Remotely Sensed Agriculture Drought Indices for Assessing the Impact on Cereal Yield

Abstract

This study aims to analyze the potential of different drought indices for identifying drought periods and predicting cereal yield in two semi-arid regions, Lleida in Catalonia and Kairouan in Tunisia, which have similar Mediterranean climates but different agricultural practices. Four drought indices, namely the Soil Moisture Anomaly Index (SMAI), the Vegetation Anomaly Index (VAI), the Evapotranspiration Anomaly Index (EAI), and the Inverse Temperature Anomaly Index (ITAI), were calculated from remote sensing data. Drought periods were identified from 2010/2011 to 2021/2022 based on the aforementioned indices. A correlation study between drought indices and wheat and barley yields was performed in order to select the most informative index and month for yield prediction. In the rainfed cereal area of Lleida, the strongest correlation was found between the EAI and VAI with barley yield (0.91 and 0.83, respectively) at the time of cereal maturity in June. For wheat, the strongest correlation was found between the EAI and VAI (0.75 and 0.72, respectively) at the time of cereal maturity in July. However, the VAI, EAI, and SMAI showed the best performance as an earlier indicator in March with a correlation with barley yield of 0.72, 0.67, and 0.64, respectively; the lowest standard deviation was for the SMAI. For wheat yield, the best earlier indicator was the SMAI in March, showing the highest correlation (0.6) and the lowest standard deviation. For the irrigated cereal zone of Kairouan, the strongest correlation (0.9) and the lowest standard deviation are found between the EAI and cereal yield in April. In terms of advanced prediction, the VAI shows a high correlation in March (0.79) while the SMAI shows a slightly lower correlation in February (0.67) and a lower standard deviation. The results highlight the importance of the EAI and SMAI as key indicators for the estimation and early estimation (respectively) of cereal yield.

1. Introduction

Drought, one of the most important natural disasters in terms of human and economic damage, is a period of prolonged dryness compared to normal for a region [1]. This hazard generally occurs as a result of two factors: (i) a medium-/long-term lack of precipitation that reduces soil moisture (SM) levels [2,3] and/or (ii) rising temperatures leading to evapotranspiration (ETP) levels that exceed water reserves [4]. The impacts of drought depend on the timing, intensity, spatial extent, and duration [5,6,7]. Their effects can range from damage caused by soil subsidence and a lack of water resources to reduced agricultural yields, increased fire hazards, increased erosion and desertification, and the spread of famine [2,8,9]. Poor harvests and declining agricultural productivity resulting from prolonged droughts have significant and far-reaching economic impacts. These adverse effects can exacerbate poverty and discourage investments in agricultural practices and technologies. The recurring nature of droughts poses a formidable threat to food security, as reduced crop yields lead to diminished food availability and, subsequently, higher food prices [10]. There are, therefore, four types of droughts: meteorological, agricultural, hydrological, and socioeconomic [11].

In most arid and semi-arid countries, water stress is worsening with high water demand and a scarcity of resources. Increased ETP due to climate change is further reducing water resources and increasing the effect of drought [12,13]. Predictions on climate change show that more than 33 countries will suffer from extremely high water stress in 2040, affecting the food security of over 896.54 million inhabitants [14]. Mediterranean countries are among the most affected by climate change and have suffered from successive droughts that affect agricultural production, increase aridity, and minimize water resources that are already scarce [10,15]. They are among the main hotspots of climate change, where droughts will become more frequent and longer [16]. In addition, the succession of drought periods causes a decrease in crop productivity, especially rainfed cereals, making the nutrition of a population that continues to increase and could reach more than 9 billion people in 2050 difficult [17]. Several periods of food insecurity have occurred in some countries as a result of successive drought periods [18,19]. Therefore, drought is classified among the most destructive causes in the agriculture sector. This hazard caused USD 37 billion in crop and livestock losses between 2008 and 2018, as a result of several drought events recorded worldwide [20], including more than USD 14 billion in production losses in Africa [21]. Other projections made by the United States Department of Agriculture show that some Mediterranean countries are probably going to enter a period of food insecurity in 2029 [22]. In this context, several studies have been focused on the study of drought and its effect on agricultural production [23,24,25,26,27,28,29,30,31].

Drought indices based on remote sensing represent a crucial tool for the study of drought in space and time. These indices allow for the study of weather conditions that influence plant water demand, including temperature, water availability to the plant through SM, water demand, and actual crop water use through ETP and vegetative development status [24,31]. Different drought indices have been developed to study and identify meteorological, agricultural, and hydrological droughts [32].

Meteorological drought indices are based mainly on precipitation and temperature, such as the Standardized Precipitation Index [33] and the Palmer Drought Severity Index [34], which can be a constraint for countries that do not have a very dense network of stations. Remote sensing allows repetitive coverage of very large areas and thus has become a crucial tool in the study of the spatiotemporal evolution of drought in remote or insufficiently monitored areas [35]. Surface and plant temperature remote sensing data have been widely investigated in several studies for detecting drought periods and assessing crop yields [36,37,38,39,40,41,42]. Among the established indices, the Temperature Anomaly Index (TAI) [43] has been largely used in the study of drought periods and their impact on yield in many regions, especially when a strong positive anomaly is recorded [40,44,45].

Regarding agricultural drought indices, optical sensor data have been extensively employed, particularly in conjunction with vegetation indices [46,47,48]. The Normalized Difference Vegetation Index (NDVI) [49] is among the main indices allowing the vegetation growth monitoring and drought identification. This index has served as a foundation for various derived drought indices, such as the Vegetation Condition Index [37], the Vegetation Health Index [47], and the Standardized Vegetation Index [48]. Notably, Amri et al. [50] proposed the Vegetation Anomaly Index (VAI) as a statistical drought index based on the NDVI, enabling the quantification of drought intensity and duration and its impact on vegetation. This index has proven to be one of the most frequently used indices of drought in different climates [46,51,52]. Moreover, ETP and potential evapotranspiration (PET) are crucial parameters in the field of operational agriculture drought monitoring [53]. Several drought indices have been developed based on these fundamental parameters [54,55,56,57,58,59]. One such index is the Evapotranspiration Anomaly Index (EAI) [60], which has demonstrated a robust correlation with meteorological drought indices, particularly the monthly precipitation. Additionally, the EAI retains a memory of past months, thereby reflecting the moisture conditions preceding a drought period when preceded by a period of precipitation [60].

