Perceptions and realities of hydroclimatic change affecting Guyanese rice farming

This study explores small farmers ’ perceptions of changes in climate across Guyana ’ s rice-producing regions. Qualitative, primary data were collected from a random sample of 189 small farmers, supplemented with 28 key informants, from across Guyana ’ s five main rice-producing regions. The most prevalent perception related to precipitation among farmers is an increase in rainfall year-round (56%), while for informants, it is an increase in rainfall intensity (81%). When considering the atmospheric conditions of temperature and humidity, farmers (88%) and informants (96%) overwhelmingly perceive warmer conditions. Considering weather and climate volatility, farmers (72%) and informants (82%) most prevalently perceive an increase in excess rainfall/flooding, but secondly, farmers (58%) and informants (71%) communicated a perceived increase in drought. Secondary quantitative hydroclimate data support the perception of a wetter climate, and to some degree, increased hydroclimatic volatility. Precipitation is critical to rice cultivation, and the data sets, combined, signal a wetter Guyanese climate, which has major economic implications for small farmers, the broader rice industry, and the economy of Guyana. However, granularity in farmers ’ perceptions suggests a need for more detailed hydro-climate monitoring across Guyana. Thus, strengthening the Guyanese Hydrometeorological Service to support improved spatial and temporal monitoring and collection of primary weather data would be a wise investment in short-and long-term climate mitigation efforts.


Introduction
Like many countries for which agriculture is a significant element of GDP, the South American country of Guyana (Fig. 1) is vulnerable to changes in the regional climate.Compounding the vulnerability, rice cultivation occurs in close proximity to the coast in Guyana, introducing risk from sea level rise.Approximately 90% of the Guyanese population is concentrated on the low coastal plain region (MoP 2015), which forms the heart of the country's economic activities (ECLAC, 2011;NAPG, 2016), the agricultural sector being key among them (Hickey and Weis, 2012;Velasco, 2014).The coastal zone ranges from 8 to 65 km in width (Velasco 2014) and runs 425 km along the Atlantic coast (Hickey and Weis 2012).The majority of this relatively flat region is located below sea level and highly susceptible to flooding due to rising seas and excess rainfall.Barr, Fankhauser, and Hamilton (2010) grouped Guyana among countries that face the highest risks related to climate change.In the Latin America and Caribbean (LAC) region, Guyana is ranked number one in terms of freshwater flood risk and faces a very high risk of droughts (Garlati 2013), experiencing both in the last two decades (NAPG, 2016).Extreme rainfall events, poor solid waste management practices, and poorly maintained physical infrastructure for drainage and irrigation, conservancies, and sea defenses are chiefly responsible for flood events (NATCOM2, 2012).A developing country, Guyana is one of the poorest countries in the LAC region.It is ranked 125th on the United Nations Development Program (UNDP) Human Development Index (HDI) (United Nations Development Program (UNDP), 2018), while the Notre Dame Global Adaptation Initiative (ND-GAIN) 1 ranks Guyana 111th in the world (ND-GAIN 2016).Thus, considering Guyana's limited capacity to adapt, a changing climate poses significant threats to the country's economic base. 1 Summarizes a country's vulnerability to climate change and other global challenges in combination with its readiness to improve resilience (ND-GAIN 2016).
O. Mahdu and A.W. Ellis While temperature affects the rate of agricultural plant development (Hatfield and Prueger, 2015;Wassmann et al., 2009), rainfall is arguably the most important controller of agricultural yields (Motha 2011).The frequency, intensity, timing, and duration of rainfall (Guan et al. 2015) may have direct impacts on yields, field conditions, irrigation systems, and transportation infrastructure.Climate change hastens the need for, and scale of, adaptation (Parry et al. 2009), and this begins with farmers' perception of the environmental changes and the observed impacts occurring in and around their farms.Perceptions (cognitive processes) of climate change are important factors in determining whether farmers are likely to undertake adaptive measures (Niles, Lubell, and Haden, 2013;Haden et al., 2012).Additionally, a better understanding of farmers' perceptions, impacts, and ongoing adaptation measures is crucial for informing policies aimed at promoting successful adaptation strategies among small-farm rice farmers2 .
The purpose of this paper is to communicate the perspectives and perceptions of a representative sample of small-farm Guyanese rice farmers regarding changes in the local hydroclimate, and then to compare these to evidence of actual hydroclimatic change.The focus is mainly precipitation, as Guyanese rice farmers are more sensitive to variability and change in the wetness of the coastal tropical climate than they are to the more conservative variable air temperature.Our basic two-part question is, do farmers perceive that a change in hydroclimate has occurred, and if so, do hydroclimatic data support the perception?The general hypothesis is that there is widespread perception of significant hydroclimatic change, and that the generalities of the perceived change are evident in precipitation data, but that granularity in the perceptions requires hydroclimate monitoring on improved spatial and temporal resolutions.

