Beyond Fixed Dates and Coarse Resolution: Developing a Dynamic Dry Season Crop Calendar for Paddy in Indonesia from 2001 to 2021

Abstract

There is valuable information that can be obtained beyond using a fixed crop calendar with coarse spatial resolution. Knowing the dynamics of the timing and location in which a particular crop is planted and harvested, with an annual temporal resolution and a fine spatial resolution, is crucial not only for monitoring crop conditions and production but also for understanding crop management under changing climates. In this study, the Normalized Difference Vegetation Index (NDVI) was utilized to develop a historical crop calendar for paddy in Indonesia with a 1 km resolution from 2001 to 2021. The result of this study is the first dynamic crop calendar that includes information about the planting, peak, and harvesting dates, as crop growth indicators, derived from the analysis of NDVI value fluctuations. Additionally, this dataset also includes the total number of cropping seasons each year. In Indonesia, there are intensive agricultural activities, including two dry cropping seasons that occur after the wet cropping season. However, this dataset is limited only to crops grown during the dry seasons, which typically begin in February and June. This dataset offers significant information at a finer spatiotemporal resolution to enable studies on agricultural fields undergoing climate change, although it is more country–specific than the other established dataset. The annual crop calendar dataset from 2001 to 2021 underscores the significance of examining the variability in cropping seasons over the years. This exploration aims to deepen our comprehension of the interplay between cropping seasons, climatic indicators, and even the social factors influencing farmers’ decisions. Furthermore, presented at a 1 km resolution, this dynamic crop calendar underscores the need for a more precise representation of diverse cropping intensities and seasons, particularly within small and fragmented agricultural areas.

1. Introduction

As food is a primary necessity in human life, the agricultural sector plays an important role in sustaining global food security as well as socioeconomic stability. However, agricultural activities rely heavily on environmental conditions, such as precipitation, temperature, humidity, sunlight, soil conditions, and other factors. This strong dependency on the environment highlights the importance of precise and timely planning to ensure optimal yield production. According to the International Rice Research Institute (IRRI), many stakeholders, researchers, and governments often provide farmers with information about crop calendars in order to help in the decision–making process and resource allocation. The crop calendar typically includes information about a particular crop in terms of the timing of the cropping season, which usually consists of three important dates, the Start of Season (SOS) or planting date, the Peak of Season (POS), and the End of Season (EOS) or harvesting date. Additionally, the crop calendars vary across regions, with details depending on the spatial resolution, whether they cover coarse areas, such as subnational, or more detailed areas, such as subdistricts. Given its rich information, researchers have used the planting and harvesting dates from crop calendars for various purposes as follows:

• To develop the historical crop yield production [1,2];
• To assess the climate impact on crop productivity [3,4,5];
• To examine the agricultural supply and demand or the food security [6,7];
• To assess the cropping system management under the changing climate [8,9];
• To assess the future projections of agricultural areas or products under the changing climate [10,11,12].

In terms of the methodology used to develop a crop calendar, Mishra et al. [13] divided it into two approaches as follows: (1) census–based, which uses multiple ground observation data, such as surveys or expert opinions, and (2) model–based, which uses remote–sensing data, such as climate data, vegetation indices to examine crop phenology, and land cover. Using these definitions,

Table 1. A summary of established crop calendar datasets.

No	Dataset	Crop Types	Spatial Coverage	Methods	Spatial Resolution
1	MIRCA2000 1 [14]	4 major crops + 22 others	Global	Census and model	0.083°
2	Crop Calendar Dataset [15]	4 major crops + 15 others	Global	Census and model	0.083°, 0.5°
3	Agricultural Growing Season Calendars [16]	All major crop rotations within a given geographical area	Global	Model	0.5°
4	SACRA 2 [17]	4 major crops + cotton	Global	Model	0.083°
5	RiceAtlas [2]	Rice	Global	Census	Subnational
6	RICA 3 [13]	Rice	Asia	Model	Subnational

¹ Monthly Irrigated and Rainfed Crop Areas around the year 2000. ² Satellite–derived CRop calendar for Agricultural simulations. ³ RIce crop Calendar for Asia.

summarizes some of the established crop calendar datasets based on the crop types, spatial coverage, spatial resolution, and methodology used to develop the datasets.

