Satellite-Based Approach for Crop Type Mapping and Assessment of Irrigation Performance in the Nile Delta

Abstract

Water scarcity, exacerbated by climate change, population growth, and competing sectoral demands, poses a major threat to agricultural sustainability, particularly in irrigated regions such as the Nile Delta in Egypt. Addressing this challenge requires innovative approaches to evaluate irrigation performance despite the limitations in ground data availability. Traditional assessment methods are often costly, labor-intensive, and reliant on field data, limiting their scalability, especially in data-scarce regions. This paper addresses this gap by presenting a comprehensive and scalable framework that employs publicly accessible satellite data to map crop types and subsequently assess irrigation performance without the need for ground truthing. The framework consists of two parts: First, crop mapping, which was conducted seasonally between 2015 and 2020 for the four primary crops in the Nile Delta (rice, maize, wheat, and clover). The WaPOR v2 Land Cover Classification layer was used as a substitute for ground truth data to label the Landsat-8 images for training the random forest algorithm. The crop maps generated at 30 m resolution had moderate to high accuracy, with overall accuracy ranging from 0.77 to 0.80 in summer and 0.87–0.95 in winter. The estimated crop areas aligned well with national agricultural statistics. Second, based on the mapped crops, three irrigation performance indicators—adequacy, reliability, and equity—were calculated and compared with their established standards. The results reveal a good level of equity, with values consistently below 10%, and a relatively reliable water supply, as indicated by the reliability indicator (0.02–0.08). Average summer adequacy ranged from 0.4 to 0.63, indicating insufficient supply, whereas winter values (1.3 to 1.7) reflected a surplus. A noticeable improvement gradient was observed for all indicators toward the north of the delta, while areas located in the delta’s new lands consistently displayed unfavorable conditions in all indicators. This approach facilitates the identification of regions where agricultural performance falls short of its potential, thereby offering valuable insights into where and how irrigation systems can be strategically improved to enhance overall performance sustainably.

1. Introduction

Irrigated agriculture is the primary water consumer in many regions of the world, accounting for approximately 70% of the world’s freshwater withdrawal [1]. Coupled with various constraints and pressures caused by population growth, competing needs across diverse sectors, and changing climate patterns, among other factors, the rate at which water resources are consumed globally outpaces their natural regeneration [2,3,4]. This situation is particularly pronounced in arid regions, where the most severe water crises occur.

Egypt, like many arid countries, faces significant challenges due to its naturally limited water resources. The nation relies heavily on Nile waters, which account for approximately 95% of its water supply through a fixed quota of 55.5 billion cubic meters per year (BCM/Y) [5]. This fixed quota is supplemented by around 1.6 BCM/Y of effective rainfall and total groundwater abstraction of 8.9 BCM/Y, with a water supply of approximately 66 BCM/Y and a demand that reached 79.5 BCM/Y in 2017 [6,7]. This imbalance directly threatens agricultural sustainability, as 62 BCM of the water resources are allocated for the surface irrigation practices [8]. Without efficiency improvements, Egypt’s agricultural water demands are expected to reach 110 BCM by 2050 in order to maintain current agricultural production levels [9]. The agricultural sector’s vulnerability is further exacerbated by concerns regarding the assurance of Egypt’s historical share of Nile water following the upstream infrastructure development [10].

At the center of Egypt’s agricultural challenges are four strategically vital crops: rice, maize, wheat, and clover. These crops collectively dominate cultivated areas, consuming approximately 50% of irrigated water [11,12] and facing a greater risk from anticipated shortages.

Consequently, the agricultural sector will require significant adjustments to improve future crop production and irrigation water management, particularly for these strategic crops [13]. To achieve effective management, it is imperative that the irrigation performance status be assessed as a preliminary step. This assessment involves defining and quantifying key indicators to establish benchmarks that could provide insights into the overall system performance in order to identify opportunities for improvement [14].

There exists a wide range of performance indicators that can be calculated based on remote sensing data [15]. Most irrigation performance assessment studies focus on indicators such as adequacy, equity, and reliability as determinants of the overall health of the irrigation scheme condition [16,17,18]. In addition, those indicators facilitate a comparison of the scheme’s spatial and temporal performance by revealing patterns of sufficiency and consistency, which can guide further diagnosis and operational management [19].

A major limitation often encountered during the evaluation of such irrigation indicators is the need for field-based observations, including meteorological data (such as rainfall and evapotranspiration), as well as crop-specific information (e.g., crop distribution, crop coefficients, and crop water demand). Such ground information is rarely collected; even when collected, it tends to be inaccessible or unreliable. Moreover, field point data does not capture spatial and temporal variability [20].

Because irrigation needs vary by crop type, accurate crop distribution data are essential for estimating irrigation demand and enabling precise, reliable performance [21]. However, crop type maps are not available and are highly needed for Egypt [9,13]. Generating such a crop pattern map is a complicated process that requires significant ground reference data. Moreover, it requires the processing and management of large amounts of geospatial data; therefore, specialized expertise and significant time investment are necessary.

Several recent studies have explored crop type mapping without direct local ground truth data; however, they typically employ hybrid approaches that combine field survey data with automated sampling strategies or rely on transferring classifiers trained in data-rich regions to data-scarce ones [22,23]. Others have adapted historical national-scale products, such as the Cropland Data Layer, as a substitute for ground truth [24], but such resources are not available for our case study area.

With the continuous development of satellite data, their application offers an efficient option with considerable spatial and temporal coverage to calculate the performance of irrigation schemes, particularly in regions with large irrigation schemes and limited data availability [25]. Assessing the performance of irrigation systems based on satellite remote sensing approaches has been carried out by various scholars worldwide, with reported successful outcomes by capturing the spatiotemporal performance variation within schemes, for instance in Brazil [26], Burkina Faso [27], Swaziland [28], Mozambique [25], Uganda [29], the USA [30], among others.

Our research here introduces a novel framework that combines crop type mapping and the subsequent assessment of their time series of irrigation performance indicators (adequacy, reliability, and equity) using open-access satellite data across the Nile Delta. The insights gained from this methodology can identify potential areas where opportunities exist for water savings or improving irrigated water management. This study is among the first to demonstrate the application of the FAO WaPOR Land Cover Classification layer as a substitute for ground truth data in crop classification. The proposed approach fills the gap by offering a replicable and scalable solution for data-scarce irrigated systems—covered by the WaPOR database or similar datasets—enabling practical, evidence-based irrigation management, even under data and resource constraints.