SM is a different indicator of drought that helps to understand the cause of vegetative stress (drought or disease) and the time delay between the EAI and the meteorological drought through soil conditions. Several remote sensing SM data play a crucial role in monitoring agriculture drought, such as Soil Moisture and Ocean Salinity (SMOS), Soil Moisture Active and Passive (SMAP), Advanced Scatterometer (ASCAT), and Advanced Microwave Scanning Radiometer for EOS (AMSR-E) [46]. Several agricultural drought indices have been developed based mainly on SM data, such as the Soil Moisture Deficit Index [61], the Soil Moisture Condition Index [62], and the Multivariate Standardized Drought Index [63]. One widely used index is the Soil Moisture Anomaly Index (SMAI) [64]. This statistical index quantifies the level of water availability in the soil relative to the long-term average, providing valuable insights into SM conditions [51,65,66,67]. In a recent study [2], agricultural drought was studied in more detail by specifying the effect of SM, soil moisture drought, the vegetative state of the plant, and vegetative drought on the classification of drought events.

Various impacts on the physical characteristics of cereals can be observed based on the duration, intensity, and timing of drought occurrence during their vegetative cycles. It has been shown that the presence of water stress during the germination stage may cause a decrease in seed germination and viability. Similarly, drought periods detected during the developmental stage could lead to reductions in cereal length and nutritional content, while water stress during the flowering stage may result in decreased size and weight [68,69,70,71]. Studying the long-term variations in cereal production and drought indices can offer valuable insights into the relationship between these factors. By analyzing trends in cereal production, including both increases and decreases, in conjunction with corresponding changes in drought indices (such as monthly variations), it becomes possible to establish meaningful connections. For instance, positive anomalies of SM or vegetative activity indices (NDVI, ETP, etc.) may indicate wetter conditions, potentially leading to higher cereal production and productivity [72]. By examining the interplay between cereal production, drought indices, and the resulting variations over time, valuable insights into the complex dynamics that affect crop yields can be obtained and inform strategies for mitigating the impacts of drought on food production and security. The vegetative cycle of cereal crops relies on water availability to facilitate the emergence and subsequent temperature, particularly during the tillering stage, to promote growth. Additionally, adequate water is crucial for grain development and filling [30]. The wheat yield is significantly affected by the timing of drought periods, leading to reduced plant density during the emergence phase, decreased plant height during tillering, and diminished grain weight during flowering and dough development [73].

Numerous studies have also highlighted the effectiveness of drought indices in predicting cereal yields. A linear model for wheat forecasting was developed in Kansas and Ukraine using NDVI data, which achieved an error rate of 10% six weeks before harvest and further improved to 7% in 2009 [74]. Additionally, an ETP-based drought index demonstrated a robust early correlation with soybean and corn yields, with an even higher correlation observed during the flowering and grain-filling stages [24]. The correlation between drought indices and rainfed winter wheat productivity in Morocco was studied by the authors of [44], who found the highest correlation with SM indices in December (germination stage) and vegetation indices in March (stem elongation stage). The high correlation between the SM index and yield was explained because the wet conditions encouraged farmers to cultivate cereals. Studies on irrigated winter wheat in India, [75,76] revealed a strong correlation between SM drought indices and winter wheat yields first in October and November, germination time, and then in February, during the stem elongation stage. These studies seem to indicate that SM-based indices would be good indicators for early yield forecasts, while the NDVI and ETP-based indices would be correlated with yield closer to harvest stages. However, on the one hand, some of these studies were only using one type of drought index and, on the other hand, some were performed over dryland cereals, whereas others were over irrigated cereals. A comprehensive analysis of the correlation between the different remote sensing drought indices (based on the NDVI, ETP, SM, and Land Surface Temperature (LST)) and yield for irrigated and dryland cereals is missing.

In this context, the objectives of this study are (1) to analyze the relationship between drought indices and the cereal vegetative status for both dryland and irrigated cereals and (2) to investigate the correlation between the different indices and the cereal yields.

Section 2 presents the study area, the database, and the applied methodology. Section 3 shows the achieved results. Section 4 is reserved for discussions and Section 5 is for conclusions.

2. Materials and Methods

2.1. Study Area

Two areas, characterized by a Mediterranean climate and different agricultural practices (rainfed, irrigated, or support irrigated crops), are studied: Lleida in Catalonia (Spain) and Kairouan in Tunisia. In this study, we refer to the agriculture year as a year beginning in September of the previous year and ending in August of the following year.

2.1.1. Lleida, Catalonia

The selected study area in Catalonia is Lleida (0°32′E to 1°5′E, 41°2′N to 42°87′N), covering an area of 12,150 km², as shown in Figure 1. This region is characterized by a Mediterranean climate with very high variability in precipitation. Lleida is located between the two isohyets, from 200 mm in the plain to over 1200 mm in the Pyrenees. The average annual rainfall at Segrià station in the Lleida plain during the last 22 years is around 355 mm. The annual precipitation in this station varies from less than 300 mm recorded in dry years (2000, 2005, 2006, 2007, 2010, 2011, and 2015) to more than 500 mm in wet years, recorded in 2003 and 2020 [77]. The four rainiest months are May, September, October, and November, with an average of more than 45 mm per month. The monthly average temperature varies from 1 °C (in January) to 26 °C (in July and August) and can exceed 38 °C in summer and be below 0 °C in winter. The annual reference ETP is around 1100 mm [78].

Lleida is an area with a very heterogeneous landscape that mixes large irrigated and rainfed areas. This area occupies the first rank in the production of cereals in Catalonia [79]. Several droughts have been recorded in Lleida, such as the drought of 1940, 1950, 1980, 1990, 1991, 1995, 2005, 2007, and 2008 [80,81,82]. These periods of drought have influenced rainfed crop yields, especially barley and wheat.

2.1.2. Kairouan, Tunisia

The Kairouan governorate (9°15′E to 10°15′E, 35°45′N to 36°55′N) is located in the center of Tunisia, covering an area of 6712 km², as shown in Figure 1. This region is characterized by a semi-arid climate and increasingly frequent drought periods. This region is characterized by highly spatially and temporally irregular rainfall and a wide range of temperatures. The Kairouan governorate is located between isohyets of 200 and 400 mm per year [83]. The annual rainfall varies between less than 200 mm during dry years (2000/2001 and 2012/2013) and more than 500 mm during wet years (1988/1989 and 1995/1996) [50,84]. The average monthly temperature ranges between 11 °C (January and February) and 30 °C (July and August). During extreme events, the temperature can exceed 40 °C. The annual reference ETP is around 1600 mm.

This area has a diversified agricultural landscape where crops are conducted in rainfed or irrigated and sometimes in complementary irrigation. Rainfed crops are predominantly concentrated in the northern section of the area, whereas most of the southern fields are irrigated. Additionally, a large number of rainfed cereal plots in the southern part have undergone a transition to irrigated cultivation in recent years due to consecutive periods of drought.