This study specifically focuses on rice, as one of the major crop types that is grown in the Asian Monsoon Region. According to Bandumula [18], around 87% of the total global rice production is produced by eleven countries in Asia. Based on the above summary, the two most recent databases, RiceAtlas and RICA, were said to have a more thorough crop calendar for rice. However, the established datasets still had several limitations including the following: (1) they were calculated based on a specific period and then provided as fixed crop calendars; therefore, they are unable to represent the diversity of the cropping seasons under changing climatic conditions, and (2) because the data were provided at a coarse spatial resolution, they were unable to capture spatial variations, particularly in dispersed agricultural regions [19]. In order to address those gaps, this study was conducted to develop a non–fixed or dynamic crop calendar, taking a model–based approach utilizing vegetation indices, particularly the Normalized Difference Vegetation Index (NDVI). The NDVI was used to monitor crop growth and extract the key crop growth stages, namely, the SOS, POS, and EOS. Additionally, this dataset covered the historical crop calendar from 2001 to 2021 and was developed with a finer resolution of 1 km. The dynamic crop calendar resulting from this study is expected to show the variation in the cropping seasons each year from 2001 to 2021 over a small and fragmented agricultural area. The result showcases the difference in the planting and harvesting dates among regions within the same district or in the same region within different periods.

According to the Food and Agriculture Organization (FAO), rice requires around 450–700 mm of water throughout its entire growth period. Considering the high reliance of rice on water, this research specifically focused on the dry season cropping period characterized by lower precipitation compared with the wet season. This limitation enables the observation and evaluation of shifts in the cropping seasons resulting from adjustments made in response to climatic conditions that may be affected by climate change. In particular, we recognize that practically, despite the advanced preparation of the crop calendar by the stakeholders, the precise dates of the cropping season may vary across regions and years due to external factors such as the actual weather conditions and farmers’ decisions. Moreover, as it uses a 1 km spatial resolution, this dataset can be used to monitor the variations in cropping management on a local scale.

2. Materials and Methods

This study was carried out in Indonesia, which is located in the tropical–humid region. According to data from the World Bank [20], agricultural areas in Indonesia account for more than 33% of the country’s total land area overall and are mostly concentrated on Java Island. Like the majority of the countries in the Asian Monsoon region, rice or paddy is the primary agricultural product in Indonesia. Figure 1 shows the map of the paddy field area in Indonesia as indicated by the green color. Supported by its climatic conditions, agricultural activities are very intensive in the country. Apriyana et al. [21] reported that there are normally three rice–growing seasons in Indonesia including the wet cropping season, which is the first and primary planting season, accounting for about 45% of total rice production, followed by two dry cropping seasons.

Regarding the season for cultivation, Figure 2, from the U.S. Department of Agriculture (USDA) [22], shows the general rice cropping seasons in Indonesia and indicates that the dry cropping seasons typically begin in February and June. Depending on when the wet cropping season is planted, the beginning of the dry cropping seasons may vary. Furthermore, the start of the wet cropping season is typically altered in accordance with the onset of the wet season, and delays at the beginning of the wet season can have a ripple effect on the subsequent two dry cropping seasons [21]. Additionally, Indonesia’s agricultural lands are widely fragmented; they are made up of numerous small areas with varied crop management practices. Mishra et al. [13] state that due to the intensive and dispersed agricultural activities, planting and harvesting dates can be varied to suit local conditions. These local conditions are not only limited to the weather conditions but also the farmers’ decisions that might be influenced by resources such as the irrigation system and its water supply, fertilizer, agricultural equipment (i.e., water pumping, tractor, etc.), as well as manpower.

The approach to developing the annual dry season crop calendar for paddy in Indonesia involved extracting the crop growth stages from vegetation indices. However, the innovation lies in the iterative process conducted annually within the study period to produce the first dynamic crop calendar. This methodology was also able to identify both single and double cropping seasons within a 1 km resolution, addressing the small and fragmented nature of agricultural areas in the country. Through a model–based approach, two datasets were utilized in this research as outlined below:

• Normalized Vegetation Difference Index (NDVI)

As one of the vegetation indices, the NDVI has often been used to quantify the biomass or greenness of vegetation. The values range from −1, indicating a high likelihood of water, to 0, representing an urban area, to +1, indicating an area of dense green vegetation. Its value has also been often used to track the growth stage of a specific crop or vegetation. The dataset used in this study was obtained from the Moderate Resolution Imaging Spectroradiometer or MODIS–Terra (MOD13A2) version 6 [23] from the following website: https://lpdaac.usgs.gov/products/mod13a2v006/ (accessed on 28 February 2023). A total of 10 granules were used in this study, which cover all the Indonesian regions from the h27v08 tile to the h31v08 tile and the h27v09 tile to the h31v09 tile. The dataset has a resolution of 1 km and is captured at 16–day intervals, with 23 images available annually from February 2000 to the present date. For this study, the MOD13A2 product from January 2001 to December 2021, with a total of 483 images, was selected.

b.