2. Materials and Methods

2.1. Study Area

The study region is the Nile Delta, with an area of about 22,000 km², hosting nearly 63% of Egypt’s agricultural land [31]. Around 93% of the Nile Delta’s total area is in the old lands, with surface irrigation from the Nile River being the main source of irrigation distributed through a network of main canals and secondary channels [32,33]. Meanwhile, new lands on the western and eastern fringes of the delta use advanced irrigation methods such as sprinkler, pivot, and drip irrigation [34]. Rainfall in the Nile Delta is limited, ranging yearly from 250 mm on the northwestern coast to 50 mm in the southern part [35]. The soil in the Nile Delta is classified as fertile, varying between sand and clay, with higher concentrations of clay in the central part gradually decreasing towards the outer edges of the delta bordering the desert [36]. The crop patterns in the region are divided into two main seasons: the summer crops, ranging from May to October, featuring rice and maize; and the winter crops, from November to April, including wheat and clover [37].

As shown in Figure 1, this study considered the eleven governorates in the delta where agriculture is predominant: Al Qalyubiyah, Al Minufiyah, Al Buhayrah, Al Iskandariyah, Kafr ash Shaykh, Al Gharbiyah, Ad Daqahliyah, Ash Sharqiyah, Dumyat, Bur Said, and Al Ismailiyah.

2.2. Input Data

Several types of satellite data were collected to support both crop type mapping and the evaluation of irrigation performance across the study area.

Table 1. Summary of datasets used for crop mapping and irrigation performance assessment.

Data	Spatial Resolution	Temporal Resolution
Crop Type Classification Data
Landsat-8 Level 2, Collection 2, Tier 1	30 m	16 days
WaPOR Land Cover Classification (Zankalon, Egypt)	30 m	Dekadal
Irrigation Performance Assessment Data
WaPOR Actual Evapotranspiration and Interception	100 m	Monthly
SSEBop Actual Evapotranspiration and Interception	≈1 km	Monthly
WaPOR Reference Evapotranspiration (ET0)	≈20 km	Monthly

Dekadal = every 10 days.

summarizes the main satellite datasets used in this study, stating their spatial and temporal coverage.

In addition to the satellite data, national agricultural statistical data from the Central Agency for Public Mobilization and Statistics (CAPMAS) [38] were also used to statistically validate the crop classification results.

Estimation of irrigation performance indicators was performed using open-access evapotranspiration (ET) products. These products were selected based on their spatial and temporal availability, as well as prior validation in the study area. For instance, the SSEBop’s actual evapotranspiration (AET) product has been validated in Egypt, showing excellent accuracy with only a slight underestimation of 0.68% when compared with lysimeter measurements in the Nile Delta [39].

WaPOR’s actual evapotranspiration has also been validated in a study conducted at the African continental level by comparing the WaPOR product with estimates from eddy covariance towers at three locations with irrigated sites in Egypt [40]. Recent research findings have shown that the WaPOR product tends to overestimate AET, particularly in irrigated fields [40,41]. This finding was further supported by [42], confirming that although the WaPOR and SSEBop products can capture the spatial distribution of evapotranspiration patterns when compared with different reference conditions, spatial analysis, and cluster analysis, WaPOR typically overestimates ET, whereas SSEBop tends to underestimate it.

Therefore, an ensemble approach was adopted by combining these two products to develop an ensemble final AET product. Generating such an ensemble was selected to balance known biases in both products and produce more balanced and realistic evapotranspiration estimates for further analysis [43,44].

It is worth noting that Google Earth Engine (GEE) was utilized for data processing and analysis. GEE offers a cloud-based platform providing access to extensive satellite data, advanced classifiers, and computational services, thereby demonstrating significant efficacy for large-scale cropland monitoring. The platform has been widely used in global studies for crop classification and land cover mapping, demonstrating its suitability for this type of application [24,45,46,47,48,49]. Using the GEE platform would also solve the inefficiency problem of handling large amounts of data that need to be analyzed.

2.3. Data Preparation and Framework for Crop Classification

The first part of Figure 2 illustrates the core framework for crop detection and mapping, which includes: satellite data preprocessing, reference data sampling, crop type classification using a proper classifier, and mapping/validation.

2.3.1. Satellite Data Pre-Processing

This study used the atmospherically corrected surface reflectance Landsat-8 data (“LAND-SAT/LC08/C02/T1/L2”) to accurately capture ground features. Landsat-8 was chosen due to its operational coverage throughout the study period, its 30 m spatial resolution—matching that of the WaPOR Land Cover Classification layer—and its proven success in previous crop classification studies [50,51]. The images from the Landsat-8 dataset were acquired in accordance with the cropping cycles within the study area, spanning from the beginning of May to the end of October to represent the summer season and from November to April for the winter season, ensuring that satellite observations captured the full phenological cycle. The acquisition covered a five-year period commencing in the summer of 2015.

Six spectral bands were selected based on their documented effectiveness for crop differentiation in remote sensing applications [24,52]: three in the visible spectrum (blue, green, and red), one in the near-infrared (NIR) range, and two in the short-wave infrared (SWIR) range. In addition to spectral bands, key vegetation indices (VIs) were calculated from Landsat-8 images, as noted, to improve classification accuracy [23,53]. The Normalized Difference Vegetation Index (NDVI) [54] was included due to its sensitivity to vegetation and phenological changes, which is widely used in crop monitoring [55]. The Green Chlorophyll Vegetation Index (GCVI) [56], which correlates with chlorophyll content, aids in differentiating crop health and types. Some water-related indices, such as the Normalized Difference Water Index (NDWI) [57] and the Land Surface Water Index (LSWI) [58], were incorporated, given their proven effectiveness in identifying water-intensive crops like rice, which is usually cultivated in a water-rich environment [59].

Monthly median composites were generated for the Landsat images with the selected bands and indices to balance the trade-off between capturing crop phenology and ensuring sufficient cloud-free observations, following the approach of [60]. The composites were then stacked into a time series Landsat image stack to capture temporal growth patterns for crop classification.

Table 2. Selected spectral bands and indices of the Landsat image for model training.