In this area, the major problem is the long duration of high temperatures, which causes a water deficit in the soil, leading farmers to switch from rainfed to irrigated agriculture. From 1990 to 2016, Tunisia declared ten years of drought (1994, 1995, 1997, 2000, 2001, 2002, 2008, 2010, 2013, and 2016), of which the most severe were the three drought years of 2000, 2001, and 2002, which cost USD 54 million [8,50,84,85]. The Kairouan governorate ranks sixth nationally and first in central Tunisia in cereal production. This region plays an important role during drought periods in ensuring cereal production, as there are more irrigated perimeters overexploiting the Kairouan aquifer [86,87,88] than in other governorates lacking groundwater resources. The variation of cereal production in Tunisia depends both on the climate conditions and on the government’s decision on the price of seeds, fertilizers, and cereal sales to encourage farmers to grow cereals.

Given the distinct agricultural practices and varied climatic conditions between the two study areas, Figure 2 shows the phenological cycles of wheat and barley, where notable differences are observed. In both study areas, wheat is sown from early November to late December in Lleida and 20 January in Kairouan. Barley, on the other hand, is sown between early November to late December in Lleida and mid-January in Kairouan. During the tillering phase, temperature plays a crucial role in accelerating the development of cereals. In the Kairouan study area, favorable temperature conditions facilitate their growth. On the other hand, in Lleida, low temperatures during December and January lead cereals to enter dormancy until they accumulate enough degree joules to resume development, resulting in a lengthened cycle. In Kairouan, farmers irrigate throughout the cereal cycle with a rotation of around 10 to 15 days, depending on the size of the field. The maximum vegetative development of cereals is reached between mid-March and mid-April. In contrast, the peak vegetative development in Lleida is observed from the end of March to early May for barley and from early April to late May for wheat. The maturation stage is reached by the end of May in Kairouan; in Lleida, this stage occurs by the end of June to mid-July.

2.2. Database

The used dataset is composed of remote sensing data (SM, NDVI, ETP, and LST) used for the calculation of the drought indexes, annual cereal masks, cereal yield data at the provincial scale (Lleida and Kairouan), and precipitation data from representative meteorological stations in the study area. The period considered for this study is from the 2010/2011 to the 2021/2022 agricultural year.

2.2.1. Annual Cereal Masks

A cereal mask is necessary to extract the corresponding drought index values for each agricultural year. For Lleida, the land cover maps are published by the Department of Climate Action, Agriculture and Rural Agenda (Departament d’Agricultura, Ramaderia, Pesca i Alimentació (DARPA) (https://agricultura.gencat.cat/ca/ambits/desenvolupament-rural/sigpac/mapa-cultius/ (accessed on 20 September 2021)) since 2009/2010. A barley and wheat mask at the field scale were extracted for each agricultural year separately.

For the Kairouan governorate, a cereal mask was obtained by applying the methodology proposed by Zribi et al. [89], which was subsequently validated in other studies [90,91]. This decision tree classification approach uses NDVI values at the beginning (e.g., December) and in the middle (e.g., April) of the cereal cycle to discriminate cereal fields. NDVI images extracted from Landsat 5 for 2010/2011, Landsat 7 for 2011/2012 and 2012/2013, Landsat 8 for 2013/2014 to 2016/2017, and Sentinel 2 data for 2017/2018 to 2021/2022 were used to obtain the cereal mask for each agricultural year. This methodology has difficulties in distinguishing wheat from barley, leading to the grouping of these two classes, wheat and barley, in the same class called cereal.

Figure 3 shows the changes in cereal cultivation within the Kairouan area for the 2012/2013 and 2020/2021 agricultural years. The data reveal a substantial increase from 23,243 ha in the 2012/2013 agricultural year to 32,667 ha in the 2020/2021 agricultural year. This difference is due to the extension of irrigated perimeters, particularly on the Kairouan plain, as an adaptation to the severity of the climate. The governorate of Kairouan can be divided into two parts, the northern part, where cereals are predominantly rainfed, which is marked by a red outline, and the southern part, where cereals have become predominantly irrigated, which is marked by a blue outline. In contrast, the annual cereal mask for the Lleida study area remains relatively stable, with barley cultivation spanning 106,761 ha and 92,785 ha and wheat cultivation covering 49,354 ha and 56,311 ha during the same agricultural years, respectively.

2.2.2. Cereal Yields

For Lleida, a database of wheat and barley yields and areas is published by the Statistical Institute of Catalonia (SIC) (https://www.idescat.cat/pub/?id=aec&n=446 (accessed on 20 September 2021)). Figure 4a shows the production and cultivated area of barley and wheat throughout the study period. The data reveal a consistently high barley production, accompanied by a nearly constant total cultivation area of approximately 100,000 ha. In comparison, wheat production remains lower than barley, despite experiencing a slight increase in the 2020/2021 agricultural year, while barley production decreased. Notably, within the Lleida region, the agricultural year 2019/2020 witnessed the highest recorded cereal production, with 549,373 tons of barley and 265,706 tons of wheat. Conversely, the lowest yields were recorded in the 2014/2015 agricultural year, with barley and wheat production amounting to 279,392 tons and 156,221 tons, respectively.

The yield data of the Kairouan governorate are taken from the Minister of Agriculture, Hydraulic Resources, and Maritime Fishing of Tunisia. These estimates are obtained by sampling crops using 1 m × 1 m frames. By taking two samples per field, the total cereal samples can exceed 3150 for the whole country. Cereal grain weight is measured separately for each governorate to obtain the average productivity for the entire governorate. An estimate of total production is then deduced from the area occupied by cereals. Figure 4b illustrates the variation in cereal production and cultivated area in the governorate of Kairouan, which encompasses both barley and wheat classes. Notably, the highest recorded cereal production occurred in the agricultural year 2011/2012 (241,400 tons).

2.2.3. Precipitation Data

Figure 5 below shows annual precipitation data from the Segrià and Urgell stations in Lleida and the Kairouan and Oueslatia stations in Kairouan. The annual precipitation is computed for each agricultural year independently, encompassing a period from September of one year to August of the following year (e.g., annual precipitation for 2011 is derived from September 2010 to August 2011). Figure 5 includes lines with the mean precipitation values during the study period for the respective stations.