Study area
Located on the northeastern coastline of South America, the nation of Guyana is bordered by the Atlantic Ocean and the countries of Suriname, Brazil, and Venezuela (Fig. 1).Comprised of 10 administrative regions, Guyana has a land area of about 215,000 km 2 and a population of approximately 757,000 people, the majority of whom live and work within the agricultural belt on the Atlantic coast.
Within Guyana, rice (Oryza sativa L.) is a major source of nutrition and rural livelihood.Primarily cultivated along the coast (administrative regions two through six3 ; Fig. 1), rice serves as the main staple in the diet of most Guyanese, with consumption of approximately 50 kg per capita annually (ECLAC 2011).The rice industry supports about 6,300 farmers and their families directly (GRDB, 2018) and 150,000 people indirectly (Ragnauth et al. 2014).As the primary agricultural sub-sector, rice cultivation represents the largest use of agricultural land, with approximately 87,400 ha (215,970 acres; 0.4% of Guyana) under cultivation (GRDB 2018).In 2018, rice contributed 3.3% to the gross domestic product (GDP), 20.3% to agriculture GDP, and 34.3% to crop agricultural GDP (BoG 2018).In 2018, approximately 75% of the rice produced in Guyana was exported, contributing 13.5% to total export earnings (BoG, 2018).
The climate and physical geography of Guyana support the country's rice production.The low-latitude climate of Guyana is evident in the small range in temperature throughout the year, with peaks in April and October (Fig. 2).The moist tropical climate of Guyana is characterized by ample rainfall (NATCOM2, 2012), segregated into a primary wet season (mid-April through July) and a drier, secondary wet season (mid-November through January) (Fig. 2).Average annual rainfall across Guyana ranges from approximately 1,400 mm to more than 4,000 mm (NATCOM2, 2012; GLSC, 2013).However, spatial variations in rainfall create three sub-climate zones: the tropical savannah, where annual rainfall is less than 1,800 mm; the highland tropical rainforest, where annual rainfall exceeds 2,700 mm; and the hilly/coastal plain, where annual rainfall ranges between 1,700 mm and 2,800 mm (NATCOM2, 2012).