Name	Band
Blue	SR_B2
Green	SR_B3
Red	SR_B4
NIR	SR_B5
SWIR-1	SR_B6
SWIR-2	SR_B7
NDVI	(NIR-RED)/(NIR + RED)
GCVI	(NIR/GREEN) − 1
NDWI	(GREEN-NIR)/(GREEN + NIR)
LSWI	(NIR-SWIR1)/(NIR + SWIR1)

lists the selected spectral features for training the model.

2.3.2. Reference Data Sampling

For the acquisition of field data, the WaPOR v2 land cover (LC) dataset was used on a sub-national scale as reference training data to train the classifier and evaluate the classification accuracy. The WaPOR LC dataset is a georeferenced land cover map covering the Zankalon region of the Nile Delta (Figure 1), which represents about 5% of the old lands under the predominant surface irrigation practices. Despite its limited spatial coverage, Zankalon is considered representative of the delta’s traditional cropping systems [13]. The WaPOR LC dataset was generated using a supervised classification approach based on a substantial set of reference data provided by the Food and Agriculture Organization (FAO) through fieldwork conducted within each level 3 geographical area [61].

This technique—using a land cover map created through supervised machine learning trained on ground samples—and its use to guide further land cover classification has been implemented in various previous studies, and its suitability has been demonstrated [23,62,63].

Training pixels were selected for four main crops in the Nile Delta: two summer crops (rice and maize) and two winter crops (wheat and clover). Furthermore, an additional class representing non-target land cover types was also included to capture irrelevant categories.

Although WaPOR LC provides an alternative to in situ reference data, its use may introduce some classification bias, with misclassified pixels. Therefore, to ensure the consistency of reference data, a quality control procedure was applied. For each sample point from WaPOR LC, a cross-check was performed against the mean NDVI signature calculated for its crop class across the growing season. Outliers were removed using a threshold-based filter around the class mean NDVI, following an approach conceptually adapted from [64]. Thresholds were determined empirically through iterative testing, balancing the removal of potentially misleading pixels while maintaining sufficient sample sizes. The final thresholds applied were ±0.01 for wheat, clover, and maize and ±0.02 for rice, based on the mean NDVI value for each crop across its respective growing season. This process ensured that reference samples accurately reflected the spectral and temporal characteristics of each crop category from the WaPOR LC dataset.

2.3.3. Model Training

Crop type mapping in this study was performed using a supervised machine learning approach, which is widely applied in remote sensing for land cover classification [23]. Among various classifiers, the random forest (RF) algorithm was selected due to its proven effectiveness in numerous crop classification studies [65,66,67] and its robustness against training data outliers [68]. RF is an ensemble learning method that combines multiple decision trees to improve accuracy and reduce overfitting [69].

The input data consisted of the stacked multi-temporal Landsat image and the reference samples described earlier. A stratified random sampling approach was applied to ensure balanced representation across crop classes. The resulting samples were then randomly split into 70% for training and 30% for validation, following standard practice.

The RF classifier was implemented using the integrated function in GEE, with parameters set to their defaults. Specifically, the ‘numberOfTrees’ parameter was set to 128, which reflects an optimal trade-off between classification accuracy and computational efficiency, supported by prior studies [24], as well as empirical trials and own observations, which involved testing different numbers of trees. Other parameters were kept at their default values, which are generally sufficient for stable and reliable performance in classification tasks.

2.3.4. Mapping and Validation

• Accuracy Assessment

After training the classifier, the accuracy of the resulting crop classification maps was assessed to ensure the reliability of the method. Creating a confusion matrix is commonly used as a tool to describe the classifier’s accuracy [70,71,72]. This matrix provides several key accuracy metrics:

-

Overall Accuracy (OA): The proportion of test samples that were correctly classified out of the total number of samples.

-

User Accuracy (UA): Indicates the probability that a pixel classified into a given class truly belongs to that class and is reflected by commission errors (instances where samples are incorrectly assigned to a class).

-

Producer Accuracy (PA): Indicates how well reference samples of a class are correctly classified, reflecting omission errors (instances where samples belonging to a class are missed) [73,74].

Additionally, the Kappa coefficient [75] was also calculated to measure the level of agreement or reliability of the assessments. A Kappa value of 1 indicates perfect agreement, while 0 indicates no agreement.

• Statistical Data Comparison

In addition to the accuracy assessment based on the confusion matrix, the classification results were further validated against official agricultural statistics obtained from CAPMAS in Egypt. For the study period, the reported crop area statistics for governorates located entirely within the delta were used for comparison. It is important to note that this comparison does not account for the spatial distribution of crops but relies solely on aggregated statistical data. The level of agreement between the classified crop areas and the reported statistics was evaluated using the coefficient of determination (R²), which measures the strength of the linear relationship between the two datasets.

2.4. Data Preparation and Framework for Assessing the Irrigation Performance

The second part of the workflow in Figure 2 displays the methodology to evaluate the irrigation performance indicators for the mapped crops using satellite-based products at the Nile Delta level.

Performance Assessment Indicators

The irrigation performance indicators selected for this study are adequacy, reliability, and equity, which are commonly used to evaluate irrigation scheme performance using remote sensing-based ET products [28]. A summary of these indicators is presented in Table 3.

Adequacy is a key metric for evaluating the sufficiency of water in an irrigation system [17,76], showing an indirect indication of the soil moisture levels in the crop’s root zone [28]. It is a sort of measure of the degree of agreement between actual water use and crop water requirements [15]. It is estimated using the seasonal relative evapotranspiration (RET), which is the ratio of seasonal actual evapotranspiration (AET) to seasonal potential evapotranspiration (PET). AET represents the actual water consumption by crops, whereas PET reflects maximum crop water demand under ideal conditions [77].

When the crop grows in an optimum state, the ratio between AETs and PETs is minimal and tends to approach 1 [78]. It is established that an RET value of 0.65 serves as the threshold for this indicator, with any value below this indicating inadequate irrigation system performance [79].

Besides adequacy, reliability is also a fundamental element in assessing water supply to cropland. Reliability, as a temporal variable, measures the consistency of water delivery [80]. It is quantified using a temporal coefficient of variation (CV) of RETs over five consecutive summer and winter seasons. A lower CV indicates a more reliable water supply, while higher values point to inconsistencies [18,81].