Figure 5 shows significant variations in precipitation patterns between different years and study areas. For instance, in Lleida, the years 2011, 2012, 2014, 2016, and 2022 stand out as the driest, while 2020 emerges as the wettest. In the Kairouan study area, the data show a pronounced variability in precipitation, with notable distinctions between the two stations. The central region experienced its driest years in 2012, 2013, 2015, 2016, 2017, and 2022, while the northern region faced particularly dry conditions in 2013, 2015, 2016, 2017, 2018, 2020, 2021, and 2022.

2.2.4. Remote Sensing Data

a.

Soil Moisture data

The SMOS mission [92], launched in November 2009, is part of the European Space Agency’s Earth Explorer missions. It is the first mission to provide global observations of SM from L-band microwave measurements every 3 days at 40 km. In this study, a 1 km SM product derived from SMOS using the Disaggregation based on Physical And Theoretical scale CHange (DISPATCH) methodology has been used. The DISPATCH algorithm uses satellite thermal MODIS LST, short-wavelength MODIS NDVI, and a digital elevation model to estimate soil evapotranspiration efficiency at a high resolution based on a semi-empirical Soil Evaporative Efficiency (SEE) model [93,94,95]. SEE is then used, in combination with the 40 km SMOS SM data, to derive an SM product at 1 km; further details can be found in [93]. The SM 1 km DISPATCH product has been validated at different study sites at different climatic conditions, such as Australia [93], Catalonia, Spain [96], Morocco [97], etc.

b.

NDVI

The vegetation index used in this study is the NDVI from the MOD13Q1 product [98], which is obtained from Terra MODIS data. This product provides NDVI values with a spatial resolution of 250 m. A sophisticated algorithm is used to select the highest-quality NDVI value from all acquisitions made over a 16-day period. These data are atmospherically corrected using a product quality assurance measurement composition technique. This technique effectively eliminates low-quality pixels that may be affected by cloud cover [99].

c.

ETP

For ETP data, the MOD16A2 product [100] was utilized. This product provides information on cumulative ETP that is generated every 8 days at a resolution of 500 m. The algorithm used in the production of the MOD16A2 product is based on the Penman–Monteith equation, [101], incorporating a daily reanalysis of weather data inputs as well as MODIS remote sensing data products such as vegetation property dynamics, albedo, and land cover.

d.

LST

To obtain LST data, the MOD11A2 product [102] was employed. This product provides average LST values with a temporal resolution of 8 days and a spatial resolution of 1 km.

Table 1. Satellite database.

Variable	Data	Time Resolution	Spatial Resolution
SM	DISPATCh SM	1 day	1 km
NDVI	MOD13Q1	16 days	250 m
ETP	MOD16A2	16 days	500 m
LST	MOD11A2	8 days	1 km

below summarizes the used satellite database in this study.

2.3. Methods

Figure 6 below outlines the methodology followed in this study. First, the DISPATCH algorithm was applied to generate a 1 km SM data series from September 2010 to August 2022. Subsequently, four statistical drought indices were calculated for 12 years using DISPATCH SM and MODIS NDVI, ETP, and LST data by applying the anomaly algorithm to take into account SM, vegetation, and temperature conditions.

These drought indices are calculated by applying a statistical anomaly algorithm [103]:

$$X_{(i,j)}=\frac{Z_{(i,j)}-{{(Z}_i)}_{mean}}{{\sigma }_i},$$

(1)

where X_(i,j) is the index (SMAI, VAI, EAI, or ITAI) for month i and year j, Z_(i,j) is the average value for the month i and the year j, and (Z_i)_mean and σ_i are the mean and the standard deviation during the month i derived from the 12 years of the SM, NDVI, ETP, and LST time series, respectively.

• Soil Moisture Anomaly Index (SMAI)

Using DISPATCH SM data with a 1 km spatial resolution, the SMAI [64,67] was calculated for the period from September 2010 to August 2022. This drought index, which has a significant correlation with precipitation patterns, plays a crucial role in the effective characterization of soil water status. The SMAI provides a global analysis of the region’s water conditions, making it a valuable tool in drought monitoring and assessment. By analyzing the SMAI results, valuable insights into the SM status can be extracted, facilitating indirectly the identification of dry periods:

-

If the SMAI < 0, i.e., the average SM of the month i is lower than the historical average SM of the same month, it can be a dry period.

-

If the SMAI > 0, i.e., the average SM of the month i is higher than the historical average SM of the same month, it can be a wet period due to rain or irrigation.

• Vegetation Anomaly Index (VAI)

The VAI [50] is calculated based on statistical parameters derived from NDVI MOD13Q1 over a 12-year period, aligned with the SMAI. This index reflects the vegetative activity of the plant and indirectly its sanitary condition and exposure to drought periods. The results obtained from the VAI provide crucial insights into the health and well-being of the plant:

-

If the VAI < 0, i.e., the average NDVI of the month i is lower than the historical average NDVI of the same month. Such conditions indicate that the plant is facing stress and may be going through a period of prolonged water stress or drought, which is characterized by limited rainfall or irrigation.

-

If the VAI > 0, i.e., the average NDVI of the month i is higher than the historical average NDVI of the same month, it is, therefore, a well-producing agricultural year. These conditions usually correspond to a wet period, suggesting that the plant is in good health.

• Evapotranspiration Anomaly Index (EAI)

The EAI [60] is derived based on statistical parameters derived from ETP MOD16A2 over the same 12-year period as the SMAI. This index provides valuable information on vegetative activity and plant health, with an emphasis on identifying periods of drought:

-

If the EAI < 0, i.e., the average ETP of the month i is lower than the historical average ETP of the same month, the plant is in a stressful situation and may correspond to a period of drought or prolonged water stress (absence of rainfall or irrigation).

-

If the EAI > 0, i.e., the average ETP of the month i is higher than the historical average ETP of the same month, which indicates a productive agricultural year with good vegetative growth. These conditions typically coincide with a wet period, signaling that the plant is in good health.

• Inverse Temperature Anomaly Index (ITAI)

The Temperature Anomaly Index (TAI) [43] provides insight into land surface temperature and confirms periods of stress. The TAI results are going to be the inverse of the SMAI, VAI, and EAI. Positive TAI anomalies represent higher-than-average temperatures. The continuation of this condition may correspond to a period of drought. To facilitate the interpretation of all drought index results, the inverse of the TAI, defined by the ITAI, was applied. The ITAI was calculated based on statistical parameters derived from LST MOD11A2 data over the same 12-year period. According to the ITAI results, we can have an idea of the state of the surface temperature and indirectly identify the dry periods:

-

If the ITAI < 0, the average temperature of the month i is higher than the historical average temperature of the same month. So, it is a period of high temperature and may correspond to a dry period. The high temperatures can also correspond to the absence of vegetation or disease in the plant.