Land cover map

For this study, the land cover map was obtained from the Ministry of Environment and Forestry upon request. The ministry’s publication schedule releases the land cover map only every five years, which resulted in the absence of annual coverage throughout our study period. Consequently, the land cover map for 2009 was selected, as it represented the temporal midpoint of the available data during the study period. To account for any potential changes in land cover that may have occurred before or after 2009, crop area identification was carried out, and further details are provided in the following section.

To develop the crop calendar dataset, the flowchart in Figure 3 outlines the methodology used in this study. The detailed approaches are described as follows:

• NDVI time series reconstruction

In order to obtain a reliable and long–-term NDVI, this study employed a process to generate a smoothed NDVI time series by eliminating atmospheric noise, which is typically caused by cloud cover. To achieve this, the raw NDVI values and information about the quality of data in each grid cell were extracted from the MOD13A2 product. The appropriate smoothing technique was chosen based on the data characteristics. The percentage of non–reliable grid cells identified as −1, 2, or 3 in the reliability layer was calculated for January 2001 to December 2021 in order to examine the non–reliable grid cells of MOD13A2 in Indonesia. Figure 4 displays the percentage of non–reliable grid cells in Indonesia, which averaged around 33%, and were primarily caused by cloud cover. Thus, the ST–Tensor method developed by Chu et al. [24] was selected for NDVI time series smoothing, which is more suitable to be used in cloudy and rainy regions with large gaps in time series. The details of the smoothing method can be found in the original paper [24]; however, in general, this methodology utilizes the correlations among the spatial neighbors, inter–annual variations, and periodic characteristics to develop the smoothed NDVI time series.

2.

Crop area identification

Given the focus of this study on the agricultural area, once the smoothed NDVI time series were obtained, the agricultural areas, specifically, paddy field areas, were isolated using the land cover map from the Ministry of Environment and Forestry for the year 2009. As mentioned previously, this specific year was selected because it represents the midpoint of the study period (2001–2021). However, land cover changes may have occurred during the study period. The most common ones are the changes in agricultural, urban, and forest areas. To minimize the impact of land cover changes on the result of this study throughout the study period compared to 2009 (as the base year of the land cover map), the agricultural areas were identified each year based on the fluctuations in the NDVI values. Figure 5 illustrates the NDVI time series for various land cover types, indicating that the NDVI values of agricultural areas, represented by the green color, fluctuated within a year, unlike forest areas with constant high NDVI values (orange color) and urban areas with constant low NDVI values (blue color). The fluctuation observed in the NDVI values can be attributed to the annual occurrence of rice cultivation, ranging from a single to a triple cropping season. In this process, for each year within the study duration, only the grid cells with fluctuating NDVI values throughout the year were chosen. This selection criterion was applied to exclude the potential inclusion of forest and urban areas in this study. From this procedure, the agricultural area for the period of 2001–2021 was determined.

3.

Number of cropping seasons assessment

Based on the NDVI time series of the agricultural area from 2001 to 2021, the determination of the number of cropping seasons in each grid cell for each year was carried out by identifying the total peak of the monthly NDVI time series within the corresponding year. For each year, the number of cropping seasons might vary within the same grid cells and usually ranges from single to triple in Indonesia, as explained earlier in the study area section. This process enabled the determination of the total number of cropping seasons from 2001 to 2021, including both the wet and dry seasons, despite the fact that this study was only limited to the dry cropping seasons. To ensure accuracy in determining the number of cropping seasons, the interval between the two peaks had to be at least two months long. Figure 6 demonstrates the NDVI time series characteristics used to distinguish the total number of cropping seasons based on the total peak, which differed among single, double, and triple cropping seasons.