Equity in irrigation management is a critical indicator that aims to ensure an equitable spatial distribution of water resources between irrigation scheme users. According to [82], it can be computed as the spatial CV of AET within each irrigation season. The system with the lowest CV values exhibits the most equitable distribution of water [83]. Thus, CV values of 0–10%, 10–25%, and >25% indicate good, fair, and poor equity, respectively [28,84].

For calculating these indicators, a pre-processing step was first required to harmonize the spatial resolution of the ET datasets. All of these products were resampled to a 30 m resolution to match the spatial resolution of the crop maps generated in the previous step.

Resampling was performed using the nearest-neighbor interpolation method available in GEE. This method was selected because it preserves the original pixel values of the dataset, ensuring that the magnitude and spatial characteristics of the data remain unchanged during the resampling process. This resampling method for the evapotranspiration products was adapted following [25].

Following resampling, the ET datasets were clipped to fit the extent of the crop maps created and subjected to a quality check to ensure geometric accuracy, data integrity, and consistency with the mapped crop boundaries. Scale factors were applied where necessary to restore the original values of each raster. Subsequently, the AETs and PETs were calculated for each product from their respective layers between the start of the season (SOS) and the end of the season (EOS), as determined.

PET = Kc × ET₀

(1)

PETs were calculated based on Equation (1), where the WaPOR seasonal reference evapotranspiration (ET₀) layer was multiplied by each crop’s average seasonal crop coefficient (Kc). These Kc values were derived from locally monthly field measurements reported in [85]. The calculated mean seasonal Kc values were as follows: rice (1.43), maize (0.76), wheat (0.47), and clover (0.385). These values were applied uniformly across the delta in the PET calculations.

Table 3. Irrigation performance indicators proposed for the study area.

Indicator	Definition	Range	Reference
Adequacy	Relative evapotranspiration (AETs/PETs)	Sufficient supply: 0.75 < A ≤ 1, Inadequate supply: A ≤ 0.65	[79]
Reliability	Temporal variation of AETs/PETs	CV ≈ 0 indicating the highest reliability	[70]
Equity	CV of actual evapotranspiration (AETs)	Good equity: 0% ≤ CV ≤ 10%, Fair equity: 10% ≤ CV ≤ 25%, Poor equity: CV ≥ 25%	[79]

Table 3. Irrigation performance indicators proposed for the study area.

Indicator	Definition	Range	Reference
Adequacy	Relative evapotranspiration (AETs/PETs)	Sufficient supply: 0.75 < A ≤ 1, Inadequate supply: A ≤ 0.65	[79]
Reliability	Temporal variation of AETs/PETs	CV ≈ 0 indicating the highest reliability	[70]
Equity	CV of actual evapotranspiration (AETs)	Good equity: 0% ≤ CV ≤ 10%, Fair equity: 10% ≤ CV ≤ 25%, Poor equity: CV ≥ 25%	[79]

3. Results

3.1. Crop Type Distribution

Maps presented in Figure 3 clearly demonstrate the effectiveness of the classification framework, which integrates the random forest model with Landsat-8 imagery within the GEE cloud computing platform to capture the crop patterns in the Nile Delta across the five-year study period. The four main crops were distributed throughout the study area with a changing crop pattern and intra-field crop heterogeneity.

Over the study period, rice was the most dominant crop, with an average area of approximately 652,000 hectares (ha), followed by clover (629,000 ha), wheat (567,000 ha), and maize (151,000 ha).

The Ad Daqahliyah governorate recorded the largest in rice cultivation, with a 5-year average of 151,000 ha, closely followed by Ash Sharqiyah with 134,000 ha. At the same time, maize was primarily scattered across the delta, with the largest concentration in Ash Sharqiyah, with around 29,000 ha.

In terms of winter crops, wheat also dominated in the Ash Sharqiyah governorate, covering an area of 126,000 ha, while the Al Buhayrah governorate had the highest clover cultivation, averaging 181,000 ha.

The detailed area statistics for each crop and season, as derived from the generated crop maps, are provided in

Table A1. Mapped rice cultivation areas in hectares in the Nile Delta governorates during the study period.

Governorates	2015	2016	2017	2018	2019	Average
Bur Said	3125	3159	3700	6590	5152	4345
Dumyat	13,295	23,615	18,175	26,440	26,749	21,655
Ad Daqahliyah	94,725	158,633	154,911	170,946	173,916	150,626
Al Qalyubiyah	15,811	10,244	11,147	17,946	6900	12,410
Al Gharbiyah	63,527	70,117	73,164	76,427	61,494	68,946
Kafr ash Shaykh	82,487	96,772	104,723	133,638	103,880	104,300
Al Buhayrah	97,294	114,557	96,933	165,457	116,780	118,204
Al Iskandariyah	1214	1285	840	4828	2043	2042
Al Isma’iliyah	1824	2730	2467	4029	3467	2903
Al Minufiyah	50,995	17,077	36,449	49,880	9428	32,766
Ash Sharqiyah	100,980	130,299	132,906	157,425	146,295	133,581
Total	525,277	628,486	635,416	813,604	656,105	651,778

,

Table A2. Mapped maize cultivation areas in hectares in the Nile Delta governorates during the study period.

Governorates	2015	2016	2017	2018	2019	Average
Bur Said	1802	2345	1206	1123	2023	1700
Dumyat	1401	2523	1193	1151	1334	1521
Ad Daqahliyah	15,277	24,206	14,230	23,088	23,455	20,051
Al Qalyubiyah	8363	14,399	9621	5503	14,040	10,385
Al Gharbiyah	14,286	20,237	14,522	29,980	34,556	22,716
Kafr ash Shaykh	13,370	12,723	15,026	15,728	28,097	16,989
Al Buhayrah	19,742	23,125	13,299	30,428	30,437	23,406
Al Iskandariyah	248	321	148	374	300	278
Al Isma’iliyah	550	532	288	200	268	367
Al Minufiyah	15,924	31,900	18,852	14,495	40,116	24,258
Ash Sharqiyah	33,269	39,743	25,566	22,105	25,142	29,165
Total	124,233	172,054	113,952	144,174	199,768	150,836

,

Table A3. Mapped wheat cultivation areas in hectares in the Nile Delta governorates during the study period.