-

If the ITAI > 0, the temperature average of the month i is lower than the temperature historical average of the same month; it is, therefore, a period of low temperature that can correspond to a wet period.

The averages of each drought index corresponding to cereal fields for each month were extracted to focus only on information relevant to the cereal sites. Drought periods were identified based on the severity and duration of SM deficits, vegetative stress, and prolonged high temperatures, with each condition lasting for a minimum of three consecutive months. For the Kairouan study area, a comparison between the performance of drought indices in relation to vegetative status and agricultural practices by comparing the average for all cereal drought indices, the average for predominantly rainfed cereals, and predominantly irrigated cereals was made.

The primary objective of this study is to evaluate the performance of the drought index in predicting cereal yield, specifically considering the total weight of cereals and accounting for the monthly intensity of each drought index. Therefore, for each month, a correlation and standard deviation analysis between the yield of each crop and the yield estimated by linear regression based on the average value of the drought index at the location of each crop over the entire region was made.

In order to select the drought indices most correlated with wheat and barley yield, Pearson’s correlation (R) and standard deviation (std) were calculated. R is among the most used correlation methods to study the relative strength of the linear relationship between two variables [104].

$$R=\frac{{\sum }_{i=1}^n\left(x_{(i,j)}-\overline{x}\right)(y_i-\overline{y})}{\sqrt{{{\sum }_{i=1}^n\left(x_{(i,j)}-\overline{x}\right)}^2}\sqrt{{{\sum }_{i=1}^n(y_i-\overline{y})}^2}},$$

(2)

where x_(i,j) and y_j are the drought index of the month i and the yield corresponding to the fields (wheat or barley) of each region for the variable year j, n is the number of studied years, $\overline{x}$ is the average of the indices x_(i,j) during the studied period (12 years), and $\overline{y}$ is the average of the yield y_i during the studied period.

The std of a regression estimation is a crucial metric that reflects the precision of predictions.

$$std=\sqrt{\frac{1}{n}{\sum }_{i=1}^n{\left(y_i-Y_i\right)}^2},$$

(3)

where y_i and Y_i are the yield (of wheat or barley) and the corresponding estimated yield for each year and n is the number of the studied data (12 years).

3. Results

3.1. Drought Periods

The following two subsections investigate the time delay between drought indices and their impact on cereal yields. Specifically, we examine drought index results for two agricultural years:

• 2014/2015 in Lleida, where record-low barley and wheat production occurred;
• 2021/2022 in Kairouan, where despite the long dry period lasting most of the agricultural year, January and March rainfall significantly improved yields for both irrigated and rainfed crops.

The third section is devoted to the presentation of the variation of the drought indices for wheat and barley during the whole study period.

3.1.1. Case Study of Drought 2014/2015 in Lleida

Figure 7 below shows an example of the temporal evolution of the different drought indices in Lleida for the 2014/2015 agricultural year, where low barley (around 30% lost to average yield) and wheat (around 25% lost to average) yields were recorded.

This season started with a heavy rainfall period from September to December 2014, with a total precipitation of 259.5 mm at the meteorological station of Lleida over these four months. June recorded above-normal rainfall of 40.9 mm and 35.7 mm in the meteorological stations of Lleida and Tarrega, respectively. Precipitation in July and August 2015 was 29.7 and 19.1 mm and 35.4 and 29.2 mm, respectively, at these two stations. From January to July 2015, temperatures were above normal reaching records of 34.2 °C and 43 °C in May and July 2015, respectively [77].

Table A1. Monthly variation of precipitation (in mm) and maximum, minimum, and average temperatures (in °C) for the 2014/2015 agricultural year at the Segrià and Urgell stations in the Lleida study area.

	Parameter	09/2014	10/2014	11/2014	12/2014	01/2015	02/2015	03/2015	04/2015	05/2015	06/2015	07/2015	08/2015
Segrià	P (mm)	118.6	21.4	108.4	11.1	16	19.6	29.6	8.6	2	40.9	29.7	19.1
	T max (°C)	27.4	23.9	16.3	10	8.9	11.4	17.4	20.9	25.9	30	33.3	30.5
	T min (°C)	15.4	10.7	6.7	0.9	−0.7	−0.3	4.9	7	10.5	15	19.1	17.1
	T mean (°C)	21.4	17.3	11.5	5.5	4.1	5.6	11.2	14	18.2	22.5	26.2	23.8
Urgell	P (mm)	29.9	32	88.5	13.1	11.4	13	20.6	23.1	3.8	35.7	35.4	29.2
	T max (°C)	28.2	24.4	16.2	8.7	9.1	11.3	17.4	21.5	27.5	30.9	35.2	31.5
	T min (°C)	16.4	12.8	7.9	1.7	0.4	0.2	5.2	7.2	11.7	16	20.1	17.7
	T mean (°C)	22.3	18.6	12.1	5.2	4.8	5.8	11.3	14.4	19.6	23.5	27.7	24.6

in Appendix A summarizes these climatic indices for the two stations (Segrià and Urgell).

Consistently, Figure 7 shows that more than 8 months have above-average temperatures, according to the ITAI. The SMAI is positive at the beginning of the period, from September to December. With respect to the VAI and EAI, similar to the SMAI, the period starts with positive values, indicating higher values than the average of the NDVI and ETP. The SMAI starts to decrease with the above-normal temperatures starting in January, showing more than five consecutive months of SM negative anomaly with extremely dry conditions in April and May. However, the effect of the initial wet period on vegetation continued in January and starts decreasing slightly in February but not so clearly or sharply as the SMAI, as detected by the VAI. The VAI clearly detects the stress period later than the SMAI, which is only slight in February and worse in May (whereas the SMAI already had clear stress conditions in April). In contrast, the EAI showed more consistency with the SMAI from January 2015 to May 2015, which showed more severe stress conditions in May. The rainfalls in June and August are well-captured by the SMAI. Similarly, these wet periods are detected by both the VAI and EAI, particularly in August. With respect to the EAI, in the winter months (January to March), the signal is similar to the SMAI, whereas in the summer period, it seems to mirror the VAI.

3.1.2. Case Study of Drought 2021/2022 in Kairouan

Figure 8 below shows the variation of drought indices of the year 2021/2022 in the Kairouan governorate, where despite a series of dry months, the precipitation in January and March led to an average cereal yield at the end of the season.

This season started with high temperatures, reaching 41.5 °C and 36.9 °C in September and November 2021, and then temperatures of around 25 °C were recorded in January, March, and April before rising to 41.9 °C, 46.6 °C, 48 °C, and 47.6 °C from May to August 2022 at the Kairouan meteorological station. Precipitation was negligible during the entire period; only 29 mm and 128.8 mm were recorded in January and March 2022, respectively, at the Kairouan station [105].