Perceptions of climate variability and change
To gain an understanding of farm-level perceptions, responses to in-person interview questions and surveys were solicited from both small farmers and key informants4 .While farmers provide a personal perspective, key informants are better positioned to comment on larger patterns seen across many farms.In addition, responses from key informants may help validate information gathered through farmer interviews and surveys.
Following Bartlett, Kotrlik, and Higgins (2001), a five percent sample size was considered sufficiently large for a homogeneous population that is likely to face similar socioeconomic, environmental, and climatic conditions.However, to satisfy the statistical convention of a robust sample, a minimum of 30 small farmers in each region was randomly selected for interview.A copy of the rice farmers' registry as of spring 2017 was obtained from the Guyana Rice Development Board (GRDB).The registry is organized by regions, with each containing the following information for each farmer: first and last name, address (name of the village only), call name 5 , and acreage sown.As the registry is stratified by region, a list of small farmers was subsequently created for each region to facilitate the sample selection.The list of farmers for each region was sorted by acreage planted, and those with more than 4.45 ha (11 acres) were removed, defining a "small farmer" for this study 6 .The list of farmers for each region was then sorted by assigned random numbers, and the first five percent were selected for interviews.Unavailable farmers (e.g., not at home, migrated) were replaced from within the next five percent of randomly selected farmers in each region.However, given limited success in locating some replacement farmers, coupled with time and cost considerations, district rice extension officers helped to identify farmers in the same village or neighboring villages.In some instances, replacement farmers were identified and interviewed from villages that were not previously captured in the random samples taken for each region.Across five regions, 189 small farmers and 28 key informants were interviewed in two phases between July and September 2017 and May 2018 (Table 1).Interviews were conducted based on farmers' availability (outside the sowing and harvesting windows).Farmers were asked their perceptions of changes in weather and climate "over the past five years", with no explicit reference to climate change.Additionally, key informants' responses were used to validate farmers' responses.
Key informants comprise an experienced farmer (+25 years of experience), a rice miller, and key employees of the Guyana Rice Development Board (GRDB) that possess diverse and in-depth current knowledge and experience of rice cultivation across Guyana.Those from the GRDB include the Chief Scientist/Plant Breeder, Plant Pathologist, Agronomist, Entomologist, Rice Extension Manager, two regional superintendents, and nineteen district rice extension officers assigned to different districts across the five rice-producing regions where farm-level interviews were conducted.
At the farm-level, a 49-question cross-sectional household survey was designed and used for interviewing farmers.Beyond the descriptors for each interviewee (questionnaire identification number, name, village, telephone number, coordinates of interview location, administrative region), the questionnaire was divided into four parts: socioeconomic characteristics; farm structure and characteristics; farmers' perceptions of climate over the past five years, the impacts on rice farming, and adaptation measures implemented; and institutional accessibility.Questions related to perceived changes in rainfall, temperature, and extreme weather events were open-ended.A second survey consisting of the same open-ended questioning regarding perceived changes of rainfall, temperature, and extreme weather events was adopted for interviews with key informants.
The questions included in the survey were theoretically defined and based on a review of the literature on the impacts of climate change at the farm-level.Further, the open-ended elements provided opportunities for farmers and informants to share information freely.Pilot testing of the survey instrument (e.g., Ruel, Wagner, and Gillespie 2015) improved measurement reliability.The lead- 6 There is no commonly accepted definition of small farmers (Morton 2007).Two criteria are usually used to defined small farmers: subsistence farmers, consuming the majority of their output within the household (Barnett, Blas, and Whiteside 1996) or scale of production (farmer size or income) (Lowder, Skoet, and Raney 2016).
author conducted face-to-face interviews with each farmer and informant in two phases, between July and September 2017 and then again in May 2018.Interviews lasted between 30 and 45 minutes and were conducted primarily at farmers' homes (Fig. 3).However, other interview locations included roadside, farm, and at off-farm employment sites.Informant interviews were conducted at their home, the Burma Rice Research Station, and regional rice extension offices (Fig. 3).
We reviewed the completed questionnaires to identify missing information or inconsistencies in the responses before data entry, and in a few instances, we resolved these by reviewing field notes and/or audio recordings if available (not all respondents agreed to recording).To help with data entry, we created a codebook containing the variable name, variable description, variable type (e.g., text, continuous, categorical, ordinal, binary), and variable codes.The codes were primarily derived from the questionnaires, which contained preassigned codes for most questions.In a few cases, however, we expanded the codes to capture additional responses.Using the codebook as a reference, we entered the survey data into a database and coded accordingly.
We cleaned the data by reverse tracing the entry from the database to individual questionnaires, making corrections where necessary, and subsequently created summary statistics for each question.For questions that generated subjective data (e.g., How did rainfall patterns change?), we used descriptive statistics to analyze each narrative response.Several recent studies have successfully employed qualitative data in part or as a whole to better understand farmers' perceptions of climate change (e.g., Zamasiya, Nyikahadzoi, and Mukamuri, 2017;Hitayezu, Wale, and Ortmann, 2017;Tripathi and Mishra, 2017;Ayanlade, Radeny, and Morton, 2017;Appiah et al., 2018).
We entered interview responses into a database and carefully identified keywords and phrases that were subsequently coded to produce descriptive statistics.We subsequently combined some codes to create categories that added depth and insight, and we then tabulated the frequency and percentage of each keyword or phrase.To validate the authenticity of the data collected and coded, we embedded quotes from farmers and informants within the analysis.We did not further stratify the data by farmer demographics, as the nature of respondent selection yielded a relatively homogenous profile across the 189 farmers interviewed.