As mentioned above, the first planting season typically begins in October. Therefore, the cropping season from October to September of the following year was considered. For instance, to determine the total number of cropping seasons in 2001, the NDVI time series from October 2001 to September 2002 was evaluated. Figure 7 shows the comparison of the total number of cropping seasons on Java Island (selected for visualization because the agricultural area is concentrated there) in 2010 and 2019. The years 2010 and 2019 were selected for comparison because they represent extremely wet and dry years in Indonesia, respectively. A difference in the total cropping seasons can be observed. Thus, this process was conducted to assess the generality and investigate any changes in the number of cropping seasons in each region.

4.

Crop calendar generation

The crop calendar, which provides data on the Start of Season (SOS), Peak of Season (POS), and End of Season (EOS), is a crucial instrument for agricultural monitoring. The smoothed NDVI values in agricultural areas obtained from the previous process were used to extract these important dates for each grid cell. NDVI values, which are a measure of vegetation greenness, can reflect the growth stages of specific crops. Using the NDVI time series, the crop calendar for the dry cropping season was obtained for this study as follows:

• Start of Season (SOS)
  • The SOS refers to the planting dates when the vegetation’s greenness first appears. Referring to Figure 8, the SOS was denoted by the green marker, which corresponds to the local minimum of the NDVI values in the time series. The SOS was retrieved by detecting the lowest NDVI value when there was a change in direction from negative to positive or when the values increased.
• Peak of Season (POS)
  • The POS corresponds to the peak of crop growth or when vegetation is at its “greenest”. In Figure 8, the POS is identified by the yellow marker, which represents the local maximum of the NDVI values in the time series. The POS can be obtained by detecting the highest NDVI value, which should be located between the SOS and EOS. As this study focused only on the dry cropping seasons, the POS was limited to between February and August, referring to the general crop calendar in Indonesia, according to the USDA [21].
• End of Season (EOS)
  • Lastly, the EOS, which marks the harvesting dates, is identified when the “greenness” of vegetation begins to diminish. In Figure 8, the EOS is identified by a red marker or another local minimum of the NDVI value in the time series. To distinguish it from the SOS, the EOS was determined by detecting the lowest NDVI value when the direction changed to negative or decreased.

It is important to note that due to the iterative process in the approach for developing an annual crop calendar, the identification of the SOS, POS, and EOS was conducted repeatedly for each grid cell spanning the period from 2001 to 2021.

3. Results

3.1. Data Records

The main result of this study is the creation of a dry season crop calendar for paddy in Indonesia from 2001 to 2021. This dataset comprises the SOS, POS, and EOS in Julian Day format for each grid cell. Due to the presence of double cropping seasons, the data were segregated into sos_1 and sos_2 to distinguish between the first and second dry cropping seasons, and this was also applicable to the POS and EOS. Furthermore, the dataset includes information about the total number of cropping seasons from 2001 to 2021. The file is freely available in NetCDF format on Figshare and can be cited as provided in the Supplementary Materials.

This dataset can be used for various purposes, including but not limited to the following:

• Monitoring the impact of the changing climate on cropping management (e.g., cropping season duration, the start of the dry cropping season) on a local scale;
• Developing the finer resolution of the crop yield dataset;
• Examining the impact of changing climate and related hazards (i.e., drought) on crop production.

To aid in understanding the result of this study, Figure 9 shows an example of the dry season crop calendar. Java Island was selected as representative for visualization because the agricultural area is concentrated in this region. As mentioned previously, the result of this study is an annual dry season crop calendar from 2001 to 2021, but Figure 9 only shows the planting and harvesting dates of the first and second dry cropping seasons in 2010. The legend indicates the Day of Year (DOY) in Julian Days.

3.2. Technical Validation

To validate the accuracy of the crop calendar result, a comparison was conducted with two established datasets, namely, the RiceAtlas [2] and RICA [13] datasets, which provide specific crop calendars for rice. However, the spatial and temporal resolution of these datasets differ from this study. Regarding the spatial resolution, as outlined in Table 1, both the RiceAtlas and RICA datasets offer crop calendars at the subnational level, whereas this dataset is accessible at a 1 km resolution. As for the temporal issue, this dataset offered (i) the SOS and EOS for two dry cropping seasons, while the RiceAtlas and RICA datasets only covered a single dry cropping season, and (ii) multiple dates for planting and harvesting from 2001 to 2021, while the RiceAtlas and RICA datasets provided a single date throughout the study period. To address the spatial and temporal resolution issue, a process to “transform” our result into a subnational level was implemented, capturing a singular date for planting and harvesting that signified the start (SOS) and end (EOS) of the dry cropping season. This was accomplished through straightforward averaging or by following these steps:

• Initially, the study area was divided into subnational levels, resulting in a total of 34 sub–study areas. Moving to the next step, West Java Province was chosen as an illustrative example to demonstrate how the results of the validation process were prepared. Nevertheless, these processes were applied separately to all sub–study areas.
• As this dataset includes an annual crop calendar for the dry cropping season, there are multiple dates recorded for the SOS and EOS. Figure 10 illustrates an example of the planting date for the second dry cropping season in West Java. Originally, there were 21 images on the left side spanning from 2001 to 2021. To address the first temporal resolution differences, these dates were averaged to obtain a single SOS date representing the entire study period, as shown on the right side of Figure 10.
• To address the second temporal resolution issue in the RiceAtlas and RICA datasets, which provide only a single date for both SOS and EOS during the dry cropping season, only one dry cropping season’s SOS and EOS (if there were two) that was closest to the dates provided by those datasets were selected.
• However, the results from the previous step still consisted of multiple dates at a 1 km resolution within the subnational levels, indicating a persisting spatial resolution issue. So, the final step was averaging all the dates from each grid cell within each subnational level, resulting in a single SOS and EOS date for each subnational level.

Finally, the result of this data preparation for the validation process was summarized, as shown in

Table A1. The summary of the validation between the result of this study and RiceAtlas.

No	Province	SOS_Result	SOS_ATLAS	EOS_Result	EOS_ATLAS
1	Kalimantan Utara	60	46	201	200
2	Kalimantan Barat	65	46	235	245
3	Kalimantan Selatan	67	46	242	245
4	Kalimantan Timur	56	46	220	200
5	Kalimantan Tengah	64	84	239	226
6	Gorontalo	71	64	244	245
7	Sulawesi Barat	64	64	228	245
8	Sulawesi Tengah	58	64	221	245
9	Sulawesi Tenggara	64	84	229	245
10	Maluku Utara	70	46	240	205
11	Maluku	63	46	220	205
12	Papua Barat	55	46	218	205
13	Papua	57	46	224	205
14	Nanggroe Aceh Darussalam	116	121	249	251
15	Sumatera Utara	123	121	257	243
16	Sumatera Barat	120	121	251	251
17	Riau	118	121	254	251
18	Jambi	109	152	240	291
19	Bengkulu	116	152	250	291
20	Sumatera Selatan	117	130	257	240
21	Kep. Bangka Belitung	140	152	268	291
22	Lampung	127	152	266	291
23	Banten	123	121	264	251
24	Jawa Barat	119	121	257	251
25	Jawa Tengah	121	121	251	251
26	DI Yogyakarta	113	121	244	251
27	Jawa Timur	117	121	246	251
28	Bali	129	135	264	261
29	Nusa Tenggara Barat	121	135	257	261
30	Nusa Tenggara Timur	126	135	261	261
31	Sulawesi Utara	113	121	255	261
32	Sulawesi Selatan	126	152	268	291

and

Table A2. The summary of the validation between the result of this study and RICA.

No	Province	SOS_Result	SOS_RICA	EOS_Result	EOS_RICA
1	Sumatera Barat	56	71	208	185
2	Bali	52	73	203	179
3	Nusa Tenggara Barat	55	67	199	180
4	Kalimantan Barat	66	89	233	194
5	Kalimantan Selatan	67	91	244	216
6	Kalimantan Timur	54	83	215	192
7	Sulawesi Utara	63	34	228	170
8	Sulawesi Tengah	57	47	220	157
9	Sulawesi Tenggara	66	93	231	206
10	Maluku Utara	70	86	242	194
11	Papua Barat	54	19	215	125
12	Papua	57	18	224	124
13	Sumatera Utara	123	159	257	299
14	Jambi	106	132	236	241
15	Sumatera Selatan	117	172	260	285
16	Kep. Bangka Belitung	141	146	270	243
17	Lampung	127	150	266	249
18	Banten	123	138	265	231
19	Jawa Barat	119	144	258	238
20	DKI Jakarta	128	147	260	277
21	DI Yogyakarta	111	100	243	200
22	Jawa Timur	114	125	224	222
23	Kalimantan Tengah	121	145	261	259
24	Sulawesi Barat	118	143	257	244
25	Sulawesi Selatan	125	107	268	261
26	Maluku	135	144	270	226
27	Papua Barat	121	194	256	295
28	Papua	131	102	274	207

in the Appendix A. It is important to note that there was a total of 32 and 26 input points for the validation process with the RiceAtlas and RICA datasets, respectively. This number of inputs was lower than the total of the 34 subnational levels in Indonesia. This was caused by the absence of agricultural areas for some regions or the missing dry season crop calendar dataset in the established datasets.