Table A2. Monthly variation of precipitation (in mm) and maximum, minimum, and average temperatures (in °C) for the 2021/2022 agricultural year at the Kairouan station in the Kairouan study area.

Parameter	09/2021	10/2021	11/2021	12/2021	01/2022	02/2022	03/2022	04/2022	05/2022	06/2022	07/2022	08/2022
P (mm)	-	8	1.2	5.4	29	9	128.8	9.4	12.6	0.6	5	5.4
T max (°C)	41.5	36.9	33.1	25	25.9	28.5	26.5	31.8	41.9	46.6	48	47.6
T min (°C)	20.4	12.8	8.1	4.7	2.9	4.8	6.4	9.7	13.8	20.4	22	23.1
T mean (°C)	29.8	22.7	18.6	14.8	12.6	14.4	15.5	19	25	31.3	32.9	32.3

in Appendix A summarizes these climatic indices for the Kairouan station.

In Figure 8, the ITAI clearly shows the above-normal temperatures for the season and also clearly captures the lower temperatures, especially in March 2022. The lower-than-average precipitation is also well captured by the SMAI, which shows negative values for the entire period, except in March and partially in January. Following the rainfall in March 2022, weeds were developed throughout the governorate, and rainfed cereals reached the end of the cycle, resulting in an increase in above-average NDVI values from April and cereal production. The EAI’s stress situation was aggravated in June by a sharp increase in temperature, combined with a lack of SM. In contrast, the VAI showed little evidence of a dry period. These above-average NDVI values could be attributed to factors such as arboriculture (mainly olive trees), weed growth following the March rains and some rainfall in previous months, and, in particular, summer vegetable crops in the Kairouan plain. In addition, Figure 8 shows that with a finer spatial resolution (between 250 m and 500 m), the irrigated areas of the Kairouan plain are well-detected by the VAI during the months of November to March or by the ITAI in July.

3.1.3. Cereal Drought Indices

Figure 9 below depicts the monthly average of the drought indices over the wheat and barley fields in Lleida together with the precipitation anomaly at Segrià station from September 2010 to August 2022.

Figure 9 shows a consistency between the different indices in the detection of the wet and dry periods. As an example, the 2019/2020 agricultural year is characterized by wet conditions extended for more than eight consecutive months. In temporal terms, the ITAI seems to precede SMAI variations, and the VAI and EAI show synchronous variations with a minimum delay of one month with respect to the SMAI, as can be seen at the end of 2013 or in the early months of 2015.

Figure 9 demonstrates the coherence between drought indices and precipitation anomalies, revealing the presence of memory effects in the SMAI, VAI, and EAI at multiple dates. Notably, after a succession of two rainy months, the subsequent month exhibits a continuation of wet conditions, even in the absence of precipitation. This pattern was distinctly observed during the 2019/2020 agricultural year when several consecutive rainy months resulted in consistently wet conditions throughout the year.

Similarly, in June, July, and August 2015, heavy precipitation in June 2015 influenced wet conditions in the following months despite the absence of further rainfall. Subsequently, the continued lack of precipitation during the subsequent months led to soil water stress.

Figure 10 below shows the variation of cereal drought indices (SMAI, VAI, EAI, and ITAI) in Kairouan. To study the performance of drought indices in relation to vegetative status and agricultural practices, a comparison between the results of the average of all cereal drought indices was made; the average for predominantly rainfed cereals and predominantly irrigated cereals was carried out separately. Figure 10 shows the indices variation of the entire area in the first row, the variation over the rainfed area in the second row, and the variation for the irrigated area in the third row.

The average evolution of the drought indices over the entire area shows, as observed in the Lleida area, a consistency between the different indices in the detection of the wet and dry periods. As an example, the 2012/2013 agricultural year is characterized by dry conditions extended for more than fifteen consecutive months. In temporal terms, the SMAI and ITAI seem to precede the variations of the VAI and EAI, which show the most delayed variations, as can be observed at the end of 2011 and 2013.

Starting in 2015 but more clearly from 2017 onwards when the majority of cereals became irrigated, there is a clear difference between the indices in the dryland and the irrigated area. In the rainfed area, Figure 10 clearly illustrates that the VAI_R shows a similar pattern as the SMAI_R, reflecting their interdependence. In contrast, the VAI_I is shown to remain positive even during the drought periods detected by the SMAI_I (e.g., the severe drought of 2021). It is clear that due to irrigation practices since 2017, the effect of meteorological drought does not impact vegetation, and the VAI remains positive in the irrigated areas. It would be expected that that SMAI is also positive during this period in the irrigated areas; however, it shows more negative values than in the rainfed area. Our hypothesis is that due to the coarser resolution of SM (compared to the NDVI) and the relatively small size of the irrigated fields, SM is not able to capture the irrigation, showing consequence negative values, such as in the dryland area. In absolute terms, the SMAI is shown to be more negative in the irrigated area compared to the dryland area. This is explained because of the lower precipitation and higher temperatures in the southern region.

3.2. Correlation Analysis between Drought Indices and Cereal Yield

3.2.1. Lleida’s Correlation Results

Figure 11 below shows, as an example, the (a) March and (b) June linear regression results (correlation and std) between the drought indices and barley yield in Lleida.

In Figure 11a, a medium to high correlation was found between barley yield and three drought indices: the VAI (0.72), EAI (0.67), and SMAI (0.64). On the other hand, the ITAI shows the lowest correlation, which means that March temperature is not a determining factor for cereal yield in Lleida. In June, as shown in Figure 11b, the highest correlation was found, with the EAI, R equal to 0.91, followed by the VAI (0.83) and ITAI (0.64). The EAI values are highly representative, indicating a positive ETP anomaly during June, correlating with an increase in barley yield. Conversely, the SMAI correlation value has decreased after May, meaning the importance of SM has decreased once cereal maturity is reached.

This analysis is performed for all months for barley and wheat separately.

Table 2. Lleida correlation and standard deviation results (in 1000 tons) between yield and drought indices for each index for each month. The highest correlations and lowest standard deviations between each drought index and wheat/barley yield are marked in bold.