Governorates	2015	2016	2017	2018	2019	Average
Bur Said	5474	7200	6911	4667	6142	6079
Dumyat	11,044	15,514	11,964	16,914	13,263	13,740
Ad Daqahliyah	97,831	115,141	92,758	113,063	98,226	103,404
Al Qalyubiyah	17,584	14,083	16,475	13,798	17,014	15,791
Al Gharbiyah	46,576	52,439	47,256	56,525	60,774	52,714
Kafr ash Shaykh	79,783	102,363	65,534	105,998	107,345	92,205
Al Buhayrah	136,514	89,012	71,788	115,802	154,130	113,449
Al Iskandariyah	8353	3983	4363	4482	9734	6183
Al Isma’iliyah	4036	3879	3234	2137	6775	4012
Al Minufiyah	47,253	29,654	32,166	31,209	30,959	34,248
Ash Sharqiyah	126,808	131,434	104,693	140,349	124,578	125,572
Total	581,257	564,703	457,143	604,944	628,940	567,397

and

Table A4. Mapped clover cultivation areas in hectares in the Nile Delta governorates during the study period.

Governorates	2015	2016	2017	2018	2019	Average
Bur Said	3624	3888	3229	1226	4853	3364
Dumyat	17,661	15,750	23,176	11,293	22,013	17,979
Ad Daqahliyah	101,371	105,050	125,341	93,797	123,954	109,903
Al Qalyubiyah	15,412	18,412	23,627	18,852	24,970	20,255
Al Gharbiyah	39,645	61,598	71,093	51,284	49,174	54,559
Kafr ash Shaykh	52,974	108,073	143,480	99,231	76,360	96,023
Al Buhayrah	80,828	189,180	299,313	215,302	120,715	181,068
Al Iskandariyah	9145	10,754	18,015	13,513	7749	11,835
Al Isma’iliyah	3617	3117	2415	2092	4264	3101
Al Minufiyah	39,879	57,980	73,182	57,099	68,501	59,328
Ash Sharqiyah	84,054	59,365	58,554	55,717	100,774	71,693
Total	448,208	633,168	841,426	619,406	603,328	629,107

in Appendix A.

3.2. Crop Classification Technical Validation

3.2.1. Accuracy Assessment of Crop Type Mapping

The classification performance of the generated crop maps was assessed using the WaPOR Land Cover (LC) dataset as reference data. A confusion matrix was constructed for each season to evaluate classification accuracy (

Table 4. The confusion matrix of the crop classified maps.

Summer season	Year	Crop	Crop			PA	UA	OA	Kappa
			Rice	Maize	Other
	2015	Rice	390	30	52	0.82	0.69	0.77	0.66
		Maize	45	428	21	0.86	0.84
		Other	126	49	277	0.61	0.79
	2016	Rice	678	60	55	0.85	0.74	0.77	0.64
		Maize	39	536	87	0.8	0.79
		Other	144	76	324	0.59	0.77
	2017	Rice	675	55	61	0.85	0.76	0.78	0.66
		Maize	83	444	28	0.8	0.81
		Other	119	47	330	0.66	0.78
	2018	Rice	675	87	54	0.82	0.77	0.77	0.65
		Maize	77	528	34	0.82	0.79
		Other	123	48	275	0.61	0.75
	2019	Rice	499	40	38	0.86	0.82	0.8	0.7
		Maize	24	344	32	0.86	0.78
		Other	82	52	290	0.68	0.8
Winter season	Year	Crop	Crop			PA	UA	OA	Kappa
			Wheat	Clover	Other
	2015/ 2016	Wheat	213	16	6	0.9	0,83	0.87	0.8
		Clover	19	155	0	0.89	0.9
		Other	24	0	94	0.79	0.94
	2016/ 2017	Wheat	170	15	8	0.88	0.88	0.88	0.82
		Clover	10	124	0	0.92	0.98
		Other	13	0	74	0.85	0.9
	2017/ 2018	Wheat	386	44	9	0.87	0.82	0.87	0.79
		Clover	65	351	0	0.84	0.88
		Other	15	0	124	0.89	0.93
	2018/ 2019	Wheat	155	4	1	0.96	0.93	0.95	0.93
		Clover	8	175	0	0.95	0.97
		Other	3	0	51	0.94	0.98
	2019/ 2020	Wheat	114	5	1	0.95	0.85	0.89	0.83
		Clover	6	111	0	0.94	0.95
		Other	13	0	101	88	0.99

PA: Producer’s accuracy (measure of omission errors). UA: User’s accuracy (measure of commission errors). OA*: Overall accuracy (total correctly classified pixels/total pixels). Kappa: Kappa coefficient (agreement between classification and ground truth).

).

The summer season maps show moderate classification accuracy, with OA ranging from 0.77 to 0.80 and Kappa coefficients ranging between 0.64 and 0.70. The highest accuracy was observed in 2019, while 2016 exhibited the lowest. This Kappa range, however, still means a high degree of agreement/matching [86].

In terms of class-specific accuracy, rice and maize had medium to moderately high PA and UA values, ranging from 0.69 to 0.86. Meanwhile, the other class exhibited the lowest PA values (0.59 to 0.68), primarily due to misclassification with rice and maize. Notably, the UA of the other class was relatively comparable with that of rice and maize, suggesting higher omission errors than commission errors, which could potentially lead to an underestimation of this class in the final maps.

While the winter seasons achieved higher results with OAs exceeding 0.87, the highest OA (0.95) was in 2018/2019. The Kappa coefficient varied from 0.79 to 0.93, with the lowest value (0.79) observed in 2017/2018 and the highest value (0.93) witnessed in 2018/2019. For the winter crops, wheat and clover, the PA values were consistently high, with averages of 0.91 and 0.90, respectively. However, for the UA, clover outperformed wheat with 0.91 to 0.86.

3.2.2. Compared with Official Statistical Data

For further validation of the classification results, the area estimates derived from our crop maps were compared with the official agricultural statistics for the six governorates that are fully located in the delta region (Bur Said, Dumyat, Ad Daqahliyah, Al Qalyubiyah, Al Gharbiyah, and Kafr ash Shaykh). As shown in Figure 4, high R² values were observed for wheat (0.94–0.99), followed by rice (0.84–0.97) and clover (0.80–0.95). Maize showed a weaker correlation, with R² values ranging from 0.52 to 0.92, indicating greater uncertainty in its classification.