Hydroclimate data and analysis
Daily station-level climate data across Guyana are sparse.To foster a robust analysis of precipitation across the rice cultivation region of Guyana, we pursued coarser data, both spatially and temporally.The aim was to mitigate data vagaries that arise when using a resolution that an observation network is unable to support due to problems such as inconsistencies in daily measurement frequency or station location that introduces anomalies that are not representative of the broader area.Upon considering several gridded data products, we chose to adopt the United States Climate Prediction Center (CPC) Merged Analysis of Precipitation (CMAP) data.The data represent a blend of precipitation gauge data and satellite-derived precipitation estimates on a 2.5 degrees latitude by 2.5 degrees longitude grid globally at pentad (five-day) and monthly temporal resolutions (Xie and Arkin 1996).We believe that the value of CMAP data for representing regional variability and change is worth the spatial and temporal sacrifices made by not pursuing station-level data.
CMAP data are available for the period January 1979 to near present.Data based on a gauge-satellite blend (referenced hereafter as CMAP1) are supplemented by a second version of the data set (CMAP2) that incorporates model forecasts of precipitation from the United States National Center for Environmental Prediction (NCEP)/National Center for Atmospheric Research (NCAR) reanalysis data set.The intent of CMAP2 is to offer a spatially complete data set, as some areas of the globe, particularly high-latitude regions, are not represented by observed data.Xie et al. (1996) provide discussion of the interpolation of gauge data to the CMAP grid resolution, while Xie and Arkin (1997) outline the application of satellite and model data into the two blended products.
Like any gridded precipitation product, CMAP data quality is highest in regions of greatest gauge density, which is a weakness of CMAP data across Guyana.A distinct strength is that satellite estimates of precipitation are most accurate in areas for which deep convection is the predominant mechanism, making that element of CMAP data most accurate in tropical climates, like that of Guyana.In fact, the data developers advise that an appropriate application of the data is identification and quantification of spatial and temporal variability of precipitation in the tropics (Arkin and Xie, 2020).Estimates of the relative percent error for both types of CMAP data are provided for each individual data value.We downloaded CMAP1 and CMAP2 data for the period 1979-2016, which represents the beginning of the CMAP data history through the period of farmer reflections on climate.We identified four CMAP grid cells that together encompass nearly all of Guyana, referencing the cells as southern, western, coastal, and north coastal (Fig. 3).For the period of study, mean relative error values range from 42 to 58% for both CMAP data sets, reinforcing the idea of refraining from focusing on the absolute magnitude of the CMAP precipitation data and instead focusing on variability and change.For either of the

Table 1
Distribution of the 189 small farmers interviewed a from across the five rice producing regions (Fig. 3).coastal cells, there is no difference in the data within the two CMAP versions.Based on these data qualities, we chose to use CMAP2, and while we include results from all four cells for comparative purposes, of greatest importance here are the two coastal grid cells, as they are most relevant to farmer perspectives while offering relatively greater data confidence.For the period 1979-2016 there are no missing data within the CMAP datasets.
We first defined the primary and secondary wet seasons for each CMAP grid cell and for each CMAP dataset.We calculated the 38year (1979-2016) mean precipitation for each pentad and subsequently the mean of the totaled 73 pentad means.We then calculated a running total of the individual pentad mean minus the annual pentad mean, beginning with pentad one (January 1-5) and ending with pentad 73 (December 27-31).Inflection points within the plot of accumulated difference from mean indicate a shift, or season change.We generated descriptive statistics seasonally and annually to quantify precipitation amount and variability for comparison of CMAP data sets and grid cells.
We next focused on precipitation change.We first segregated the seasonal data into the early and late halves of the 38-year study period for which we generated descriptive statistics and conducted a two-sample t-test to determine if the two data populations are significantly different.We then calculated 38-year seasonal precipitation trends using Sen's slope estimator and tested the significance with the Mann-Kendall test.We also plotted the seasonal precipitation time series to visually assess the degree to which any trends are monotonic and linear.To examine change in precipitation at around the time of seasonal transition, we repeated the analyses for the 15-day period (3 pentads) at the beginning and end of each seasonal period.The idea was to reveal any obvious expansion or contraction of the seasonal periods.Lastly, we quantified volatility in precipitation by converting the raw precipitation data for each pentad within each CMAP grid cell to percentiles and creating time series of the annual frequency with which precipitation was above/ O. Mahdu and A.W. Ellis below the 75th/25th percentiles for the pentads comprising each season.