Cereal	Month	SMAI		VAI		EAI		ITAI
		R	std	R	std	R	std	R	std
Barley	November	−0.18	32.37	0.22	56.57	−0.12	35.76	−0.41	43.76
	December	−0.05	33.43	0.05	58.00	0.10	63.35	−0.08	47.58
	January	0.50	29.12	0.36	51.88	0.41	39.11	0.07	82.40
	February	0.58	28.65	0.72	31.59	0.68	23.69	0.14	31.59
	March	0.64	26.69	0.72	34.70	0.67	29.65	−0.17	28.12
	April	0.51	28.88	0.60	51.12	0.53	43.23	−0.07	44.63
	May	0.68	25.83	0.73	31.43	0.84	28.87	0.67	27.94
	June	0.44	32.08	0.83	30.08	0.91	18.18	0.64	28.67
	July	0.10	37.52	0.80	30.78	0.74	32.39	0.20	41.92
	August	0.16	36.26	0.78	33.11	0.55	30.68	0.31	39.24
Wheat	November	−0.15	16.04	−0.05	31.46	0.01	17.36	−0.19	22.31
	December	−0.03	16.25	0.01	28.92	0.01	32.53	−0.51	20.36
	January	0.53	13.67	0.26	26.07	0.41	19.98	−0.03	38.49
	February	0.49	14.35	0.48	18.18	0.59	12.75	0.34	14.12
	March	0.60	13.36	0.55	21.22	0.55	16.45	−0.06	14.18
	April	0.41	14.78	0.45	27.00	0.20	23.86	−0.27	18.08
	May	0.52	15.25	0.45	21.57	0.59	21.38	0.44	16.50
	June	0.63	13.92	0.60	24.13	0.71	17.10	0.48	15.69
	July	0.28	17.48	0.72	19.83	0.75	16.78	0.41	17.65
	August	0.19	18.55	0.64	22.38	0.39	18.41	0.20	19.82

shows a summary of both the correlation and the standard deviation at each month for all the indices.

Figure 12 below illustrates the correlation variation of the different indices over the different months for (a) barley and (b) wheat.

Regarding barley, in Figure 12a, the EAI and VAI displayed a continuous gradual increase in the correlations from February until the end of the growth cycle, where the highest correlations are in June (R equal to 0.91 and 0.83, respectively). In this month, the standard deviation (see Table 2) is the lowest std (18,180 tons) for the EAI. Similarly, the SMAI exhibited a gradual increase in correlation with yield starting from January (R = 0.47) and reaching its maximum in May (R = 0.68). In May, the SMAI also showed its lowest value of std (25,830 tons), two months before harvest. In contrast, the ITAI showed insignificant correlations at the beginning of the cycle but recorded R equal to 0.67 and 0.64 in May and June, respectively.

Similar correlation results were obtained for wheat, in Figure 12b, but with lower correlation values compared to barley. The strongest correlation was found for all the indices at the end of the growth cycle in July; the highest value was for the EAI (R = 0.75) followed by the VAI (R = 0.72), and in June, the SMAI (R = 0.63) and the ITAI (R = 0.48). For the months before harvest, the highest correlation and lowest standard deviation are found in March for the SMAI (R = 0.60 and std = 13,360 tons) and in February for the EAI (R = 0.59 and std = 12,750 tons). The VAI index shows similar correlations to the SMAI index in March (R = 0.55) but has a higher standard deviation than the SMAI and EAI.

3.2.2. Kairouan’s Correlation Results

The correlation results for Kairouan are detailed in

Table 3. Kairouan correlation and standard deviation (in 1000 tons) results between yield and drought indices for each index for each month. The highest correlations and lowest standard deviations between each drought index and cereal yield are marked in bold.

Month	SMAI		VAI		EAI		ITAI
	R	std	R	std	R	std	R	std
November	0.63	14.03	−0.19	30.41	−0.11	37.01	0.65	18.99
December	0.44	15.88	0.39	28.24	−0.06	39.69	0.72	26.47
January	0.64	15.24	0.35	29.64	0.15	64.13	0.55	35.14
February	0.67	13.88	0.54	27.57	0.32	50.14	0.15	39.18
March	0.51	16.23	0.79	20.32	0.66	64.19	−0.07	19.29
April	0.11	18.50	0.71	19.10	0.90	13.43	0.11	26.18
May	0.18	18.54	0.37	23.54	0.53	18.26	0.30	17.65
June	0.42	16.54	0.40	28.05	0.14	18.03	−0.08	17.69

and illustrated in Figure 13.

Figure 13 shows a very peculiar behavior of the correlation between yield and the different indices. During the agricultural season, each index follows an increased correlation followed by a decreased correlation. The moment of maximum correlation is different for each index. The first maximum is observed for the ITAI at the beginning of the cereal cycle, in December (R of 0.72). The SMAI shows the highest correlation (R of 0.67) and lowest std (13,880 tons) in February. The VAI has two similar values in March and April (R equal to 0.79 and 0.71, respectively) and a low std (std equal to 20,320 and 19,100 tons in March and April, respectively). The last maximum in the season is found with the EAI in April (R of 0.9) and its lowest std (13,430 tons). The SMAI also exhibits a maximum in November (R = 0.63).

4. Discussion

The SMAI, VAI, EAI, and ITAI drought indices have demonstrated an efficient performance in identifying drought periods and predicting cereal yields (wheat and barley) in two Mediterranean regions. Rainfed cereals in the Lleida region show a strong sensitivity to rainfall, as evidenced by the remarkable yields recorded during the two wettest years of our study period (2012/2013 and 2019/2020). However, in Kairouan, the sensitivity to rainfall was primarily observed during the early stages of the period when cereals relied predominantly on rainfall to get watered. Over time, with the expansion of irrigated perimeters in Kairouan, cereal yields have become less dependent on rainfall. This trend is evident from the data in 2021/2022, where despite below-average annual rainfall, cereal yields remained at an average level, highlighting the reduced impact of precipitation on yields. Notably, the timing of rainfall also plays a pivotal role in determining cereal yields. Rainfall during critical growth periods significantly improves cereal yields, which is evident in the success of the rainfed crops during the wettest years in Lleida. On the contrary, drought periods, especially if they occur during the crucial stages of the growth cycle, negatively influence yields in both regions. In the following sections, we will discuss, in detail, the difference found between these two study areas in terms of 1) the detection of drought periods and the time delay of stress detection between the ITAI, SMAI, EAI, and VAI and 2) the correlation between the aforementioned drought indices and cereal yields.

In temporal terms, it was observed both in Lleida and Kairouan that the ITAI and SMAI variations seem to precede the VAI and EAI, which show synchronous variations with a minimum delay of one month with respect to the SMAI. The synchronicity between the ITAI and SMAI can be explained by the fact that wet soils are colder than dry soils, and thus a positive SMAI is correlated timewise with a positive ITAI. A delay of one to three months is noted between SM stress and vegetative stress (see Figure 9 and Figure 10), as demonstrated in [60]. This finding also aligns with the distinction between vegetative drought and soil moisture drought, as mentioned in [2].

