point.
These results were supported by the performed significance tests (Table 4), which highlight that the year 2018 turns out significantly different for a representative sample of crops.
This study also presented the first approaches to the evaluation possibilities of the VPP product. Crop species specifics and regional differences could be carried out in any combination, down to the farm and field level. In conjunction with IACS data, a wide variety of crop-specific and regional cartographic and statistical analyses could be performed (Figures 4–8).
Figures 6 and 8 illustrate the drastic impact 2018 has had on VPP productivity. Figure 8, compared to Figure 6, shows the additional disadvantage of applicants compared to the non-applicant total of farms once again. Unfortunately, these results could not be validated because the dimensionless units of the VPP were difficult to communicate, for example in the field of agricultural management. When working with VPP data, it was important that this study dealt with productivity derived from the satellite image alone. Thus, there was no yield model behind the output product as in other cases, such as for field crops (Migdall et al. 2009) or for grassland (Reinermann et al. 2020). The HR-VPP remained in its own world as a remote sensing product. Such remote sensing products must be translated and anchored for other disciplines, i.e. generally accepted, in order to develop their surplus value.
For agriculture, practitioners as well as decision-makers in politics, values like VPP were not tangible. For them, the yield data or yield losses caused by weather were the units of communication to be chosen. When working with VPP data to provide information to stakeholders and decision-makers, VPP will have a hard time being seen as a suitable parameter. Table 5 shows that comparisons between official harvest statistics (StaLa 2018) and losses in VPP productivity are difficult to nearly impossible. Reasons for this are that the periods of comparison are different and therefore do not replicate the same growing conditions. Only a small number of years were available for the VPP product, while the yield losses or the yield level were always compared with the yield data of the previous years (mostly 5–10 years on average). Furthermore, the VPP measured the productivity of the whole plant, while for yield did this only for the usable part of a plant (e.g. rapeseed or wheat grains). When the grain matured quickly, as in 2018, the productivity losses then overestimated the actual yield losses by several percentage points (Table 5). For crops such as sugar beets or potatoes, the situation was aggravated by the fact that yield formation occurs below the earth's surface, and thus below the area that can be detected by the satellite alone (without a connected growing model), which was why yield losses can be underestimated.
Crop . | Harvest losses in 2018 compared with the 10-year average . | Difference in VPP productivity 2018 to mean of 2017, 2019 and 2020 . | Difference in harvest losses to losses in VPP productivity . |
---|---|---|---|
Winter rapeseed | −19% | −22% | −3 |
Winter wheat | −13% | −22% | −9 |
Winter barley | −12% | −23% | −11 |
Potatoes | −24% | −16% | +8 |
Sugar beets | −24% | −12% | +12 |
Field grass | −43% | −26% | +17 |
Pastures | −37% | −17% | +20 |
Silage maize | −20 to −30% | −17% | min. +3 |
Crop . | Harvest losses in 2018 compared with the 10-year average . | Difference in VPP productivity 2018 to mean of 2017, 2019 and 2020 . | Difference in harvest losses to losses in VPP productivity . |
---|---|---|---|
Winter rapeseed | −19% | −22% | −3 |
Winter wheat | −13% | −22% | −9 |
Winter barley | −12% | −23% | −11 |
Potatoes | −24% | −16% | +8 |
Sugar beets | −24% | −12% | +12 |
Field grass | −43% | −26% | +17 |
Pastures | −37% | −17% | +20 |
Silage maize | −20 to −30% | −17% | min. +3 |
The VPP was also suitable as a time series to trace the impact of the 2018 drought. This was even possible precisely for fields or other land covers (Figure 10), although such analyses were only appropriate if the land cover was constant, as was the case for grassland, for example. For cropland, such a comparison could not in good conscience be undertaken from the time series, since crop rotation ensured that one would be comparing different crop types. Overall, the VPP time series was still very short, now 5 years (2017–2021), so time series analyses reached their limits here. In this study, the problem was circumvented by using the VPP community statistics per crop as a local basis for comparison. Figure 5 shows that this approach of using local growing conditions was justified.
The FTLCI per farm is calculated via the methodology in Table 3 proved not to be useful for the application case of declaring applications for drought aid (Figures 9 and 11). A threshold value in the FTLCI, below which the underperforming farms could be classified as most likely to apply, could not be derived.
An FTLCI indicator value remained that indicates whether the farm was more productive or not in 2018 compared to the local crop-specific mean than in the reference period (mean 2017, 2019, and 2020) as seen in Figure 11. This value may have an important place in the descriptive statistics of such drought events in the future, but it remains unusable for now. This is partly due to the short HR-VPP time series, and partly due to the lack of data and methods of validation. A new drought event could help to check, validate and improve what has already been achieved in the form of the development of such an indicator FTLCI. However, this study was at the very beginning of this process and needs a longer VPP dataset and, unfortunately, more drought events to expand the knowledge in the use of the HR-VPP for this use case.
CONCLUSION
This work has shown that there was much potential in combining IACS data with CLMS HR-VPP data. It was possible to investigate targeted questions about crop productivity in detail on a species-specific and spatial level. Also, this potential was far from fully exploited in this study, as the VPP product could be considered for diverse use cases (nature conservation, forestry, urban green). Since the HR-VPP data were provided free of charge, this increased the potential for future applications, for example, the detailed analysis of extreme events, such as an agricultural drought in 2018, as demonstrated in this work. However, the long waiting time for a new HR-VPP year product proved to be an obstacle, for example, the year 2021 was not published on the WEkEO platform until November 2022.
While much potential could be seen in event reanalysis with VPP data, an even greater challenge was whether it would be possible to use this or similar remote sensing products to explain drought aid applications. In 2018, millions of euros were distributed as drought aid. Thousands of hectares of agricultural land have been associated with the applicants. If it is possible to figure out the patterns, including using remote sensing data, according to which probabilities of application per farm can potentially be predicted, one would create a valuable tool to assist policymakers. The target here was to be able to predict, as early as May and June, how many applicants and what application amount would be expected in the summer if a drought aid program were to be relaunched. The VPP data provided important starting points for this, but as an annual total product, they will ultimately not be able to play a role in accompanying monitoring. A variety of reasons were possible why a farm ultimately made such an application for drought aid. Remote sensing data alone could not provide the sole parameters for explanation. From a large feature space of combinations, the pattern or patterns, if any, must first be found. Farm structure, number of livestock, location of land, soils and local climate, in an infinite number of combinations, can help explain whether there were systematic bases for drought aid applications. Machine and deep learning technologies can help extract these patterns. Recognized patterns can then greatly improve the predictive accuracy of future models, so this is an important research need that can be derived from this study. It is also worth pointing out the possibility that no patterns can ever be found at all, and the rationale is different: Applying for drought aid involves bureaucracy. Farm data must be disclosed and the application submitted. The farm manager always decides individually whether this makes sense for his farm or is feasible at all. Even the best model cannot adequately reflect such individual decisions. Despite high yield losses, there need not automatically be a threat to the farm's existence. Remote sensing can never provide this insight into the bookkeeping of a farm, which is why a farm-specific derivation of applications for drought aid will be extremely difficult. In the future, however, remote sensing can be used to identify regions at particular risk using HR-VPP. Combined with IACS data, this can be used to make statistical predictions of where high probabilities of future drought-related climate impacts will occur in agriculture.
FUNDING
This work received no funding.
DATA AVAILABILITY STATEMENT
Data cannot be made publicly available; readers should contact the corresponding author for details.
CONFLICT OF INTEREST
The author declares there is no conflict.