Perceived climate change
When asked about specific changes perceived7 "over the past five years", 106 (56.1%) farmers reported an increase in total rainfall in and out of the traditional rainy sowing seasons of May through June and December through January (Table 2).Contrasting this, 80 (42.3%) farmers perceived an increase in out-of-season rainfall alone, while 56 (29.6%) reported an increase in rainfall intensity (Table 2).The perception of increased precipitation throughout the year is evident in responses that seasons have become less distinct or that weather patterns have shifted as indicated by 43 (22.8%)farmers (Table 2).One small farmer related, "we normally expect rain in May-June, but now you don't get rain….rain is out of season as rainfall patterns have shifted….thetotal amount of rain you get is similar but out of season."The results indicate a clear perception of a wetter climate recently, with most farmers feeling that it has increased throughout the year.Still, many farmers believe that it has increased only outside of the typical rainy sowing seasons.
While small farmers provide a local-farm perspective, key informants are positioned to better comment on larger patterns seen across many farms.While also perceiving wetter conditions, informants predominantly (n = 23, or 81.2%) perceived an increase in rainfall intensity (Table 2).Increased rainfall intensity is problematic for rice cultivation in Guyana, as much of the farmland is below sea level, and draft8 relies on the change in tidal flow and gravity to facilitate the movement of excess water from inundated fields via drainage canals and eventually into a river and/or the Atlantic Ocean.Given that tides change every six hours, kokers [sluices] cannot be opened if intense rainfall occurs during high tide.Factoring-in existing drainage capacity and a poorly maintained drainage infrastructure, more rainfall over a short period of time could easily overwhelm the current drainage system even during low tide.
Some informants (n = 14, or 50%) also perceived a temporal shift in annual weather patterns (i.e., wet seasons timing), while 12 (42.9%)informants reported an increase in total rainfall in and out of the traditional rainy sowing seasons (Table 2).Some informants (n = 13, or 46.4%) feel that the distinction between the relatively wet and dry seasons has faded, with the rainy period for sowing beginning earlier and ending later (Table 2).The results indicate a limited feeling of a change toward perpetual wetness, with less distinction between seasons.However, the majority of farmers and key informants do not perceive this change.
Some farmers and informants also perceive that rainfall has become more unpredictable − 25 (13.2%)farmers and eight (28.6%)informants noted that it has become difficult to predict rainfall (Table 2).In the words of one farmer, "yo kyant predict de weather any mo" [you can't predict the weather anymore] while another mentioned, "it is not that rain has stopped falling, it is you don't know when it will fall."In other words, rain falls when it is least expected.
Alluding to increased intra-annual variability, a small number of farmers (n = 15, or 7.9%) indicated that in recent years rainfall greatly fluctuated between seasons and/or years.One farmer noted, "sometimes yo get mo rain in de small crop than de big crop."In other words, rainfall during the spring crop (November-April), or short rainy season, sometimes exceeds that of the autumn crop (May-October), or long rainy season, when rainfall is historically greatest.Rao et al. (2012) found that rainfall variability was greater during the secondary rainy season (December-January) than the primary rainy season (May-July).
While the perceptions of farmers and informants are not perfectly aligned, both groups perceive recent changes in rainfall.The finer detail of increased rainfall intensity is important; in a practical sense, it presents a flooding risk for small farmers, while in a theoretical sense, it underscores the importance of obtaining the perspectives of the key informants, who more directly communicated this hydroclimatic anomaly that is often difficult to characterize when measured data are spatially and temporally coarse.
Given the pervasive nature of climate warming, it is not surprising that farmers and informants perceived changes in temperature "over the past five years".Table 3 summarizes the perceptions of small farmers and key informants with regard to changes in temperature, humidity, and climate volatility (i.e., extreme weather events).When asked how temperature has changed9 , 167 (88.3%) farmers indicated that the days have become hotter or more heated [sic] (Table 3).An overwhelming majority of informants 27 (96.4%)also feel that the climate has become warmer in recent years.As one key informant recalled, "field work used to be quite pleasant.You can go out without an umbrella in the sun and you won't feel the heat or you won't feel your skin burning.But for the last five years, that has changed; it is not the same.You have to go with a hat, an umbrella, a long sleeve shirt.The kind of sweating [perspiration] you do now is far more than you use to do 10, 15 years ago."Informants (n = 14, or 50.0%) also mentioned that there has been an increase in humidity, particularly during daytime (Table 3).
In crop agriculture, extreme weather or climate events can cause physical damage and affect the timing and conditions of field operations (Powell and Reinhard 2015).When asked to describe how extreme weather events (flooding, drought, wind) have changed10 , 136 (71.9%) farmers noted that occurrences of excess rainfall have led to more flooding (Table 3).It must be noted that excess rainfall, poorly maintained drainage, and the tidal flow collectively impact the magnitude of flooding.However, farmers in some areas noted that even with well-maintained drainage, flooding still occurs because of the intensity of rainfall events.The sheer volume of water coupled with inadequate infrastructure to discharge excess water leads to an overflow of the waterways.As a result, there is no outlet for water in inundated rice fields to flow since often times the water in the field and in the nearby trenches and canals are at the same level.
Perhaps indicating a perceived increase in volatility, 110 (58.2%) farmers experienced a drought in the last five years (Table 3).One farmer stated, "meh had to pay people fa pump wata because ah drought and den when de drought done de rain tek off and duck out the whole place" [he paid for water to be pumped into his field because of drought only to experience flooding soon after].In comparison, 23 (82.1%) informants reported excess rainfall led to flooding while 20 (71.4%) reported an increase in drought (Table 3).This is not surprising given that Guyana experienced drought as recent as 2015-2016(NAPG, 2016)).
Possibly a further indication of perceived volatility, 56 (29.6%) farmers and 16 (57.1%)informants reported an increase in high winds (Table 3).High winds either occurred independent of or in concert with heavy rainfall.Although the agricultural belt along Guyana's Atlantic coast normally experiences a northeastern trade wind, it appears that the combination of rainfall and high winds have become more noticeable.But while greater emphasis is placed on reports of flooding and drought, high winds, especially as it relates to agriculture in general, and rice specifically, are often overlooked and/or ignored.This is because high winds are not likely seen as a separate threat but as part of heavy downpours where flooding due to excess water is of greater concern.High winds are often only acknowledged if there is a human toll and/or property damage.As the data in Table 3 suggest, the perceived changes in extreme weather events reported by small farmers and informants are very similar.Independent observations support perceptions of weather volatility.For instance, in the last two decades, Guyana has experienced floods (2005, 2006, 2008, 2010, 2011, 2013, 2014, and 2015) and droughts (1997-1998, 2009-2010, and 2015-2016) (NAPG, 2016).
Although analyzing farmers' perceptions is a practical approach to understanding recent changes in climate, one caveat is that the results presented here are predicated on the recall strength of the interviewees.However, given the life importance of rice cultivation in Guyana, and the importance of weather and climate to rice cultivation, recall among farmers and informants may be strong.While qualitative data have been used to better understand farmers' perception of recent changes in climate, an important distinction of the methods employed in this study relates to the open-ended nature of the perception questions.In addition to testing the consistency of observed weather variables, this approach provides a more descriptive account of how climate variables have changed.Regardless, the general findings here are consistent with studies in other regions where past experience plays a crucial role in influencing farmers' beliefs and perceptions (Takahashi et al. 2016); farmers perceptions are correlated with observed climate parameters or