Complementing the R² values, error metrics—mean absolute error (MAE) and root mean squared error (RMSE)—were also calculated (

Table 5. RMSE and MAE for crop area estimates compared with official statistics.

Crop	Year	RMSE (ha)	MAE (ha)
Rice	2015	18,243	15,826
Rice	2016	20,126	17,569
Rice	2017	15,875	13,329
Rice	2018	13,477	11,159
Rice	2019	16,103	13,710
Maize	2015	10,154	8944
Maize	2016	11,297	9923
Maize	2017	9411	7968
Maize	2018	8306	6861
Maize	2019	10,125	8734
Wheat	2015/2016	11,050	9774
Wheat	2016/2017	12,279	10,948
Wheat	2017/2018	10,167	8566
Wheat	2018/2019	9511	7604
Wheat	2019/2020	10,838	9311
Clover	2015/2016	13,382	11,413
Clover	2016/2017	14,917	12,518
Clover	2017/2018	12,301	10,195
Clover	2018/2019	11,241	9063
Clover	2019/2020	13,074	10,961

). Across all crops, wheat and clover showed consistently moderate RMSE values (~9,500–14,900 ha), while rice had the highest RMSE (up to 20,100 ha), reflecting the larger cultivated area and potentially greater classification variability. Remarkably, maize had the lowest RMSE values among all crops (as low as ~8,300 ha), yet exhibited the lowest R² values, indicating that although the total classified area was relatively close to the reported area in absolute terms, the year-to-year variability and spatial misclassification were higher. The most significant spatial discrepancy was observed in the Al Qalyubiyah governorate, where rice was overestimated by 573.7% and maize was underestimated by 61.8%.

3.3. Irrigation Performance Assessment

In this case study, seasonal differences in irrigation performance indicators are expected, primarily due to differences in ET between summer and winter, driven by seasonal climatic variations. To illustrate these seasonal dynamics, Figure 5 presents the multi-year average maps of AET and ET₀ across the Nile Delta.

During the summer seasons (2015–2019), ET₀ values were consistently high throughout the delta—often exceeding 1200 mm/season—reflecting intense atmospheric demand. Meanwhile, the ensemble AET showed more spatial variability, with higher values concentrated in the central and northern parts of the delta and lower values toward the periphery.

In contrast, winter seasons (2015–2020) exhibited more uniform ET₀ values, typically ranging between 600 and 700 mm/season. The AET in winter also showed more consistent and moderate values, with fewer spatial disparities than in summer. These patterns confirm that climatic and hydrologic conditions during winter are more favorable for irrigation adequacy and equity.

3.3.1. Adequacy

Figure 6 and Figure 7 present the seasonal adequacy results for the four main crops. Rice consistently exhibited adequacy values below the critical threshold of 0.65 across all summer seasons, with a mean spatial adequacy of approximately 0.4, indicating significant irrigation deficits. Spatially, rice adequacy was lowest in the central and southern schemes, typically around or below 0.4, but improved as we moved northward, reaching up to around 0.5 before declining again at the fringes. Maize showed moderately better results overall, with a mean adequacy of 0.63, borderline for acceptable irrigation performance. The spatial trend for maize was more scattered; however, the central regions showed noticeable deficits (below 0.65), while northern areas exhibited higher adequacy, approaching 0.7–0.8, before declining again at the coastal boundary.

In contrast, winter crops experienced an oversupply of irrigated water. Clover fields showed higher values of adequacy, generally ranging from 1.3 to 1.7, and even approaching 2.0 in some northern zones before decreasing at the far north. Wheat results were more consistent, with most areas exhibiting adequate supply between 1.2 and 1.3 in the central delta, increasing to over 1.5 in the north. It should be noted that regions with inadequate water supply during the winter months were largely confined to the eastern and western margins of the study area, particularly within the newly reclaimed lands.

3.3.2. Reliability

Reliability was assessed using the CV of RETs over the five study years (Figure 8). Rice and wheat showed steady water use patterns, with a low CV of 0.02, indicating a consistent irrigation supply across years. In comparison, maize had a slightly variable irrigation pattern, as demonstrated by fluctuating RET values, ranging from 0.6 to the highest point recorded at 0.7 in 2019, and a CV of 0.06. Clover displayed the highest interannual variability in water use, with a CV of 0.08, indicating a greater susceptibility to fluctuations and variability in irrigation practices.

3.3.3. Equity

Equity in irrigation performance was assessed using the spatial CV of AET across crop pixels. The results indicated generally good levels of equity (CV < 10%) for all four crops, although some spatial heterogeneity in water distribution was evident.

Among the crops, maize was associated with the highest variability, with an average CV of 6.6%, followed by rice (6.2%), clover (5.7%), and wheat (5.6%) across the study period.

As visualized in Figure 9 and Figure 10, the best equity values are in the downstream region of the delta, particularly in the upper-western zones. Equity conditions were generally more favorable during the winter seasons; however, the most pronounced disparities were also observed in this season, particularly in the eastern and western fringes of the delta—areas corresponding to newly reclaimed lands.

4. Discussion and Recommendations

Section 4.1 presents an overview of the crop mapping results, including associated limitations and uncertainties, while in the Section 4.2, we analyze the irrigation performance indicators and their spatiotemporal patterns, also addressing relevant limitations and offering practical recommendations for future management.

4.1. Overview of the Crop Mapping Results

The classification methodology introduced in this study showed promising results for crop type mapping in data-scarce regions, with a key novelty of using spatiotemporal crop data derived from the WaPOR LC product as a proxy for ground truth data. The availability of the WaPOR LC layer enhances the framework’s scalability to other regions, utilizing this ground truth-free mapping approach, which simplifies the sampling process and results in significant savings in human effort and financial resources. However, the study also encountered challenges that influenced classification performance, particularly for summer crops, and revealed areas where the methodology could be strengthened.

4.1.1. Reference Data Limitations

Although satellite imagery is becoming increasingly available with unprecedented temporal, spatial, and spectral resolutions, the availability of ground data has not kept pace with this trend, particularly in developing regions. This difference posed challenges to accurately train and validate the supervised classification approaches that are typically used [22,63,87]. However, due to the lack of in situ ground truth records for our study area, the WaPOR LC layer was used as a substitute. Nonetheless, it is essential to note that the WaPOR LC dataset itself is generated through a decision tree classification process, and therefore, it may have possibly misclassified pixels already included in it. Those mislabeled pixels may have subsequently affected the classification accuracy of our model by creating a misleading pattern.