Following the preparation of the results for the validation process, the evaluation focused on examining the concurrence between the findings of this study and the recognized datasets, specifically, from RiceAtlas and RICA. The evaluation of dataset agreement was conducted using Pearson correlation and the Coefficient of Determination (R²) for both the Start of Season (SOS) and End of Season (EOS).

The validation results are presented in Figure 11, with the upper row showing the agreement with the RICA dataset and the bottom row showing the agreement with the RiceAtlas dataset. The left and right columns indicate the validation results for the Start of Season (SOS) and End of Season (EOS), respectively. The X–axis represents the already established datasets (RiceAtlas and RICA), while the Y–axis represents the results from this study. The agreement with the RICA dataset showed a relatively lower R² value as written using red color in the upper row of the figure, accompanied by a large Pearson correlation value (R_SOS: 0.88 and R_EOS: 0.77). This suggests that the model showed strong alignment with the RICA dataset; however, it tends to undervalue the End of Season (EOS) because the forecasted harvesting dates were predicted to occur later than those recorded in the RICA dataset. Additionally, the RICA dataset provided more information about the wet cropping season than the RiceAtlas dataset, but it did not include the Central Java crop calendar despite it being a large proportion of the agricultural area. On the other hand, the agreement with the RiceAtlas dataset resulted in a higher R² value as written using red color in the bottom row of the figure, followed by a large Pearson correlation value (R_SOS: 0.94 and R_EOS: 0.74). This suggests that the model had a good agreement with the RiceAtlas dataset both for the planting and harvesting dates. However, the limitation is that the RiceAtlas dataset only provided one dry cropping season, unlike the result of this study, which provided two.

4. Discussion

This study introduces an innovative approach by creating the first dynamic crop calendar, which builds upon the conventional model–based method using vegetation indices data. Using remote sensing data, this dataset becomes a valuable tool for crop monitoring, particularly in the context of climate change and human influences that impact agricultural practices [25]. By focusing on the annual variations in cropping patterns, this innovation allows for a more thorough understanding and examination of changes in cropping intensity and shifts in planting dates. This aligns with the statement made by Laborte et al. [2] that annual variations can be observed by using time series of satellite images; in this study, vegetation indices were employed. The significance of a dynamic crop calendar is particularly evident in tropical humid regions, where cropping intensity can double or triple within a season. In these areas, alterations in planting dates are typically contingent on the onset of the wet season [21,26]. Consequently, the shift in the onset of the wet season has a cascading effect on the commencement of the dry cropping season [21]. Hence, gaining insights into the association between climatic indicators and cropping seasons becomes imperative as the phenology of crops is greatly influenced by climatic processes [17,27].

An illustration of this necessity is found in a study by Sawano et al. [26], which modeled how crop calendars depend on rainfall patterns in rainfed rice cultivation in northeast Thailand. Their research reveals that the availability of water plays a crucial role in determining the transplanting date, as early submergence allows for early transplanting. Consequently, transplanting may commence sooner in regions with abundant rainfall compared with those with limited or delayed precipitation. Imran et al. [28] examined the spatial–temporal variations in climate indicators that affected the rice crop calendar in Pakistan from 2005 to 2015. Their research suggests that reduced rainfall in specific villages contributes to stress in the NDVI cycles. Interestingly, spatial variations in temperature do not appear to exert a significant impact. However, when examining inter–annual variations, it becomes evident that the temporal fluctuations in temperature play a role in influencing NDVI throughout the crop growing seasons. In contrast, precipitation does not exhibit notable temporal changes over the study period.

Developed in a finer spatial resolution of 1 km² grid cells, this dataset aims to fill the gap in addressing spatial variations in small and fragmented agricultural areas [19]. The utilization of a finer spatial resolution in this crop calendar dataset enables a more accurate representation of diverse cropping intensities and seasons across agricultural regions. As depicted in Figure 9, even when agricultural areas are located in the same region or close to each other, there are noticeable variations in the starting dates for planting and harvesting. These subtle differences cannot be captured when using a crop calendar with a broader spatial resolution.