Table 2
Perceptions of changed rainfall among farmers and informants.The number of respondents and percentage of total respondents that conveyed each perception are reported, and the rank (most-to-least common) of each perception is noted., Walker, and Botha, 2013;Ayanlade, Radeny, and Morton, 2017;Elum, Modise, and Marr, 2017); and farmers are cognizant of long-term changes in temperature and rainfall, and they are conscious of the risk related to climate variability and extreme weather events (Tripathi and Mishra 2017).

Measured hydroclimatic change
Inflection points within the accumulated difference between mean precipitation for each pentad through the year and annual mean precipitation illustrates the bounds between the shorter, primary wet season and the longer, secondary wet season (Fig. 4).The beginning (end) of the primary (secondary) wet season progresses temporally when moving from south-to-north across Guyana.The primary (secondary) season begins (ends) on April 11 (pentad 21) for the southern CMAP cell, April 21 (pentad 23) for the western and coastal cells, and May 6 (pentad 26) for the northern coastal cell.The primary (secondary) season ends (begins) on August 28 (pentad 48) for each of the cells.Likely related to season length, the primary wet season is wettest for the southern CMAP cell (longest primary season) and driest for the northern cell (shortest primary season), while the secondary season is wettest along the coast and north coast and driest across the south and west (Table 4).The proportion of mean annual precipitation attributed to the primary wet season decreases from south-to-north, from 65% for the southern cell, to 59% and 54% for the western and coastal cells respectively, to 46% for the northern coastal cell.While there are regional differences in season length and seasonal precipitation magnitude, the covariability of seasonal precipitation through the 38-year study period among the CMAP grid cells is very high.For the annual timeframe, and for the two seasons, all cell-to-cell correlations are highly statistically significant (p-value less than 0.01) with correlation coefficients greater than 0.83, except for southern-northern coastal correlations of 0.78 annually and 0.71 for the primary wet season.
A two-sample t-test comparing seasonal precipitation during the early (1979)(1980)(1981)(1982)(1983)(1984)(1985)(1986)(1987)(1988)(1989)(1990)(1991)(1992)(1993)(1994)(1995)(1996)(1997) and late (1998-2016) halves of the 38-year period of study reveal significantly wetter seasons over the latter one-half of the record for both the primary and secondary wet seasons for each CMAP grid cell (Table 5).While the data populations consist of only 19 years, the magnitude of the differences and the level of statistical significance strongly indicate a change to wetter conditions.Similarly, the annual standard deviation in precipitation among the pentads that define each of the wet seasons increased significantly from the early to late halves of the record for each of the CMAP grid cells and significantly so for all but the secondary wet season for the southern grid cell (Table 5).The change to wetter seasons from the early to late halves of the record is evident in the structure of the two data populations for each season and for each CMAP grid cell studied (Fig. 5).
Reinforcing the apparent increase in seasonal precipitation from early to late halves of the study period are positive trends in seasonal precipitation and the standard deviation of within-season precipitation through the 38-year period (Table 6).Large magnitude trends in primary wet season precipitation and standard deviation are apparent for each of the four grid cells studied, and each trend is statistically significant.For the secondary wet season, precipitation for only the northern coastal grid cell is characterized by a statistically significant positive trend (Table 6).While the secondary wet season for the southern, western and coastal cells is wetter for the more recent one-half of the study period than for the earlier half (Table 5), trend results indicate that the change is not Fig. 4. For each of the four CMAP grid cells studied, graphical depiction of season bounds using the accumulated sum of the difference between mean precipitation for each pentad and the annual mean precipitation.
O. Mahdu and A.W. Ellis the result of a long-term linear trend through the study period (Table 6).Time series of precipitation reveal that the statistically significant trends in seasonal precipitation are largely predicated on a positive trend from the early-1990 s through the early-to-mid-2000 s, with little change since (Fig. 6).For the two coastal grid cells, secondary wet season precipitation is frequently greater than primary wet season precipitation since the early 2000 s.While a change in primary seasonal precipitation is evident through the 38year record, the positive trend does not appear to be ongoing.Trend analysis of precipitation for the periods consisting of the three pentads (15 days) at the start and end of the primary and secondary wet seasons revealed no statistically significant change in precipitation, suggesting that the length and or timing of the seasons has not changed appreciably.Upon converting the historical record of precipitation for each pentad to percentiles and summing the annual frequency of percentiles within the upper (wet) and lower (dry) quartiles within the primary wet season, it is evident that the frequency of the wettest days has increased while the frequency of the driest days has decreased (Table 7).This indicates a change toward wetter conditions, but not necessarily an increase in volatility, which would be evident if the frequency of both the wettest and driest days increased.While this change toward a greater frequency of the wettest pentads and lesser frequency of the driest pentads is also evident for the secondary wet season (Table 7), the signal is weaker with lesser statistical significance across the four CMAP cells.Increases in the annual standard deviation in precipitation percentile among pentads that define the primary wet season portrays increased volatility for all four CMAP cells studied (Table 8).While similar trends are evident for the secondary wet season, the change is much less clearly indicated by statistical significance (Table 8).