Furthermore, the spatial coverage of the WaPOR LC layer used was limited to the Zankalon region, representing only around 5% of the cultivated area in the Nile Delta. This restricts the geographic representativeness of the training samples and may have limited the classifier’s ability to generalize across diverse land use patterns throughout the delta. To mitigate this, a quality control procedure based on crop-specific NDVI thresholds was applied to filter unreliable samples, as described in Section 2.3.2.

Still, despite applying a stratified sampling approach, the limited spatial coverage and inherent imbalance in crop prevalence—where dominant crops, such as rice, clover, and wheat, are more extensively represented than minor crops like maize—highlight the need for a more geographically diverse reference dataset. This imbalance may have affected the model’s ability to distinguish spectrally similar or spatially fragmented crops. Future improvements could involve incorporating participatory field mapping or utilizing crowdsourced agricultural observations, such as farmer-contributed data through mobile platforms. Such efforts would help enhance the quality, coverage, and representativeness of the training data, which could in turn improve the overall classification performance, particularly for underrepresented crop types.

4.1.2. Challenges from Intercropping and Field Fragmentation

In Egypt, agricultural fields are characterized by their sparse distribution and relatively small size, with an average plot size of about 0.2 ha in regions such as Zankalon [40]. This presents a major challenge for remote sensing classification, as even satellite imagery with a pixel resolution of 30 m often exceeds the dimensions of individual plots. Consequently, a single pixel may cover multiple plots with different crops. This issue is further compounded by intercropping practice, which is a common farming practice in the region, involving growing several crops concurrently in the same field [35].

To account for this spectral heterogeneity, an ‘other’ class was incorporated into the classification scheme to capture crops not among the four primary classes of interest. However, this class inherently lacked a consistent phenological or spectral profile, increasing confusion with other crop categories. This was reflected in the classification results, where the other class recorded the lowest producer’s accuracy (0.59–0.68), indicating frequent misclassifications, particularly with maize and rice.

4.1.3. Statistical Data Uncertainties

To evaluate the classification results, crop area estimates from the classified maps were compared with official statistics from CAPMAS. However, the use of these official statistics as validation data introduces its own sources of uncertainty. National crop area estimates in Egypt are generally based on farmer surveys rather than direct field measurements, which can also create an additional potential source of error in the findings [88].

Moreover, a significant mismatch often exists between the cropping patterns prescribed by the Egyptian Ministry of Agriculture and the actual cropping decisions made by farmers. These deviations arise due to several factors, most notably economic incentives. For instance, rice typically offers higher profitability than other summer crops like maize, prompting farmers to grow rice even in areas where other crops are officially recommended [89]. This may partially explain the discrepancies observed between the classified maps and official statistics, such as the persistent underestimation of maize cultivation areas; however, further evidence is required.

4.2. Irrigation Performance

While this study offers valuable insights into the irrigation performance across the Nile Delta, highlighting distinct spatial and temporal patterns, certain methodological limitations should be noted to contextualize the findings and inform future applications.

First, the use of a relatively coarse-resolution ET₀ product, resampled from 20 km to 30 m, may introduce a potential source of uncertainty in the assessment estimation, particularly in heterogeneous smallholder fields.

Constrained by the lack of ground-based weather station data, we relied on the credibility of the WaPOR ET₀ estimates, which are based on the FAO Penman–Monteith approach under standardized conditions. According to the FAO WaPOR Data Quality Assessment Report (v1.0) [90], comparisons between WaPOR ET₀ and Penman–Monteith-based ET₀ from the Port Said weather station in Egypt revealed a strong correlation (R² = 0.94), indicating that this product is reliable in the Egyptian context. These findings, along with the validated workflow demonstrated by [25], who applied a similar approach using WaPOR ET₀ resampled via nearest-neighbor interpolation to 30 m resolution, support the general reliability of WaPOR ET₀ for large-scale irrigation assessments, despite the coarse resolution.

Despite these constraints, our findings reveal distinct spatial patterns across the delta. While equity and reliability indicators performed favorably overall, several clusters exhibited extremes in water adequacy—either under-irrigation in summer or oversupply in winter. Low adequacy values, below the threshold of 0.65, were consistently observed during summer, particularly in rice fields, which had mean values around 0.4, highlighting widespread water deficits. Maize fields exhibited slightly better adequacy values (~0.63), especially in downstream regions, but they were still running close to the water shortage limit.

The water shortage during the summer season can be attributed to the naturally limited water supply coupled with relatively high evapotranspiration rates, making this season reach a water peak demand [26]. These outcomes are comparable with studies like [91], which reported similar deficits in summer 2003 and 2004 in fields of the northern delta, with adequacy values ranging from 0.5 to 0.7. In our findings, rice and maize crops in the central and southern delta exhibited adequacy values below the critical threshold (approximately 0.4–0.65), while values improved northward—reaching up to 0.5 for rice and ~0.8 for maize—before declining again at the northern extremes (introduction of coastal areas), and in the eastern and western zones, likely due to their location in newly reclaimed lands.

On the other hand, the overall adequacy trend in the winter crops was adequate to oversupply, aligning with previous work by [21,81]. Both observed seasonal over-irrigation during winter 2008–2009, with results ranging from 0.64 to greater than one at all the levels assessed, which strengthens the credibility of our results. In our case, wheat fields typically recorded values between 1.2 and 1.5 in central regions and exceeded 1.5 in the north, while clover often reached values approaching 2.0 before declining at the edges.

However, notable exceptions were identified in peripheral eastern and western schemes, most of which are located in the new lands bordering the delta, exhibiting low adequacy even in winter. These newly reclaimed areas have lower cropping intensities and undergo deficit irrigation, mainly served by central pivots and drip irrigation [35]. The reason behind this observation could be the unfavorable soil properties, such as sandy soil, on marginal lands located directly adjacent to the desert. This winter’s inadequacy is consistent with [92], who identified early-season water stress in wheat schemes in the El-Salhia project (eastern delta new lands).