Furthermore, the crop calendar stands as a crucial element in the estimation of crop yield and production across agricultural regions. It has been noted that variations in crop yield occur on a significantly smaller scale than what is typically addressed by geographical information systems [26]. Consequently, adopting a finer crop calendar offers distinct advantages for generating a more precise and accurate crop yield model or estimation within a specific agricultural area, leading to improved accuracy in crop production estimates. By integrating production and area data with the crop calendar, we can better understand the spatial and temporal changes in rice production. For instance, combining the information provided in the crop calendar on climate impacts and rice production allows for a more accurate evaluation of seasonal and regional differences in rice availability, helping to manage potential shortages during specific periods. Additionally, this approach strengthens our capacity to address concerns related to food security [2] and to provide timely and precise information on current and future crop production [29].

Numerous studies have extensively documented the influence of climate variability and hazards (e.g., drought and floods) on crop yield, underscoring rice’s significant reliance on the water supply. Research conducted by Rowhani et al. [30] reveals that in Tanzania, an increase in precipitation variability from January to June negatively affects rice yields, while an increase in temperature variability during the same period appears to have a positive impact on yields. In a separate study, Tao et al. [31] observed an increase in rice yield at the county level in China, correlated with climate trends spanning from 1980 to 2008, which encompassed precipitation, temperature, and sunshine duration data. Conversely, Yang et al. [32] identified a notable negative impact of an increased number of rainy days on rice yield in the middle and lower reaches of the Yangtze River Basin. This effect was attributed to a reduction in the solar radiation and temperature essential for optimal rice growth. Furthermore, Kim et al. [33] and Hendrawan et al. [34] unveiled the global–scale impact of drought on major crop yields, including rice, while Irawan et al. [35] specifically assessed the local–scale impact of drought on rice yield. Global–scale assessments from studies [33,34] have shown that rice is not the primary crop most severely affected by drought, with semiarid regions proving more vulnerable to such conditions. However, when viewed on a more local scale in tropical humid regions, it is reported that drought does have a detrimental impact on rice yield loss [35].

As described in the methodology section, the construction of this dataset involved analyzing the MODIS NDVI time series with a 16–day interval and a 1 km² temporal resolution. This approach facilitated the development of a crop calendar with finer spatial resolution, encompassing both dry cropping seasons for each year within the study period. The code, available in the Supplementary Materials, allows users to generate an annual crop calendar for any specified period and region. However, the outcome is contingent on the quality of the spatial and temporal resolution of the NDVI data employed. In essence, using NDVI data with coarse spatial resolution results in a corresponding coarse spatial resolution in the crop calendar dataset, while employing NDVI data with low temporal resolution (e.g., a monthly interval) diminishes the accuracy of the crop calendar dataset.

Furthermore, the code is adaptable for extracting the phenological stages of crop growth occurring within the same year. As illustrated in Figure 2, the dry cropping seasons in Indonesia, spanning from February to October, can be assessed using the provided code because they occur within the same year. However, generating the crop calendar for the wet cropping season (usually occurring between October and February) is not feasible with the existing code, as that would span two different years. However, with modifications and enhancements to the code using the same methodology, the crop calendar dataset can be generated even when spanning across different years. Nevertheless, the validation of this study poses a challenge and limitation due to its pioneering nature as the first dynamic crop calendar with a 1 km² resolution. This is attributed to the absence of any established dataset that aligns with both the temporal and spatial resolutions employed in this investigation.

5. Conclusions

In this study, the development of an annual dry cropping season calendar was demonstrated using the MODIS NDVI time series, which effectively represents crop phenological states. The critical dates, including the Start of Season (SOS) or planting dates, Peak of Season (POS), and End of Season (EOS) or harvesting dates, all in 1 km² resolution, were presented and stored in NetCDF format for both the first and second dry cropping seasons in Indonesia. This dataset serves as a valuable resource for investigating annual variations in cropping intensity and shifts in cropping seasons. The methodology to develop the crop calendar can be applied to another specified region and period. However, the quality of the crop calendar produced will depend on the spatial and temporal quality of the NDVI dataset. Additionally, it holds potential for future research, offering insights into the influence of climate indicators on the initiation of cropping seasons and cropping duration. Moreover, it can be utilized to develop historical crop production records, assess the impact of climate variability and hazards on crop yield, and make projections for future crop yields under changing climate conditions.