Since the adoption of irrigation practices in the Kairouan plain, the SMAI seems to behave in the opposite way as the VAI, EAI, and ITAI, particularly during the extremely dry period from 2020 to 2022. During this period, the VAI and EAI remain high, indicating that cereals are well-watered. However, the SMAI is not able to capture these wet conditions. This might be related to the low spatial resolution of the SM product (1 km) compared to the average field size (0.5 ha). Indeed, the average size of an irrigated field is 400 smaller than a soil moisture pixel. In these conditions, a substantial percentage of the fields must be irrigated in order for the average SM measured by the satellite to be dominated by irrigation plots. As can be seen in Figure 1, irrigated cereal plots in the Kairouan region are scattered and surrounded by non-cultivated plots. Our hypothesis is that even if cultivated plots are irrigated (and well-watered), the overall SM at 1 km is dominated by the non-cultivated (non-irrigated) plots, and thus SM is not able to capture irrigation.

Several studies have shown that the timing of the dry period during the growing cycle of cereal has an impact on yields [6,7,106]. Sowing conditions, such as rainfall, cumulative germination temperatures, and sowing date [107,108], along with the timing of the drought period [73], play a significant role in determining cereal yields. A clear relationship between drought periods, particularly when marked by negative SMAI values for consecutive periods of at least three months (as depicted in Figure 9 and Figure 10), and cereal yields, with a distinct focus on the vegetative cycle of barley and wheat is found. This influence is most conspicuously observed when the stress period is recorded at the beginning of the cycle or in the middle, such as in the years 2011/2012, 2013/2014, 2014/2015, 2016/2017, and 2018/2019 in Lleida, where a substantial reduction in yield is notably registered. Nevertheless, since 2017, the impact of drought periods on cereal production in Kairouan has notably diminished, as illustrated in Figure 4b, which is attributed to consistent irrigation throughout the vegetative growth cycle. In our study, we aimed to identify the most effective drought index and month for predicting cereal yields. Due to the different climate patterns and agricultural practices in the two study regions, we observed a distinct correlation between cereal yield and drought indices.

In Lleida, the cereals are mostly cultivated in rainfed conditions. The sowing of cereals starts in November, which explains the low correlation between the VAI and EAI and cereal yield (barley and wheat) during November and December. The correlation between the SMAI and cereal yields begins to increase from January to June, with a maximum correlation in May and June (R equal to 0.68 and 0.63 with barley and wheat yields, respectively). Additionally, the correlation between cereal yield and the VAI and EAI showed a gradual increase from February, with the maximum recorded in June. These correlation results are similar to the results corresponding to the province of Lleida reported in [109], where the correlation between wheat yield and precipitation drought index starts to increase from January to May.

In the cereal region of Lleida, the correlation of the SMAI, EAI, and VAI with yield (see Figure 12) follows a similar pattern during the vegetative cycle, where a positive anomaly of any of those indices leads to an increase in yield. The temperature has only an impact in May a June where a lower-than-average temperature leads to increased yield, indicating that average temperatures are sufficient for development and lower temperatures allow for less ET. In this sense, the authors of [26] also showed the influence of high temperatures in spring (April and May) on barley and wheat yield in some regions in Spain. At the end of the cereal cycle, SM decreases its impact from May onwards, whereas the NDVI and ET are still relevant.

In contrast, in the irrigated area of Kairouan, the period of the highest correlation of the different drought indices with yield is distinct and strongly related to the phenology cycle (see Figure 13). Initially, there is a strong correlation between the SMAI and cereal yields in November (R = 0.63), as the presence of precipitation in November encourages farmers to sow cereals. In February, a strong correlation between cereal yield and the SMAI is noted (R = 0.67). In March and April, which correspond to the vegetative peak of cereals, [50,90,108], the VAI and EAI reach their maximum correlation with cereal yield. As the NDVI decreases in May, the correlation between yield and the VAI decreases. The ITAI does not seem to play an important role during the vegetative cycle. Only a positive correlation between yield and the ITAI is found at the beginning of the cereal cycle, R = 0.72 in December, indicating that the decrease in temperature during this period is a consequence of the positive SMAI. The findings in the study area of Kairouan are coherent with the results found by the authors of [44], who also found a correlation between an SM-based index and yield in December and an NDVI-based index in Morocco. Moreover, a strong correlation between winter wheat yield and drought indices based on SM was also found in October, November, and February in studies [75,76]. In this study, we have been able to go further and analyze the impact of different agriculture indices on cereal yield both in rainfed and irrigated areas.

In addition, the results of the correlation between yield and drought indices are consistent with the interpretation of the timing of drought occurrence in the cereal growing year. This study confirms that when wet periods were recorded, particularly in November, February, March, and April, cereal yields increased. The SMAI, VAI, EAI, and ITAI drought indices have shown that in addition to their good performance in identifying drought events, they are also suitable for predicting cereal yields.

5. Conclusions

Drought indices, including the SMAI, VAI, EAI, and ITAI, were calculated to study the SM, vegetation, and temperature conditions over the last 12 years in the rainfed cereal area of Lleida and the irrigated and rainfed cereal zone of Kairouan. This study analyzed the correlation and standard deviation between cereal yield and agriculture drought indices to identify the most informative month and indices for cereal yield prediction. In the dry cereal area of Lleida, the strongest correlation was found between the EAI and VAI with barley yield (0.91 and 0.83 respectively) at the time of cereal maturity in June. For wheat, the strongest correlation was found between the EAI and VAI (0.75 and 0.72 respectively) at the time of cereal maturity in July. However, the VAI, EAI, and SMAI showed the best performance as earlier indicators in March, with correlations with barley yield of 0.72, 0.67, and 0.64, respectively; the lowest standard deviation was for the SMAI. For wheat yield, the best earlier indicator was the SMAI in March, showing the highest correlation (0.60) and the lowest standard deviation.

For the irrigated cereal zone of Kairouan, the strongest correlation (0.9) is found between the EAI and cereal yield at the time of dough development in April, which corresponds to the minimum standard deviation. In terms of advanced prediction, the VAI shows a high correlation in March (0.79), while the SMAI shows a slightly lower correlation in February (0.67) and a lower standard deviation. The SMAI showed its limitation to detect irrigation in the Kairouan area, where irrigated fields have a much smaller size than the SM spatial resolution. Despite this limitation, the SMAI was a good earlier predictor of cereal yield in that area. These results provide valuable information for decision-making on food security through early yield prediction and the implementation of more efficient water management strategies.