A relatively reliable water supply, regardless of its adequacy, was detected in the summer crops. The stable variability of the RET in summer seasons, causing this reliability, could be linked to the fact that the water management system in summer adopts a supply-driven approach to ensure a consistent water supply for rice cultivation, prioritizing continuous availability rather than responding to fluctuating demand [91]. In addition, controlled pump operation in summer follows the regulations established by the Water User Associations (WUAs), which means that water is always provided in a regulated manner, whilst in winter, pump operation is influenced by the availability of water for delivery [91]. This previous factor may explain the variable reliability of the clover crop but not the consistency of wheat.

In line with previous studies [32,81,91], the overall equity indicator consistently demonstrated favorable performance, which means water was equally and promptly distributed among growers. Still, there is potential to improve equity, as some cultivators usually clustered in the northwest areas have achieved significantly better equity within the study area. This flow of enhanced equity towards the northwestern areas, primarily observed in the winter seasons, could be attributed to a number of factors. Firstly, the northern areas benefit from slightly more rainfall during winter, marginally improving local water availability for users compared with other parts of the delta. Additionally, the uneven water distribution along canals frequently causes downstream farmers to manually pump water from nearby drainage channels or shallow groundwater to irrigate their fields [93,94]. As a result, a surplus of drainage water accumulates at the lower end of the study area.

Meanwhile, the highest CV obtainable in the equity indicator, indicating the lowest equitable water consumption patterns, was consistently detected during winter. These patterns were primarily linked to schemes in the new lands, which were also accompanied by lower adequacy values. This irregularity suggests chronic issues related to soil characteristics, irrigation infrastructure, and access to water. These areas, which underperform across all indicators, should be treated as intervention hotspots for future improvement.

Given these findings, multiple actionable insights emerge for future water management. Oversupply in winter, sometimes reaching 170% of crop demand, indicates a strong potential for water savings through reallocation or storage for later use in summer. However, implementing this strategy would require addressing the lack of downstream storage capacity in the Nile River system. Therefore, a focused study is needed to identify where and how water can be effectively stored and managed in the delta system.

Significant savings can also be achieved through enhanced on-farm techniques, with further improvement in traditional irrigation systems and practices. For example, to address the challenge of higher water consumption associated with the summer season, more specifically rice cultivation, planting rice using wide furrows can significantly reduce rice water consumption (saving around 0.7 BCM by 2040) [95]. To save more water in the winter, wheat can be cultivated using raised beds [96], which has been approved to reduce the water needed to irrigate wheat by 15%, and subsequently, this saved water can be further allocated for other crops/uses.

Another water-saving strategy is to reduce water loss by minimizing evaporation from soil and plants. This can be performed through several methods, such as mulching and using piped networks for water distribution. While some of these measures have already been implemented on a localized scale by the Ministry of Agriculture and farmers, wider systemic adoption and standardized evaluation remain lacking [35].

To address the issue of the difference in water distribution between the head-and-tail schemes of the delta, collaborative efforts are needed from the WUAs to manage water distribution and regulate the reuse of drainage water. This should ensure fair allocation across the irrigation system, promoting uniform irrigation opportunities and equitable access to water in the upstream and downstream fields.

While there is a noticeable horizontal expansion of agricultural projects into the new lands of the delta, their performance remains suboptimal. As highlighted in [97], sustained support and optimized agriculture practices could enhance productivity over time. These areas, having scored lowest across all indicators, should be prioritized for research and long-term development planning.

This study’s method—integrating crop mapping with irrigation performance metrics through open-source satellite data—is both economical and scalable. As WaPOR LC and ET products remain accessible, the framework enables near-real-time monitoring and can assist water authorities such as the Ministry of Water Resources and Irrigation (MWRI) in coordinating seasonal water distribution with real crop needs. Such predictive capabilities are crucial for adaptive, sustainable irrigation planning across the Nile Delta and beyond.

The framework is particularly well suited to regions lacking high-resolution weather or irrigation field data, yet satellite-based ET and land cover products are accessible. However, for areas with extremely fragmented plots or where ET₀ inputs deviate significantly from the FAO Penman–Monteith assumptions, the results should be interpreted with caution. Ongoing improvements, such as enhanced calibration with local weather stations or integration of finer-scale meteorological inputs, would further enhance the framework’s applicability.

Future applications can leverage this methodology to produce timely, seasonally updated assessments. Coupled with policy and operational tools, this framework offers decision-makers the ability to anticipate shortages, plan water allocations more equitably, and target underperforming areas for improvement. Ultimately, this can support more sustainable and resilient irrigation practices in Egypt and similar semi-arid agricultural systems.

5. Conclusions

This study presents a comprehensive framework for crop type mapping and the evaluation of irrigation performance assessment across the Nile Delta using satellite data. The framework consists of two key components: first, seasonal crop type mapping was conducted from 2015 to 2020, targeting the four main crops in the region: rice, maize, wheat, and clover. The WaPOR Land Cover Classification layer served as a proxy for ground truth data in training a random forest classifier, enabling large-scale mapping under limited in situ data conditions. The resulting 30 m crop maps showed moderate to high classification performance, with overall accuracies ranging from 0.77 to 0.80 in the summer and 0.87 to 0.95 in the winter, showing notable agreement with official agricultural statistics. These results demonstrate the method’s potential to effectively capture crop patterns even in the absence of field-based training data.

The second component evaluated irrigation performance through three key indicators—adequacy, reliability, and equity—applied to the mapped crops. This analysis revealed clear seasonal and spatiotemporal patterns in irrigation performance, identifying areas for improvement. Overall, the results indicate that a satisfactory level of equity is accompanied by an acceptable degree of reliability. However, several clusters display significant variations in water adequacy, with insufficient water supply during summers and excesses during winters. A visible northwestward gradient of improvement was observed across both seasons. Moving horizontally to the new lands of the delta, it can be observed that those fields are underperforming in all performance indicators, thus making them clear hotspots for intervention.

Despite reliance on resampled coarse-resolution ET₀ data, previous validations of WaPOR products and the adoption of proven workflows support the reliability of the results at the regional scale. Nevertheless, future studies should consider integrating finer-resolution meteorological data and local calibration to improve accuracy.

In a country like Egypt—where water scarcity is intensifying and uncertainty about future water availability remains high—these findings can play a critical role in evaluating irrigation system performance. The insights can support policy-makers and water managers in designing informed, sustainable strategies for water allocation and agricultural planning.