Summary and conclusions
A recent change in the regional climate of Guyana is clearly perceived by small-farm rice farmers and key informants from within the Guyanese rice industry.Not surprisingly, a warmer climate is perceived by nearly all surveyed.Important to rice cultivation, particularly during the two annual sowing seasons, there exists a clear perception that the recent climate became wetter.Farmers foremost perceived an increase in rainfall throughout the year, while informants most predominantly communicated a perceived increase in precipitation intensity.There is a strong perception of increased volatility, as both farmers and informants predominantly communicate an increase in the frequency of excessive rainfall/flooding, while the majority within both groups also perceived increased occurrence of drought.However, mindful of recall strength and the focus of the survey instrument on the most recent five years, the signal of hydroclimatic volatility within survey responses may reflect impressions of pronounced drought that occurred within the past few years.
While not investigated here, perceptions of a warming Guyanese climate are plausible, if not absolute, given the nearly universal climate warming globally.The perception of a wetter climate is supported by analysis of gridded pentad-level precipitation data, as is, to some degree, increased hydroclimatic volatility.For both the primary (April/May-August) and secondary (September-March/April) wet seasons, both precipitation amount and within-season variability in precipitation increased from the early to late halves of the 1979-2016 period.Increases during the primary wet season were supported by positive linear trends through the 38-year period, largely predicated on an increase from the early 1990s through the middle 2000s, but with little change since.The increased wetness of the secondary wet season from the early to late halves of the study period is not supported by a statistically significant positive trend.Supporting a wetter recent climate, the frequency of the wettest pentads increased and the frequency of the driest pentads decreased through the study period for both seasons.Representing increased volatility, intra-seasonal variability in pentad-level precipitation as a percentile increased within each of the wet seasons.
A change toward a wetter climate, and particularly more intense precipitation events, is problematic for rice farming in Guyana, as farmlands are generally below sea level, and draft 11 relies on the change in tidal flow to facilitate the movement of excess water from inundated fields via drainage canals and eventually into a river or the Atlantic Ocean.Given that tides change every six hours, kokers [sluices] cannot be opened if intense rainfall occurs during high tide.Considering existing drainage capacity and a poorly maintained drainage infrastructure, increased rainfall and more intense events could frequently overwhelm the current drainage system even during low tide.It must be noted that excess rainfall, poorly maintained drainage, and tidal flow collectively impact the magnitude of flooding.However, farmers in some areas noted that even with well-maintained drainage, flooding still occurs because of the intensity of rainfall events.The sheer volume of water coupled with inadequate infrastructure to discharge excess water leads to an overflow of the waterways.As a result, there is no outlet for water in inundated rice fields to flow since often times the water in the field and in the nearby trenches and canals are at the same level.
Despite the economic implications for both farmers and the rice industry in Guyana, there was little understanding of how farmers perceive changes in climate prior to this study.Farmers' perceptions of climate change play an essential role in their decisions to adapt and subsequently inform policies that promote successful climate adaptation strategies.The results of this study provide a better understanding of farmers' perceptions supported by hydroclimate data.However, the granularity within the perceptions requires hydroclimate monitoring on fine spatial and temporal resolutions, which could be a useful goal for the Guyanese rice farming industry.
It would be beneficial to strengthen the capacity of the Guyanese Hydrometeorological Service to monitor and collect weather/ climate data.Monitoring would aid rice growers in real-time, while accumulated data would further aid climatological analyses and improvements in indirect measurement through reliable ground-truthing across space and through time.Improved monitoring would benefit from a coupling with effective outreach in which weather information is communicated by region, includes early warnings triggered by evolving conditions, and is easy for farmers to interpret regardless of education level.
As this research concentrated on small farmers across the primary rice-producing regions of Guyana, future research could extend the analysis by expanding to medium-and large-scale farmers.This would allow for comparison across groups at the national and regional levels regarding climate change perceptions, impacts, and adaptation.Further research could focus on specific areas that have

Table 7
For the frequency of 75th percentile (wet quartile) and 25th percentile (dry quartile) occurrences among annual pentads during the primary and secondary wet seasons, the period-of-record trends (number of pentads yr − 1 ) and mean values (number of pentads) from the early and late halves of the record.Asterisks indicate the significance of the trends and differences, per a two-sample t-test, in the frequencies during the early and late halves of the record (p-value ≤ 0.01 (**), p-value ≤ 0.05 (*)).historically suffered from floods, droughts, and saltwater infiltration.Such research findings could help policymakers to devise specific policies that target differences at the district and/or regional levels.Given the importance of rice cultivation to the Guyanese people and the nation's economy, and given the uncertainty of the climate upon which rice cultivation depends, the rice-climate relationship in Guyana is deserving of further study that transitions to a greater reliance on measured data rather than human perception.

Declaration of Competing Interest
The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.

Fig. 3 .
Fig. 3. Locations of farmer and key informant interviews and the four CMAP grid cells (inset map).

Fig. 5 .
Fig. 5. Box and whisker plots for seasonal precipitation data from the early and late halves of the 38-year record 1979-2016 for each of the four CMAP grid cells studied.Data are segregated by primary (a) and secondary (b) wet season, and by early (left plot) and late (right plot) halves of the record for each grid cell.

Fig. 6 .
Fig. 6.Seasonal precipitation for the primary and secondary wet seasons for each of the four CMAP grid cells studied, 1979-2016.Farmer and informant interviews asking about conditions over "the past 5 years" were conducted between July and September 2017 and during May 2018.
189 small farmers and 28 key informants were interviewed in two phases between July and September 2017 and in May 2018 aO.Mahdu and A.W. Ellis Small farmers' open-ended responses are based on perceived changes occurring over multiple seasons and/or years, over the past 5 years, where rainfall may have increased "in-season", "out of season", and/or "in and out of season". a

Table 3
Perceptions of changed temperature, humidity, weather, and climate volatility among farmers and informants.

Table 4
Season length (days), and median and standard deviation of precipitation (CMAP2; mm) during the primary (April/May-August) and secondary (September-March/April) seasons for each of the CMAP grid cellsstudied, 1979-2016.

Table 5
For the early and late halves of the period 1979-2016, mean seasonal precipitation (CMAP2; mm) and standard deviation in precipitation among the pentads comprising each season (SD; mm).Asterisks indicate the significance of the difference per a two-sample t-test of the two data populations (pvalue ≤ 0.01 (**), p-value ≤ 0.05 (*)).

Table 6
Trends in total seasonal precipitation and standard deviation in precipitation among the pentads comprising each season (CMAP2; mm yr − 1 ) through the period 1979-2016 during the primary and secondary seasons for each of the CMAP grid cells studied.Asterisks indicate the significance of the trend (p-value ≤ 0.01 (**), p-value ≤ 0.05 (*)).

Table 8
For the annual standard deviation in precipitation percentile among pentads, the period-of-record trends (percentile yr − 1 ) and mean values (percentile) from the early and late halves of the record for the primary and secondary wet seasons.Asterisks indicate the significance of the trends and differences, per a two-sample t-test, in the frequencies during the early and late halves of the record (p-value ≤ 0.01 (**), p-value ≤ 0.05 (*)).
11Draining of the land via the trenches and canals.